Radar Remote Sensing: Applications and Challenges 0128234571, 9780128234570

Citation preview

RADAR REMOTE SENSING

This page intentionally left blank

Earth Observation

RADAR REMOTE SENSING

Applications and Challenges Edited by

PRASHANT K. SRIVASTAVA Remote Sensing Laboratory, Institute of Environment and Sustainable Development, Banaras Hindu University, India

DILEEP KUMAR GUPTA Remote Sensing Laboratory, Institute of Environment and Sustainable Development, Banaras Hindu University, India

TANVIR ISLAM NASA Jet Propulsion Laboratory, Pasadena, California, United States

DAWEI HAN Professor of Hydroinformatics, Department of Civil Engineering, University of Bristol, United Kingdom

RAJENDRA PRASAD Professor, Department of Physics, Indian Institute of Technology (Banaras Hindu University), Varanasi, India

Elsevier Radarweg 29, PO Box 211, 1000 AE Amsterdam, Netherlands The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, United Kingdom 50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States Copyright © 2022 Elsevier Inc. All rights reserved. No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher. Details on how to seek permission, further information about the Publisher’s permissions policies and our arrangements with organizations such as the Copyright Clearance Center and the Copyright Licensing Agency, can be found at our website: www.elsevier.com/permissions. This book and the individual contributions contained in it are protected under copyright by the Publisher (other than as may be noted herein).

Notices

Knowledge and best practice in this field are constantly changing. As new research and experience broaden our understanding, changes in research methods, professional practices, or medical treatment may become necessary. Practitioners and researchers must always rely on their own experience and knowledge in evaluating and using any information, methods, compounds, or experiments described herein. In using such information or methods they should be mindful of their own safety and the safety of others, including parties for whom they have a professional responsibility. To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors, assume any liability for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions, or ideas contained in the material herein. ISBN: 978-0-12-823457-0 For information on all Elsevier publications visit our website at https://www.elsevier.com/books-and-journals

Publisher: Candice Janco Acquisitions Editor: Peter Llewellyn Editorial Project Manager: Howi M. De Ramos Production Project Manager: Paul Prasad Chandramohan Cover Designer: Mark Rogers Typeset by TNQ Technologies

Contents

Contributors Foreword

xiii xix

SECTION 1 Basis of radar remote sensing

1

1. Introduction to RADAR remote sensing

3

Dileep Kumar Gupta, Shivendu Prashar, Sartajvir Singh, Prashant K. Srivastava and Rajendra Prasad 1. Brief history of RADAR remote sensing 2. Optical versus RADAR remote sensing 3. Fundamentals of RADAR 4. Types of RADAR 5. Operational frequencies of RADAR 6. Backscatter mechanisms 7. Radar image characteristics 8. Application of microwave-based remote sensing References

2. Microwave components and devices for RADAR systems

3 4 6 11 14 16 20 21 24

29

Vikram Kumar and Dileep Kumar Gupta 1. Introduction 2. Transmission line 3. Antennas 4. Microwave filters 5. Absorbers 6. Microwave sources 7. Mode converter 8. Network analyzer 9. Some other important microwave components 10. Summary References

29 30 34 37 38 40 42 44 45 46 46

v

vi

Contents

3. Theory of monostatic and bistatic radar systems

49

Suraj A. Yadav, Dileep Kumar Gupta, Rajendra Prasad, Jyoti Sharma and Prashant K. Srivastava 1. 2. 3. 4.

Introduction Bistatic and monostatic radar system configuration Radar equation Radar cross-section per unit area/scattering coefficient system and measurement concepts 5. Measurement procedures 6. Procedure of bistatic specular scatterometer measurement and its calibration over natural terrain 7. Summary References

4. Review of microwave fundamentals and its applications

49 50 51 53 57 58 62 62

65

Shivendu Prashar, Umesh Kumar Tiwari and Sartajvir Singh 1. Introduction 2. Theory of radiative transfer 3. Electromagnetic interaction with discrete objects 4. Interaction with inhomogeneous media 5. Interaction with a smooth surface 6. Interaction with rough surfaces 7. Microwave interaction with natural surfaces 8. Summary Funding References

65 67 70 76 77 80 81 86 87 87

SECTION 2 Conventional methods for radar remote sensing

91

5. Comparative flood area analysis based on change detection and binarization methods using Sentinel-1 synthetic aperture radar data

93

Bikash Ranjan Parida, Arvind Chandra Pandey, Sourav Kumar and Gaurav Tripathi 1. 2. 3. 4. 5. 6.

Introduction Study area Materials and methods Results Discussion Conclusions

93 96 97 99 101 104

Contents

Acknowledgments References

6. Subsurface feature identification using L Band Synthetic Aperture Radar (SAR) data over Jaisalmer, India

105 105

109

Monika, Himanshu Govil and Shashi Kumar 1. Introduction 2. Study area 3. Data used 4. Methodology 5. Result 6. Conclusion Acknowledgments Author contributions References

7. Terrestrial water budget through radar remote sensing

109 110 112 114 117 119 120 120 120

123

J. Indu, Akhilesh S. Nair, Ankita Pradhan, Rohit Mangla, Sooraj Krishnan, Kaushlendra Verma and Vinayak Huggannavar 1. Introduction 2. Precipitation from radar remote sensing 3. Soil moisture from radar remote sensing 4. Water levels from radar altimetry 5. Summary and conclusions Acknowledgments References Further reading

8. Application of synthetic aperture radar remote sensing in forestry

123 125 129 138 139 139 139 148

149

Arjun G. Koppad, Syeda Sarfin and Anup Kumar Das 1. Introduction 2. Polarimetric matrix generation 3. Polarimetric speckle filtering 4. Orientation angle correction 5. Polarimetric decomposition 6. Terrain correction 7. Polarimetric classification 8. Summary and final remarks References Further reading

149 156 156 156 157 159 160 172 173 174

vii

viii

Contents

9. Classification of Radar data using Bayesian optimized two-dimensional Convolutional Neural Network

175

Achala Shakya, Mantosh Biswas and Mahesh Pal 1. Introduction 2. Background 3. Dataset and ground data collection 4. Dataset preparation for classification 5. Methodology 6. Results and discussion 7. Conclusion Acknowledgment References

10. Modeling and simulation of synthetic aperture radar dataset for retrieval of soil surface parameters

175 176 179 179 181 182 184 184 185

187

Sayyad Shafiyoddin and Ajit Kumar 1. Introduction 2. Study area and collection of field data 3. Collection and processing of satellite data 4. Soil moisture modeling 5. Results and discussion 6. Conclusion References

11. Flood inundation mapping from synthetic aperture radar and optical data using support vector machine: a case study from Kopili River basin during Cyclone Amphan

187 189 190 193 196 199 200

203

Prasad Balasaheb Wale, Thota Sivasankar, Varun Narayan Mishra and Ratna Sanyal 1. Introduction 2. Study area 3. Material and methods 4. Result and discussion 5. Conclusion References

12. Performance assessment of phased array type L-band Synthetic Aperture Radar and Landsat-8 used in image classification

203 206 207 210 216 216

219

Swati Suman, Prashant K. Srivastava, George P. Petropoulos, Ram Avtar, Rajendra Prasad, Sudhir Kumar Singh, S.K. Mustak, Ioannis N. Faraslis and Dileep Kumar Gupta 1. Introduction

219

Contents

2. Datasets 3. Methodology 4. Results and discussion 5. Conclusions and future work Acknowledgments References

13. Evaluation of speckle filtering methods using polarimetric Sentinel-1A data

221 224 229 239 241 241

245

Varun Narayan Mishra and Thota Sivasankar 1. Introduction 2. Study site and data used 3. Methodology 4. Results and discussion 5. Conclusion Acknowledgment References

SECTION 3 Advanced methods for radar remote sensing 14. Emerging techniques of polarimetric interferometric synthetic aperture radar for scattering-based characterization

245 246 248 249 253 253 254

257

259

Bhanu Prakash and Shashi Kumar 1. 2. 3. 4. 5. 6. 7. 8.

Introduction Synthetic aperture radar polarimetry Polarimetric decomposition Polarization orientation angle Probability distributions Polarimetric synthetic aperture radar interferometry Polarimetric synthetic aperture radar interferometry coherence-based decomposition Polarimetric synthetic aperture radar interferometry decorrelation-based decomposition model Acknowledgment References

15. Advanced method for radar remote sensing: circularly polarized synthetic aperture radar

259 261 263 269 270 271 274 277 284 284

287

Josaphat Tetuko Sri Sumantyo 1. Introduction 2. Circularly polarized scattering for remote sensing 3. Specification of circular polarized synthetic aperture radar for microsatellite

287 289 295

ix

x

Contents

4. Radio-frequency system of circular polarized synthetic aperture radar 5. Flight test and images 6. Summary and future research References

16. A processing chain for estimating crop biophysical parameters using temporal Sentinel-1 synthetic aperture radar data in cloud computing framework

299 303 305 306

309

Dipankar Mandal, Vineet Kumar, Juan M. Lopez-Sanchez, Y.S. Rao, Heather McNairn, Avik Bhattacharya and Scott Mitchell 1. Introduction 2. Methodology 3. Results and discussion 4. Conclusion Acknowledgments Code availability References

17. Fuzzy logic for the retrieval of kidney bean crop growth variables using ground-based scatterometer measurements

309 311 316 322 322 323 323

327

Dileep Kumar Gupta, Rajendra Prasad, Pradeep Kumar and Prashant K. Srivastava 1. Introduction 2. Method and observations 3. Fuzzy inference system 4. Results and discussion 5. Conclusion References

18. Monitoring tropical peatlands subsidence by time-series interferometric synthetic aperture radar (InSAR) technique

327 329 331 332 337 338

341

Deha Agus Umarhadi, Ram Avtar, Pankaj Kumar, Ali P. Yunus, Tonni Agustiono Kurniawan, Ali Kharrazi, Mamoru Ishikawa and Wirastuti Widyatmanti 1. Introduction 2. Interferometry synthetic aperture radar for tropical peatlands 3. Case study: Sintang, Indonesia 4. Summary Acknowledgments References

341 342 345 352 353 353

Contents

19. Toward a North American continental wetland map from space: wetland classification using satellite imagery and machine learning algorithms on Google Earth Engine

357

Masoud Mahdianpari, Brian Brisco, Bahram Salehi, Jean Granger, Fariba Mohammadimanesh, Megan Lang and Souleymane Toure 1. Introduction 2. Wetland classification systems 3. Wetland field data 4. Remote sensing data 5. Cloud computing platforms and machine learning algorithms 6. Wetland classification results for Canada 7. Conclusion References Further reading

SECTION 4 Future challenges in radar remote sensing 20. Challenges in Radar remote sensing

357 359 362 363 365 366 369 369 373

375 377

Prashant K. Srivastava, Rajendra Prasad, Sumit Chaudhary Kumar, Suraj A. Yadav, Jyoti Sharma, Swati Suman, Varsha Pandey, Rishabh Singh and Dileep Kumar Gupta 1. Introduction 2. Conclusion References

21. The study of Indian Space Research Organization’s Ku-band based scatterometer satellite (SCATSAT-1) in agriculture: applications and challenges

377 385 385

389

Ravneet Kaur, Reet Kamal Tiwari, Raman Maini, Sartajvir Singh and Vishakha Sood 1. Introduction 2. Background of SCATSAT-1 3. Applications in agriculture 4. Summary and conclusions Acknowledgments Funding References

389 391 393 398 400 400 400

xi

xii

Contents

22. Radar remote sensing of soil moisture: fundamentals, challenges & way-out

405

Hari Shanker Srivastava and Parul Patel 1. 2. 3. 4. 5.

Introduction Effect of target parameters on SAR sensitivity toward soil moisture Addressing the effect of target parameters on SAR sensitivity toward soil moisture Effect of the sensor parameters on SAR sensitivity toward soil moisture To identify sensitive polarimetric parameters derived from fully and hybrid polarimetric SAR for soil moisture 6. Addressing the various challenges involved in ground truth planning and ground truth data collection for radar remote sensing of soil moisture 7. Addressing the challenges involved in development of a soil moisture retrieval model using radar remote sensing 8. Addressing challenges involved in SAR data processing due to a huge data volume 9. Addressing the issue of interval and scale of a soil moisture map 10. Conclusion References

Index

405 407 410 416 427 432 435 440 441 443 444

447

Contributors

Ram Avtar Graduate School of Environmental Science, Faculty of Environmental Earth Science, Hokkaido University, Sapporo, Japan; Institute for the Advanced Study of Sustainability, United Nations University (UNU-IAS), Shibuya City, Tokyo, Japan Avik Bhattacharya Microwave Remote Sensing Lab, Centre of Studies in Resources Engineering, Indian Institute of Technology Bombay, Mumbai, Maharashtra, India Mantosh Biswas Computer Engineering Department, National Institute of Technology, Kurukshetra, Haryana, India Brian Brisco The Canada Centre for Mapping and Earth Observation, Ottawa, ON, Canada Sumit Chaudhary Kumar Indian Institute of Remote Sensing, Dehradun, Uttarakhand, India Anup Kumar Das Space Application Centre, Ahmedabad, Gujarat, India Ioannis N. Faraslis Department of Environmental Sciences, University of Thessaly, Thessaly, Greece Himanshu Govil Department of Applied Geology, National Institute of Technology, Raipur, Chhattisgarh, India Jean Granger C-CORE, St. John’s, NL, Canada Dileep Kumar Gupta Remote Sensing Laboratory, Institute of Environment and Sustainable Development, Banaras Hindu University, Varanasi, Uttar Pradesh, India; Department of Physics, Indian Institute of Technology, Banaras Hindu University, Varanasi, Uttar Pradesh, India Vinayak Huggannavar Department of Civil Engineering, Interdisciplinary Centre for Climate Studies, Indian Institute of Technology Bombay, Powai, Maharashtra, India J. Indu Department of Civil Engineering, Interdisciplinary Centre for Climate Studies, Indian Institute of Technology Bombay, Powai, Maharashtra, India Mamoru Ishikawa Graduate School of Environmental Science, Faculty of Environmental Earth Science, Hokkaido University, Sapporo, Japan; Faculty of Environmental Earth Science, Hokkaido University, Sapporo, Japan

xiii

xiv

Contributors

Ravneet Kaur Apex Institute of Technology, Department of Computer Science and Engineering, Chandigarh University, Gharuan, Punjab, India; Department of Computer Science and Engineering, Punjabi University, Patiala, Punjab, India Ali Kharrazi Advanced Systems Analysis Group, International Institute for Applied Systems Analysis, Laxenburg, Austria; Euro-Mediterranean Center for Climate Change, Ca’ Foscari University of Venice, Venice, Italy; Faculty of Global Studies, Akita International University, Akita, Japan Arjun G. Koppad Department of Natural Resource Management (NRM), College of Forestry, UAS Dharwad, Karnataka, India Sooraj Krishnan Department of Civil Engineering, Interdisciplinary Centre for Climate Studies, Indian Institute of Technology Bombay, Powai, Maharashtra, India Ajit Kumar Microwave and Imaging Spectroscopy Research Laboratory, Milliya Arts, Science, and Management Science College, Beed, Maharashtra, India Pankaj Kumar Natural Resources and Ecosystem Services, Institute for Global Environmental Strategies, Hayama, Kanagawa, Japan Pradeep Kumar School of Environmental Sciences, Jawaharlal Nehru University, New Delhi, India Shashi Kumar Photogrammetry and Remote Sensing Department, Indian Institute of Remote Sensing, Indian Space Research Organisation, Dehradun, Uttarakhand, India Sourav Kumar Department of Geoinformatics, School of Natural Resource Management, Central University of Jharkhand, Ranchi, Jharkhand, India Vikram Kumar Department of Electronics and Communication Engineering, National Institute of Technology, Patna, Bihar, India Vineet Kumar Department of Water Resources, Delft University of Technology, Delft, the Netherlands Tonni Agustiono Kurniawan Faculty of Social Work, Health, and Nursing, Ravensburg-Weingarten University of Applied Sciences, Weingarten, Germany; College of the Environment and Ecology, Xiamen University, Xiamen, PR China Megan Lang U.S. Fish and Wildlife Service, National Wetlands Inventory, Falls Church, VA, United States Juan M. Lopez-Sanchez Institute for Computer Research, University of Alicante, Alicante, Spain

Contributors

Masoud Mahdianpari C-CORE, St. John’s, NL, Canada; Department of Electrical and Computer Engineering, Memorial University of Newfoundland, St. John’s, NL, Canada Raman Maini Department of Computer Science and Engineering, Punjabi University, Patiala, Punjab, India Dipankar Mandal Microwave Remote Sensing Lab, Centre of Studies in Resources Engineering, Indian Institute of Technology Bombay, Mumbai, Maharashtra, India; Department of Agronomy, Kansas State University, Manhattan, KS, United States Rohit Mangla Department of Civil Engineering, Interdisciplinary Centre for Climate Studies, Indian Institute of Technology Bombay, Powai, Maharashtra, India Heather McNairn Ottawa Research and Development Centre, Agriculture and Agri-Food Canada, Ottawa, ON, Canada Varun Narayan Mishra Centre for Climate Change and Water Research, Suresh Gyan Vihar University, Jaipur, Rajasthan, India Scott Mitchell Geomatics and Landscape Ecology Laboratory, Carleton University, Ottawa, ON, Canada Fariba Mohammadimanesh C-CORE, St. John’s, NL, Canada Monika Department of Applied Geology, National Institute of Technology, Raipur, Chhattisgarh, India S.K. Mustak Department of Geography, Central University of Punjab, Bathinda, Punjab, India Akhilesh S. Nair Department of Civil Engineering, Interdisciplinary Centre for Climate Studies, Indian Institute of Technology Bombay, Powai, Maharashtra, India Mahesh Pal Civil Engineering Department, National Institute of Technology, Kurukshetra, Haryana, India Arvind Chandra Pandey Department of Geoinformatics, School of Natural Resource Management, Central University of Jharkhand, Ranchi, Jharkhand, India Varsha Pandey Remote Sensing Laboratory, Institute of Environment and Sustainable Development, Banaras Hindu University, Varanasi, Uttar Pradesh, India Bikash Ranjan Parida Department of Geoinformatics, School of Natural Resource Management, Central University of Jharkhand, Ranchi, Jharkhand, India

xv

xvi

Contributors

Parul Patel Space Applications Centre (SAC), ISRO, Ahmedabad, Gujarat, India George P. Petropoulos Department of Geography, Harokopio University of Athens, Kallithea, Athens, Greece Ankita Pradhan Department of Civil Engineering, Interdisciplinary Centre for Climate Studies, Indian Institute of Technology Bombay, Powai, Maharashtra, India Bhanu Prakash Photogrammetry and Remote Sensing Department, Indian Institute of Remote Sensing, Indian Space Research Organisation, Dehradun, Uttarakhand, India Rajendra Prasad Department of Physics, Indian Institute of Technology, Banaras Hindu University, Varanasi, Uttar Pradesh, India Shivendu Prashar Chitkara University School of Engineering and Technology, Chitkara University, Solan, Himachal Pradesh, India; Department of Civil Engineering, Indian Institute of Technology, Ropar, Punjab, India; Central Scientific Instruments Organization, Chandigarh, India Y.S. Rao Microwave Remote Sensing Lab, Centre of Studies in Resources Engineering, Indian Institute of Technology Bombay, Mumbai, Maharashtra, India Bahram Salehi Environmental Resources Engineering, College of Environmental Science and Forestry, State University of New York, New York, NY, United States Ratna Sanyal Computer Science and Engineering Area, NIIT University, Neemrana, Rajasthan, India Syeda Sarfin NISAR Project, Department of NRM, College of Forestry, UAS Dharwad, Karnataka, India Sayyad Shafiyoddin Microwave and Imaging Spectroscopy Research Laboratory, Milliya Arts, Science, and Management Science College, Beed, Maharashtra, India Achala Shakya Computer Engineering Department, National Institute of Technology, Kurukshetra, Haryana, India Hari Shanker Srivastava Indian Institute of Remote Sensing (IIRS), Indian Space Research Organisation (ISRO), Dehradun, Uttarakhand, India Jyoti Sharma Department of Physics, Indian Institute of Technology, Banaras Hindu University, Varanasi, Uttar Pradesh, India

Contributors

Sartajvir Singh Chitkara University School of Engineering and Technology, Chitkara University, Solan, Himachal Pradesh, India; Department of Civil Engineering, Indian Institute of Technology, Ropar, Punjab, India Sudhir Kumar Singh KBCAOS, University of Allahabad, Allahabad, Uttar Pradesh, India Rishabh Singh Remote Sensing Laboratory, Institute of Environment and Sustainable Development, Banaras Hindu University, Varanasi, Uttar Pradesh, India Thota Sivasankar Geographic Information Systems (GIS) Area, NIIT University, Neemrana, Rajasthan, India Vishakha Sood Aiotronics Automation, Palampur, Himachal Pradesh, India Josaphat Tetuko Sri Sumantyo Center for Environmental Remote Sensing, Chiba University, Chiba, Japan; Fakultas Teknik, Universitas Sebelas Maret, Solo, Indonesia Prashant K. Srivastava Remote Sensing Laboratory, Institute of Environment and Sustainable Development, Banaras Hindu University, Varanasi, Uttar Pradesh, India; Institute of Environment and Sustainable Development, Banaras Hindu University, Varanasi, Uttar Pradesh, India Swati Suman Remote Sensing Laboratory, Institute of Environment and Sustainable Development, Banaras Hindu University, Varanasi, Uttar Pradesh, India Reet Kamal Tiwari Department of Civil Engineering, Indian Institute of Technology, Ropar, Punjab, India Umesh Kumar Tiwari Central Scientific Instruments Organization, Solan, Chandigarh, India Souleymane Toure Environmental Resources Engineering, College of Environmental Science and Forestry, State University of New York, New York, NY, United States Gaurav Tripathi Department of Geoinformatics, School of Natural Resource Management, Central University of Jharkhand, Ranchi, Jharkhand, India Deha Agus Umarhadi Graduate School of Environmental Science, Faculty of Environmental Earth Science, Hokkaido University, Sapporo, Japan Kaushlendra Verma Department of Civil Engineering, Interdisciplinary Centre for Climate Studies, Indian Institute of Technology Bombay, Powai, Maharashtra, India

xvii

xviii

Contributors

Prasad Balasaheb Wale Geographic Information Systems (GIS) Area, NIIT University, Neemrana, Rajasthan, India Wirastuti Widyatmanti Department of Geographic Information Science, Faculty of Geography, Universitas Gadjah Mada, Yogyakarta, Indonesia Suraj A. Yadav Department of Physics, Indian Institute of Technology, Banaras Hindu University, Varanasi, Uttar Pradesh, India Ali P. Yunus Department of Earth and Environmental Science, Indian Institute of Science Education and Research Mohali, Mohali, India; Center for Climate Change Adaptation, National Institute for Environmental Studies, Tsukuba, Japan

Foreword

The microwave region of the electromagnetic spectrum has some special properties for remote sensing. The wavelength range of microwave radiation is 1 m to 1 mm. Microwaves are able to penetrate clouds, dust, and rain to some extent. Remote sensing at microwave frequencies gathers useful information about the earth’s atmosphere, land, and oceans under almost all weather conditions at any time. Electromagnetic radiation (EMR) generated by the sun is not required in radar remote sensing. It generates its own EMR that is transmitted to the terrain. Radar remote sensing is popular for earth observations in the scientific community using different types of radar geometry such as monostatic, bistatic, and multistatic. Experimental investigations and their applications for backscattering (monostatic) measurements of the natural terrain have been performed using synthetic aperture radar (SAR) onboard, groundbased, airborne, and spaceborne platforms including RADARASAT, Advanced Land Observing Satellite (ALOS) Phased Array type L-band Synthetic Aperture Radar (PALSAR), ERS-1 and 2, ENVISAT, RISAT, Sentinel, and TERRASAR-X. SAR data are useful for retrieving many geophysical parameters, ocean variables, and meteorologic variables. Monostatic SAR imagery contains information only in backscattered signals from the terrain. Bistatic and multistatic SAR systems are useful for generating three-dimensional images of the terrain. Information gathered by SAR from different azimuthal and elevation angles is more valuable than only backscattered information. Consequently, it is interesting to increase information about the different types of radar geometry in radar remote sensing for potential applications. Therefore, the primary aim of this book is to advance knowledge about the scientific understanding, developments, and applications of radar remote sensing in different radar geometries such as monostatic, bistatic, and multistatic. In essence, a book needs to be compiled that is devoted to collecting developments in and rigorous applications of radar remote sensing using different radar geometries and platforms at different scales including local, regional, and global. Radar Remote Sensing: Application and Challenges includes broad coverage for earth’s observation. By linking the entire system, this book is the first handbook promoting the synergistic and multidisciplinary activities in and challenges to the earth’s observation using different types of radar geometry and platforms among scientists and users who work in the field of radar remote sensing for the earth’s observation. The book was made possible because of extensive and valuable contributions from interdisciplinary experts from across the world in the field of radar remote sensing and applications. The book is divided into four sections, in which Section I contains the basics of radar remote sensing, Section II details conventional methods for radar remote sensing,

xix

xx

Foreword

Section III provides advanced methods for radar remote sensing, and Section IV provides future challenges to radar remote sensing. Chapter 1 provides a brief overview of microwave remote sensing and reviews its different applications. Chapter 2 provides essential components of the microwave system that are necessary to design a microwave radar measurement system. The second section of the book contains chapters related to optical/infrared methods. Chapter 3 of this section provides the details about different radar bands and configurations used for various applications. Chapter 4 deals with basic phenomena and parameters related to wave propagation and interactions with distinct media and objects. Section II starts with Chapter 5, revealing the significance of comparing and analyzing different polarizations for flood mapping and assessment. Chapter 6 provides techniques for identifying subsurface features using SAR data. Chapter 7 discusses various radar remote sensing techniques that aid in monitoring the terrestrial water budget elements using the Integrated Multisatellite Retrievals product of Global Precipitation Measurement constellation satellites. Chapter 8 of this section provides an overview of SAR for forest monitoring using L-band frequency. In Chapter 9, the authors provide detailed information about machine learning techniques for SAR image classification. In Chapter 10, the modeling and retrieval of soil surface parameters based on microwave remote sensing are discussed using SAR datasets. In Chapter 11, the fusion of SAR data with optical datasets is reviewed to study the flood situation in a Kopili river basin. Chapter 12 provides a detailed account of artificial intelligence techniques such as kernel-based support vector machine and artificial neural networks for a classification and performance assessment of Landsat-8 and ALOS-2 PALSAR, whereas Chapter 13 examines the effectiveness of various filtering methods to suppress speckles in dual-polarization Sentinel 1A SAR images. Section III of the book starts with Chapter 14, which provides an overview of different concepts and techniques related to target characterization using SAR remote sensing. Chapter 15 explains the scattering characteristics of C-band circularly polarized SAR. In Chapter 16, the cloud computing potentials of the Google Earth Engine are demonstrated as a unified processing pipeline for water cloud model inversion. Chapter 17 explores the potential of backscattering and fuzzy logic algorithms to retrieve kidney bean crop growth variables using ground-based multiangular, multitemporal, and dualpolarized bistatic scatterometer data. Chapter 18 reviews studies that used traditional and time-series InSAR techniques to map peat subsidence in the tropical region, whereas Chapter 19 provides state-of-the-art satellite remote sensing for wetland classification. Section IV starts with Chapter 20, which highlights various challenges involved in the radar remote sensing of soil moisture. Chapter 21 provides a systematic review of the applications, challenges, and future requirements of SCATSAT-1 over land-cover applications. This study is essential to understand vegetation dynamics with active microwave

Foreword

sensors and explore the potential applications of SCATSAT-1 in emerging fields. The last chapter deals with the challenges and future prospects of radar remote sensing. We believe that this book may be read by people with a common interest in radar remote sensing in ground-based scatterometers and SAR systems, airborne scatterometers and SAR systems, and spaceborne scatterometers and SAR systems at different frequencies and polarizations with monostatic, bistatic, and multistatic geometry. This book is also useful for readers from other diverse backgrounds like earth and environmental, meteorological sciences, etc. Last but not the least, the editors are grateful to all contributing authors and anonymous reviewers for their time, talent, and energies and for adhering to a strict time line; and to the staff at Academic Press, Elsevier, for their patience and support throughout. Prashant K. Srivastava Varanasi, India Dileep Kumar Gupta Varanasi, India Tanvir Islam Los Angeles, United States Dawei Han Bristol, United Kingdom Rajendra Prasad Varanasi, India

xxi

This page intentionally left blank

SECTION 1

Basis of radar remote sensing

1

This page intentionally left blank

CHAPTER 1

Introduction to RADAR remote sensing Dileep Kumar Gupta1, 2, Shivendu Prashar3, 4, Sartajvir Singh3, Prashant K. Srivastava1 and Rajendra Prasad2

1 Remote Sensing Laboratory, Institute of Environment and Sustainable Development, Banaras Hindu University, Varanasi, Uttar Pradesh, India; 2Department of Physics, Indian Institute of Technology, Banaras Hindu University, Varanasi, Uttar Pradesh, India; 3Chitkara University School of Engineering and Technology, Chitkara University, Solan, Himachal Pradesh, India; 4Central Scientific Instruments Organization, Chandigarh, India

1. Brief history of RADAR remote sensing In the beginning of the 19th century, the development of remote sensing as a scientific field was closely tied to developments in photography. Photographic cameras were used for remote sensing purpose to record information in the visible and near-infrared frequency spectrum. Daguerre is the first person, who has taken the first image of the moon in the first month of the year 1839 using the daguerreotype technique (https:// www.aps.org/publications/apsnews/201301/physicshistory.cfm). After that, the director of the Paris Observatory suggested the use of photography for topographic purposes. By the end of the 18th century, photographic cameras were deployed above the earth’s surface on kites and balloons to take oblique aerial photographs of the landscape. The first airplane aerial photography was conducted in 1909. These aerial photographs taken by airplanes had an important role in collecting information about the position and movement of enemy troops during the World War I. Seven years after proven of electromagnetic wave Heinrich Hertz experiment, Italian scientist Guglielmo Marconi (1874e1937) transmitted radio waves over the Atlantic. This event laid the foundation for current remote sensing and RADAR. Developments in the field of RADAR arrived in World War II. Secretly, Britain, Germany, and the United States were developing and implementing RADAR technology. In 1904, Christian Hulsmeyer developed a radio wave-based system to detect obstacles. In 1922, Marconi demonstrated a similar technique to the American Institute of Electrical Engineers to know the determine of a ship in poor weather conditions. With the continuous effort of researchers, the well-established concept of the early RADAR system came about in the 1930s. The Scottish scientist Robert Watson-Watt gets the credit for the development of real operational pulsed radar systems. He was in touch with Prof. E.V. Appleton, who had discovered the reflection of radio signals from the ionized layer of the upper atmosphere in 1924. Appleton found that the distance between the antenna and the Appleton layer (ionized layer) could be estimated by measuring the time interval between the transmission and return of signals. Watson-Watt had used the same mechanism to locate a storm and provide the direction and range of the storm. April 2, 1935, was a memorable day for Watson-Watt Radar Remote Sensing ISBN 978-0-12-823457-0, https://doi.org/10.1016/B978-0-12-823457-0.00018-5

© 2022 Elsevier Inc. All rights reserved.

3

4

Radar Remote Sensing

when he received a patent for his RADAR device. The device designed by Watson-Watt was efficient for locating and ranging aircraft using pulses of microwaves instead of continuous waves. He proposed the idea of developing a system to locate enemy planes and send an alert to forces. Similarly, researchers in the US Naval Research Laboratory were exploring pulsed RADAR to detect objects (Jensen, 2009). On the other side of the Atlantic, the United Kingdom was concerned about the increasing arial power of Nazi Germany in late 1930. It asked Watson-Watt to invent a microwave system that could destroy aircraft. Such an idea had been proposed by Serbian-American scientist Nikola Tesla in 1924. Tesla claimed that death rays invented by him could stop an aircraft. Similarly, British scientists G.H. Matthews and T.F. Wall applied for a patent on death rays but failed. In 1934, Tesla again proposed a system that could hunt 10,000 planes from a distance of 250 miles. For this purpose, he proposed a plan to construct the 12-power plant worth $2 million each in the United States to generate 12 death rays. The cost was very high, so no one paid attention to his proposal. The idea of death rays was also not supported by Watson-Watt. However, he expressed his view about developing an early warning system to locate aircraft even in a cloudy or nighttime scenario. Work done by Watson-Watt and coresearchers resulted in the installation of a series of tall towers with transmitters and receivers (Chain Home) in Britain during the World War II outbreak. This is why a major development was seen in microwave remote sensing technology during World War II. RADAR had become a major technological; weapon to locate enemy aircraft. Correspondingly, in 1940, with the advancement in cavity magnetrons, the antenna size was reduced and a wave of up to 10 cm with high power could easily be generated. It helped to mount small antennas on aircraft to locate ships and enemy aircraft as air-toair RADAR. RADAR-like H2S had been developed, which aided navigation to and from the target surface. With the end of World War II, the application of RADAR diversified. Meteorologists tried to find applications for RADAR in the observation of weather forecasting.

2. Optical versus RADAR remote sensing Electromagnetic (EM) radiation is available with various frequency spectrums in the universe, from radio waves to gamma rays. Available EM radiation, which can easily travel through the earth’s surface to space without distortion, is useful for earth remote sensing. EM radiation used in remote sensing travels through some distance in the atmosphere before reaching the earth’s surface. EM radiation is scattered and absorbed by gases and particles present in the atmosphere. EM radiation is scattered when it interacts with particles or large gas molecules present in the atmosphere. Thus, EM radiation is redirected from its original path. EM radiation is strongly affected by atmospheric gases such as molecular nitrogen and oxygen, as well as by other constituents including

Introduction to RADAR remote sensing

methane, hydrogen, helium, and nitrogen compounds. The strongest absorption occurs at wavelengths shorter than 0.3 mm, mainly owing to ozone. The strength of the scattering that takes place depends on several factors: the wavelength of the radiation, the abundance of particles or gases, and the distance the radiation travels through the atmosphere. The optical remote sensing system (Fig. 1.1) is composed of the sun, the targeted surface, the atmosphere, and the sensor installed on the satellite (Kulkarni and Rege, 2020). Light emitted from the sun interacts with the atmosphere before its interaction with the targeted surface on the earth. Some part of this light energy reflects from the targeted object surface and transmits back to the sensor. Here, the reflection is of two types: diffuse and specular. Specular reflection is uncommon in remote sensing; however, it may be found on water’s surface, for instance (Ulaby and Long, 2015). Diffuse reflection is mostly found with natural surfaces. Often, the characterization of diffuse reflectance is done through a bidirectional reflectance distribution function. Radar remote sensing is also called active remote sensing (Singh et al., 2013). Radar remote sensing provides a unique view for gathering information about the earth’s environment from space, which cannot be achieved any other way (Fig. 1.2). Radar signals at higher wavelengths (greater than 1 cm) have the ability to penetrate clouds and rain to some extent. In RADAR remote sensing, the RADAR antenna acts as a source of energy in place of the sun, which is used in optical remote sensing. These qualities of RADAR system allow earth observation independently in any time of the day and under almost all weather conditions. The clouds are dense enough to obscure the ground surface completely in optical imagery, but they have little effect on microwaves. Ice clouds have almost little effect on microwave or EM radiation with wavelengths greater than 1 cm. However, water clouds have less effect on microwaves and a significant effect on EM radiation with wavelengths less than 2 cm. Rain has a greater effect on microwaves than clouds, but it is effective for wavelengths less than 4 cm. Otherwise, at greater than 4 cm wavelength, rain also has a minimal effect. Radar systems in remote sensing are coherent. Therefore, RADAR remote sensing executes measurements within a resolution cell through interference (constructive and destructive) amid signals scattered from separate targeted objects (Foster et al., 2011; Tedesco and Jeyaratnam, 2016; Singh and Singh, 2020). The beam of radiation emitted from RADAR in satellites illuminates the targeted surface; in turn, the target scatters the radiation in multiple directions. A part of this scattered radiation returns to a small RADAR antenna. Generally, emission and reception in this type of remote sensing are done by the same antenna. Therefore, it is called monostatic. To take earth observations, the spaceborne and airborne implements an aperture synthesis technique called synthetic aperture RADAR (SAR) (Rahman et al., 2010; Du et al., 2015). SAR uses the Doppler effect, owing to which high spatial resolution images can be obtained from earth observatory satellites.

5

6

Radar Remote Sensing

3. Fundamentals of RADAR Radar detects an object by analyzing alterations in the transmitting frequency and time duration. Primary RADAR consists of a frequency generator and a time control unit; a modulator with a transmitter to produce a modulated pulse; an antenna to transmit the pulse; a duplexer that helps bidirectional communications between the transmitter and the receiver; another antenna to collect reflected pulses from the target; a receiver to analyze this reflection; and some other units such as signal processing, data processing, and a display unit. The processing unit is to process the received signal to detect the target’s location, relative velocity, and direction. The target location can be calculated with the help of the travel time taken by the signal from the transmitter to the receiver. The relative velocity of the target can be measured by analyzing changes in the carrier frequency of the arrived signal. The direction or relative angle of the target can be determined by analyzing the reflected wave’s angle. 3.1 Radar block diagram and operation The RADAR operation can be described using a block diagram, as shown in (Fig. 1.3). The RADAR system has two sections: a transmitter and a receiver. Both sections have their respective operations and consist of different units. The RADAR operation is described next, along with the function of each unit. 3.1.1 Transmitter section (a) Waveform generator: The transmitter section’s first component is a waveform generator used to generate a repetitive chain of pulses. Usually, a magnetron is used as a microwave waveform generator for RADAR. A typical RADAR generates a low power on the order of megawatts, a pulse width on the order of microseconds, and a pulse frequency of 100 pulses per second for a target located at the 100 to 200nm range. This generated waveform travels through a transmission line and then is radiated by the antenna into space. (b) Transmitter: The generated waveform is then fed to the transmitter. The transmitter can be a magnetron, a traveling wave tube, or a transistor amplifier according to the required conditions. Usually, the magnetron is used as a transmitter for most RADAR systems, but if high power is needed, amplifiers are used. (c) Pulse modulator: A pulse modulator is used to build the synchronization of the waveform generator with the transmitter. The pulse modulator’s primary function is to switch the power amplifier on and off according to the generated pulses. (d) Duplexer: Usually, a single antenna is used for both the transmission and receiver of RADAR signals. The receiver antenna needs a lot of protection from damage because of the high power of the transmitter. Therefore, a duplexer is used to isolate the transmitter and receiver section. Thus, a duplexer is used to provide signals to the

Introduction to RADAR remote sensing

Figure 1.1 Basic diagram of the optical remote sensing system.

transmitter and channel the reflected echo signals to the receiver instead of the transmitter. The duplexer contains two gas-discharge devices: transmit-receive (TR) and antitransmit-receive (ATR). The function of TR is to protect the receiver at the time of transmission, and ATR directs echo signals toward the receiver at the time of reception. 3.1.2 Receiver section (a) Low-noise radio-frequency amplifier: The receiver of the RADAR must be of the superheterodyne type. This unit acts as the input stage or first stage of the receiver section; the first stage should be a low-noise radio-frequency (RF) amplifier such as a lownoise transistor or parametric amplifier. The primary function of this stage is to generate an RF pulse that should be proportional to the echo of the transmitted signal. (b) Mixer and local oscillator: It converts RF signals into the intermediate-frequency (IF) pulse, and then these IF pulses are fed to the IF amplifier. Usually, the input stage of the receiver section is an RF amplifier, but sometimes the mixer is also used as the input stage instead of the RF amplifier. This leads to less sensitive functioning of the receiver section because of the high noise figure.

7

8

Radar Remote Sensing

Figure 1.2 Basic diagram of RADAR remote sensing system.

Figure 1.3 Block diagram for RADAR operation. IF, intermediate frequency; RF, radio frequency.

Introduction to RADAR remote sensing

(c) Intermediate-frequency amplifier: Its function is to amplify IF pulses generated by the mixer and maximize the signal-to-noise ratio (SNR) of the received signal. It is designed to act as a matched filter. Its frequency response function should have a magnitude equal to the echo signals to maximize the peak signal-to-mean-noise power ratio. The phase spectrum behaves like a negative of the phase spectrum of the reflected signals. It also enhances the receiver section’s ability to detect echo signals by removing the effects of unwanted signals. The center frequency of a typical IF amplifier for RADAR used as air surveillance is 30e60 MHz, and the bandwidth must be on the order of 1 MHz. The receiver’s bandwidth is completely associated with the bandwidth of the IF amplifier. (d) Second detector or demodulator: After maximizing the SNR in the IF amplifiers, signals are transferred to the second detector to demodulate the pulses. This unit consists of a crystal diode that simply extracts the pulse modulation. (e) Video amplifier: It amplifies the received signals to the level where they can be appropriately displayed on the screen. The screen is usually a cathode ray tube (CRT). (f) Threshold decision: This unit is used to decide the existence of the object in space. Generally, some threshold limits are set for comparison with the magnitude of the received signals. If the output signal surpasses the threshold limit, it indicates the object’s existence in the space; otherwise, only noise is present. (g) Display: It shows the final output of the RADAR. Plan position indication is the most common display unit used in a RADAR system with a CRT. It displays the range and location of the object in polar coordinates. The receiver outputs modulate the CRT’s electron beam intensity to make the electron beam sweep in the outward direction from the center of the tube. This beam rotates at an angle with the axis of the antenna.

3.2 The RADAR equation The RADAR uses RF EM radiation to detect a target. Radar’s waveform is a pulsemodulated narrow sinusoidal carrier required to capture target information. The target’s location or range as well as distance can be determined by the time (T) taken by the signal transmission from the transmission to return. Because the EM wave always travels at the speed of light, c ¼ 3  108 m/s, range R is: R¼

cT 2

(1.1)

9

10

Radar Remote Sensing

where T is the total elapsed time taken by the two-way propagating RADAR wave. Therefore, half of this time is used as one-way propagation. The wavelength of the propagating signal: c l¼ (1.2) f where f is the frequency of a sinusoidal wave. The RADAR equation simply represents the RADAR range’s physical dependence on transmit power characteristics and echo signals. It is helpful for measuring the maximum distance of the target from the RADAR and helps to comprehend the characteristics of the RADAR system and RADAR operation. First, consider that EM energy propagates in the ideal condition of no dispersion. If the radiator used in the transmitting antenna is isotropic, EM energy propagates uniformly in all directions. Energy is spread out in the same amount through an imaginary spherical surface. Power density Ss at maximum distance R from the RADAR can be represented as the ratio of the transmitted power to the surface area of the sphere (4pR2): Ss ¼

Pt 4pR2

(1.3)

Sometimes RADAR uses a directive antenna to employ radiation in a particular direction; this increases the power density in the direction of radiation. This increased power in one direction is measured as gain G of the directive antenna. The gain of the antenna is defined as the directional power of the radiation. Thus, directional power density Sd at distance R with transmitting gain G is: Sd ¼

Pt G 4pR2

(1.4)

Target detection depends on the reflected power from the target at the receiver of the RADAR. The reflected power can be determined with the help of the RADAR crosssection (s). Because the amount of reflected radiation depends on the target area, a larger area will produce a larger RADAR cross-section. The power of the echo signal is a result of power density Sd and variable RADAR cross-section s: Pr ¼ Sd $s ¼

Pt Gs 4pR2

(1.5)

Hence, the power density of the reflected signal is: Sr ¼

Pr Pt s ¼ G 2 2 4pR2 4pR 4pR

(1.6)

The unit of the RADAR cross-section is the unit of area. It represents the size of the target as observed by the RADAR. The receiver antenna of the RADAR collects some

Introduction to RADAR remote sensing

part of the reflected signal and depends on the effective area of the antenna, Ae. Thus, the power received by the RADAR antenna is: Pe ¼

Pt s Pt GsAe G Ae ¼ 2 2 4pR 4pR ð4pÞ2 R4

(1.7)

Solving it for range R,  R¼

Pt GsAe 2

ð4pÞ Pe

1 4

(1.8)

The maximum range up to where a target can be detected is Rmax, and Pemin is the minimum detectable power of the signal:  1 Pt GsAe 4 Rmax ¼ : (1.9) ð4pÞ2 Pe min This is the basic RADAR equation. It illustrates how RADAR characteristics can influence the RADAR range. Essential RADAR parameters are the transmitting gain and effective area of the receiving antenna. The relation between the transmitting gain and effective area of the receiving antenna is: G¼

4pAe l2

(1.10)

Eq. (1.9) can be written in two other forms by substituting the values of G and Ae from the earlier equation, one by one:  14 Pt sA2e Rmax ¼ (1.11) 4p l2 Pe min 1  Pt G 2 l2 s 4 Rmax ¼ (1.12) ð4pÞ3 Pe min Eqs. (1.9)e(1.12) are three different forms of RADAR equation and show the different dependencies of RADAR range on wavelength l. In Eq. (1.9), the range is independent of the wavelength; in Eqs. (1.11) and (1.12), it is inversely and directly proportional at l1/2, respectively.

4. Types of RADAR Radio detection and ranging, or RADAR, is an active remote sensing sensor that can work day and night (Richards, 2013). By active means, it works similar to taking a photo

11

12

Radar Remote Sensing

in the dark using the flash unit of a camera. Radar follows the principle of echolocation. Echolocation is used by bats and sea mammals. It can be called a natural active remote sensing system. According to this principle, the RADAR executes three primary functions: (1) it transmits an EM signal of RF range toward a target, (2) it receives part of the backscattered signal from the target, and (3) it measures the strength and time delay of the backscattered signal (Richards, 1999). Based on the application, RADAR is classified as an imaging RADAR (used to obtain images of about 10 m to 1 km), an altimeter (to measure surface height variations), and a scatterometer (for reflectivity as a function of the incident angle, illumination direction, and polarization). Based on the process, RADAR is classified as real aperture RADAR (RAR) and SAR. Within the category of RAR, we have altimeters, scatterometers, and side-looking airborne RADAR. This RADAR typically focuses on the azimuth resolution, which should be directly proportional to the equivalent distance of the sensor from the surface. The human eye and an optical sensor has a similar angle format, and RAR has an angle time delay format. These are discussed subsequently. 4.1 Radar altimeter An altimeter uses the time delay relationship to determine the range distance with accuracy. A spaceborne altimeter has great precision (Queffeulou et al., 1999; R
emy and Parouty, 2009; Adodo et al., 2018). It can precisely measure down to a few centimeters even from a height of 800 km. These are popular RADAR systems that can be used on aircraft and terrestrial satellites. In addition, they are used in lunar missions and on interplanetary spacecraft. 4.2 Imaging RADARs Imaging RADAR takes photographs, but the image formation is completed with the RADAR waves and not by waves in the visible region. In remote sensing systems, we frequently use the term “image.” What is this imaging and how it is different from the nonimaging? Microwave altimeter data can be gridded to make a graphical representation of the spatial variability of RADAR measurements (Dubois et al., 1995; Landmann et al., 2013). However, we cannot call the use of these methods imaging. An imaging system measures properties associated with spatial variability directly without gridding independent measurements (Rahman et al., 2010). We also cannot call optical scanning imagers an imaging system. In an image, a hologram appears as an image when it is properly illuminated; however, upon a closer view of the photographic plate, only interference fringes of dark and bright lines appear, and not a picture. A hologram is an actual recording of scattered light energy. During reconstruction with the help of proper illumination, the actual

Introduction to RADAR remote sensing

object can be observed in this hologram. In a RADAR system, each echo is used to generate plenty of data points, and the obtained data points are associated with the spatial dimensions of the target scene. The collection of echoes sequence the outcomes of a two-dimensional RADAR image of the captured scene. Compared with optical imagery, an imaging RADAR provides nothing about the direction to the target. 4.3 Scatterometers Scatterometers provide an accurate estimation of the RADAR cross-section of the targeted object (Gtjymer and Zecchetto, 1993; Singh et al., 2021c). However, they compromise the estimation of range accuracy and spatial resolution. As usual, the RADAR principal time delay is used to calculate the range but only to locate the cross-section measurement. They have a vital role in research associated with the study of microwave interactions with surface features (Hillard, 2003; Long, 2017; Tripathy et al., 2019; Singh et al., 2020b). These systems can be flown by satellites and aircraft as well as by helicopters to measure cross-sections at different observation angles. Spaceborne scatterometers fall into one of two types: wind scatterometers and rain RADAR. Scatterometers are classified into two categories, bistatic and monostatic, according to the location of the antenna that transmits and receives. 4.4 Sideways-looking airborne radar Sideways-looking airborne RADAR (SLAR) is a satellite mounted on airborne imaging RADAR, developed in 1950. It always pointed in the perpendicular direction of the flight; therefore, it is called sideways-looking. SLAR is used to create low-resolution images to measure the distances of scattering targets on the earth’s surface. In SLAR, the RADAR is mounted on an aircraft flying at an altitude (H) and its movement is in a straight line above a reference surface. The antenna points sideways as well as down at the ground. Because of this, the RADAR beam illuminates a small part of a continuous swath across the imaged areas. The area of illumination is called the antenna footprint. However, this system cannot measure the range by similar constraints as the altimeter. It is able to map all natural resources such as the sea and ocean, flood monitoring, glacier, water, and all calamites. In it, the echo signal is used to modulate the beam of the oscilloscope instead of displaying it on the screen. 4.5 Synthetic aperture RADAR SAR works on the principle of SLAR but applies image processing techniques to generate high-resolution images. SAR synthesizes a large antenna 1e2 to 600 m long. SAR maps the earth’s surface at very high spatial resolution. Most civilian and military RADAR is dependent on SAR. It can provide deep information about soil moisture

13

14

Radar Remote Sensing

content, vegetation growth parameters, oceans, and other land surface parameters. The advantage of using a larger antenna is that it improves the azimuthal resolution; this aperture can be further synthesized. Inverse SAR (ISAR) and interferometric SAR (IfSAR) are two types of conventional SAR. ISAR generates a two-dimensional high-resolution image of a target. For this, ISAR analyzes the movement of targeted objects instead of the emitter to create the so-called synthetic aperture. As the name suggests, IfSAR uses interferometry on two SAR images to generate X, Y, and Z. The transparency of the atmosphere to RADAR’s operating wavelengths (1 cm to several meters) makes it more useful compared with instruments operating at optical and infrared frequencies.

5. Operational frequencies of RADAR Radar transmits and receives signals in the microwave region of the EM spectrum and at a particular polarization. The frequency range of the signal in this region is 1e90 GHz (or 0.3e100 cm in wavelength) and could be grouped into some kind of logical band name. Traditionally, RADAR bands used in earth observatories are divided into X, C, L, and P frequency bands (Tables 1.1 and 1.2). However, the names of frequency bands were randomly chosen as alphabet letters for defense security reasons in wartime. Generally, spaceborne RADAR systems operate with bands listed in the tables.

5.1 Characteristics of RADAR frequency 5.1.1 Atmospheric effect Microwave radiation is insensitive to aerosols in the atmosphere and is weakly affected by water vapor in the atmosphere. Clouds and precipitation in the atmosphere may affect some frequencies (Singh et al., 2021b,d). Moreover, frequencies below the X-band used in RADAR operations are generally independent of the effect of the atmosphere. However, the phase of the signal is affected by the components of the atmosphere. Table 1.1 Radar systems operating under a particular frequency band range. S. No.

Band

Wavelength (cm)

Satellite

1. 2. 3.

S-band P-band L-band

w94.5 w69.0 w23.5

4. 5.

C-band X-band

w5.6 w3.1

NovaSAR, NISAR-S Biomass ALOS-2, Palsar-2, SAOCOM-1, NISAR-L Sentinel-1, Radarsat-2, RCM TerraSAR-X, TanDEM-X, COSMO-SkyMed

Introduction to RADAR remote sensing

Table 1.2 Brief overview of radio and microwave bands. Band designation

Radar frequency range

Typical use

HF

3e30 MHz

VHF UHF L

30e330 MHz 330e1000 MHz 1e2 GHz

S

2e4 GHz

C

4e8 GHz

X

8e12 GHz

Ku

12e18 GHz

K

18e27 GHz

Ka

27e40 GHz

mm

40e300 GHz

Mobile or fixed voice communication Very long-range surveillance Very long-range surveillance Long-range surveillance, route traffic control, satellite communications Moderate range surveillance, air traffic control, weather prediction Airborne weather, satellite communications transmission, wi-fi devices Tracking, missile guidance, marine RADAR, airborne intercept, wireless computer networks High-resolution mapping, satellite altimetry, very small-apertureterminal systems on ships Few uses (absorbed by H2O vapors) Very high-resolution mapping, airport surveillance Experimental, high-speed wireless communications (5G)

5.1.2 Wave polarization Polarization is a major characteristic of an EM signal that influences backscattering. In a RADAR remote sensing system, linear polarization is more popular than circular polarization. Transmitted and received RADAR signals are polarized horizontally (H) or vertically (V). A signal is said to be linearly polarized in H polarization when the electric field of the EM signal is vibrating perpendicular to the plane of incidence, and if it is vibrating parallel to the plane of incidence, it is V polarization. In physics, these are called perpendicular and parallel polarization. To define the polarization of SAR imagery, only two alphabets are used. The transmitted polarization is represented by the first letter and the received polarization is represented by the second letter: (i) HH: Transmission horizontally; reception vertically (ii) HV: Transmission horizontally; reception vertically (iii) VH: Transmission vertically; reception horizontally (iv) VV: Transmission vertically; reception vertically

15

16

Radar Remote Sensing

(v) Quad-pol: H and V transmission; H and V reception (HH þ HV þ VH þ VV) HV and VH are called cross-polarization backscatter components, whereas HH and VV are referred to as co-polarization. Radar that is not fully polarimetric measures only the signal amplitude. However, fully polarimetric RADAR receives and emits SAR information in two orthogonally polarized states and simultaneously calculates the relative phase difference between the states. According to this process, we can get a scattering matrix that responds to the target to any combination of emitted and received polarizations. Radar sensitivity to an object of a specific size depends on the selection of the operating wavelength. For example, objects with a size less than the operating wavelength of RADAR are undetectable by that RADAR; however, these objects may cause partial attenuation to the signal. Consequently, RADAR signals of longer wavelengths (from the L-band and P-band) are not much affected by small leaves of vegetation in a forest. However, the signals interact with larger objects such as the trunks and bigger branches of trees. This outcome has a positive correlation with the large biomass on the earth’s surface. Compared with this, RADAR signals of shorter wavelength (X-band and C-band) give a positive correlation with the small and sparse biomass vegetation.

6. Backscatter mechanisms Before starting an actual content interpretation of an SAR image, we must have some understanding of RADAR signal interactions with different types of land covers. Here, some backscatter mechanisms are briefly described. 6.1 Direct backscatter g0D




This type of scattering occurs when the emitted signal is directly reflected back to the sensory unit after a single reflection from a surface perpendicular to the RADAR’s direction of illumination (Long and Drinkwater, 2000). Finally, we get a strong copolarization reflection and a bright SAR image. The case in which RADAR is oriented toward bare mountain slopes and leaves of a dense vegetation canopy can cause such scattering (Long and Hardin, 1994; Macelloni et al., 2003; Chaube et al., 2019). In this type of scattering, the emitted signal by the RADAR is sensed by the sensor after a single reflection. Here, the reflecting surface is perpendicular to the direction of RADAR illumination. It creates a strong polarization (HH or VV) reflection and gives the impression of brightness in the SAR image. Such a type of backscattering can be found in a situation in which the slope of a mountain is oriented toward the RADAR (Oza et al., 2019). A similar scattering effect can be observed in the case of leaves on a dense vegetation canopy surface (Fig. 1.4).

Introduction to RADAR remote sensing

Figure 1.4 Representation of direct backscatter.

6.2 Forward scattering In this case, the interacting surface results in little or no scattering in reverse to the RADAR antenna. This happens when the transmitted signal continues its path away from the RADAR antenna after reflection from the target surface (Cui et al., 2016). It appears dark in an SAR image for both co-polarization and cross-polarization (Rahman et al., 2010). Here, the targeted surface can be calm water or bare soil (at longer wavelengths) (Moran et al., 1998) (Fig. 1.5).

6.3 Diffuse scattering In this scattering, the RADAR transmitted signal scatters in a different direction after interaction with the targeted surface. The amount of signal reflected directly back to the RADAR is measured. The targeted surface is rough (such as a ploughed field or waves on water) relative to the RADAR wavelength (Vanonckelen et al., 2014). In addition, here, backscattering (co-polarization) increases with the roughness of the surface (Fig. 1.6).

17

18

Radar Remote Sensing

Figure 1.5 Representation of forward scattering concept.

Figure 1.6 Concept of diffusion scattering.

6.4 Double-bounce scattering g0DB




This type of scattering is observed when the targeted surface is perpendicular to the RADAR antenna orientation. Tree stems on a flat ground are an example of such a surface. Here, the corner of the structure provides a double-bounce scattering effect that reflects data in the direction similar to that of their origin (i.e., RADAR antenna). The good thing is that all reflected waves are coherent (same phase), because all multiple reflected wavelengths travel the same distance here. Coherence is possible because the polarization direction of the RADAR signals does not alter after reflections from the vertical target, and the same is possible only at co-polarization (Yueh et al., 2009). There is

Introduction to RADAR remote sensing

Figure 1.7 Concept of double-bounce scattering.

another type of double-bounce scattering called the specular type. This case arises when the forest is flooded and in irrigated rice fields, for ships in water, and for bridges and liquid platforms. In this type, the scatterer is replaced with lossless specular reflections on the liquid surface (Fig. 1.7). 6.5 Volume scattering g0V




This case arises when a three-dimensional targeted object causes multiple reflections to the RADAR signal after interaction with it (Wismann, 2000). It happens when the interacting targeted objects with RADAR signal (L-band) are twigs and branches of forest vegetation. However, the bushes, shrubs, or crops also create volume scattering, but at shorter C-band wavelengths. There is random orientation of the main targeted object, and so is the polarization (co-polarization and cross-polarization) of the backscattered wavelength (Fig. 1.8). The presence of all of these scattering mechanisms is not individual in a RADAR image. However, the image contains the collective effect of all of the scattering
mechanisms. Therefore, the total backscatter g0T of the effect can be expressed as: g0T ¼ g0D þ g0DB þ g0V þ g0others :

19

20

Radar Remote Sensing

Figure 1.8 Concept of volume scattering.

7. Radar image characteristics 7.1 Sigma-nought The brightness of the RADAR image is generally expressed in s0 (sigma-nought) and is defines as the RADAR backscatter per unit area (Singh et al., 2021c). The unit of s0 is the decibel (dB). The expression used to calculate s0 from backscatter amplitude is:   2 s0 ¼ 10log10 DNAmp þ K: where DNAmp is the image pixel digital number measured in SAR amplitude image, and K is the calibration factor and depends on the SAR sensor and the processor system used. 2 ), Eq. (1.6) If the data are in terms of backscatter power (where DNP ¼ DNAmp becomes: s0 ¼ 10log10 ðDNP Þ þ K: 7.2 Gamma-nought For homogeneous targets, the s0 varies with theincidence angle between the ground normal and the sensorlook angle (higher in the near range part of the image and lower

Introduction to RADAR remote sensing

in the far range of the image). To obtain g0 (gamma-naught), some range-dependent factors can be removed. Then, the expression becomes: g0 ¼

s0 : cos4

where 4 is the angle of incidence. 7.3 Spectral signature A spectral signature is a function of the wavelength and is defined as the ratio of reflected radiation energy ½Er ðlÞ to incident radiation energy ½Et ðlÞ on an object. All matter on earth’s surface has separate values of spectral reflectance characteristics. The reflectance is directly related to the object’s color and tone in an image (Jensen, 2009). The color of an object can be described based on the wavelength reflected by it. The spectral reflectance of an object is averaged over separately defined wavelengths, which helps to differentiate it from others. The reflectance ðrðlÞÞ value varies with the wavelength and terrain features. It can be expressed in mathematical expression as:   rðlÞ ¼ Er ðlÞ=E ðlÞ  100: t

8. Application of microwave-based remote sensing Radar has a lot of applications and can be employed on the ground, in the air, on the sea, and in space. The main goal of ground-based RADAR is to detect, locate, and track aircraft or a space target. Radar on the sea or shipboard RADAR is used for safety and navigation, for detecting other ships, buoys, and objects on the sea surface, and for observing aircraft. Airborne RADAR is used for mapping or observing the land’s surface, for detecting other aircrafts, ships, and vehicles on the ground, for terrain and storm avoidance, and for navigation. Radar in space is used to guide spacecraft and in remote sensing satellites to map the land surface and sea. 8.1 Cryosphere The cryosphere, including the sea, ice, and snow, is an important part of the climate system that has a considerable impact on the surface energy balance, and so is a sensitive indicator of climate change (Rajak et al., 2015; Oza et al., 2019). Microwave remote sensing was extensively explored in the many cryospheric applications such as snow cover estimation, snow water equivalents over the Himalayas using a scatterometer, and sea ice estimation over Antarctica (Long and Drinkwater, 2000; Stuart and Long, 2008; Baumhoer et al., 2018; Singh et al., 2020a,b, 2021a,c). The maximum and minimum

21

22

Radar Remote Sensing

sea ice extent were observed in 2013 (Reid et al., 2015) and 2016 (Turner et al., 2017), respectively, over Antarctica. Such drifts in the ice over coastal regions are of great interest for addressing critical scientific issues and a better understanding of the leading causes of ice shelf breakdown and their related impacts on the rise in sea levels (Timmermann et al., 2001; Srivastav and Kumar, 2021; Srivastav et al., 2021). Moreover, the study of snow cover in the Himalayas is important for South Asian countries, because the Himalayas range is an important source of water for irrigation, power generation, and many other applications (Singh et al., 2016, 2018a,b, 2019). Satellite-based remote sensing has an important role in analyzing time-series variability over large and inaccessible Antarctica. It is of fundamental importance for tracking and predicting global rises in sea levels (Zibordi and Van Woert, 1993; Dierking et al., 2012). Remote sensing methods are used to monitor and map ice sheets, ice dynamics, mass balance estimation, and valuable information regarding the extent of sea ice concentration (R
emy and Parouty, 2009; Sood et al., 2020). 8.2 Biomass estimation Forests are an essential part of the environment ecosystem. They have a strong impact on the local and global environment (Yueh et al., 2009; Guo et al., 2019). Forests act as a carbon sink and have a vital role in estimating the global carbon budget. The canopy height, species, vegetation structure, and density need an accurate estimation after a fixed period (Paloscia et al., 2018). The exercise of periodic estimation is significant because short-time events such as seasonal and forest fires affect vegetation in forests (Naeimi and Wagner, 2010; Paloscia et al., 2018). Therefore, data on forests become more important, and RADAR remote sensing can provide that wealth of data. The imaging RADAR is mostly used to map forest biomass (Yueh et al., 2009). It helps in conservation studies, mapping and modeling of vegetation dynamics, and climate processes. The longer wavelengths are suitable for mapping the forest canopy and have high sensitivity to standing biomass. L-Band wavelengths are sensitive to biomass up to 70e80 ha1 and Pband wavelengths are sensitive to biomass of approximately 100 tha1. 8.3 Soil moisture The imaging RADAR is the most relevant system for estimating soil moisture remotely (Macelloni et al., 2003; Mladenova et al., 2009; Zhao et al., 2014). In precision agriculture, mapping soil with high resolution is important; medium-resolution (circle 500-m to 1-km) mapping is impossible with passive sensors (Ulaby et al., 1982). Mediumresolution mapping provides information about the forecasting of weather, climate, and hydrology. However, RADAR backscatter is limited because it is sensitive to the roughness of the surface (Mladenova et al., 2009; Oveisgharan et al., 2018). Therefore, difficulty persists in developing a robust model to estimate soil moisture over agricultural fields.

Introduction to RADAR remote sensing

8.4 Hydrology and oceanography The source of physical activities related to sea surfaces is the motion system of wind pressure and wind-driven oceanic activity from surface water waves to bottom water waves (Bartsch, 2010; Lin et al., 2017; Mandal et al., 2018; Ratheesh et al., 2019; Sikhakolli, 2019). Global and local climates are controlled by coupling between the ocean and atmosphere. This coupling depends on ocean surface winds (Singh et al., 2012). Therefore, information about ocean wind velocity is important for predicting activities related to oceanography, meteorology, and climate (Drinkwater et al., 2001; Liu, 2002; Stoffelen et al., 2016). It will help to improve the numerical ability of weather prediction systems for more precision in predicting future weather. Wind vector measurements through shipborne systems are biased geographically, as is the method it uses. Measurements done for wind velocity by shipborne systems are inaccurate; some reasons for this are the wrong positioning of anemometers, nonuniform ship motion, less skilled observers, poor instrumentation, and transmission errors. However, RADAR-based satellite-borne remote sensing can provide wind data globally with higher resolution and frequent sampling (Bhowmick et al., 2019). In satellite-borne systems, altimeters and microwave radiometers can calculate all weather wind speed; however, they do not estimate the direction of the wind (Bhowmick et al., 2013; Li and Shen, 2015; Lin et al., 2017). The wind’s direction is a crucial parameter for estimating ocean surface air momentum fluxes to fetch information about the atmosphere. To estimate the ocean surface air direction, a satellite-borne scatterometer is able to measure it (Jaiswal et al., 2019; Sikhakolli, 2019). Satellite-borne scatterometers have high accuracy and high resolution for wind speed and direction estimation even under cloudy conditions. Wind scatterometers have the speciality of measuring near-surface wind velocity over oceans. Rain RADAR is another type of scatterometers used to estimate the properties of rain clouds; it accurately measures the clouds’ RADAR cross-section. When the operating RADAR frequencies are above 10 GHz, scattering from water droplets in the atmosphere becomes significant because the wavelength is small and approximately equal to the size of the water droplet. When the size of the droplets becomes large, they are sustained in the cloud and will fall to the earth’s surface as rain. 8.5 Air traffic control and navigation RADAR is the most important system in airports all over the world to control the air traffic route as well as the area of airports. The departure and arrival of aircrafts and the airport ground vehicle traffic are controlled by high-resolution RADAR. It is also used to guide aircraft to safe landing during bad weather. Ground-based RADAR with the high-resolution mapping is used for aircraft navigation. Marine and aviation RADAR systems are generally used for aircraft navigation. When a craft or

23

24

Radar Remote Sensing

object comes within the range of a RADAR navigator, the distance and angular direction of the object is mapped. Spaceborne RADAR is used for guidance during landing on the moon or other planets and to analyze the earth’s surface. Ground-based RADAR is used for detecting satellites. 8.6 Military The uses of RADAR systems are larger in the military as the civilian application. It is used to control and guide weapons and troops in the field for navigation, for surveillance, and to detecting any object, force, or vehicle in prohibited regions. In the military, not only ground-based, but also airborne and spaceborne RADAR is employed.

References Adodo, F.I., Remy, F., Picard, G., 2018. Seasonal variations of the backscattering coefficient measured by radar altimeters over the Antarctic Ice Sheet. Cryosphere 12, 1767. https://doi.org/10.5194/tc‒12‒ 1767‒2018. Bartsch, A., 2010. Ten years of sea winds on QuikSCAT for snow applications. Rem. Sens. 2, 1142e1156. https://doi.org/10.3390/rs2041142. Baumhoer, C.A., Dietz, A.J., Dech, S., Kuenzer, C., 2018. Remote sensing of antarctic glacier and ice-shelf front dynamics-a review. Rem. Sens. 10, 1e28. https://doi.org/10.3390/rs10091445. Bhowmick, S.A., Kumar, R., Kumar, A.S.K., 2013. Cross calibration of the OceanSAT-2 scatterometer with QuikSCAT scatterometer using natural terrestrial targets. IEEE Trans. Geosci. Rem. Sens. 52, 3393e3398. Bhowmick, S.A., Cotton, J., Fore, A., Kumar, R., Payan, C., Rodríguez, E., Sharma, A., Stiles, B., Stoffelen, A., Verhoef, A., 2019. An assessment of the performance of ISRO’s SCATSAT-1 Scatterometer. Curr. Sci. 117, 959e972. https://doi.org/10.18520/cs/v117/i6/959-972. Chaube, N.R., Chaurasia, S., Tripathy, R., Pandey, D.K., Misra, A., Bhattacharya, B.K., Chauhan, P., Yarakulla, K., Bairagi, G.D., Srivastava, P.K., Teheliani, P., Ray, S.S., 2019. Crop phenology and soil moisture applications of SCATSAT-1. Curr. Sci. 117, 1022e1031. https://doi.org/10.18520/cs/ v117/i6/1022-1031. Cui, Y., Xiong, C., Lemmetyinen, J., Shi, J., Jiang, L., Peng, B., Li, H., Zhao, T., Ji, D., Hu, T., 2016. Estimating snow water equivalent with backscattering at X and Ku band based on absorption loss. Rem. Sens. 8. https://doi.org/10.3390/rs8060505. Dierking, W., Linow, S., Rack, W., 2012. Toward a robust retrieval of snow accumulation over the Antarctic ice sheet using satellite radar. J. Geophys. Res. Atmos. 117, 1e17. https://doi.org/10.1029/ 2011JD017227. Drinkwater, M.R., Long, D.G., Bingham, A.W., 2001. Greenland snow accumulation estimates from satellite radar scatterometer data. J. Geophys. Res. Atmos. 106, 33935e33950. Du, P., Samat, A., Waske, B., Liu, S., Li, Z., 2015. Random Forest and Rotation Forest for fully polarized SAR image classification using polarimetric and spatial features. J. Photogramm. Rem. Sens. 105, 38e53. https://doi.org/10.1016/j.isprsjprs.2015.03.002. Dubois, P.C., Zyl, J. van, Engman, T., 1995. Measuring soil moisture with imaging radars. IEEE Trans. Geosci. Rem. Sens. 33, 915e926. https://doi.org/10.1109/36.406677. Foster, J.L., Hall, D.K., Eylander, J.B., Riggs, G.A., Nghiem, S.V., Tedesco, M., Kim, E., Montesano, P.M., Kelly, R.E.J., Casey, K.A., Choudhury, B., 2011. A blended global snow product using visible, passive microwave and scatterometer satellite data. Int. J. Rem. Sens. 32, 1371e1395. https://doi.org/10.1080/ 01431160903548013. Gtjymer, T.H., Zecchetto, S., 1993. Applications of scatterometer winds in coastal areas. Int. J. Rem. Sens. 14, 1787e1812. https://doi.org/10.1080/01431169308954002. Guo, Y., Senthilnath, J., Wu, W., Zhang, X., Zeng, Z., Huang, H., 2019. Radiometric calibration for multispectral camera of different imaging conditions mounted on a UAV platform. Sustainability 11, 978. https://doi.org/10.3390/su11040978.

Introduction to RADAR remote sensing

Hillard, U., 2003. Assessing snowmelt dynamics with NASA scatterometer (NSCAT) data and a hydrologic process model. Rem. Sens. Environ. 86, 52e69. https://doi.org/10.1016/S0034-4257(03)00068-3. Jaiswal, N., Kumar, P., Kishtawal, C.M., 2019. SCATSAT-1 wind products for tropical cyclone monitoring, prediction and surface wind structure analysis. Curr. Sci. 117, 983e992. https://doi.org/10.18520/cs/ v117/i6/983-992. Jensen, J.R., 2009. Remote sensing of the environment: An earth resource perspective 2/e. Pearson Education India. Kulkarni, S.C., Rege, P.P., 2020. Pixel level fusion techniques for SAR and optical images: a review. Inf. Fusion 59, 13e29. https://doi.org/10.1016/j.inffus.2020.01.003. Landmann, T., Schramm, M., Huettich, C., Dech, S., 2013. MODIS-based change vector analysis for assessing wetland dynamics in Southern Africa. Rem. Sens. Lett. 4, 104e113. https://doi.org/10.1080/ 2150704X.2012.699201. Li, D., Shen, H., 2015. Evaluation of wind vectors observed by HY-2A scatterometer using ocean buoy observations, ASCAT measurements, and numerical model data. Chin. J. Oceanol. Limnol. 33, 1191e1200. https://doi.org/10.1007/s00343-015-4136-4. Lin, C.-C., Lengert, W., Attema, E., 2017. Three generations of C-band wind scatterometer systems from ERS-1/2 to MetOp/ASCAT, and MetOp second generation. IEEE J. Sel. Top. Appl. Earth Obs. Rem. Sens. 10, 2098e2122. https://doi.org/10.1109/JSTARS.2016.2616166. Liu, W.T., 2002. Progress in scatterometer application. J. Oceanogr. 58, 121e136. Long, D.G., 2017. Polar applications of spaceborne scatterometers. IEEE J. Sel. Top. Appl. Earth Obs. Rem. Sens. 10, 2307e2320. https://doi.org/10.1109/JSTARS.2016.2629418. Long, D.G., Drinkwater, M.R., 2000. Azimuth variation in microwave scatterometer and radiometer data over Antarctica. IEEE Trans. Geosci. Rem. Sens. 38, 1857e1870. https://doi.org/10.1109/36.851769. Long, D.G., Hardin, P.J., 1994. Vegetation studies of the Amazon basin using enhanced resolution Seasat scatterometer data. IEEE Trans. Geosci. Rem. Sens. 32, 449e460. Macelloni, G., Paloscia, S., Pampaloni, P., Santi, E., 2003. Global scale monitoring of soil and vegetation using SSM/I and ERS wind scatterometer. Int. J. Rem. Sens. 24, 2409e2425. https://doi.org/ 10.1080/01431160210154830. Mandal, S., Sil, S., Shee, A., Swain, D., Pandey, P.C., 2018. Comparative analysis of SCATSat-1 gridded winds with buoys, ASCAT, and ECMWF winds in the Bay of Bengal. IEEE J. Sel. Top. Appl. Earth Obs. Rem. Sens. 11, 845e851. https://doi.org/10.1109/JSTARS.2018.2798621. Mladenova, I., Lakshmi, V., Walker, J.P., Long, D.G., Jeu, R. De, 2009. An assessment of QuikSCAT Kuband scatterometer data for soil moisture sensitivity. Geosci. Rem. Sens. Lett. IEEE 6, 640e643. Moran, M.S., Vidal, A., Troufleau, D., Inoue, Y., Mitchell, T.A., 1998. Ku-and C-band SAR for discriminating agricultural crop and soil conditions. IEEE Trans. Geosci. Rem. Sens. 36, 265e272. https:// doi.org/10.1109/36.655335. Naeimi, V., Wagner, W., 2010. C-band scatterometers and their applications. In: Geoscience and Remote Sensing New Achievements. InTech, pp. 229e246. https://doi.org/10.5772/9102. Oveisgharan, S., Haddad, Z., Turk, J., Rodriguez, E., Li, L., 2018. Soil moisture and vegetation water content retrieval using QuikSCAT data. Rem. Sens. 10, 636. https://doi.org/10.3390/rs10040636. Oza, S.R., Bothale, R.V., Ram Rajak, D., Jayaprasad, P., Maity, S., Thakur, P.K., Tripathi, N., Chouksey, A., Bahuguna, I.M., 2019. Assessment of cryospheric parameters over the Himalaya and Antarctic regions using SCATSAT-1 enhanced resolution data. Curr. Sci. 117, 1002e1013. https:// doi.org/10.18520/cs/v117/i6/1002-1013. Paloscia, S., Pampaloni, P., Santi, E., 2018. Radiometric microwave indices for Remote Sensing of land surfaces. Rem. Sens. 10. https://doi.org/10.3390/rs10121859. Queffeulou, P., Chapron, B., Bentamy, A., 1999. Comparing Ku-band NSCAT scatterometer and ERS-2 altimeter winds. IEEE Trans. Geosci. Rem. Sens. https://doi.org/10.1109/36.763283. Rahman, M.M., Sumantyo, J.T.S., Sadek, M.F., 2010. Microwave and optical image fusion for surface and sub-surface feature mapping in eastern Sahara. Int. J. Rem. Sens. 31, 5465e5480. https://doi.org/ 10.1080/01431160903302999. Rajak, D.R., Singh, R.K., Jayaprasad, P., Oza, S.R., Sharma, R., Kumar, R., 2015. Sea ice occurrence probability data and its applications over the Antarctic. J. Geomatic. 9, 193e197. Ratheesh, S., Chaudhary, A., Agarwal, N., Sharma, R., 2019. Role of ocean dynamics on mesoscale and sub-mesoscale variability of Ekman pumping for the Bay of Bengal using SCATSAT-1 forced ocean model simulations. Curr. Sci. 117, 993e1001. https://doi.org/10.18520/cs/v117/i6/993-1001.

25

26

Radar Remote Sensing

Reid, P., Stammerjohn, S., Massom, R., Scambos, T., Lieser, J., 2015. The record 2013 Southern Hemisphere sea-ice extent maximum. Ann. Glaciol. 56, 99e106. https://doi.org/10.3189/ 2015AoG69A892. R
emy, F., Parouty, S., 2009. Antarctic ice sheet and radar altimetry: a Review. Rem. Sens. 1, 1212e1239. https://doi.org/10.3390/rs1041212. Richards, J.A., 1999. Remote Sensing Digital Image Analysis. Springer-Verlag, Berlin. Richards, J.A., 2013. Remote Sensing Digital Image Analysis. Springer-Verlag, Berlin Heidelberg. https:// doi.org/10.1007/978-3-642-30062-2. Sikhakolli, R., 2019. A study on SST dependent biases in SCATSAT-1 retrieved winds. In: Proceedings of the 2019 IEEE Recent Advances in Geoscience and Remote Sensing: Technologies, Standards and Applications. TENGARSS, pp. 137e140. https://doi.org/10.1109/TENGARSS48957.2019.8976049, 2019. Singh, D.K., Gusain, H.S., Mishra, V., Gupta, N., Das, R.K., 2018a. Automated mapping of snow/ice surface temperature using Landsat-8 data in Beas River basin, India, and validation with wireless sensor network data. Arabian J. Geosci. 11. https://doi.org/10.1007/s12517-018-3497-3. Singh, D.K., Gusain, H.S., Mishra, V., Gupta, N., 2018b. Snow cover variability in North-West Himalaya during last decade. Arabian J. Geosci. 11, 579. https://doi.org/10.1007/s12517-018-3926-3. Singh, K.K., Mishra, V.D., Singh, D.K., Ganju, A., 2013. Estimation of snow surface temperature for NW Himalayan regions using passive microwave satellite data. Indian J. Radio Space Phys. 42, 27e33. Singh, K.K., DewaIi, S.K., Singh, D.K., Mishra, V.D., Kaur, M., 2016. Monitoring of snow surface temperature in North-West Himalaya using passive microwave satellite data. Indian J. Radio Space Phys. 45, 20e29. Singh, R., Kumar, P., Pal, P.K., 2012. Assimilation of oceansat-2-scatterometer-derived surface winds in the weather research and forecasting model. IEEE Trans. Geosci. Rem. Sens. 50, 1015e1021. https:// doi.org/10.1109/TGRS.2011.2164410. Singh, S., Sood, V., Kaur, R., Prashar, S., 2019. An efficient algorithm for detection of seasonal snow cover variations over undulating North Indian Himalayas, India. Adv. Space Res. 64, 314e327. https:// doi.org/10.1016/j.asr.2019.04.016. Singh, S., Tiwari, R.K., Sood, V., 2020a. Estimation and validation of enhanced resolution brightness temperature products of SCATSAT-1. In: 2020 IEEE 5th International Conference on Computing Communication and Automation (ICCCA). IEEE, pp. 758e762. https://doi.org/10.1109/ ICCCA49541.2020.9250718. Singh, S., Tiwari, R.K., Gusain, H.S., Sood, V., 2020b. Potential applications of SCATSAT-1 satellite sensor: a systematic review. IEEE Sensor. J. 20, 12459e12471. https://doi.org/10.1109/ JSEN.2020.3002720. Singh, S., Tiwari, R.K., Sood, V., Gusain, H.S., 2021a. Detection and validation of spatiotemporal snow cover variability in the Himalayas using Ku-band (13.5 GHz) SCATSAT-1 data. Int. J. Rem. Sens. 42, 805e815. https://doi.org/10.1080/2150704X.2020.1825866. Singh, S., Tiwari, R.K., Sood, V., Prashar, S., 2021b. Fusion of SCATSAT-1 and optical data for cloud-free imaging and its applications in classification. Arabian J. Geosci. 14, 1978. https://doi.org/10.1007/ s12517-021-08359-7. Singh, S., Tiwari, R.K., Sood, V., Prashar, S., 2021c. In: Reddy, V.S., Prasad, V.K., Wang, J., Reddy, K.T.V. (Eds.), Unsupervised Snow Cover Classification Using Dual-Polarized SCATSAT-1 Satellite Data BT - Soft Computing and Signal Processing. Springer Singapore, Singapore, pp. 627e635. https://doi.org/10.1007/978-981-33-6912-2_57. Singh, S., Tiwari, R.K., Sood, V., 2021d. Cloud removal for satellite image using fusion of SCATSAT-1and MODIS data. In: 3rd Conference of the Arabian Journal of Geosciences. Singh, U.S., Singh, R.K., 2020. Application of maximum-likelihood classification for segregation between Arctic multi-year ice and first-year ice using SCATSAT-1 data. Rem. Sens. Applicat.: Soci. Environ. 18, 100310. https://doi.org/10.1016/j.rsase.2020.100310. Sood, V., Gusain, H.S., Gupta, S., Singh, S., Kaur, S., 2020. Evaluation of SCATSAT-1 data for snow cover area mapping over a part of Western Himalayas. Adv. Space Res. 66, 2556e2567. https://doi.org/ 10.1016/j.asr.2020.08.017. Srivastav, A.L., Kumar, A., 2021. An endeavor to achieve sustainable development goals through floral waste management: a short review. J. Clean. Prod. 283, 124669. https://doi.org/10.1016/ j.jclepro.2020.124669.

Introduction to RADAR remote sensing

Srivastav, A.L., Dhyani, R., Ranjan, M., Madhav, S., Sillanp€a€a, M., 2021. Climate-resilient strategies for sustainable management of water resources and agriculture. Environ. Sci. Pollut. Control Ser. https:// doi.org/10.1007/s11356-021-14332-4. Stoffelen, A., Li, Z., Kloe, J. de, 2016. Expected performance of the wind retrieval from the CFOSAT rotating fan-beam scatterometer. In: 2016 IEEE International Geoscience and Remote Sensing Symposium (IGARSS). IEEE, pp. 5804e5807. https://doi.org/10.1109/IGARSS.2016.7730516. Stuart, K.M., Long, D.G., 2008. Analysis of antarctic iceberg and sea ice melting patterns using quikscat. Int. Geosci. Rem. Sens. Symp. 5, 192e195. https://doi.org/10.1109/IGARSS.2008.4780060. Tedesco, M., Jeyaratnam, J., 2016. A new operational snow retrieval algorithm applied to historical AMSRE brightness temperatures. Rem. Sens. 8. https://doi.org/10.3390/rs8121037. Timmermann, R., Beckmann, A., Hellmer, H.H., 2001. The role of sea ice in the fresh-water budget of the Weddell Sea, Antarctica. Ann. Glaciol. 33, 419e424. https://doi.org/10.3189/172756401781818121. Tripathy, R., Bhattacharya, B.K., Tahlani, P., Gaur, P., Ray, S.S., 2019. Rice grain yield estimation over some Asian countries using ISRO’s SCATSAT-1 Ku-band scatterometer data. Int. Archiv. Photogramm., Rem. Sens. Spat. Informat. Sci. ISPRS Archiv. 42, 257e262. https://doi.org/10.5194/isprsarchives-XLII-3-W6-257-2019. Turner, J., Phillips, T., Marshall, G.J., Hosking, J.S., Pope, J.O., Bracegirdle, T.J., Deb, P., 2017. Unprecedented springtime retreat of Antarctic sea ice in 2016. Geophys. Res. Lett. 44, 6868e6875. https:// doi.org/10.1002/2017GL073656. Ulaby, F., Long, D., 2015. Microwave Radar and Radiometric Remote Sensing. Artech House. Ulaby, F.T., Moore, R.K., Fung, A.K., 1982. Microwave Remote Sensing: Active and Passive. Volume 2Radar Remote Sensing and Surface Scattering and Emission Theory. Vanonckelen, S., Lhermitte, S., Balthazar, V., Rompaey, A. Van, 2014. Performance of atmospheric and topographic correction methods on Landsat imagery in mountain areas. Int. J. Rem. Sens. 35, 4952e4972. https://doi.org/10.1080/01431161.2014.933280. Wismann, V., 2000. Monitoring of seasonal snowmelt on Greenland with ERS scatterometer data. IEEE Trans. Geosci. Rem. Sens. 38, 1821e1826. https://doi.org/10.1109/36.851766. Yueh, S.H., Dinardo, S.J., Akgiray, A., West, R., Cline, D.W., Elder, K., 2009. Airborne Ku-band polarimetric radar remote sensing of terrestrial snow cover. IEEE Trans. Geosci. Rem. Sens. 47, 3347e3364. https://doi.org/10.1109/TGRS.2009.2022945. Zhao, T., Shi, J., Lin, M., Yin, X., Liu, Y., Lan, H., Xiong, C., 2014. Potential soil moisture product from the Chinese HY-2 scanning microwave radiometer and its initial assessment. J. Appl. Remote Sens. 8, 083560. https://doi.org/10.1117/1.JRS.8.083560. Zibordi, G., Woert, M.L. Van, 1993. Antarctic sea ice mapping using the AVHRR. Rem. Sens. Environ. 45, 155e163. https://doi.org/10.1016/0034-4257(93)90039-Z.

27

This page intentionally left blank

CHAPTER 2

Microwave components and devices for RADAR systems Vikram Kumar1 and Dileep Kumar Gupta2 1

Department of Electronics and Communication Engineering, National Institute of Technology, Patna, Bihar, India; Remote Sensing Laboratory, Institute of Environment and Sustainable Development, Banaras Hindu University, Varanasi, Uttar Pradesh, India 2

1. Introduction During World War II, experimental methods were first achieved for aircraft detection. Along with microwave, other techniques were used, such as for the acoustic noise of aircraft engines, detecting electrical noise generated at the time of ignition, and performing thermal imaging with infrared sensors on aircraft. However, none of these techniques were as effective as microwave. The development of microwave systems made it possible to produce short electromagnetic (EM) pulses. This was a key advance that allowed modern radio detection and ranging (RADAR) systems to come into existence. The principle of RADAR was revealed at US Naval Research Laboratory in 1930, when scientists observed that an aircraft flying through the beam of a transmitting antenna caused a fluctuation in the received signal (RADAR). RADAR emits microwave signals designed to propagate and reflect back from the targeted object to the receiver. The change in time delay of the wave propagation gives information about the object when the microwave reflects back from the target. The object to be identified can be anything, such as an enemy aircraft, habitable zone, mountain, missile, or cloud. RADAR mainly detects distance, speed, and the specific direction of a required object. However, the distance of an object should be within an unambiguous range. Its main applications are in the military, remote sensing, air traffic control, law enforcement and highway safety, aircraft safety and navigation, ship safety, and space engineering. It works on the principle of microwave propagation and requires specific sources, a transmission line, antennas, and transmitter and receiver units. At the end of 19th century, Hertz’s thorough investigations validated the Faradaye Maxwell theory of electromagnetism (Bryant, 1988). This research opened up the EM spectrum between DC and visible light for scientific analysis and experiments. After few years, J.C. Bose reported his microwave experiments to the Royal Society in England (Bose, 2016). Bose invented the crystal radio detector, waveguide, horn antenna, polarizer, and some other apparatus for microwave frequencies. Several discoveries then led to a new scope of technology using the EM spectrum as a tool. In the early decades of the 20th century, one metallic box was a curious-looking device with some Radar Remote Sensing ISBN 978-0-12-823457-0, https://doi.org/10.1016/B978-0-12-823457-0.00012-4

© 2022 Elsevier Inc. All rights reserved.

29

30

Radar Remote Sensing

Figure 2.1 Block diagram of a typical microwave system.

grooves around its central cylinder and sandwich between magnetic plates called a magnetron. This cavity magnetron produced high-power microwaves (HPMs) and was the most important item used for war (News, 1970). In World War II, it was used for RADAR to detect enemy fighter planes. From that time, microwave has become part of engineering and research. Basic microwave systems have a block diagram, as shown in Fig. 2.1. Fig. 2.1 shows a typical block diagram of an HPM system (Benford et al., 2007; Barker and Schamiloglu, 2001; Kesari and Basu, 2018; Giri, 2004). Prime Power is the subsystem generates relatively low power electrical signal in a long pulse or continuous mode. The pulse power generators deliver short and intense electrical pulses by storing low power and long electrical pulse of prime power. HPM sources are a combination of microwave source along with modified the approach to connect with the pulse power system supplies higher input voltage and current. Whenever there is a change or a discontinuity in the geometry, there will be a local change in the field distribution and this property of discontinuity can be usage by mode converter sub-system. This subsystem uses to radiate microwave power with spatial compression, high directivity and high efficiency. All the system based on transmission lines and waveguide is basic structure of HPM source and mode converter. Waveguide consists of a metallic pipe with any sort of cross-section that could be used as a waveguide. This metallic waveguide carries an EM wave above the cutoff frequency, depending on its cross-sectional parameters and the material inside it.

2. Transmission line In radio-frequency (RF) engineering, a transmission line is a metallic structure along with some dielectrics designed to propagate an EM signal over some distances with minimum losses and distortion. A transmission line is a specialized cable, waveguide, or other

Microwave components and devices for RADAR systems

structure that mostly carries microwaves, or currents with a frequency high enough so that the nature of their wave must be considered (Collin, 1990; Montgomery, 1948). Transmission lines have been used to connect radio transmitters and receivers with antennas, distribute cables for television signals, route trunk lines calls, function as a microwave source for antennas for RADAR or satellite transmitters, as receiving antennas for microwave power meters or spectrum analyzers, as testing microwave components with network analyzers, and so on (Collin, 1990; Montgomery, 1948; Schmitt, 2002). There are various types of transmission lines for microwave, as discussed subsequently. 2.1 Coaxial cable Coaxial cables have a center conductor on which there is a dielectric insulator; on the top of the conducting wire is braid mesh. The coaxial cable can be defined as a capacitor, owing to the center conductor and metal mesh over it. It is defined as coaxial pffiffiffiffiffiffiffiffiffiffi capacitance. The characteristic impedance of a lossless line is given as L=C . If it is a lossy cable, the expression will be something like (R þ juL)/(G þ juC) (Collin, 1990; Montgomery, 1948; Schmitt, 2002). Thus, if R ¼ 0 and G ¼ 0, ju will cancel, pffiffiffiffiffiffiffiffiffiffi resulting in Z0 ¼ L=C , where L is the expression of inductance and C is the expression capacitance. The capacitance and inductance per unit length of a coaxial cable are given by Eq. 1. 2pε   F=m; ln D=d

L ¼

m D ln H m 2p d =

C ¼

(1)

Here, ε ¼ εr ε0 , where εr is the relative permittivity and ε0 is the absolute permittivity    ε0 ¼ 8:854 1012 F m ; also, m ¼ mr m0 , where mr is the relative permeability and m0 is the absolute permeability ðm0 ¼ 4p 107 H =mÞ(Collin, 1990; Montgomery, 1948; Schmitt, 2002) (Fig. 2.2).

Figure 2.2 Schematic of a coaxial waveguide with one outer hollow cosnducting waveguide and one inner central conductor positioned parallel to the axial direction.

31

32

Radar Remote Sensing

Figure 2.3 Schematic of a strip line with two parallel metallic plates of higher width and one metallic central conductor.

2.2 Strip lines Strip lines have a metallic strip inserted in the middle of a dielectric plate, which is sandwiched between two wide metallic planes, as shown in Fig. 2.3. Electric fields are confined within the substrate material, but magnetic field lines encircle the strip line. In the case of a microstrip line, a dielectric plate is sandwiched between the conducting ground plane and a metallic patch or line of suitable design. Therefore, the difference between a microstrip line and a strip line is that in the microstrip line, there is only one metallic plane, whereas in a strip line, there is another metallic plane on top of the dielectric. Thus, in the strip line, it is necessary to align the top metallic planes mechanically with the bottom metallic planes to leave no air gap between them. There is an advantage of strip line in that owing to two metallic ground planes, radiation leakage is small. Microstrip lines have some radiation leakage, which is useful when designing microstrip antennas. This leaky radiation is known as the fringing field. A microstrip line is a conductor of width W printed on a thin grounded dielectric substrate of thickness h and relative permittivity εr . Microstrip lines are used as a feeding technique for an inset feed. Moreover, the electrical properties can vary by the varying permittivity and permeability of the dielectric material. A diagram of a microstrip line is shown in Fig. 2.6. The effective dielectric constant of a microstrip line is given by Bhat and Koul (1989), Poole and Darwazeh (2015) as (Eq. 2.2):   εr þ 1 εr  1 12h 1 2 εeff ¼ (2) þ 1þ 2 2 W

Microwave components and devices for RADAR systems

The values of characteristic impedance used are mostly 50 and 75 U. The value that I use here is a 50U transmission line. The characteristic impedance, Zo, can be calculated as (Eq. 3):  8 60 8h w w > > þ for  1 ln pffiffiffiffiffiffi > < εeff w 4h h Zo ¼ (3) hw i i > 120 p hw w > > þ 1:393 þ 0:667ln þ 1:444 for  1 : pffiffiffiffiffiffi εeff h h h 2.3 Waveguide A waveguide is a conducting pipe with a rectangular cross-section used to guide the propagation of microwaves. A rectangular waveguide is commonly used to transport RF signals for transverse electric or transverse magnetic modes only. At any axial position, owing to the cross-section of waveguide forms metallic rectangular ring and is electrically shorted, therefore transverse EM modes are not supported to propagate through it (Collin, 1990; Montgomery, 1948; Schmitt, 2002). However, these waveguides have the advantage of the transmission of HPM with no leakage. Some important subcomponents of a waveguide system are: (a) Flanges: A metallic collar joined on the edges of a microwave waveguide with other microwave components to avoid losses caused by the leakage of microwaves. Flanges have desirable electric characteristics and add mechanical strength to the system. (b) Bend waveguide: Bends are a form of waveguide used to change the specific field of an EM wave. E-bends distort the electric field and H-bends distort the magnetic field to enable the waveguide to be bent in the required direction. (c) Twisted waveguide: These are used to twist the polarization of a source-generated mode to the required mode. The polarization twist is performed mostly for an orthogonal direction or for a 45-degree angle. (d) Attenuators: These consist of a resistive wane inside the waveguide to absorb microwave power according to its position. The volume occupied by the wane inside the waveguide results in attenuation, which is maximum if the wane is placed at the middle of waveguide. (i) Fixed attenuators: The position of the resistive wane is fixed in such a design, and it absorbs a constant percentage of power for particular frequencies. (ii) Variable attenuators: In this design, the position of the resistive wane can be changed to vary the percentage of attenuation. (e) Waveguide tee: Tees are junctions that require two microwave signals of the same frequency to be combined or split. Different types of tees are: • E-plane tee: This lies in the plane of the electric field, so the design consists of one arm connected vertically over a waveguide. It is a voltage or series junction.

33

34

Radar Remote Sensing

In an E-plane tee, the signal is divided into two parts with the same magnitude but in opposite phase, or two signals of the same magnitude and of opposite phase combined into one signal. • H-plane tee: The entire arm of the H-plane tee lies in the plane of the magnetic field and is divided among the horizontal plane. It is considered the current or parallel junction. (f) Magic tee: The combined form of H-plane and E-plane tee are used to form the magic tee. The four-port junction magic tee is also known as the hybrid tee and is a 3-dB coupler used in microwave systems. The magic tee may be used as a power combiner or divider, depending on the needs of the application. (g) Slide screw tuner: The sliding screw is placed at the top of the waveguide and is parallel to the E-plane. The role of probe attached with a sliding screw is to develop susceptance magnitude and control by the depth of penetration inside the waveguide. In addition, it can move along the axis of propagation (Fig.2.4). (h) Movable short: The adjustable load is able to move along the axial position of waveguide and is used to adjust the standing wave ratio. Microwave engineers commonly use small pieces of transmission line, usually as a printed planar form, arranged in certain patterns to build. Such printed planar forms, such as microstrip lines and strip lines, are used for low-power microwave transmission. However, for defense or ground-penetrating RADAR, waveguides are useful because they have the ability to handle high power (Giri, 2004) (Fig. 2.5).

3. Antennas A monopole and a dipole are two different antennas. A monopole is fed from one end and has a ground plane, whereas a dipole is fed on the nearest end of two metallic poles.

Figure 2.4 Microstrip line with width W and thickness h.

Microwave components and devices for RADAR systems

Figure 2.5 Diagram of (A) flanges, (B) E-bend and H-bend, (C) twisted waveguide, (D) variable attenuator, (E) E-plane tee and H-plane tee, (F) magic tee, (G) slide screw tuner, and (H) movable short.

Monopole antennas may perform as a dipole antenna by using a ground plane of equal length of monopole vertical radiator. Monopole antennas are half the size of their dipole counterparts; hence, they are attractive when a smaller antenna is desirable. Monopoles are mounted above the ground plane; they can also be grounded to an infinite conductor. However, mostly a ground plane of half the radiator in electrical length is sufficient. A dipole requires the same electrical length to be mounted vertically above the samegrounded conductor underneath. There are different types of dipole and monopole antennas with different wavelength vertical monopole antennas, such as l/2, l/4, l/8, or l/10 lengths, and many more (Balanis, 2009; Kraus and Marhefka, 2002; Chang, 2005). Fig. 2.6A and B shows typical diagrams of monopole and dipole antennas. The antenna consist of a helix or spring-shaped metallic structure connected to a feed and another pole connected with a ground plane, known as a helical antenna. Its polarization and radiation properties depend on the diameter, pitch, number of turns,

Figure 2.6 Schematic of (A) monopole antenna, (B) dipole antenna, and (C) helical antenna.

35

36

Radar Remote Sensing

wavelength, excitation, and spacing between helical loops. Fig. 2.6C shows a typical diagram of a helical antenna. A helical antenna works in two modes: normal and axial. • In normal mode, the antenna dimension is small compared with the wavelength. Helical antennas in normal mode behave like monopole antenna. Hence, their radiation patterns are similar. In this mode, the antenna can radiate in a broadside, omnidirectionally, and linearly polarized pattern. • An axial mode helical antenna can produce circularly polarized wavefronts, but the radiation pattern will be similar to any other directional antenna. Major lobes of the radiation pattern are along the axis of the helix, like a pointed dumbbell shape (Balanis, 2009; Kraus and Marhefka, 2002; Chang, 2005; Nakano et al., 1986; Kraus, 1949; Angelakos and Kajfez, 1967). Because this has to be a directional antenna, the radiating major lobe should be in one direction. For this reason, the other end of the helix is terminated with a flat metal sheet or ground reflector to reflect the waves in a forward direction. Circular polarization is often used in which the linear polarized wave of the transmitting and receiving antennas cannot easily be controlled (Balanis, 2009; Kraus and Marhefka, 2002; Chang, 2005). They are used in RADAR, transmission or reception of VHF signals through the ionosphere, animal tracking, satellite applications, and spacecraft communications. They can receive signals from rotating log-periodic antenna antennas and hence are preferred as good receivers. The component that radiates and receives EM energy from a fixed direction without a back lobe, and which has an open-ended waveguide structure in which the open end is flared so that it looks like a horn, is known as a horn antenna. There are several types of horns: sectional E-plane, sectional H-plane, and pyramidal horn. A schematic diagram of a pyramidal horn antenna is shown in Fig. 2.7. Horn antennas are significant for achieving higher gain and are used to transmit and receive microwave signals. The name is derived from the design of the flared metallic walls. The flared aperture of the output side is always greater than the aperture size of the input waveguide. The flaring helps match the impedance between the input waveguide and the output medium. The flared portion can be square, rectangular, or conical. The maximum radiation and response correspond with the axis of the horn in case of dominant mode radiation.

Figure 2.7 Schematic of pyramidal horn antenna.

Microwave components and devices for RADAR systems

4. Microwave filters Microwave filters are a two-port component used to provide frequency selectivity. They are designed to operate on signals in the megahertz to gigahertz frequency ranges. Microwave filters are a low-frequency electrical filter but differ in their implementation because circuit dimensions are about the electrical wavelength of microwave frequencies. In the design of a filter, one must first decide where to place the cutoff and infiniteattenuation frequencies, and what value of terminating impedance is required. Most microwave filters are bandpass or band-stop filters that absorb the energy of the unwanted frequency range and dissipate it in heat. The amount of attenuation depends on the resistive element property. Only occasional filter designs have the feature of channeling unwanted frequencies differently. Planner microwave filters are designed with distributed circuit elements along with transmission lines in place of lumped elements such as as inductors and capacitors that are used at lower frequencies. These distributed circuit elements are designed with varying transmission line design parameters and adding some untouched patch with transmission line. Untoched patch work on capacitive nature and transmission line length parameter work on inductive nature. These distributed circuit elements nature can make microwave filter design more difficult, but it also introduces a variety of useful coupling and transmission effects that are not possible at lower frequencies. Conventional microstrip lowpass filters such as stepped-impedance, open-stub, and end-to-end filters are widely used in many microwave circuits. The design layout is shown in Fig. 2.8 Microwave filters are designed using different methods (Hong, 2011; Pramanick and Bhartia, 2016; Cohn, 1950; Matthaei et al., 1980):

Figure 2.8 Layout of (A) a three-pole stepped-impedance microstrip lowpass filter, (B) a three-pole microstrip lowpass filter using open-circuited stubs, and (C) an end-coupled half-wavelength resonator filter.

37

38

Radar Remote Sensing

• Image parameter method: The impedance of an infinite chain of identical sections is known as image impedance. The image parameter method yields a usable filter, but precise design improvement is impossible in this method. • Insertion loss method (ILM): This method allows a systematic way to design and synthesize a filter with varying frequency response. ILM also allows filter performance to improve in a straightforward manner at the expense of a higher-order filter. A waveguide filter is a subsystem of microwave systems constructed by modifying the waveguide. Waveguides are hollow metal conductors inside which an EM wave is transmitted. In the waveguide placing different metallic plates having slots of different kinds forms different window structures and is used to design microwave filters. These slot cuts are a waveguide iris and are used to create an obstruction sitting within the waveguide. Such an obstruction provides capacitive or inductive elements, as shown in Fig. 2.9. A waveguide iris places a shunt capacitance or inductance across the waveguide. The magnitude of the reactance is directly proportional to the size of the opening of the iris. Hence, irises are manually designed and welded inside the waveguide.

5. Absorbers Microwave is an EM signal, so such waves have two components, electric and magnetic field traveling in the same direction on the orthogonal plane. To absorb radiations an absorbing material should have the capability of canceling both the electric and magnetic field of the EM wave. Microwave absorbers are used in a wide range of applications to decay stray or unwanted radiation that could interfere with a system’s operation (Tirkey and Gupta, 2019; Microwave, 2018). RADAR is composed of the transmission of EM into the atmosphere, which is then reflected back from the target. Conventional aircraft such as passenger airliners are easily detected by RADAR because of their shape and material properties. Metals used in aircraft are also strong reflectors of EM waves and can easily be detected using RADAR. Composite materials are also detected using RADAR, although they are not as easily detected as metals (Jayalakshmi et al., 2018). Microwave absorbers have uses in antenna testing labs, too, to provide test antennas in a noninterfering environment similar to open space. A typical layout of a microwave absorber with cone-shaped foam is shown in Fig. 5. Cone-shaped or pyramidal-shaped foam provides a large surface area for the absorption of waves.

Figure 2.9 Layout of a waveguide filter: (A) inductive iris, (B) capacitive iris, and (C) parallel L-C iris.

Microwave components and devices for RADAR systems

Microwave-absorbing materials (MAMs) are classified into two categories: magnetic and dielectric. An excellent MAM should be lightweight, thin, and able to cover a broad frequency range. Magnetic absorbers have excellent absorptivity but their densities are usually too high, whereas dielectric absorbers are much lighter in weight but they are not efficient in terms of absorptivity. Since the absorption loss is a function of conductivity, dielectric permittivity, and magnetic permeability of the material. The absorption loss in the material is therefore caused by the heat loss due to the alignment of electric and/or magnetic dipole in the presence of the EM field. The contribution to the absorption mainly comes from the magnetic losses (m) and dielectric losses (ε). Therefore, composite MAM prepared by integrating both magnetic and dielectric materials in the polymer matrix could lead to a better absorbing material. Conventionally, MAMs are fabricated in the form of sheets consisting of insulating or conducting polymers, such as rubber, ferrite, carbon black, graphite mixtures, and carbon fiber. The loss tangent defined as in Eq. 4 determines the MAM capability: tan de ¼

ε'' ε'

(4)

Parameter ε00 is an imaginary part of the dielectric constant that varies with frequency for the absorber; it encapsulate EM energy in the material. Parameter ε0 is the dissipative factor, which quantifies the amount of EM energy dissipated in heat energy. In a dielectric material, a higher loss tangent results in higher attenuation as the wave propagates through the material. However, a very high value of dissipation facto, will lead to an increase in conductivity and thus increasing reflection from the aireabsorber interface. Similarly, permeability for absorbers is complex. The magnetic loss tangent is defined in Eq. (5) (Microwave, 2018): tan dm ¼

m00 m0

(5)

The magnetic loss tangent is directly proportional to attenuation of the magnetic field as EM waves travel through the medium. In addition, the attenuation of a wave traveling through an absorber is measured by evaluating the complex permittivity and permeability of the material. The higher the losses, dielectric and magnetic, the better the wave attenuates in the material. When an EM wave navigates through lossy material, it decays exponentially by factor eax, where attenuation factor (a) of the wave. The absorber attenuation factor is given in Eq. (6) and depends on dielectric loss and magnetic loss: rffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi   dB 2pð8:686Þ m0 ε0  ð1 þ tan2 de Þð1 þ tan2 dm Þ  ð1  tande tandm Þ (6) a ¼ cm lo 2

39

40

Radar Remote Sensing

In addition to electric and magnetic loss tangents, electrical conductivity has an important role in dissipating the EM wave. Power dissipated in the absorber is denoted as A and can be expressed by Eq. (7): 1 1 1 A ¼ sjEj2 þ uε0 ε00 jEj2 þ um0 m00 jHj2 2 2 2

(7)

where E and H are the electric and magnetic field strength, respectively, of the incident EM wave. In addition, ε0 and ε are the permittivity of the free space and dielectric; m0 and m are the permeability of the free space and magnetic material; s is the conductivity of absorber; and u is the angular speed of the EM wave. The aircraft can be covered in materials that absorb RADAR signals to achieve stealth technology. The absorbing materials attenuate the incident microwave signals on aircraft and reflect back the very low power of the same microwave frequency. This idea is used to keep aircraft invisible from RADAR.

6. Microwave sources Microwave tubes use electron bunching or plasma oscillation to generate coherent EM radiation. This microwave radiation is produced when electron plasma oscillation generates polarity between slow wave structures (SWSs) or forms bunches of electrons that correlate with the input frequency (Fig. 2.10), therefore it produces a spontaneous emission with random phase and radiates it from the output port (Barker and Schamiloglu, 2001; Benford et al., 2007; Buck; Giebeler; Hull; Joshi et al., 2019; Kesari and Basu, 2018; Vintizenko, 2014).

Figure 2.10 Layout of a microwave absorber with cone-shaped foam.

Microwave components and devices for RADAR systems

Figure 2.11 Schematic of a relativistic magnetron.

6.1 Relativistic magnetron In a magnetron, depicted in Fig. 2.11, DC voltage is applied between the cathode and anode to enable electron emission from the cylindrical cathode. In the presence of a radial electric field and an external applied axial DC magnetic field, electrons revolve around the cathode (Hull, 1921; Vintizenko, 2014). At the cavities’ gap, electron bunching takes place, resulting in the formation of electron spokes. The kinetic energy of the electron beam is transformed into EM radiation and is often extracted from one of the anode cavities. A relativistic magnetron is a stable, compact, and reliable HPM source, and its oscillation frequency is determined by the depth of the anode cavities, so it is hard to tune this magnetron during its operation (Benford et al., 2007; Barker and Schamiloglu, 2001; Kesari and Basu, 2018; Hull, 1921; Vintizenko, 2014). The difference between an ordinary magnetron and a relativistic magnetron depends on the applied DC voltage and current. Principal characteristics of the relativistic magnetron are its tunability, manufacturability, and low efficiency. 6.2 Relativistic klystron amplifier Klystrons are an electron gun that generates an electron beam of desired parameters and electron velocity controlled by the DC potential applied to the anode (Benford et al., 2007; Barker and Schamiloglu, 2001; Kesari and Basu, 2018; Giebeler, 1969). As the electron beam enters grids of the buncher cavity, some of the electrons in the group accelerate while others decelerate, forming electron bunches with different velocities. At the catcher cavity of the device, the electron bunch releases kinetic energy to the RF waves. Finally, the unspent electrons are collected at the collector subassembly of the klystron. Operation of a relativistic klystron amplifier (RKA) is similar to that of conventional klystron amplifiers, but it uses a large-diameter annular beam instead of a solid pencil beam. This thin annular electron beam can propagate near the outer wall of the structure and can carry a much higher current at a given DC voltage. A RKA schematic is shown in Fig. 2.12.

41

42

Radar Remote Sensing

Figure 2.12 Schematic of a relativistic klystron amplifier.

6.3 Relativistic backward wave oscillator A backward wave oscillator (BWO) is a Cerenkov-type microwave oscillator with excellent tuning capability and a broad frequency coverage range (Benford et al., 2007; Barker and Schamiloglu, 2001; Kesari and Basu, 2018; Giebeler, 1969). It has a hollow cylindrical electron beam that interacts with the backward RF wave space harmonics of the waveguide mode, which causes absolute instability along the periodic SWS. During interaction of the electron beam with the RF waves at the SWS, the space charge bunch initiates a wave propagating in the backward direction; however, to protect the cathode, a reflector is used. If the velocity of the bunch electrons and of the RF phase are such that the total phase delay is in a loop, backward oscillation of electrons starts ðQn ¼ N ,2pÞ. The BWO structure that works with relativistic electron beams that can produce high-power coherent radiation is called the relativistic BWO (Fig. 2.13).

7. Mode converter Whenever there is a change or discontinuity in the geometry, there will be a local change in the field distribution. This means that at the change, new modes are generated. Far from the bend, the new modes may disappear or attenuate out. To achieve the highest

Figure 2.13 Schematic of a typical relativistic backward wave oscillator (BWO).

Microwave components and devices for RADAR systems

impact on the target, the radiated EM wave must be of the proper mode. Some of the HPM sources generate azimuthally symmetrical modes with a null in the direction of propagation. Such azimuthally symmetrical modes are converted to the required modes with maximum power in the direction of propagation by using a suitable mode converter (Benford et al., 2007; Kumar et al., 2019a,b; Chittora et al., 2015). Transforming the mode pattern of the RF waves into the desired modes can be achieved by adding deformation to the propagating waveguide. This is termed as the mode transducer. The mode converter is a controllable mode switching device used to transform RF energy selectively into the desired mode of propagation (Benford et al., 2007). Some important mode conversions are: • Converting TE mode to higher-order TE modes, and vice versa. • Converting TM to TE mode, because TM mode has a null at its axis propagation. • Converting TE to TM mode, because TM is the mode; this interacts with the azimuthal component of electric field electron plasma. • Converting circular waveguide modes to rectangular waveguide modes, and vice versa. Fig. 2.14 shows the requirements of the mode converter in microwave systems. Fig. 2.14A illustrates that the radiation pattern of the TM01 mode has a null at the axis of propagation, whereas the radiation pattern of the TE11 mode has highest gain at the axis of propagation, as shown in Fig. 2.14B. RADAR require directed energy propagation to target the object that needs to be exposed to EM radiation. Hence, mode converters need to convert source-generated TM modes to TE11 mode.

Figure 2.14 Comparison of microwave systems: (A) without a mode converter, and (B) with a mode converter.

43

44

Radar Remote Sensing

Figure 2.15 Ray optics.

8. Network analyzer In ray optics, EM waves incident on a surface result in reflected or transmitted waves, or both. As shown in Fig. 2.15, an incident wave is divide into reflected and transmitted waves. The microwave measurement setup using a network analyzer to measure a device under test (DUT) has the same reflection and transmission features as ray optics (Fig. 2.16). A generalized block diagram of a network analyzer is shown with the input microwave signal and its reflected and transmitted wave through DUT. To measure the incident, reflected, and transmitted wave, a network analyzer has certain properties (Keysight; VNA; Tektroix, 2016): • Measures the microwave components, devices, circuits, and subassemblies • Contains a source and receiver • Displays normalized amplitude and phase (for a particular frequency or power sweeps) • Offers calibration to fix noises Network analyzers measure the impedance or the reflection factor, and a sine generator stimulates the DUT. Two receivers take the place of the combination of a voltmeter and current meter at input port 1 and output port 2. These receivers characterize the

Figure 2.16 Microwave measurement setup using a network analyzer.

Microwave components and devices for RADAR systems

Figure 2.17 Types of test performed using a network analyzer. Measurements related to reflection performed at port 1, with measurements related to transmission performed at port 2.

response of the device by measuring the phase and amplitude of signals. Fig. 2.17 shows various measurement operations of a network analyzer. In addition, calibration capabilities are required to eliminate systematic errors and compute the appropriate ratios to produce reflection loss. Two basic types of network analyzers are: • Scalar network analyzer: measures the amplitude of the microwave signal • Vector network analyzer: measures the amplitude and phase of the microwave signal

9. Some other important microwave components •

•

•

•

•

Power sensor: A microwave power meter based on a sensor that measures electrical power at microwave frequencies. Usually, a microwave power sensor consists of a measuring block that contains a power-sensing element, connected to the meter via a USB cable. This measuring block is referred to as a power sensor (Anritsu; Absorption, 2016). Isolators: An isolator is a microwave device that allows microwaves to pass through them in one direction while microwaves in the reverse direction are absorbed by a polarized filter (Collin, 1990; Chang, 2005). Here, ferrite materials are used for the gyration of propagating mode polarization. Ferrites are ceramic-like materials and are a blend of metallic oxides and some small proportions of one or more additional metallic elements such as barium, manganese, nickel, and zinc. Circulators: Circulators are three matched port devices designed to allow microwave beam to flow in a clockwise direction and almost reject transmission in the anticlockwise direction. A microwave circulator is a multiport junction device. Power usually flows in the direction from port 1 to port 2, port 2 to port 3, and so on (Collin, 1990; Chang, 2005). Direction coupler: Power delivered to a load or an antenna can be measured using a sampling technique in which a known fraction of power is measured so that the total may be calculated. A number of coupling units used for this purpose is known as a directional coupler. Matched termination: This is a termination producing no reflected wave at any transverse section of the waveguide. It absorbs the entire incident wave. This is equivalent to connecting the line with its characteristic impedance.

45

46

Radar Remote Sensing

• Slotted section: This is a length of waveguide in which a nonradiating slot is cut on the broader side. This subsection of microwave system is used to measure the voltage standing wave ratio (VSWR) Measuring VSWR provides data about how efficiently RF power is transmitted from the power source, through a transmission line, into the load, e.g. from a magnetron through the waveguide to an antenna. For an ideal system, 100% of the energy is transmitted through it and there is no reflection therefore VSWR is 1.

10. Summary Microwaves are used for RADAR, heating food, targeting objects as weaponry, jamming electronics devices, finding precious elements inside the earth’s surface, biomedical imaging, and so forth. Various components are useful for operating a microwave system, depending on the frequency, power, and application. This chapter discusses various transmission lines in which waveguides are able to handle HPMs with no leaks. For low power, microstrip lines and strip lines are useful for designing RADAR circuitry, but for defense or mining, RADAR should contain waveguides as a transmission line. Waveguides are useful because they can handle HPMs. In the section on antennas, monopole, dipole, and helical antennas were discussed. In addition, the horn antenna was explained; and its flaring helps to match impedance between the input waveguide and output medium. Many microwave planar circuits contain conventional microstrip lowpass filters, such as stepped-impedance, open-stub, and end-to-end filters. Waveguide filters are also available in continuity to design waveguide-based microwave systems. Microwave environment tests in anechoic chambers have MAMs. MAMs are classified as magnetic and dielectric absorbing materials according to their material properties. An excellent MAM should be a lightweight thin sheet able to cover a broad frequency range. Aircraft can be covered in materials that absorb RADAR signals to achieve stealth technology and keep them invisible from RADAR. RADAR requires microwave sources and different high-power sources designed using microwave tubes, as discussed in this chapter. HPM sources are used to generate coherent high-power EM radiation. Because most sources generate azimuthally symmetrical modes, to obtain a directed radiation pattern at the axis of propagation, mode converters are used. Microwave components are characterized using a vector network analyzer and are used only for cold analysis. Some other microwave components are discussed that are widely use in RADAR systems.

References “Absorption RF Power Sensor Types » Electronics Notes,” Electronics Notes: reference site for electronics, radio & wireless. 2016 https://www.electronics-notes.com/articles/test-methods/rf-microwavepower-meter/absorption-power-sensor-types.php (accessed December 28, 2021). Angelakos, D.J., Kajfez, D., April 1967. Modifications on the axial-mode helical antenna. Proc. IEEE 55 (4), 558e559. https://doi.org/10.1109/PROC.1967.5583. “Power Sensors | Anritsu India,” Anritsu Home | Anritsu Asia Pacific. https://www.anritsu.com/en-in/ test-measurement/rf-microwave/power-sensors (accessed December 28, 2021).

Microwave components and devices for RADAR systems

Balanis, C.A., 2009. Antenna Theory: Analysis and Design, third ed. John Wiley & Sons. 3rd ed (With CD ). Barker, R.J., Schamiloglu, E., 2001. High-Power Microwave Sources and Technologies. Wiley-IEEE Press. Benford, J., Swegle, J.A., Schamiloglu, E., 2007. High Power Microwaves, second ed. CRC Press. Bhat, B., Koul, S.K., 1989. Stripline-like Transmission Lines for Microwave Integrated Circuits. New Age International. “Bose Institute | Founder: Sir J C Bose,” Bose Institute | Welcome. 2016 http://www.jcbose.ac.in/founder (accessed December 30, 2021). Bryant, J.H., May 1988. The first century of microwaves-1886 to 1986. IEEE Trans. Microw. Theor. Tech. 36 (5), 830e858. https://doi.org/10.1109/22.3602. Buck, H.W., Jan. 1902. The education of the electrical engineer. Trans. Am. Inst. Electr. Eng. 1165e1168. https://doi.org/10.1109/t-aiee.1902.4764043. Chang, K., 2005. Encyclopedia of RF and Microwave Engineering, 6 Volume Set. Wiley-Interscience. Chittora, A., Mukherjee, J., Singh, S., Sharma, A., Aug. 2015. Dielectric loaded TM01 to TE11 mode converter for S-band applications. IEEE Trans. Dielectr. Electr. Insul. 4, 2057e2063. https://doi.org/ 10.1109/tdei.2015.005038. Cohn, S.B., Jul. 1950. Design relations for the wide-band waveguide filter. Proc. IRE 799e803. https:// doi.org/10.1109/jrproc.1950.233777 no. 7. Collin, R.E., 1990. Field Theory of Guided Waves. John Wiley & Sons. Giebeler, R.H., Jan. 1969. High-power microwave generation. J. Microw. Power (2), 79e85. https:// doi.org/10.1080/00222739.1969.11688708. Giri, D.V., 2004. High-power Electromagnetic Radiators. Harvard University Press. Hong, J.-S., 2011. Microstrip Filters for RF/Microwave Applications. Wiley. Hull, A.W., Sep. 1921. The magnetron. J. Am. Inst. Electr. Eng. 9, 715e723. https://doi.org/10.1109/ joaiee.1921.6594005. Jayalakshmi, C.G., Inamdar, A., Anand, A., Kandasubramanian, B., Dec. 2018. Polymer matrix composites as broadband RADAR absorbing structures for stealth aircrafts. J. Appl. Polym. Sci. 47241. https:// doi.org/10.1002/app.47241. Joshi, N.R., Ramirez, A.D., Russell, S.D., Brock, D.W., 2019. Acoustic emission technology for high power microwave RADAR tubes. In: Acoustic Emission Technology for High Power Microwave RADAR Tubes. IntechOpen. Kesari, V., Basu, B.N., 2018. High Power Microwave Tubes. Morgan & Claypool Publishers. Keysight. Network Analyzers | Keysight. 2016 https://www.keysight.com/in/en/products/networkanalyzers.html (accessed December 28, 2021). Kraus, J.D., Marhefka, R.J., 2002. Antennas. Kraus, J.D., March 1949. The helical antenna. Proc. IRE 37 (3), 263e272. https://doi.org/10.1109/ JRPROC.1949.231279. Kumar, V., Dwivedi, S., Jain, P.K., 2019a. Circular sectoral waveguideTM01toTE11mode converter. Microw. Opt. Technol. Lett. 7, 1697e1701. https://doi.org/10.1002/mop.31789. Feb. Kumar, V., Dwivedi, S., Jain, P.K., 2019b. Experimental investigation and design of sectoral waveguide TM01 to TE11 mode converter. J. Microw. Power Electromagn. Energy 4, 276e295. https:// doi.org/10.1080/08327823.2019.1677428. Oct. Matthaei, G.L., Young, L., Jones, E.M.T., 1980. Microwave Filters, Impedance-Matching Networks, and Coupling Structures. Artech House. Theory and Application of Rf/Microwave Absorbers, Nov. 2018. Laird Tech Notes. https://www.laird. com/sites/default/files/2018-11/ABS-CS-RF%20Microwave%20Absorbers-Laird_081214.pdf. Montgomery, C.G., 1948. Principles of Microwave Circuits. McGraw-Hill Book Company, London. Nakano, H., Samada, Y., Yamauchi, J., September 1986. Axial mode helical antennas. IEEE Trans. Antennas Propag. 34 (9), 1143e1148. https://doi.org/10.1109/TAP.1986.1143944. News, Jan. 1, 1970. From World War II Radar to Microwave Popcorn, the Cavity Magnetron Was Thered Newsroom | News Details. Techno Professional Network for Engineers - ENGGtalks. December 30, 2021. https://www.enggtalks.com/news/94832/from-world-war-ii-radar-to-microwave-popcornthe-cavity-magnetro?c¼2859. Poole, C., Darwazeh, I., 2015. Microwave Active Circuit Analysis and Design. Academic Press, p. 103.

47

48

Radar Remote Sensing

Pramanick, P., Bhartia, P., 2016. Modern RF and Microwave Filter Design. Artech House. NRL HistorydRADAR > U.S. Naval Research Laboratory > NRL News. U.S. Naval Research Laboratory. https://www.nrl.navy.mil/Media/News/Article/2577147/nrl-history-november-1930/ (accessed December 30, 2021). Schmitt, R., 2002. Electromagnetics Explained. Elsevier. What Are Vector Network Analyzers | VNAs Explained | Tektronix. Test and Measurement Equipment | Tektronix. 2016 https://www.tek.com/en/document/primer/what-vector-network-analyzer-andhow-does-it-work (accessed December 28, 2021). Tirkey, M.M., Gupta, N., 2019. Electromagnetic absorber design challenges. IEEE Electromagn. Comp. Mag. 8 (1), 59e65. Vintizenko, I.I., Jan. 2014. Modifications of a relativistic magnetron. Tech. Phys. 1, 113e118. https:// doi.org/10.1134/s1063784214010204. Vector Network Analyzers(VNA) | Anritsu India. Anritsu Home | Anritsu Asia Pacific. https://www.anritsu. com/en-in/test-measurement/network-analyzer/vector-network-analyzers (accessed December 28, 2021).

CHAPTER 3

Theory of monostatic and bistatic radar systems Suraj A. Yadav2, Dileep Kumar Gupta1, 2, Rajendra Prasad2, Jyoti Sharma2 and Prashant K. Srivastava1

1 Remote Sensing Laboratory, Institute of Environment and Sustainable Development, Banaras Hindu University, Varanasi, Uttar Pradesh, India; 2Department of Physics, Indian Institute of Technology, Banaras Hindu University, Varanasi, Uttar Pradesh, India

1. Introduction Radar has been widely used for earth exploration with applications in geophysics, mineral exploration, hydrology, oceanography, topographic mapping, and agriculture, forest, and land use monitoring. Radar uses short-wavelength electromagnetic waves to sense radar reflectance in the frequency region from 300 MHz to 30 GHz. This region is called the microwave region. The most common radar bands used is listed in Table 3.1. Radar’s data are complementary with data collected with passive systems using the infrared, visible, and ultraviolet regions. Because of the higher penetrating capability of microwaves, the earth’s atmosphere acts as a transparent layer, leading to day and night data acquisition whenever required. Radar responses are related to received scattered and reflected electromagnetic waves. The relationship between radar responses is complex because it involves scattering phenomena affected by the detailed composition of the target, angle of incidence or transmittance wave, transmitted and received polarization, and radar geometry (Schaubert et al., 1981). Radar responses are quantified in terms of the radar cross-section (RCS) or the scattering coefficient of the illuminated target. Active remote sensing deals with the theoretical simulation of RCSs or scattering coefficients and their experimental validation (Blacksmith et al., 1965). Traditional radar systems operate in a monostatic manner consisting of a transmitting and receiving antenna close to each other, or sometimes the signal antenna acts as the both transmitting and receiving operations. However, in the case of bistatic and multistatic, the transmitter and receiver are usually separated by a distance comparable to the target distance (Bachman, 1965). More than one receiver used with one transmitter in radar systems is called a multistatic radar system. The development of theoretical modeling has numerous reasons: (1) to study sensitivity in terms of target, geometric, and electromagnetic wave parameters by developing a relationship between those modeled with measured scattering coefficient; (2) to build an understanding of the scattering mechanisms between microwaves and the volume

Radar Remote Sensing ISBN 978-0-12-823457-0, https://doi.org/10.1016/B978-0-12-823457-0.00019-7

© 2022 Elsevier Inc. All rights reserved.

49

50

Radar Remote Sensing

Table 3.1 Designation of radar bands. Bands

Frequency

Wavelength

P L S C X Ku K Ka

300e1000 MHz 1000e2000 MHz 2000e4000 MHz 4000e8000 MHz 8000e12,500 MHz 12.5e18.0 GHz 18.0e26.5 GHz 26.5e40 GHz

30e100 cm 15e30 cm 7.5e15 cm 3.75e7.5 cm 2.4e3.75 cm 1.67e2.4 cm 1.1e1.67 cm 0.75e1.1 cm

layer of the target; (3) to provide a tool for time series data extraction from the radar system; and (4) to use the simulated prediction for the experimental development of future advances in the radar system.

2. Bistatic and monostatic radar system configuration A radar system is an instrument that uses a transmitter antenna for target illumination and a receiver antenna for scattered or reflected energy measures. Fig. 3.1A shows the transmitter and receiver at different locations; it is called a bistatic radar configuration system. Fig. 3.2A represents a monostatic configuration (backscatter configuration) that shares a colocated transmitter and receiver using duplexers. The forward scatter alignment convention and backward scatter alignment convention were used to solve the scattering calculation of bistatic and monostatic radar systems, as shown in Fig. 3.1B and Fig. 3.2B, respectively.

Figure 3.1 (A) Bistatic radar; (B) forward scatter alignment convention.

Theory of monostatic and bistatic radar systems

Figure 3.2 (A) Monostatic radar; (B) backscatter alignment convention.

3. Radar equation This section presents the derivation of the radar equation, which develops an understanding of the interaction of polarized electromagnetic waves with the point and distributed targets and how radar has a vital role in capturing the fully polarimetric response from these targets. The target is commonly defined as an object seen by transmitting an antenna in a particular direction. On the other hand, the point target is self-contained, such as a car, drones, and aircraft. In comparison, a terrain-like object is defined as the distributed target. 3.1 Point target bistatic radar equation The bistatic radar configuration system for observing the point target is depicted in Fig. 3.1A. The transmitter antenna in a particular direction radiates q polarized power, represented by Pqt at distance Rt from the target. However, a receiver antenna at distance Rr receives the p polarized scattered power and is denoted by Ppr (Ulaby et al., 2014; Walker, 1980). Illuminated power density (riq ) on the target at distance Rt in a given direction is:
 Pqt Gt W i rq ¼ (3.1) 4pRt2 m2 Pqt

represents the power per unit area from the transmitting antenna and Gt is 4pRt2 the directive transmitting antenna gain. The reradiated/scattered power from the target is further received by the receiver at the angle of interest. The p polarized scattered power Ppr is:

where

Ppr

¼

riq spq

¼

Pqt Gt 4pRt2

spq ðW Þ

(3.2)

51

52

Radar Remote Sensing

where spq represents the RCS of the target. The RCS of the target is defined as the power intercepted by the target, which is isotopically scattered in all directions, and which produces an echo equal to that received from the target. The reradiated power is scattered in the form of spherical wavefronts over a spherical surface. By using Eqs. (3.1) and (3.2), we obtained the expression of scattered power density incident upon the receiving antenna from distance Rr :
 Ppr Pqt Gt W s rp ¼ ¼ spq 2 (3.3) 2 2 m 4pRr ð4pRt Rr Þ Upon incorporating the receiver intercept with an effective area ðAr Þ, radiation efficiency (§r ), and directivity (Dr ), the value of Ppr will be expressed as: Ppr ¼ §r Ar rsp ¼

Pqt Gt §r Ar ð4pRt Rr Þ2

spq

(3.4)

The effective area ðAr Þ and radiation efficiency (§r ) are given as: Ar ¼

l2 Gr Dr and §r ¼ 4p Dr

(3.5)

Upon substituting Eq. (3.5), we obtain the point target bistatic radar equation as: Ppr Pqt

¼

Gr Gt l2 ð4pÞ3 Rt2 Rr2

spq

(3.6)

3.2 Point target monostatic radar equation The bistatic radar equation shown in Eq. (3.6) can be simplified under the monostatic condition to obtain a monostatic radar equation for the point target. For a monostatic configuration, the same antenna is used to transmit and receive microwave leads to Gt ¼ Gr ¼ G . The range from target to the transmitter and receiver antenna will be the same (i.e., Rt ¼ Rr ¼ R ). Hence, the point target monostatic radar equation can be obtained by simplifying Eq. (3.6): Ppr Pqt

¼

G 2 l2 ð4pÞ3 R4

spq

(3.7)

In general, the point target bistatic and monostatic radar equation shown in Eqs. (3.6) and (3.7) is valid for the target whose physical dimension is smaller than the radar beam’s solid angle.

Theory of monostatic and bistatic radar systems

3.3 Distributed target radar equation for monostatic and bistatic configuration Eqs. (3.6) and (3.7) could be extended by integrating bistatic power and backscattered power over a distributed target of area A, respectively: ZZ t Pq Gt Gr ðqa ; 4a Þl2 0 r Pp ðqÞ ¼ spq dA; for bistatic configuration (3.8) ð4pÞ3 R2t R2r A

ZZ Ppr ðqÞ

¼ A

Pqt G 2 ðqa ; 4a Þl2 ð4pÞ3 R4

s0pq dA; for monostaticðbackscatteredÞ configuration (3.9)

where q is the incidence angle relative to normal incidence and s0pq is the pq-polarized bistatic/backscattering cross-section (bistatic/backscattering coefficient) and is defined as ensemble average (< >) over many measurements of RCS normalized to the distributed area (A) of the antenna beam. It is a dimensionless quantity: s0pq ¼

Cspq D A

(3.10)

The value of the scattering coefficient (s0pq ) could have been written in terms of incident electric fields and scattered electric fields in a particular direction under consideration. It is not a simple ratio of scattered power to the incident power. Therefore, the value of s0pq also depends on a direction (i.e., it might be larger in one direction compared with others): s0pq ¼

4pRr2 CjE r j2 D A jEt j2

(3.11)

where jEt j is the magnitude of an incident electric field; jEr j represents the magnitude of the scattered electric field in the direction of the receiver; and Rr2 is the distance from target to a receiver.

4. Radar cross-section per unit area/scattering coefficient system and measurement concepts 4.1 Fundamentals The basic aim of any radar system is to measure the scattering coefficient as defined in Eq. (3.11). The value of s0pq is determined by measuring reflected power from the point target

53

54

Radar Remote Sensing

or distributed target using the radar equation derived in Section 2.3. Such a type of reflected power might be accounted for by the integration of propagation factor (F). So, we can reformulate the radar equation as: Ppr Pqt

¼

Gr Gt l2 Ft2 Fr2 ð4pÞ3 Rt2 Rr2

spq

(3.12)

In contrast to Eq. (3.12), there are many measurement systems for which losses owing to data processing, system loss, and atmosphere propagation must be accounted for separately by introducing loss factors into the radar equation’s denominator. The radar equation can be reformulated by including the dependency of reflected power with the incidence/scattering angle, polarization, and frequency used in the radar system and expressed as the polarization scattering matrix (S): 2 pffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffi j4 # j4 sHH eHH sHV eHV S ¼ 4 pffiffiffiffiffiffiffiffiffi (3.13) p ffiffiffiffiffiffiffiffiffi j4 j4 sVV eVV sVH eVH Within this formulation, sHH is the RCS for horizontal transmission and horizontal reception. Similarly, sVV is the RCS for vertical transmission and vertical reception. However, sHV and sVH are the horizontal transmission and vertical reception, and vice versa. RCS can be written in terms of the normalized parameter in the case of surface clutter: sc ¼ so A

(3.14)

where, sc is the RCS of the clutter and so RCS per unit illuminated area (A). The illuminated area (A) for a low incidence angle can be approximated as: bRt Gt cs sec qi Ay 2

(3.15)

where b is a constant value that represents the beam shape and often is close to 1. However, the more precise value of RCS surface clutter at a higher angle for volumetric clutter is defined as: sc ¼ sv V

(3.16)

Here, sv represents radar reflectivity per unit illuminated volume (V). The illuminated volume (V) is approximated as: V ¼

bURt2 cs 2

where U is defined as a solid angle defined by antenna 3-dB beam-widths.

(3.17)

Theory of monostatic and bistatic radar systems

The selection of reflectivity measurement using a radar system depends on various other important factors, such as the resolution of the system, range sampling, target description, system stability, data objective, interference factor, and system calibration. These aspects will be discussed in the following sections. 4.2 Calibration and characterization of the radar system With the help of a radar system, data are commonly gathered by pulsed, single-frequency, and moderately short-pulse radar systems. They are used to illustrate many factors involved in characterization. A block diagram of a pulsed radar system shown in Fig. 3.3. To obtain a meaningful value for RCS, the radar system should be characterized and calibrated. Characterization of the radar system includes the measurement of optimal parameters defined in terms of their high sensitivity at an incidence/scattering angle, frequencies, and polarization. Radar measurement is carried out in two modes: a coherent or incoherent system (Currie et al., 1978; Mensa et al., 1983). Most radar systems work in the incoherent mode (i.e., the only amplitude of the target’s scattered signal). In contrast to coherent scattering measurement, both magnitude and phase measurements are carried out. The output of the phase channel is separately recorded to obtain coherent scattering measurement. Thus, types of measurement that occur in measuring radar responses are amplitude and phase, radar characterization, calibration, and validation for coherent and incoherent radar systems (Currie, 1989; Long, 1975). 4.3 Amplitude calibration There are various well-known methods for calibrating the amplitude of a radar system. However, no convention is universally accepted. The four most common methods are (Long, 1975; Skolnik, 1962): ➢ Absolute calibration In absolute calibration, the target’s radar responses are compared with the standard target’s radar responses (i.e., reference target) of the same cross-section at the same range.

Figure 3.3 Pulsed radar block diagram. PRF, pulsed radio frequency; RF, radio frequency.

55

56

Radar Remote Sensing

Some common marks used to calibrate radar systems are sphere, corner reflector, and flat aluminum plate. ➢ Relative calibration In relative calibration, the received signal’s transfer function is developed to define the transfer of power from the input to output. This type of calibration can be accomplished by varying input signals at known intervals and recording the resulting signal at those intervals to establish the transfer function. Furthermore, intermediate values between those intervals can be obtained to train and test the established function using the interpolation method. ➢ Mixed calibration Mixed calibration is the combination of absolute calibration and relative calibration. Absolute calibration can be used to establish the starting point of the transfer function. After that, relative calibration can be used to complete the transfer function curve within the dynamic receiver range at the same RCS. ➢ Indirect calibration In this calibration, one form of the radar equation and measured radar parameter is used to determine the relationship between radar responses from the target and RCS. A relative calibration is also incorporated within indirect calibration as part of the careful measurement of scattered radar responses from the target at a known reference point in the system.

4.4 Subsystem characterization Measurement of the radar system that requires characterization for meaningful values of a target response includes (1) a receiver, (2) a transmitter, (3) an antenna, (4) a waveguide path, and (5) a data acquisition system (Barton, 1976; Mensa, 1981). The radar receiver parameters required for characterization: • Bandwidth • Tx/Rx tube (duplexer) recovery • Preamplifier recovery • Gain variation with a range • Dynamic range and transfer function • Signal-to-noise ratio Significant features of a radar transmitter needed for characterization include: • Frequency • Peak power • Pulse width • Amplitude and phase stability

Theory of monostatic and bistatic radar systems

The characterization of antenna and waveguide parameters also influences the performance of radar during data acquisition: • Antenna gain • Side lobe level • Copolarization and cross-polarization isolation and levels • Losses • Antenna beam width Finally, the characteristics of data acquisition system that strongly affect radar response include: • Amplitude resolution • Stability • Time jitter and time aperture width • Slew characteristics

5. Measurement procedures Every RCS measurement system is unique and has its requirements for data acquisition. For the successful measurement of RCS, one should have a basic understanding of how to use data, how much data are required, and how to process the collected data. A detailed effect of the RCS from the target is analyzed once we finish radar measurements. The basic operation of the measurement system is divided into: (1) Exercise of radar responses against the simple target with known characteristics. (2) Characterization of radar system from recorded data to measure radar parameters. (3) Calibration and validation of data acquisition. (4) Data analysis. By keeping these procedures in mind, one should have reasonable confidence regarding radar data acquisition. To collect qualitative data from the target, we should follow several guidelines to ensure radar response from the target of interest: • Before starting an experiment, warm up your equipment. • Preliminary performance checks such as setting the knob, checking that all connections and antenna gain selections are within range, etc. • Measure transmitted power. • Use relative calibration to check minimal detected power by transmitted power. • Measure absolute data from the calibrated target. • Check the accuracy of calibration by measuring it many times. • Check relative calibration. • Record system control setting. • Record data. • Absolute calibration. • Turn off radar

57

58

Radar Remote Sensing

The final step of any radar system is to examine and interpret the measured data to establish correlations and dependencies with the target parameters (Skinner et al., 1998).

6. Procedure of bistatic specular scatterometer measurement and its calibration over natural terrain Fig. 3.4 shows the geometry of the bistatic scatterometer system. In the bistatic scatterometer configuration, the transmitter and receiver are placed opposite each other during measurement of the microwave response from the natural terrain. The transmitter consists of a high-power signal generator (E8257D, 10 MHz to 20 GHz), pyramidal dual-polarized horn antenna, and antenna support tower. The receiver consists of an EPM-P series power meter (E4416A), peak and average power sensor (E9327A, 50 MHz to 18 GHz), pyramidal dual-polarized horn antenna, and antenna support tower (Gupta, 2017; Gupta et al., 2015, 2016, 2018; Yadav et al., 2018, 2020). A 20-dB signal was transmitted. Polarization of the horn antenna was changed by using 90-degree E-H twisters. The portable iron antenna support towers were specially made in the workshop to carry the transmitting and receiving antennas. The antenna support tower had the ability to change the incidence angle, vertical height of the transmitting and receiving antennas from the ground surface, and linear distance of the transmitting and receiving antennas from the center of the test bed. The incidence angle, vertical height of transmitting and receiving antennas from the ground surface, and linear distance of the transmitting and receiving antennas from the center of test bed could be measured by the pointer

Figure 3.4 Geometry of the bistatic scatterometer system.

Theory of monostatic and bistatic radar systems

provided on the circular scale and linear scale, respectively. The position of the transmitting and receiving antennas were always adjusted because the focus on both antenna beams coincided at the center of the test bed for each incidence angle of bistatic scatterometer measurement. The laser pointer was used to match the focus of the transmitting and receiving antenna beams at the center of the test bed. Adjustment of the footprint of the antenna beam was needed to focus the common area of the test bed by adjusting the transmitting and receiving antennas beam at every bistatic scatterometer measurement. It tried to receive the coherent component of the scattered waves from the target in the specular direction. The antennas were placed in a far-field region from the center of the target to minimize near-field interactions. Calibration of system was done regularly during the experiment to ensure the integrity of the system. All observations were carried out by changing the incidence (qi ) and receiving angles (qr ) from 20 to 70 degrees in the elevation direction for azimuthal angle (4 ¼ 0). The incident wave became scattered in all directions owing to the interactions of the target. The scattered wave may be presented in terms of diffuse and coherent components. Most probably, the coherent component of the scattered wave was received in the specular direction. Fig.3.5 shows the aluminum sheet used during the calibration of the bistatic scatterometer system (Curie and well, 1989). This system has the ability to measuring reflected and transmitted power an angular range of 20e70 degrees at HH and VV polarizations for the X-band if l is the wavelength of the transmitted wave, Pt is the transmitted power, Gr and Gt are gain of receiving and the transmitting antennas, and R1 and R2 are the distance of the transmitting and receiving antenna, respectively, from the center of the illuminated area.

Figure 3.5 Aluminum sheet used during the calibration of the system (Pandey, 2011).

59

60

Radar Remote Sensing

For the average measurement of the bistatic scattering coefficient of the target (crop and soil in our case), the bistatic scatterometer system was calibrated for the target of a known radar cross-section. The power received from a standard target (i.e., a perfectly flat and smooth aluminum plate) is written as: PrAl ¼

Pt Grm Gtm l2 2

ð4pÞ

ðR1 þ R2 Þ2

(3.18)

where Gtm and Grm represent the maximum gain of the transmitting and receiving antennas, respectively. If the reflectivity of a reflecting target is jR0 j2 , the received power from the target can be expressed as: Pr ¼

Pt Gr Gt l2 jR0 j2 2

ð4pÞ

ðR1 þ R2 Þ2

(3.19)

The reflectivity of the target may be obtained by: r ¼ jR0 j2 ¼

Pr PrAl

(3.20)

For the Fraunhofer zone observation, range (R ¼ R1 ¼ R2 ) can be taken to be large enough so that R could be considered constant over the surface (A0). The radar equation (Ulaby, 1981, 1982) reduces to: I 0 2 Gr Gt s ds Pr ¼ Pt l (3.21) ð4pÞ2 R4 A0

From Eqs. (3.18) and (3.21), we get: Pr I ¼ Al pR2 Pr where:

(3.22)

I I ¼

s0 Gtn Grn ds A0

and Gtn ¼

Gt Gt ; Gtn ¼ Gtm Gtm

(3.23)

Theory of monostatic and bistatic radar systems

Assuming the scattering coefficient is constant over a 3-dB bandwidth of the antenna beam, we have: I 0 2 Pr s ¼ pR Al Gtn Grn ds (3.24) Pr A0

From Eqs. (3.20) and (3.23), we have: s0 ¼ pR2

jR0 j2 I

(3.25)

where I is the illuminated area of the target. An antenna beam falls on the target surface in the form of an ellipse. Fig. 3.6 shows the geometry of the illuminated area on the target by the incidence of a reflected beam. Its minor axis and major axis are given by: 4  Minor axis ¼ 2R2 tan az (3.26) 2 4 h   4  4 i Major axis ¼ R2 sin el sec q  el þ sec q þ el (3.27) 2 2 2

Figure 3.6 Geometry of illuminated area by the beam on the target surface.

61

62

Radar Remote Sensing

where 4el , 4az , and q are the elevation, azimuth, and look angle, respectively, of the antenna beam. Therefore, the illuminated area is given by: 4  4 h   p 4  4 i I ¼ R2 tan az sin el sec q  el þ sec q þ el (3.28) 2 2 2 2 2 Now, substituting the value of I into Eq. (3.25), the value of bistatic scattering coefficient is obtained as: 4  4  2jR0 j2 cot az cosec el 2  2  s0 ¼  (3.29) 4  4 sec q  el þ sec q þ el 2 2 Therefore, the bistatic scattering coefficient in unit dB can be written as: 2 4  4  3 2 az 2jR0 j cot cosec el 6 2  2 7 s0 ðdBÞ ¼ 10log10 4  4el  4 5 sec q  þ sec q þ el 2 2

(3.30)

Therefore, knowing the value of the elevation, the azimuth, the look angle of an antenna beam, and the reflectivity of the target, we can compute the average bistatic scattering coefficient of the illuminated area by the antenna beam over the target.

7. Summary In this chapter, an attempt was made to cover a broad area of monostatic and bistatic radar systems along with the fundamentals and measurements procedures of RSCs.

References Bachman, C., 1965. Some recent developments in RCS measurement techniques. Proc. IEEE 53, 962e972. Barton, D.K., 1976. Radar System Analysis. Artech House, Dedham, Massachusetts. Blacksmith, P., Hiatt, R., Mack, R., 1965. Introduction to radar cross-section measurements. Proc. IEEE 53, 901e920. Currie, N., Scheer, J., Holm, W., 1978. mm-wave instrumentation radar systems. Microw. J. 21, 35e38. Curie, N.C., Well, G.W., 1989. Radar cross section measurement concepts. Radar Reflectivity Measurement: Technique and Applications. Artech House, Norwood, MA, pp. 61e91. Currie, N.C., 1989. Radar reflectivity measurement: techniques and applications. Norwood. Gupta, D.K., 2017. Estimation of Crop Growth Variables and Soil Miostures by Microwave Remote Sensing Using Artificial Neural Networks. Gupta, D.K., Kumar, P., Mishra, V., Prasad, R., Dikshit, P., Dwivedi, S., Ohri, A., Singh, R., Srivastava, V., 2015. Bistatic measurements for the estimation of rice crop variables using artificial neural network. Adv. Space Res. 55, 1613e1623. Gupta, D.K., Prasad, R., Kumar, P., Mishra, V.N., 2016. Estimation of crop variables using bistatic scatterometer data and artificial neural network trained by empirical models. Comput. Electron. Agric. 123, 64e73.

Theory of monostatic and bistatic radar systems

Gupta, D.K., Prasad, R., Kumar, P., Vishwakarma, A.K., Srivastava, P.K., 2018. Vegetation water content retrieval using scatterometer data at X-band. Geocarto Int. 33, 602e611. Long, M.W., 1975. Radar Reflectivity of Land and Sea. Lexington. Mensa, D.L., 1981. High Resolution Radar Imaging. Artech House. Inc., Dedham, MA. Mensa, D.L., Halevy, S., Wade, G., 1983. Coherent Doppler tomography for microwave imaging. Proc. IEEE 71, 254e261. Pandey, A., 2011. Monitoring of Soil/Crops through Microwave Remote Sensing Using Soft Computational Techniques. PhD Thesis, Institute of Technology, Banaras Hindu University, Varanasi. Schaubert, D., Farrar, F., Sindoris, A., Hayes, S., 1981. Microstrip antennas with frequency agility and polarization diversity. IEEE Trans. Antenn. Propag. 29, 118e123. Skinner, J.P., Kent, B.M., Wittmann, R.C., Mensa, D.L., Andersh, D.J., 1998. Normalization and interpretation of radar images. IEEE Trans. Antenn. Propag. 46, 502e506. Skolnik, M.I., 1962. Introduction to radar. Radar Hand. 2, 21. Ulaby, F.T., 1981. Microwave Remote Sensing. Active and Passive. Ulaby, F.T., 1982. Microwave Remote Sensing Active and passive. Rader Remote Sensing and Surface Scattering and Emission Theory, pp. 848e902. Ulaby, F.T., Long, D.G., Blackwell, W.J., Elachi, C., Fung, A.K., Ruf, C., Sarabandi, K., Zebker, H.A., Van Zyl, J., 2014. Microwave Radar and Radiometric Remote Sensing. University of Michigan Press, Ann Arbor, p. 6. Walker, J.L., 1980. Range-Doppler imaging of rotating objects. IEEE Trans. Aero. Electron. Syst. 23e52. Yadav, S.A., Prasad, R., Vishwakarma, A.K., Sharma, J., Verma, B., Srivastava, P.K., 2020. Optimization of dual-polarized bistatic specular scatterometer for studying microwave scattering response and vegetation growth parameters retrieval of paddy crop using a machine learning algorithm. Comput. Electron. Agric. 175, 105592. Yadav, S.A., Prasad, R., Vishwakarma, A.K., Yadav, V.P., 2018. Random forest regression for the estimation of leaf area index of okra crop using ground based bistatic scatterometer. International archives of the photogrammetry. Rem. Sens. Spat. Informat. Sci.

63

This page intentionally left blank

CHAPTER 4

Review of microwave fundamentals and its applications Shivendu Prashar1, 3, Umesh Kumar Tiwari2 and Sartajvir Singh1 1

Chitkara University School of Engineering and Technology, Chitkara University, Solan, Himachal Pradesh, India; Central Scientific Instruments Organization, Solan, Chandigarh, India; 3Department of Civil Engineering, Indian Institute of Technology, Ropar, Punjab, India

2

1. Introduction Remote sensing is the process of collecting information about various aspects of objects on or above earth’s surface using satellites in space and airborne vehicles (Jensen, 2009). The spaceborne or airborne vehicles are equipped with active or passive sensors to sense data from earth. Fig. 4.1 shows the remote sensing system. The remote sensing community believes in the myth that the history of remote sensing is associated with the progress of photography (optical sensing) in the mid1990s. However, current microwave remote sensing does not allow us to believe this. The history of optical remote sensing shares little space with microwave remote sensing. Nevertheless, progress in the 1990s in the field of electricity and magnetism can be considered a root cause of microwave remote sensing. The beginning of modern physics and equations given by James Clerk Maxwell in that era brought more information about the physical world. Maxwell summarized all of the experimental concepts of electricity and magnetism given by Faraday and Oersted. He also explained their behavior in a medium with the help of four simple equations (Feynman et al., 1963): V $ E ¼ r=ε 0

(4.1)

V$B ¼ 0

(4.2)

V  E ¼ vB=vt

(4.3)

V  B ¼ m0 J þ m0 ε0 vE=vt

(4.4)

in which E and B are electric and magnetic field vectors, respectively. r is the electric charge density, J is electric current density, m0 is free space permeability, and ε0 is free space permittivity. He validated the existence of these equations in free space and in a situation in which there is no static or dynamic electric energy and magnetic energy. Maxwell explained that it is the wave nature of the electric and magnetic fields owing to which they propagate in free space as well. The oscillating electric field can induce

Radar Remote Sensing ISBN 978-0-12-823457-0, https://doi.org/10.1016/B978-0-12-823457-0.00010-0

© 2022 Elsevier Inc. All rights reserved.

65

66

Radar Remote Sensing

the oscillating magnetic field, and vice versa. Therefore, Maxwell named them electromagnetic (EM) waves. The theoretical speed of these EM waves is calculated as: qffiffiffiffiffiffiffiffiffiffiffiffiffi v0 ¼ 1=m ε (4.5) 0 0 Maxwell found that the speed of EM waves was approximately equal to the visible light speed calculated by Fizeau. With this fact, it was clear that the phenomena of electricity and magnetism are related to the wave nature of light. The theoretical description of EM waves was published by Maxwell in 1868 and confirmed experimentally by Heinrich Hertz after almost 20 years, in 1886. Hertz studied the phenomena of optics such as polarization, reflection, refraction, and interference to create standing waves for wavelength measurement on these EM waves. Interestingly, the measured wavelength (a little less than 1 m) is near the operating wavelength of low-frequency microwave radars. In the early part of the 20th century, it was proved that EM radiation behaves the same as visible light, heat radiation, and radio waves. Therefore, the total range of possible wavelengths was named the EM spectra. With time, the discrete ranges of wavelengths in EM spectra were named X-ray, ultraviolet, visible, infrared, microwave, radio, and so on. Fig. 4.2 shows the different wavelength ranges of the EM spectra.

Figure 4.1 Remote sensing system.

Review of microwave fundamentals and its applications

Figure 4.2 Electromagnetic spectrum.

In the EM spectrum, the frequency range from radio frequency and microwave frequency to visible frequency shows nominal transmissivity through the atmosphere. However, transparent frequencies in the regions of radio waves, microwaves, infrared, and visible ranges are able to analyze the surfaces, lower atmosphere characteristics, and aerosol particles from satellites, because their attenuation from the atmosphere is minimal (Long and Hardin, 1994). For microwave frequencies less than 10 GHz, the impact of attenuation caused by atmospheric factors can be ignored. However, with an increase in frequency, attenuation caused by the atmosphere increases, and when the frequency becomes greater than 100 GHz, the atmosphere becomes almost opaque for them. However, in this scenario, we can perform significant measurements of atmospheric characteristics because the lower portion of the atmosphere becomes opaque, and then the upper atmospheric layers can be analyzed (Paloscia et al., 2018). It is a curiosity for researchers who study the behavior of radiation when they travel through a homogeneous medium. The theory associated with this phenomenon is named the radiative transfer theory.

2. Theory of radiative transfer According to this theory, energy emitted by the stellar atmosphere was described by Chandrasekhar through the equation of radiative transfer. The theory explains the intensity of EM waves propagating through a medium that absorbs, radiates, and scatters them. Researchers using remote sensing for studies on the terrestrial atmosphere deal with media that are hot enough to emit an ample amount of microwave radiation. These media in a lower and medium atmosphere have a temperature above 200K, which is enough for microwave emission. At a longer wavelength (small frequency), the medium is assumed to be in thermal equilibrium and nonradiating (Feynman et al., 1963). These wavelengths can be ignored in models for precipitable water vapors, because their size is much smaller than these vapors. However, higher microwave frequencies are applicable for studying backscattering from these water vapors. The backscattered approach can be used for both types of imaging applications (i.e., active and passive)

67

68

Radar Remote Sensing

(Woodhouse, 2006; Ulaby et al., 1981). However, to study the interaction of these radiations with layers of snow and vegetation, local backscattering has to be considered. In the case of active sensors, local backscattering is small compared with incident microwaves; thus, backscattering is ignored there. The question is, how much energy emerges from part of the atmosphere when radiation is incident on it. For this, one has to evaluate the intensity of the radiation at arbitrary point p along the path of radiation (see Fig. 4.3). However, the radiation incident at the point is not the only radiation; the atmosphere itself radiates energy throughout the path from 0 to p according to Planck’s function. Furthermore, owing to absorption, both radiations (emitted and initial) of atmosphere decay exponentially (Ulaby et al., 1982). In a radiative transfer, it is a property of atmospheric volume and can vary within the medium. Therefore, we can use the term volume absorption coef  ficient af ðp0 Þ at frequency f and position p0 to define this process (Fung, 1994). For all types of continuous media (e.g., atmosphere), only absorption is considered; other types of energy losses (e.g., scattering) are ignored (Fung, 1994). In radiative transfer theory, energy absorption from an arbitrary point (i.e., p) to an instrument position ðp0 Þ is also measured. This measurement is called the optical depth or opacity (Elachi, 1987, 1988) and is represented by function sf ðp0 ; pÞ, Zp

0

sf ðp ; pÞ ¼

af ðp00 Þ dp00

(4.6)

p=

where p00 is points along the path. The optical depth behaves exponentially in an atmosphere volume. Therefore, the initial radiation can be written as esf ð0;pÞ . This term now represents the total absorption from 0 to point p. The radiation emitted at each point p00 along the path from p0 to p can then be defined according to the radiative transfer theory as (Fung, 1994; Ulaby et al., 1982): If ðpÞ ¼ If ð0Þe

sf ð0;pÞ

Zp þ

0

af ðp0 Þ Bf fT ðp0 Þgesf ðp ;pÞ dp0

(4.7)

0

where, If ðpÞ is the radiation intensity at position P along the path, If ð0Þ is the intensity at the initial point, and Bf fT ðp0 Þg is Plank’s constant, representing blackbody radiation. The term af ðp0 Þ is added to modify the blackbody radiation, because we cannot say that the object is a pure blackbody in the current scenario. 2.1 Microwave brightness temperature A practical illustration of the radiative transfer equation is shown in the brightness temperature. The brightness temperature gives an observable entity to the radiation

Review of microwave fundamentals and its applications

intensity characteristics. It is measured in Kelvin (Singh et al., 2016). The description of the radiative transfer equation can be given in terms of the brightness temperature using a linear conversion as:  If ¼ 2k l2 Tbt ð f Þ (4.8) where Tbt ð f Þ is the brightness temperature at frequency f. Now, the fraction factor can be written from Eq. (4.7) as: Tbt ¼ Tbt ð0Þesf ð0;pÞ þ

Zp

0

af ðp0 Þ Bf fT ðp0 Þgesf ðp ;pÞ dp0

(4.9)

0

Zp Tbt ¼ Tbt ð0ÞU þ

af ðp0 ÞT ðp0 Þesf ðpÞ dp0

(4.10)

0

where Tbt ð0Þ represents brightness temperature of the background. The term esf ð0;pÞ representing total absorption is replaced by transmissivity from point 0 to s. Factor sf ð0; pÞ gives information about the opacity of the medium along the total path and defines U in a range from 0 (fully opaque) to 1 (completely transparent) (Singh et al., 2013, 2016, 2020a). Eq. (4.10) is a more simplified expression that defines the radiative transfer equation while using microwaves. Let us discuss a case in which the atmosphere is homogeneous. Then, T(p) ¼ T, af ðpÞ ¼ a and sf ¼ ap throughout, and then Eq. (4.10) becomes: Zp Tbt ¼ Tbt ð0ÞU þ

aTeap dp ¼ Tbt ð0ÞU þ Tð1  UÞ

(4.11)

0

In Eq. (4.11), the contribution of radiation behind the atmosphere is represented by the first term and the contribution owing to the atmosphere is represented by the second term. For now, this equation can be observed as a base to model the response of microwaves that are backscattered from the surfaces or atmosphere (Singh et al., 2013, 2020a). 2.2 Faraday rotation Faraday rotation influences the polarized EM wave traveling through the ionosphere. In remote sensing, the ionosphere is important because microwave radiation has to pass through it, which may add a chance of error in measurement. Faraday said that whenever light passes through an applied magnetic field, the polarization of light rotates. Similarly, when microwaves pass through the ionosphere, the local magnetic field of it applies the Faraday rotation effect on microwaves. The waves of lower frequency (higher

69

70

Radar Remote Sensing

wavelength) are highly affected by this event (Ulaby et al., 1981, 1982). Faraday rotation is a problem in the case of linear polarization measurement. Therefore, the effect on linear polarization is significant in the L-band with a maximum of 100 degrees rotation. However, the P-band can have the worst limitation, experiencing hundreds of degrees of rotation. In the case of passive microwave sensing, errors in brightness temperature caused by this limitation can be up to 10 K (Singh et al., 2013, 2016). It decreases the credibility of the data taken by daytime satellites for various applications. These applications include an estimation of salinity and soil moisture measurements. However, estimation through full polarimetry is useful because only polarization is rotated by this effect, but the nature of polarization remains the same (Feynman et al., 1963).

3. Electromagnetic interaction with discrete objects In radiation transfer theory, it is assumed that the variation in dielectric properties amid layers is gradual and smooth (see Fig. 4.3). First, we need to know the difference between a surface and an object through EM interaction. If the dimensions of interacting mediums are large compared with the size of the wavelength, the boundary is called a surface; however, if the dimensions of one of the media are smaller than or the same as the interacting wavelength, it is called an object (Ulaby et al., 1986). For example, assume a wavelength interacts with a smooth sphere with a diameter approximately 100 times larger than the wavelength. In this case, the boundary becomes an infinite plane for a small-scale wavelength and acts as a surface rather than an object. However, with an increase in wavelength, the size of the sphere becomes important, and the whole boundary of the object will be considered (Janssen, 1993; Cui et al., 2016). Furthermore, if the boundary is smooth, an ordered field will be obtained, and if the boundary is rough or small objects are randomly located, an unpredictable field will be obtained. There are some phenomena in execution whenever the wavelength interacts with edges or as a whole with different objects and surfaces; we will discuss them in a later section. 3.1 Diffraction Diffraction explains the interaction of EM waves with an object rather than a surface. This term is generally used to define wave interaction at an aperture. We can consider the aperture here as a space between objects or a hole in a bigger object. The diffraction can be visualized as interference caused by a large number of superimposed waves. In diffraction, it seems that wavefronts remain the same in the center and bent at the sharp edges of an object. It is a feature of waves that whenever they meet a barrier or aperture, they create so-called phenomena of diffraction. Moreover, the width of the aperture decides the shape of the wave pattern after diffraction. Whenever the incident wave’s wavelength is comparable to the width of the aperture, the pattern after diffraction at sharp edges is almost circular and appears as a point source (see Fig. 4.4).

Review of microwave fundamentals and its applications

Figure 4.3 Description of radiative transfer equation through atmosphere volume.

Figure 4.4 Parallel electromagnetic waves arriving at an obstruction.

Diffraction in remote sensing is important for two reasons: (1) an antenna behaves like a hole in the measuring wave field during the reception mode and as a linear wavefront during transmission, which means the antenna becomes an aperture when it collects microwaves (Cui et al., 2016; Singh et al., 2021b). The larger the antenna (aperture), the narrower is the wave pattern transmitted. (2) The size of targeted objects (tree branches, wheat stalks, ocean waves, etc.) on earth’s surface are comparable to the operating wavelength of the sensor. However, scattering from an object defines the object’s small size and position at a large distance from the source of the EM wave (an incident EM wave can have straight and parallel wavefronts). The same situation can be occur with remote sensing in which instruments are at a large distance from the small-appearing targeted object at earth’s surface (Rehman et al., 2010; Singh et al., 2020b).

71

72

Radar Remote Sensing

3.2 Scattering Redirection of incident waves from an object is called scattering. Both scattering and diffraction are associated with coherent distortion of an incident EM wave. Reflection, refraction, and diffraction can be assumed to be types of scattering. However, essentially, scattering is a random distortion of an EM wave by an element of a size similar to or less than an incident EM wave. On the other hand, ordered scatterings from surfaces less than the wavelength and discrete boundaries are known as reflection and diffraction, respectively. The scattering cross-section ðsÞ quantifies the effectiveness of a scatterer. Once light is incident on an object or boundary, it may scatter in any direction (isotropic). Moreover, in remote sensing, it is difficult to analyze scattering in all directions (Long, 2017; Long and Drinkwater, 2000). Therefore, we define the association of the angle of observation with the directional scattering cross-section as:   Scattered powerper unit solid angle into direction 4 W U1   sð4Þ ¼ (4.12) Intensity of the incident plane wave=4p W m2 U1 where U represents a solid angle. To normalize the plane wave to U, the term 4p is added in the denominator. The integration of sð4Þ over all directions around the target will give the total scattered power. In the case of a perfect scatterer, the total energy scattered by it is equal to the incident energy backscattered by the total cross-section (Singh et al., 2021c). Therefore, the absorbance capacity of an object can be defined by its absorption cross-section. In addition, energy coming back in the direction of incidence after scattering is more important in microwave radar. It is also called backscattering or radar cross-section (RCS) (Peckham, 1991; Oza et al., 2019). 3.3 Radar cross-section Radar is an active system, and the amount of incident energy on target can be controlled by it. Moreover, in the radar system, backscattered energy to the sensor is of more concern (Ploscia et al., 2018; Long and Drinkwater, 2000). Therefore, energy arriving back to the radar system at range (r) defines RCS as: star ¼

Irec  2  4pr Iinc

(4.13)

where star is the targeted area, depending on the intensity received by the sensor after redirection of energy isotropically from the target. Eq. (4.13) is similar to Eq. (4.12) with addition of extra quantity r 2 . Scattering depends on many parameters of the target, such as the dimensions, dielectric properties, orientation, and nature of the surface. Moreover, these parameters vary with the modulation in the angle of observation, frequency, and polarization. The RCS approaches zero if the target

Review of microwave fundamentals and its applications

backscatters very low power toward the radar antenna. This case arises when the target object is small, absorbs all energy, is transparent, or scatters the wave in a direction other than the direction of the antenna Long and Hardin (1994); Macelloni et al. (2003); Chaube et al. (2019). On the other hand, the RCS may be much larger when the target has backscattered more energy than an isotropic scatterer. The case of a larger RCS can be found in Mie scattering from different scatterers or Bragg scattering from the target surfaces. For a perfectly conducting surface, the RCS depends on the frequency or wavelength of the wave. In active microwave remote sensing, one has to deal with scattering from discrete objects (water droplets), distributed targets (bare ground), and a combination of the two (vegetation and earth’s surface) (Ulaby et al., 1986; Du et al., 2015; Dubois et al., 1995). Interestingly, for an individual target, the area on which the measurement is taken does not affect the calculated RCS, because the returned power is mostly constant. Conversely, in the case of a distributed target, with an increase in the measurement area, the total reverse scattered energy increases in the same fashion, and so the RCS varies. If the measurement area is doubled, the backscattered energy as well as the RCS will also double; this case is not admissible in remote sensing. Therefore, to observe natural surfaces more conveniently, generalized methods are required that can measure all scatterers and distributed targets. Generally, microwave remote sensing is used to observe energy backscattered by a distributed area compared with a specific object. Mostly, the area concerned is the size of the instrument’s footprint. However, in image analysis, it is the area of a surface related to a single resolution cell or pixel of an image. It may cause a problem if for an application a comparison is made between estimations done through different instruments or through estimations obtained after combining nonimaging and imaging techniques (Oveisgharan et al., 2018). Therefore, there is a need to normalize measurements not based on the dimensions of footprints or pixels. To deal such situations, we need a quantity or expression that can relate the target area estimated from star to real geometrical area A of the earth’s surface; it is defined as the backscatter coefficient (sigma-nought) (Adodo et al., 2018; Singh et al., 2019, 2021c) (see Fig. 4.5): star s0 ¼ (4.14) A where s0 is also called the differential RCS or normalized RCS. The unit of s0 is the decibel. The expression used to calculate s0 from the backscatter amplitude is (Singh et al., 2021a; Sood et al., 2020):  2 s0 ¼ 10log10 DNAmp þK (4.15)

73

74

Radar Remote Sensing

where DNAmp is the image pixel digital number measured in the synthetic aperture radar (SAR) amplitude image; and K is the calibration factor and depends on the SAR sensor and the processor system used. Different types of surfaces (forest canopy or snow layer) illuminated by microwaves behave like a volume scatterer, which means the equal scattering of energy in all directions (Long and Hardin, 1994). In this case, s0 over a volume modulates according to the area projected by incident energy and becomes a function of the cosine of the incidence angle ð4in Þ, as in Eq. (4.15): s0V ¼ k cos4in

(4.16)

where k is constant and is estimated by some particular target properties. With the increase in 4in , the incident energy will be distributed over a larger surface area compared with cos 4in . The backscattered energy per unit target area is reduced owing to reduced incident energy. This means that for homogeneous targets, s0 varies with the angle between the ground normal and the sensor, or the incident angle (higher in the near range part of the image and lower in the far range of the image). To obtain g0 (gamma-naught), some of the range-dependent factors can be removed, and then the expression becomes: g0 ¼

s0 cos 4in

(4.17)

where g0 remains approximately constant for all 4in values. However, this quantity is again associated only with ground surface estimation and is not instrument specific. On the other hand, imaging radar systems directly measure brightness temperature, b0 . The precise value of earth’s area that is illuminated cannot be estimated because ground topography affects the viewing geometry and the real area is different from the calculated one (i.e., A) (Adodo, 2018). In the absence of topographic data, A is calculated using a plane reference area, and it gives an incorrect measurement of s0 . If we have topographic data, a more realistic and correct value of s0 can be estimated based on the average slope target ground area (see Fig. 4.5). Then, s0 can be defined as:

Figure 4.5 Three different ways to normalize the cross-section values by area over distributed targets.

Review of microwave fundamentals and its applications

s0 ¼

b0 sin 4in

(4.18)

Scattering is important for both types of microwave remote sensing (i.e., active and passive). In passive remote sensing, incident energy is scattered by particles (water droplets) or objects (tree canopies that come between the sensor and target) (Oveisgharan et al., 2018; Foster et al., 2011). Along the same lines, microwave energy from the targeted object may scatter out of the field of view of the sensor. In active remote sensing, the important thing is the directionality of the cross-section. Generally, in active remote sensing we are interested in backscattered energy; however, in some cases forward scattering is also considered. This case arises when microwaves interact with snow, the earth’s surface, or vegetation. Research related to scattering was carried out in 1870 and 1915. We know that the size of objects matters when an EM wave of comparable wavelength interacts with them. Microwaves are also EM waves; therefore, the quantitative theory of optical scattering is equally applicable to microwaves. There are two main regimes in scattering: (1) the size of the target object is at least 10 times larger than the wavelength, and (2) the size of the target object is smaller by a factor of 10 with respect to the wavelength. In the first case, the scattering properties can be calculated through geometrical optics; this is called nonselective scattering. The physical cross-section is directly associated with the scattering cross-section (Singh et al., 2021b). A common example of this phenomenon is opaqueness and whiteness shown by milk and clouds. In both scenarios, the size of droplets is larger than the incident wavelength. The second case is associated with Rayleigh scattering and is an example of microwave scattering by spherical water droplets present in the atmosphere. Based on these studies, the size and shape of the target are not the only factors affecting the scattering cross-section object; the target’s constituent material also affects the cross-section. At the optical region, the incident wave’s frequency and polarization become dominant factors influencing the scattering cross-section.

3.4 Spectral signature The spectral signature is a function of the wavelength and is defined as the ratio of reflected radiation energy ½Er ðlÞ to incident radiation energy ½Et ðlÞ on an object. All matter on the earth’s surface has a separate value of spectral reflectance characteristics. Reflectance is directly related to the object’s color and tone in an image ( Jensen, 2009). The color of an object can be described based on the wavelength reflected by it. The spectral reflectance of an object is averaged over separately defined wavelengths,

75

76

Radar Remote Sensing

which helps to differentiate it from others. The reflectance ðrðlÞÞ value varies with the wavelength and terrain features. It can be expressed by the mathematical expression:

 rðlÞ ¼ Er ðlÞ=E ðlÞ  100 (4.19) t

4. Interaction with inhomogeneous media Inhomogeneous media are a composition of arbitrarily distributed distinct elements that have significant cross-sections. In addition to radiation absorption and emission as done by continuous media, inhomogeneous media (volume scatterer) have constituents that scatter radiation. Therefore, the wavelength size and scatterers within a media differentiate it from homogeneous media. Such media are also called sparse media in which distinct elements in the medium have a lower volume fraction (100 GHz) when the waves travel through the troposphere and the lower stratosphere.

81

82

Radar Remote Sensing

These particles have a moderate physical temperature (not warm or cool); therefore, they emit microwaves. Thus hydrometers have ability to change the brightness temperature through absorption, emission, and scattering. Therefore, cloud and precipitation characteristics can be easily estimated through observations taken by passive microwave sensors in remote sensing (Cui et al., 2016). In active remote sensing, hydrometers attenuate radar pulses and generate unwanted signals (radar clutter) during observations through the atmosphere. On the other hand, this phenomenon can explore hydrometers through an analysis of scattering characteristics at high frequency. These sensors are called rain radars; they give weather forecasting images. Moreover, hydrometeors lower the efficiency of terrestrial and satellite radio communications. The properties of hydrometeors depend on the location, time of the year, and air temperature. Nonprecipitating cloud droplets are generally around 0.1 mm; however, the size of fog and haze (hydrometeors) is in the range of 50 mm or less than this. Such hydrometeors absorb microwave efficiently and provide less scattering. When cloud droplets become closer to raindrops (0.1e5 mm), their size is comparable to higher-frequency microwaves. Therefore, measuring microwave backscattering helps to estimate rainfall intensity as well as its distribution. There are two types of frozen hydrometeors: plate-shaped or needle particles (10,000 km2) make a significant contribution to the total precipitation budget, whereas small systems (70% crown density], moderately dense forest [40%e70%], and open/sparse forest [10%e40%]). In each transect, the name of the tree species, tree height, and diameter at breast height (DBH) for all trees were recorded. The crown density was estimated using a densitometer. Photographs of the canopy density of the forest are shown in Fig. 8.8. The densitometer is a small handy instrument that has a concave mirror with 24 grids. The instrument is

Figure 8.8 Photos showing different percentages of crown cover.

Application of synthetic aperture radar remote sensing in forestry

held in the right hand and the number of grids covered by the canopy is counted. Each grid is counted as 4 if it is fully covered by the canopy reflection. If it is half covered, it is counted as 2, and if it is partly covered, it is counted as 1. If all 24 grids were covered by the canopy, the crown density would be 100% (i.e., 24  4 ¼ 96) with a correction factor of 1.042. The number of grids covered by the tree crown was counted and the crown density was estimated. Tree growth parameters such as DBH were recorded using a diameter tape and Ravi altimeter, respectively. The wood volume was estimated using a standard formula given subsequently. Based on the wood volume, the amount of carbon sequestered in each of the forest density classes was estimated by a standard formula (47% of the biomass was considered to be carbon content). Based on the area assessed using ALOS 2 PALSAR data, the total biomass of Joida taluka was estimated according to the crown density classes. Fig. 8.9 shows field photos demonstrating the collected data: V ¼ BHF where V ¼ tree wood volume (m3) B ¼ basal area (m2) F ¼ form factor ¼ Volume of tree w.r.t diameter at base/Volume of tree w.r.t dia at breast height. H ¼ Tree height (m) basal area (m2) ¼ 3.143  d2/4 d ¼ diameter at breast height (m) Field data were collected from a square plot of 20  20 m. The details of latitude and longitude were recorded using a GPS. The crown density was recorded using a densitometer. The slope of the plot was determined with a tape using Pythagorus’ theorem, and the aspect of the forest was recorded with a prismatic compass. Details of ground truth data for 17 plots are given in Table 8.2. The ground truth data were used for LULC classification. The first step in classification is the generation of vector data containing different classes. It is usually done by selecting the area by squares or rectangles in the image. Different classes were identified based on their reflectance and ground truth data. The specific training samples were unique for each class. The supervised classification was done for three decompositions; maps indicating the different LULC classes under different decompositions are given in Figs. 8.10e8.12. The area of LULC classes from different decompositions in SNAP software was obtained as a percentage, based on the percent area. The LULC class area in terms of hectares was calculated for all decompositions; results are given in Table 8.3. The results indicate that among the different LULC classes according to the supervised classification, Pauli decomposition shows dense forest, moderately dense forest, sparse forest, and plantation as having similar percentages of area cover (21%e23%). Classification done from Yamaguchi decomposition indicated that the dense forest area (46%)

163

164

Radar Remote Sensing

Figure 8.9 Forests and field data recording.

Table 8.2 Ground truth data for the study area. Latitude

Longitude

Crown density

Slope %

Aspect

Type of forest

Vol/plot m3

1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16. 17.

15 090 79.900 15 80 48.400 15 080 47.100 15 110 38.300 15 090 57.600 15 110 72.400 15 100 13.700 15 090 46.600 15 080 59.900 15 110 03.200 15 110 89.400 15 100 34.400 15 110 89.300 15 100 90.900 15 90 64.200 15 00 11.2000 15 50 6.7200

74 280 17.000 74 320 16.200 74 310 35.200 74 320 23.200 74 290 46.700 74 310 84.200 74 230 14.700 74 180 ,25.100 74 310 88.600 74 330 64.100 74 310 30.900 74 340 09.900 74 310 32.000 74 330 83.200 74 370 26.000 74 320 56.9900 74 350 2.5200

83% 80% 92% 82% 83% 90% 92% 96% 90% 80% 87% 90% 69% 50% 50% 32% 35%

8.44 6.6 9.23 6.67 8.33 7.55 17.8 4.83 16.36 3% 9.33 10.91 5.6 4.38 3.80% 4.70% 4.10%

N N N NE NE NE NW NW W E E SW NW SW W NE W

Teak forest Mixed forest Teak forest Natural forest Teak forest Mixed forest Teak forest Mixed forest Natural forest Natural forest Mixed forest Natural forest Teak forest Natural forest Mixed forest Natural forest Natural forest

6.888 12.836 7.327 4.510 9.614 41.416 6.637 6.712 12.071 8.388 8.499 8.232 8.397 12.710 17.064 6.250 9.630

Application of synthetic aperture radar remote sensing in forestry

Sl. No.

165

166

Radar Remote Sensing

Figure 8.10 Land use land cover classification of Pauli decomposition. M Dense, medium dense.

Application of synthetic aperture radar remote sensing in forestry

Figure 8.11 Land use land cover classification of Freeman-Durden decomposition. M Dense, medium dense.

167

168

Radar Remote Sensing

Figure 8.12 Land use land cover classification of Yamaguchi decomposition. M Dense, medium dense.

Application of synthetic aperture radar remote sensing in forestry

Table 8.3 Area in different land use land cover (LULC) classes under different decompositions. Classification from Pauli decomposition

Decomposition Sl. No

1. 2. 3. 4. 5. 6. 7.

LULC classes

Dense forest Moderately dense forest Sparse forest Plantation Open\Agriculture Water body Settlements Total area

Classification from Yamaguchi decomposition

Classification from Freeman-Durden decomposition

Area (%)

Total Area (ha)

Area (%)

Total area (ha)

Area (%)

Total area (ha)

21.00 23.34

60,284 67,002

46.59 14.96

133,725 42,952

37.85 4.03

108,647 11,589

22.17 22.86 6.36 2.30 1.93 100

63,639 65,628 18,279 6624 5545 287,000

23.68 3.45 5.96 2.85 2.48 100

67,964 9910 17,120 8185 7143 287,000

17.15 20.14 5.59 4.31 10.88 100

49,238 57,822 16,061 12,393 31,251 287,000

cover is followed by sparse forest (23.68%), moderately dense forest (14.96%), and plantation (3.45%). Similarly, classification from Freeman-Durden decomposition shows the dense forest area cover is most (37.85%), followed by plantation (20.14%), sparse forest (17.15%), and moderately dense forest (4.03%). Among these three decompositions, Yamaguchi decomposition is best for classifying the forest compared with Pauli and Freeman-Durden decompositions. The Yamaguchi decomposition model has four scattering components; they include helix scattering, which in turn helps classify forest much better than other decomposition models, which have only three scattering mechanisms. The LULC classes obtained in all three decompositions were used to calculate the producer accuracy, user accuracy, and kappa coefficient. The confusion matrices for assessing the overall accuracy for LULC classes of all three decompositions are given in Tables 8.4e8.6. Table 8.4 Confusion matrix of land use land cover classes (Freeman-Durden decomposition). Classes

1

2

3

4

5

6

7

Total

User accuracy

Producer accuracy

Dense forest Moderately dense forest Sparse/scrub forest Plantation Agriculture Settlements Water body Total

32 6

8 48

0 2

0 7

0 0

0 0

0 0

40 63

80.00 76.19

59.25 77.41

16

2

28

0

0

0

0

46

60.86

56.00

0 0 0 0 54

0 4 0 0 62

10 10 0 0 50

24 0 0 0 31

2 0 0 36 40 0 0 54 4 18 0 22 6 0 25 31 52 18 25 292 kappa coefficient

66.66 74.07 81.81 80.64

77.41 76.92 100 100 0.68

169

170

Radar Remote Sensing

Table 8.5 Confusion matrix of land use land cover classes (Yamaguchi decomposition). Classes

1

2

3

4

5

6

7

Total

User accuracy

Producers accuracy

Water body (1) Settlement (2) Agriculture (3) Plantation (4) Dense forest (5) Sparse/open forest (6) Moderately dense forest (7) Total

39 0 2 0 0 0

0 26 4 0 0 0

6 7 28 0 0 7

0 0 0 28 3 0

0 0 0 7 32 0

0 0 6 0 0 38

0 0 0 9 4 4

45 33 40 44 39 49

86.00 78.00 70.00 63.00 82.00 77.00

95.00 86.00 58.00 90.00 68.00 86.00

0

0

0

0

8

0

33

41

80.00

66.00

41

30

48

31

47 44 50 291 kappa coefficient

0.70

Table 8.6 Confusion matrix of land use land cover classes (Pauli decomposition). Classes

1

2

3

4

5

6

7

Total

User accuracy

Producer accuracy

Dense forest Moderately dense forest Sparse/scrub forest Plantation Agriculture Settlements Water body Total

19 2

5 15

6 6

0 1

0 0

0 0

0 0

30 24

59.25 57.14

61.53 70.58

8

0

13

3

2

0

0

26

43.47

27.77

0 0 0 0 29

0 0 0 0 20

8 2 0 0 35

19 0 0 0 23

0 25 2 7 36

0 0 27 55.55 0 8 35 67.74 18 0 20 90 0 14 21 66.66 18 22 183 kappa coefficient

41.66 65.62 100 63.63 0.67

The formula for user accuracy, producer accuracy, and kappa coefficient is: Overall accuracy ¼ Users accuracy ¼ Producer accuracy ¼ K ¼

Total number of training sample correctly classified Total number of training samples used

Number of correct classified samples row wise Total number of samples in each categoryðRow totalÞ Number of correct classified samples column wise Total number of samples classified in categoryðRow totalÞ NA  B NN  B

Application of synthetic aperture radar remote sensing in forestry

where K ¼ kappa coefficient N ¼ total number of pixels A ¼ sum of correctly classified pixels B ¼ sum of products of column total and row total The overall accuracy of Freeman-Durden decomposition is 73.63 and the kappa coefficient is 0.68. The overall accuracy of Yamaguchi decomposition is 76.97 and the kappa coefficient is 0.7, and the overall accuracy of Pauli decomposition is 67.21 and the kappa coefficient is 0.61. The accuracy of Yamaguchi decomposition is higher it is because it uses scattering models with four scattering mechanisms (Kumar et al., 2017; Bhardwaj et al., 2017). Among the LULC classes, only forest classes such as dense forest, moderately dense forest, and sparse forest were used to assess the biomass and carbon sequestration. The transects ground truth data were used to estimate the total biomass in each forest class. The biomass and carbon sequestered in different forest classes according to the three decomposed data are given in Tables 8.7e8.9. AGB ¼ AVG=1000000 AGB ¼ aboveground biomass (Million tonnes) A ¼ total area in hectares V ¼ average volume (m3/ha) G ¼ wood specific gravity Carbon sequestrationðm.tÞ ¼ AGB*0.47 The average biomass and carbon sequestered per hectare according to Yamaguchi decomposition are 263 m3/ha and 123.61 t/ha, respectively. The average biomass and carbon sequestered per hectare according to Freeman-Durden decomposition

Table 8.7 Aboveground biomass (AGB) (m3) according to Yamaguchi decomposition by supervised classification classes. Sl. no.

1. 2.

Land use land cover classes

Total area (ha)

Average (vol/m3/ha)

Dense forest 133,725 277 Moderately 42,952 318 dense forest 3. Sparse forest 67,694 199 Total 244,371 Average Carbon sequestration (37.352 3 0.47)

Total AGB m m3

Average specific gravity

Total AGB (Mt)

37.042 13.659

0.62 0.57

22.966 7.785

13.471 0.49 64.172 263 m3/ha

6.601 37.352 123.61 t/ha 17.555 Mt

171

172

Radar Remote Sensing

Table 8.8 Aboveground biomass (AGB) (m3) according to Freeman-Durden decomposition by supervised classification classes. Sl. no.

Land use land cover classes

Total area (ha)

Average (vol/m3/ha)

1. 2.

Dense forest 108,647 277 Moderately 11,589 318 dense forest 3. Sparse forest 49,238 199 Total 169,474 Average Carbon sequestration(25.55 3 0.47)

Total AGB m m3

Average specific gravity

Total AGB (Mt)

30.09 3.68

0.62 0.57

18.66 2.10

9.79 0.49 43.56 256.23 m3/ha

4.80 25.55 151.18 t/ha 12.00 Mt

Table 8.9 Aboveground biomass (AGB) (m3) according to Pauli decomposition by supervised classification classes.

Sl. no.

1. 2. 3.

Land use land cover classes

Dense forest Moderately dense forest Sparse forest Total

Average (vol/m3/ha)

Total AGB m m3

Average specific gravity

Total AGB (Mt)

60,284 67,002

277 318

16.69 21.30

0.62 0.57

10.35 12.14

63,639 190,925

199

12.66 0.49 50.65 266.28 m3/ha

Total area (ha)

Average Carbon sequestration(28.69*0.47)

6.20 28.69 151 t/ha 13.48 Mt

are 256.23 m3/ha and 151.18 t/ha, respectively. Similarly, the average biomass and carbon sequestered per hectare according to Pauli decomposition are 266.28 m3/ha and 151 t/ha, respectively. The total carbon sequestered in the study area is 17.55, 12, and 13.48 Mt, according to Yamaguchi, Freeman-Durden, and Pauli decomposition, respectively. Among these three decomposition, Yamaguchi decomposition shows the highest biomass and carbon sequestration. This was because of the scattering mechanisms, including helix scattering, which enables better forest classification.

8. Summary and final remarks 8.1 Summary Remote sensing is most suited for retrieving forest aboveground biomass. Among the different methods of remote sensing, SAR remote sensing has the best advantage because it is most suited to retrieve biomass parameters of the forest throughout the year. Because

Application of synthetic aperture radar remote sensing in forestry

it works day and night and in all weather conditions, it most advantageous compared with optical and thermal remote sensing. SAR remote sensing has several applications. One such application explained in this chapter is LULC classes and estimation of AGB. SAR is also most suited for other applications such as change detection, flood mapping, and landslide mapping. Among the different decompositions, Yamaguchi decomposition is good for classifying LULC because four scattering mechanisms operate in this decomposition, including helix scattering, which is not found in other decompositions. In other decompositions, such as Pauli and Freeman-Durden, there are only three scattering mechanisms such as surface, double-bounce, and volumetric. Using L-band data, AGB and carbon sequestration were estimated, but because of the nonavailability of repeat data for the study area, interferometry was not done. This is important for determining the forest tree height profile, which is lacking in this chapter. L-band data are most suitable for forest mapping and estimating AGB and carbon sequestration apart from LULC classification.

References AbdurahmanBayanudin, A., Jatmiko, R.H., 2016. Orthorectification of sentinel-1 SAR (synthetic aperture radar) data in some parts of South-eastern Sulawesi using sentinel-1 toolbox. In: IOP Conference Series: Earth and Environmental Science 47, pp. 1e13. Alparone, L., Facheris, L., Baronti, S., Garzelli, A., Nencini, F., 2004. Fusion of multispectral and SAR images by intensity modulation. In: Proceedings of the 7th International Conference on Information Fusion, pp. 637e643. Arii, M., van Zyl, J.J., Kim, Y., 2010. A general characterization for polarimetric scattering from vegetation canopies. IEEE Trans. Geosci. Rem. Sens. 48, 3349e3357. Arii, M., van Zyl, J.J., Kim, Y., 2011. Adaptive model-based decomposition of polarimetric SAR covariance matrices. IEEE Trans. Geosci. Rem. Sens. 49, 1104e1113. Cantalloube, H., Nahum, C., 2000. How to Compute a Multi-Look SAR Image. European Space AgencyPublications-EsaSp 450, pp. 635e640. Cloude, S.R., Pottier, E., 1996. A review of target decomposition theorems in radar polarimetry. IEEE Trans. Geosci. Rem. Sens. 34, 498e518. Esha, S., Jayaprasad, P., James, M.E., 2019. Image fusion of SAR and optical images for identifying Antarctic ice features. J. Indian Soc. Remote Sens. 47 (12), 2113e2127. Freeman, A., Durden, S.L., 1998. A three-component scattering model for polarimetric SAR data. IEEE Trans. Geosci. Rem. Sens. 36 (3), 963e973. https://doi.org/10.1109/36.673687. In this issue. Kulkarni Samadhan, C., Rege, P.P., 2020. Pixel level fusion techniques for SAR and optical images: a review. Inf. Fusion 59, 13e29. https://doi.org/10.1016/j.inffus.2020.01.003. Kumar, V., Bhardwaj, A., Haldar, A.L., 2017. Land Use Land Cover (Lulc) Classification using Fusion of Multi-Sensor Optical and Microwave Data from Sentinel Mission. Lee, J.S., Ainsworth, T.L., 2011. The effect of orientation angle compensation on coherency matrix and polarimetric target decompositions. IEEE Trans. Geosci. Rem. Sens. 49, 53e64. Lee, J.S., Pottier, E., 2009. Polarimetric Radar Imaging: From Basics to Applications, vol. 142. CRC, Boca Raton, FL, USA. Lee, J.-S., Pottier, E., 2017. Polarimetric Radar Imaging: From Basics to Applications. CRC press. Marco, L., Wright, T., 2009. Absolute Radiometric and Polarimetric Calibration Of Alos Palsar Products, 1, p. 3. Meneghini, A., 2019. An Evaluation of Sentinel-1 and Sentinel-2 for Land Cover Classification. Clark University.

173

174

Radar Remote Sensing

Neumann, M., Ferro-Famil, L., Pottier, E., 2009. A general model-based polarimetric decomposition scheme for vegetated areas. In: Proceedings of POLinSAR 2009 Workshop, Frascati, Italy, 26e30 January. Riedel, T., Thiel, C., Schmullius, C., 2007. Fusion of optical and SAR satellite data for improved land cover mapping in agricultural areas. In: Proc. Envisat Symposium. Wei, H., Naoto, Y., 2018. Multi-temporal sentinel-1 and-2 data fusion for optical image simulation. ISPRS Int. J. Geo-Inf. 7 (10), 389 doi:10.3390. Yamaguchi, Y., Moriyama, T., Ishido, M., Yamada, H., 2005. Four-component scattering model for polarimetric SAR image decomposition. IEEE Trans. Geosci. Rem. Sens. 43, 1699e1706. Yamaguchi, Y., Yajima, Y., Yamada, H., 2006. A four-component decomposition of polsar images based on the coherency matrix. Geosci. Rem. Sens. Lett. IEEE 3, 292e296. Yamaguchi, Y., Sato, A., Boerner, W.M., Sato, R., Yamada, H., 2011. Four-component scattering power decomposition with rotation of coherency matrix. IEEE Trans. Geosci. Rem. Sens. 49, 2251e2258.

Further reading Cloude, S.P., 2010. Applications in Remote Sensing. Oxford University Press, New York, NY, USA.

CHAPTER 9

Classification of Radar data using Bayesian optimized two-dimensional Convolutional Neural Network Achala Shakya1, Mantosh Biswas1 and Mahesh Pal2 1

Computer Engineering Department, National Institute of Technology, Kurukshetra, Haryana, India; 2Civil Engineering Department, National Institute of Technology, Kurukshetra, Haryana, India

1. Introduction Classification means setting class labels to image pixels based on similar properties. Image classification in remote sensing refers to grouping pixels that are similar in characteristics into different land cover types (Lillesand et al., 2015). Several image classification methods have been widely used to extract information from remote sensing images to distinguish and identify land cover types based on their pixel values. Land cover maps are crucial for several purposes such as urbanization (Liu and Weng, 2013), biodiversity (Falcucci et al., 2007), agriculture (Bargiel and Herrmann, 2011), hazard assessment (van der Sande et al., 2003), and the natural environment (Ullmann et al., 2014). Because of developments in the field of remote sensing, satellite images are widely used to monitor land cover types and their spatial distribution (Congalton et al., 2014). Radar images (i.e., Synthetic Aperture Radar [SAR] images) are used efficiently when weather conditions are unsuitable for acquiring optical data. Contrary to optical data, SAR data contain cloud-free images (Balzter et al., 2015). Several studies reported their use in various applications, and it has been proven that SAR data can be effective for land use/land cover monitoring in areas where obtaining cloud-free images is not easy (Niu and Ban, 2013; Balzter et al., 2015). A study by Niu and Ban (2013) analyzed multitemporal and fullpolarimetric RADARSAT-2 for land cover mapping. The intensity information was used with polarimetric parameters to improve the accuracy of land cover classification. Baltzer et al. (2015) presented Sentinel-1A satellite and Shuttle Radar Topography Mission (SRTM) Digital Elevation Model (DEM) data for European Coordination of information on the environment (CORINE) land cover mapping. Classification accuracy of up to 70% was achieved by dual polarimetric data, DEM textures, and DEM-related products. A study by Abdikan et al. (2016) investigated land cover mapping using Support Vector Machine (SVM) classification for Sentinel-1A satellite data. Machine learning techniques such as SVM are extensively used in remote sensing to achieve high classification accuracy for small training datasets (Abdikan et al., 2016).

Radar Remote Sensing ISBN 978-0-12-823457-0, https://doi.org/10.1016/B978-0-12-823457-0.00008-2

© 2022 Elsevier Inc. All rights reserved.

175

176

Radar Remote Sensing

Deep learning (DL) algorithms have evolved as effective algorithms for land cover classification because of their ability to extract higher-level spatial information through feature extraction (Song et al., 2019; Shakya et al., 2020, 2021). Work using DL techniques such as Convolutional Neural Networks (CNNs), deep belief networks, deep autoencoders, and recurrent neural networks has been extensively reported for land use/land cover classification using remote sensing datasets (Hamida et al., 2018; Paoletti et al., 2019; Cao et al., 2020). Although various DL algorithms were proposed and reported in the literature, the CNN is the most desirable algorithm owing to its high computational efficacy and ability to recognize image patterns (Krizhevsky et al., 2012). Two-dimensional (2D) and three-dimensional (3D)CNN are extensively used for satellite images because of their inherent ability to extract spatial and spectral features; thus, they provide high classification accuracy (Cheng et al., 2018; Paoletti et al., 2019; Cao et al., 2020). CNN classifiers are comparatively more robust and effective in extracting spatial features of remote sensing data because of their local connections and weight sharing structures. Use of different DL algorithms requires various hyperparameters, known as adjustable parameters, to be set that must be tuned to obtain a model with optimal performance for the considered datasets. To select and optimize these hyperparameters of DL algorithms, various optimization techniques such as grid search, random search, and Bayesian optimization (Sameen et al., 2019) are used. Optimization defines the loss function/cost function and minimizes it using one or another optimization routine. Bayesian optimization methods always find the best-optimized hyperparameter setting compared with the grid and random searches (Sameen et al., 2019). Keeping this in view of improved performance by 2D-CNN for land cover classification studies, this chapter investigated the effectiveness of Bayesian optimized 2D- CNN to classify a SAR (S1) dataset over an agricultural area (Central State Farm, Hissar) in India.

2. Background 2.1 Sentinel-1A data After the launch of the Sentinel-1A satellite mission in April 2014, global coverage of SAR data became easily available at higher spatial and temporal resolution, and was widely used for land cover classification. The S1 satellite sensor operates in the Cband at a frequency of 5.405 GHz and acquires information about the earth’s surface in selectable single (HorizontaleHorizontal [HH] or VerticaleVertical [VV]) and dual polarization (HH þ VerticaleHorizontal [VH]). These data are freely available and are widely used in various applications including land cover classification. In Sentinel-1A, C-band SAR records backscatter signals day and night, independent of the illumination and weather conditions. It acquires microwave images in four exclusive modes: Stripmap, Interferometric Wide Swath (IW), Extrawide Swath (EW), and Wave (WV). It can

Classification of Radar data using Bayesian optimized two-dimensional Convolutional Neural Network

capture images (in terms of polarization) using the same set of transmitted pulses by using its antenna. Depending on the acquisition mode, S1 can acquires images in dual polarization modes (VV and VH) or in single polarization (HH or VV) at a 10 m  10 m cell size with a swath of 250 km. It has a temporal resolution of 12 days (combined with Sentinel-1B, the temporal resolution is 6 days). Microwave signals at C-band can penetrate up to 5 cm deep below the soil surface. Sentinel-1A level-1 data are categorized into two product types: Ground Range Detected (GRD) and single look complex. The wide range of applications requires Sentinel-1A GRD products with standard corrections. After applying standard corrections, Sentinel-1A (S1) GRD images will have square pixels (consisting of the amplitude) with reduced speckle. Radar (SAR [S1]) data have various useful applications because SAR data can provide complementary information to infrared and visible remote sensing data. Because of its strong penetrating ability in bad weather conditions, SAR images can provide information about biomass, forest canopy, trunks, and different forest types. SAR images allow the identification of several land cover types, including water bodies, urban areas, and agricultural fields. In geomorphology and geology, SAR data provide information about roughness and surface texture, which have an important role in geological mapping and detection. In crop identification, SAR images acquired using different polarizations may be effective. Various other successful applications of SAR include soil moisture estimation, flood mapping, and hydrological modeling. 2.2 Classification Various methods based on statistical and machine learning theories have been proposed for land use/land cover classification. Statistical approaches assume that input data are normally distributed and include methods such as the distance measure (Du and Chang, 2001), clustering (Huang, 2002), maximum-likelihood classifier (Shakya and Kumar, 2019; Tripathi et al., 2020), and logistic regression (Etter et al., 2006). Machine learning algorithms such as decision trees, neural networks (Lu et al., 2016), random forest (Gislason et al., 2006), and SVM (Pal and Mather, 2005) have been well-used to classify land use and land. To classify remote sensing data, two methods such as pixel-based and object-based classification have widely been employed. In pixel-based classification, individual image pixels are analyzed using spectral information as they consider only the spectral values of class information which makes their implementation easier and faster. Despite the popularity and improved performance of pixel-based classifiers (Mountrakis et al., 2011), these algorithms provide salt-and-pepper noise (speckles) in the classified image. Furthermore, traditional pixel-based classifiers do not consider the spatial contexts of an image, whereas in object-based image classification, pixels are grouped into representative shapes and sizes. Object-based classifiers use geometric, contextual, textual, spectral, brightness, and backscattered values from the image to classify land cover features

177

178

Radar Remote Sensing

into their respective classes (Petropoulos et al., 2013). Spatial-spectral classifiers using spectral and spatial information simultaneously have also been proposed, and their classification accuracy is better than traditional pixel-based classifiers (Shakya et al., 2020). Developments in CNN-based DL algorithms and their applications for remote sensing image classification suggest significant improvements in classification accuracy compared with spectral-spatial classifiers. The main difference between a traditional neural network and CNN is that CNN benefits from the properties of natural signals, such as the use of multiple layers, local connections, shared weights, and pooling (LeCun et al., 2015). The CNN model is created in a sequence of layers, the very first of which is convolutional and pooling layers. Units in convolutional layers are structured to the feature maps and connected through filter sequences to local patches inside the feature map of the preceding layer. The obtained output is then passed through a non-linear activation function i.e., Rectified Linear Unit (ReLU). In this manner, CNN recognizes invariant and correlated information to the location within local groups. On the other hand, pooling layers calculate peak values of local unit patches within feature maps to combine similar features semantically. Different stages of convolution, nonlinearity, and pooling are stacked together, followed by fully connected layers using a Softmax function that involves classification. Like other neural networks, CNN is trained using backpropagation and stochastic gradient descent algorithms.

2.3 Bayesian optimization Bayesian optimization optimizes decision making concerning which parameter needs to be set next for iteration (Sameen et al., 2019). Within this context, an objective function is applied to understand how prior settings were carried out. Bayesian optimization methods always find the best-optimized hyperparameter setting faster compared with the grid and random searches. Training DL algorithms such as CNN can be timeconsuming according to amount of data as well as the computational density. In these cases, Bayesian optimization can be an effective approach to find optimal hyperparameters of CNN. The procedure for optimization is based on Bayes’ theorem using the equation with Z as the model and Y as the observation (Kramer et al., 2011): PðY Þ ¼ ðPðZÞPðZÞÞ=PðZÞ

(9.1)

where PðY Þ defines the posterior probability of Z given Y , PðZÞ is known as the likelihood of Y given Z, PðZÞ is the prior probability of Z, and PðZÞ is known as the marginal probability of Z. Bayesian optimization finds the minimum value of a function f(x) on a bounded set Y .

Classification of Radar data using Bayesian optimized two-dimensional Convolutional Neural Network

3. Dataset and ground data collection A Sentinel-1A image was acquired March 23, 2019 for Central State Farm, Hissar, Haryana. This area consists of agricultural farmland with large experimental crop fields where various seasonal crops (winter and summer) are grown mainly for seed generation to be distributed later to state farmers (Fig. 9.1). In the study area, 12 land cover types (fallow land, built-up area, dense vegetation, fenugreek, fodder, gram, mustard, oats, peas, sparse vegetation, spinach, and wheat) were identified after a field visit to the study area on April 6, 2019. VV and VH polarized images from S1 at 10-m resolution were used for the study (Fig. 9.2).

4. Dataset preparation for classification In the literature, the combination (VH, VV, [VV  VH], VH/VV, [VV þ VH]/2) achieved high classification accuracy owing to the combined intensity values during layer stacking compared with various other combinations (i.e., [VH], [VV], [(VV þ VH)/2],

Figure 9.1 False color composite image map (near-infrared, red, and green bands) of study area using Sentinel-2 data acquired March 24, 2019.

179

180

Radar Remote Sensing

Figure 9.2 Original (A) Sentinel-1 (verticalevertical) and Sentinel-1 (verticalehorizontal) polarized images for Mar. acquired March 23, 2019.

[VH/VV], [VH, VV, VH/VV], [VH, VV, (VV þ VH)/2], and [VH, VV, (VV e VH)]). Thus, the best combination (VH, VV, [VV e VH], VH/VV, [VV þ VH]/2), as suggested in the literature along with their textures (a total of 25 bands), were used for land cover classification using 2D-CNN in this study. Texture is a key element of human visual perception and is used in many computer vision systems. Texture refers to the spatial variation of image gray-scale levels (tone) as a function of scale. Combined with spectral data, texture provides another level of information to interpret features in pixel-based classification. In this study, different texture measures were computed using a window or kernel of the same size (i.e., 3  3). The size of the kernel partially determines the success of texture-based image classification. If the window size is too small, not enough spatial information can be extracted to distinguish among different land features. If the window size is too large, it could overlap different features and introduce spatial errors. The three-texture metrics adopted in this study are: 1. Occurrence metrics Occurrence metrics (measures) apply a texture filter based on the first-order occurrence measure. This includes the texture measures of data range, mean, variance, entropy, and skewness. Of these occurrence measures, only the data range was used for the study, making a total of two bands for both VH and VV polarization. 2. Co-occurrence metrics Co-occurrence measures apply second-order texture filters based on the cooccurrence matrix. These filters include mean, variance, homogeneity, contrast, dissimilarity, entropy, second moment, and correlation. A setting for gray-scale quantization levels allows one to reduce the number of shades of gray required to represent

Classification of Radar data using Bayesian optimized two-dimensional Convolutional Neural Network

the image. This option also reduces processing time. Of various co-occurrence measures, correlation, dissimilarity, homogeneity, mean, and variance were used for the study, making a total of 10 bands for both VH and VV polarization. 3. Morphology metrics Morphologic filters that use structuring elements in an image include dilate, erode, opening, and closing. Thus, a total of eight bands were used for both VH and VV polarization. Therefore, the total texture feature of 20 bands was used in this study.

5. Methodology The methodology proposed in this chapter is provided in Fig. 9.3. It is composed of preprocessing Sentinel-1 (S1) data, data combination, texture feature extraction, and other combinations ([VH, VV, (VV  VH), VH/VV, (VV þ VH)/2]), layer stacking, classification using Bayesian optimized 2D-CNN followed by an accuracy assessment in terms of the classified image, overall accuracy, and kappa value. 5.1 Data preprocessing Several operations available in the SNAP tool of the Sentinel-1 toolbox were used to preprocess S1 data: 1. The first operation is to calibrate sdata to normalize the radar cross-section and desired phase of surface features. 2. A deburst operation was performed to merge three subswaths (i.e., IW1, IW2, and IW3) for further operations. 3. Next, a multilooking operation improved phase fidelity to feed the output to the next operation.

Figure 9.3 Proposed methodology for the classification.

181

182

Radar Remote Sensing

4. Speckle filtering using a refined Lee 3  3 filter was done to remove speckle noise, because this filter performed better. 5. Furthermore, geometric correction (terrain correction) was performed, because interferometric phase data need to be projected to a geographic coordinate system using a DEM-assisted geocoding step. 6. Range Doppler terrain correction was done using the Universal Transverse Mercator (UTM) World Geodetic System 1984 (WGS84) projection and 10-m resampling and reprojection was made to fit the fusion requirements. The whole preprocessing was executed. 5.2 Classification using Bayesian-optimized two-dimensional convolutional neural network To classify datasets used in this study, spectral and spatial properties of pixels were used. For patch-based CNN, spatial properties of a pixel are extracted using neighboring pixels from a window with a fixed size. To train and test the 2D-CNN classifier, the image patch size (say, p  p) was extracted using ground reference images, and patches that contained central pixels with a nonzero value were considered during classification using ground truth. The patches were processed by 2D-CNN to extract textures and were forwarded to fully connected layers after pooling and flattening. For all randomly selected samples, 75% of data were used for training and rest (25%) were for testing. The performance of all classifiers was measured in terms of the overall classification accuracy and kappa value.

6. Results and discussion The experimental work has been performed using Python with the Keras framework on a computer with 16 GB of GPU with the Ubuntu 14.04 Operating System. Table 9.1 provides the results of 2D-CNN in terms of the optimal values of different hyperparameters and accuracy metrics for the considered combinations of S1 datasets. Results in Table 9.1 indicate that the combination of VV and VH polarized data with their textures using of patch size of 5 improved classification performance by w5% compared with the other scenarios considered in this chapter. Keeping in view the improved performance by the Bayesian-optimized 2D-CNN classifier for the considered S1 dataset, classified images are provided in Fig. 9.4. Results from Fig. 9.4 indicate that the boundary regions in Fig. 9.4A overlap with few missing classes and few classes are not classified accurately, whereas in Fig. 9.4B, all classes are accurately classified.

Table 9.1 Optimal values of different hyperparameters and accuracy metrics for considered datasets. Values of hyperparameters for considered dataset Dataset

(VH, VV, [VV e VH], VH/VV, [VV þ VH]/2) þ (textures of VH and VV polarization)

Convolution and pooling layer

Learning rate and stride

Batch size

Epochs

Fully connected layer

Two convolution layers • First convolution layer with 600 filters of size 3  3 is followed by max pooling with 3  3 filter • Second convolution layer with 600 filters of size 5  5 is followed by average pooling with 3  3 filter Two convolution layers • First convolution layer with 600 filters of size 5  5 is followed by average pooling with 5  5 filter • Second convolution layer with 600 filters of size 3  3 is followed by average pooling with 3  3 filter

0.1 and 1

32

1000

0.1 and 2

32

1000

VH, verticalehorizontal; VV, verticalevertical.

Overall accuracy (%)

Kappa value

Two fully connected layers with (50 and 200) filters

94.11

0.93

Three fully connected layers with (50, 200, and 200) filters

99.40

0.99

Classification of Radar data using Bayesian optimized two-dimensional Convolutional Neural Network

(VH, VV, [VV e VH], VH/VV, [VV þ VH]/2)

Accuracy metric

183

184

Radar Remote Sensing

Figure 9.4 Classified images of S1 dataset for (A) (VH, VV, [VV e VH], VH/VV, [VV þ VH]/2); and (B) (VH, VV, [VV e VH], VH/VV, [VV þ VH]/2) þ texture of VH and VV polarizations. VH, verticalehorizontal; VV, verticalevertical.

7. Conclusion In this chapter, the classification performance of S1 imagery and its potential for classification was investigated using a Bayesian-optimized 2D-CNN classifier. To improve the classification accuracy of S1 imagery, the high-performing combination of S1 (VV and VH) from the literature was considered and then layer-stacked with the texture of both VH and VV polarized data. The experimental results indicated that the incorporation of texture was able to achieve high classification accuracy. The results also emphasized that selecting different variables with 2D-CNN is important for achieving the desired accuracy. Results also indicated that dual-polarized S1 data can be suitable for extracting land cover classes for a particular area.

Acknowledgment The authors are thankful to the Space Applications Center, Ahmedabad, for financial support under the project entitled “Fusion of Optical and Multi-frequency Multi-polarimetric SAR data for Enhanced Land cover Mapping,” to carry out this work. The authors would like to acknowledge the National Institute of Technology, Kurukshetra, India, for providing the required computing facilities to carry out this study.

Classification of Radar data using Bayesian optimized two-dimensional Convolutional Neural Network

References Abdikan, S., Sanli, F.B., Ustuner, M., Cal o, F., 2016. Land cover mapping using sentinel-1 SAR data. Int. Arch. Photogram. Rem. Sens. Spatial Inf. Sci. 41, 757 (Prague, Czech Republic). Balzter, H., Cole, B., Thiel, C., Schmullius, C., 2015. Mapping CORINE land cover from Sentinel-1A SAR and SRTM digital elevation model data using random forests. Rem. Sens. 7 (11), 14876e14898. Bargiel, D., Herrmann, S., 2011. Multi-temporal land-cover classification of agricultural areas in two European regions with high resolution spotlight TerraSAR-X data. Rem. Sens. 3 (5), 859e877. Cao, X., Yao, J., Xu, Z., Meng, D., 2020. Hyperspectral image classification with convolutional neural network and active learning. IEEE Trans. Geosci. Rem. Sens. 58 (7), 4604e4616. Cheng, G., Yang, C., Yao, X., Guo, L., Han, J., 2018. When deep learning meets metric learning: remote sensing image scene classification via learning discriminative CNNs. IEEE Trans. Geosci. Rem. Sens. 56 (5), 2811e2821. Congalton, R.G., Gu, J., Yadav, K., Thenkabail, P., Ozdogan, M., 2014. Global land cover mapping: a review and uncertainty analysis. Rem. Sens. 6 (12), 12070e12093. Du, Q., Chang, C.I., 2001. A linear constrained distance-based discriminant analysis for hyperspectral image classification. Pattern Recogn. 34 (2), 361e373. Etter, A., McAlpine, C., Wilson, K., Phinn, S., Possingham, H., 2006. Regional patterns of agricultural land use and deforestation in Colombia. Agric. Ecosyst. Environ. 114 (2e4), 369e386. Falcucci, A., Maiorano, L., Boitani, L., 2007. Changes in land-use/land-cover patterns in Italy and their implications for biodiversity conservation. Landsc. Ecol. 22 (4), 617e631. Gislason, P.O., Benediktsson, J.A., Sveinsson, J.R., 2006. Random Forests for land cover classification. Pattern Recogn. Lett. 27 (4), 294e300. Hamida, B.A., Benoit, A., Lambert, P., 2018. 3-D deep learning approach for remote sensing image classification. IEEE Trans. Geosci. Rem. Sens. 5 (8), 4420e4434. Huang, K.Y., 2002. A synergistic automatic clustering technique (SYNERACT) for multispectral image analysis. Photogramm. Eng. Rem. Sens. 68 (1), 33e40. Kramer, O., Ciaurri, D.E., Koziel, S., 2011. Derivative-free optimization. In: Computational Optimization, Methods and Algorithms. Springer, Berlin, Heidelberg, pp. 61e83. Krizhevsky, A., Sutskever, I., Hinton, G.E., 2012. ImageNet classification with deep convolutional neural networks. Adv. Neural Inf. Process. Syst. 1097e1105. LeCun, Y., Bengio, Y., Hinton, G., 2015. Deep learning. Nature 521 (7553), 436e444. Lillesand, T.M., Kiefer, R.W., Chipman, J.W., 2015. Remote Sensing and Image Interpretation. John Wiley, New York. Liu, H., Weng, Q., 2013. Landscape metrics for analysing urbanization-induced land use and land cover changes. Geocarto Int. 28 (7), 582e593. Lu, D., Chen, Q., Wang, G., Liu, L., Li, G., Moran, E., 2016. A survey of remote sensing-based aboveground biomass estimation methods in forest ecosystems. Int. J. Digital Earth 9 (1), 63e105. Mountrakis, G., Im, J., Ogole, C., 2011. Support vector machines in remote sensing: a review. ISPRS J. Photogram. Rem. Sens. 66 (3), 247e259. Niu, X., Ban, Y., 2013. Multi-temporal RADARSAT-2 polarimetric SAR data for urban land-cover classification using an object-based support vector machine and a rule-based approach. Int. J. Rem. Sens. 34 (1), 1e26. Pal, M., Mather, P.M., 2005. Support vector machines for classification in remote sensing. Int. J. Rem. Sens. 26 (5), 1007e1011. Paoletti, M.E., Haut, J.M., Plaza, J., Plaza, A., 2019. Deep learning classifiers for hyperspectral imaging: a review. ISPRS J. Photogram. Rem. Sens. 158, 279e317. Petropoulos, G.P., Vadrevu, K.P., Kalaitzidis, C., 2013. Spectral angle mapper and object-based classification combined with hyperspectral remote sensing imagery for obtaining land use/cover mapping in a Mediterranean region. Geocarto Int. 28, 114e129. Sameen, M.I., Pradhan, B., Lee, S., 2019. Self-learning random forests model for mapping groundwater yield in data-scarce areas. Nat. Resour. Res. 28 (3), 757e775.

185

186

Radar Remote Sensing

Shakya, A., Kumar, A., 2019. Noise clustering-based hypertangent kernel classifier for satellite imaging analysis. J. Indian Soc. Rem. Sens. 47 (12), 2009e2025. Shakya, A., Biswas, M., Pal, M., 2020. CNN-based fusion and classification of SAR and Optical data. Int. J. Rem. Sens. 41 (22), 8839e8861. Shakya, A., Biswas, M., Pal, M., 2021. Parametric study of convolutional neural network based remote sensing image classification. Int. J. Rem. Sens. 42 (7), 2663e2635. Song, J., Gao, S., Zhu, Y., Ma, C., 2019. A survey of remote sensing image classification based on CNNs. Big Earth Data 3 (3), 232e254. Tripathi, G., Pandey, A.C., Parida, B.R., Kumar, A., 2020. Flood inundation mapping and impact assessment using multi-temporal optical and SAR satellite data: a case study of 2017 flood in Darbhanga district, Bihar, India. Water Resour. Manag. 34 (6), 1e22. Ullmann, T., Schmitt, A., Roth, A., Duffe, J., Dech, S., Hubberten, H.W., Baumhauer, R., 2014. Land cover characterization and classification of arctic tundra environments by means of polarized synthetic aperture X-and C-Band Radar (PolSAR) and Landsat 8 multispectral imagerydRichards Island, Canada. Rem. Sens. 6 (9), 8565e8593. Van der Sande, C.J., De Jong, S.M., De Roo, A.P.J., 2003. A segmentation and classification approach of IKONOS-2 imagery for land cover mapping to assist flood risk and flood damage assessment. Int. J. Appl. Earth Obs. Geoinf. 4 (3), 217e229.

CHAPTER 10

Modeling and simulation of synthetic aperture radar dataset for retrieval of soil surface parameters Sayyad Shafiyoddin and Ajit Kumar Microwave and Imaging Spectroscopy Research Laboratory, Milliya Arts, Science, and Management Science College, Beed, Maharashtra, India

1. Introduction Soil moisture is an important parameter for many applications, including life process, hydrology cycle, and agricultural ones (Lecomte et al., 2001; Aubert et al., 2011). Soil moisture estimation from remote sensing is a complex method. One can directly measure soil moisture at the field site using appropriate sensors. In active remote sensing, soil moisture retrieval from synthetic aperture radar (SAR) is challenging because of the dynamic nature of soil surface parameters (dielectric constant, surface roughness, and incidence angle). SAR sensitivity to soil moisture has an important role in water management, sustainable agriculture, drought monitoring, and flood forecasting. Because it is free, opensource software and open access to high-resolution Sentinel-1 SAR data offers the large potential to estimate different physical parameters (soil surface parameters). Microwave backscatter measurements from SAR images are a function of the physical and electrical properties of the target and configuration of the SAR sensor (frequency, incidence angle, and polarization). Soil surface roughness and soil moisture are the most important parameters affecting the backscattering signal. This problem has been reported because of complex calculations and theoretical solutions. Soil moisture controls various parameters owing to airewater interactions and the temporary retention of water (Lecomte et al., 2001; Aubert et al., 2011; Kite and Pietroniro, 1996; Jackson, 1993; Entekhabi, 1993). Information about soil moisture is required to predict unwanted problems such as floods, low agricultural productivity, droughts, and hydrologic issues, and for climate studies (Jackson et al., 1987; Martin and Aber, 1997; Sivapalan et al., 2003; Field et al., 1995; Seneviratne et al., 2010; Kerr and Ostrovsky, 2003). Because soil moisture is one of the most important parameters for defining various processes, it needs to be quantified regularly. For a particular application, soil moisture is needed in different spatial and temporal domains because of its large variations. One can measure soil moisture at its local area to understand variations in different climates using time domain reflectometry or gravimetric methods. However, it is difficult to collect soil moisture information by that technique for a large Radar Remote Sensing ISBN 978-0-12-823457-0, https://doi.org/10.1016/B978-0-12-823457-0.00004-5

© 2022 Elsevier Inc. All rights reserved.

187

188

Radar Remote Sensing

area or specific domain. Collecting information about soil moisture using that method is tedious and time-consuming, although it will give the correct value about soil moisture for a particular area. It is difficult to implement this method of soil moisture collection for global or regional areas, though. A well-known alternative method for collecting soil surface parameters is to use SAR data. SAR equipment is based on microwave remote sensing in which the SAR sensor transmits a microwave signal that interacts with different objects and collects the backscatter signals. The SAR signal is sensitive to surface parameters (surface roughness, soil moisture, and dielectric constant) (Mulla, 2013). For the permittivity of a soilewater mixture, a permittivity (ε) gradient exists between dry soil (ε ¼ 2) and water (ε ¼ 80), which affects the SAR signal accordingly (Walker, 1999; Narvekar et al., 2015; Ulaby, 1982; Altese et al., 1996). Because of the high sensitivity toward permittivity, different soil moisture models were proposed in the literature. To retrieve soil moisture from SAR backscatter, various theoretical, empirical, and semiempirical models were proposed (Verhoest et al., 2008; Barrett et al., 2009; Oh et al., 1992, 1994, 2002; Dubois et al., 1995; Oh, 2004; Attema and Ulaby, 1978). Most models are based on full polarization (quad-polarization data) However, other models are available. For instance, the water cloud and modified Dubois (MDB) and Dubois models can also be used successfully to estimate soil moisture over different lands (Fung et al., 1992; Fung and Cheng, 1994; Baghdadi et al., 2012; El Hajj et al., 2016; Dabrowska-Zielinska et al., 2018; Wagner et al., 2008). Many papers concluded that quad-polarization data are more useful for estimating soil surface parameters. For C-band, some researchers (Zribi et al., 2019; Yang et al., 2019; El Hajj et al., 2017; Bousbih et al., 2018) explained that Sentinel1 data can be used to calculate soil surface parameters with high accuracy. Many researchers have used a neural network (NN) model to estimate soil moisture by inverting radar signals. An artificial NN model uses training samples provided by simulated backscatter values employing different models (i.e., integral equation model, Oh model, MDB model). The different models do not require many parameters; they are almost independent of them (Qiu et al., 2019; Hachani et al., 2019). Models on C-band, and L-band data used to estimate soil surface parameters (Hachani et al., 2019) can be applied. Some authors (Ezzahar et al., 2020; Hosseini and McNairn, 2017) have used the MDB model and C-band data to estimate soil moisture and other surface parameters. The Sentinel-1A satellite operates in the C-band at 5.405 GHz and acquires information about the surface at selectable single (HH or VV) and dual polarization (HH þ HV or VV þ VH) (Dave et al., 2019; Alexakis et al., 2017). Sentinel-1 data are freely available for various applications including soil surface parameters and moisture (Wagner et al., 2009; Bauer-Marschallinger et al., 2018; Rao et al., 2013). In this chapter, we focus dual-polarized Sentinel-1A satellite images to estimate the soil surface in India. We used a different method and its potential to find the soil moisture. We also used a combination of the MDB and Oh models to calculate soil moisture and Topp’s model for relative soil permittivity from SAR backscatter values.

Modeling and simulation of synthetic aperture radar dataset for retrieval of soil surface parameters

2. Study area and collection of field data The study area in the Beed district is situated in the central west part of Aurangabad, Maharashtra between 18.28 and 19.28 degrees longitudinally, and between 74.54 and 76.57 degrees latitudinally (Fig. 10.1). The total area of the Beed district is 1061.53 km2. The Beed district has various types of soil with lots of variation in its surface parameters. The main crop of Beed during the kharif season are soybeans, pigeon peas, pearl millet, maize, black gram, sorghum, groundnut, sunflower, cotton, and sugarcane. Soil surface parameters (soil surface roughness and soil moisture) were measured at the same time as SAR data acquisition. Soil surface moisture (Mv in vol.%) was measured in 55 different bare soil land parcels, the area of which ranged from 2600 to 9600 m2. Measurements were taken with a soil moisture meter sensor (Lutron PMS-714 Moisture Meter) based on a microprocessor. The Lutron PMS-714 Moisture Meter is designed to measure the moisture content of soil; it has a measurement range of 0%e50% with 0.1% resolution. The microprocessor circuit ensures high accuracy, and the durable ABS plastic housing means it is ideal for use in the field. The built-in LCD screen makes the meter easy to read, and a data hold function allows you to freeze the current value on the display. The meter shows max/min functions and has a low battery indicator. These measurements were taken from the field for the top of the soil (0- to 5-cm depth), as given in the specifications and working notes of the sensor. The roughness (s) is also measured using in the house-fabricated pin profilometer needle with a length of 1 m and resolution of 2 cm (Fig. 10.2). Measurements of surface roughness in field measures

Figure 10.1 Geographic location of Beed, Maharashtra, India.

189

190

Radar Remote Sensing

Figure 10.2 In-house profile meter needle with a length of 1 m and a resolution of 2 cm.

are parallel and perpendicular to the field furrow. On average, five parallel and perpendicular measurement values are taken for each land parcel. Finally, the nonuniform roughness value of each area is computed using Eq. (10.1), and the images are processed in the ENVI software package: sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  PN  2 2 i¼1 zi  z s¼ (10.1) N 1 Where N is the number of profile points, zi is the surface elevation at the point i in cm, and z is the average surface elevation in cm (Table 10.1). Some images of the field data collection are shown in (Fig. 10.3); they were taken at the time of satellite accusation path.

3. Collection and processing of satellite data The method is explained by the flowchart shown in (Fig. 10.4). First, georeferencing was done on SAR images with a collected ground control point. After georeferencing, the backscattering coefficient was calculated according to different algorithms and formulas provided by the different SAR image provider agencies (i.e., Indian Space Research Organisation and European Space Agency). After that, speckle noise was removed by Lee-Sigma filtering using the Sentinel Application Platform (SNAP) and ENVI software tools. As explained in Fig. 10.4, field measurements were taken

Moisture (%)

Soil surface roughness (cm)

Date

Incidence angle q (degrees) over study area [nearefar]

No. of field samples

Minimum

Maximum

Mean

Minimum

Maximum

October 2, 2019 September 20, 2019 September 8, 2019 August 21, 2019 August 14, 2019

[36e37] [36e37] [36e37] [37e38] [37e39]

15 12 12 12 15

11.23 16.4 22.7 7.5 15.8

16.78 24.9 28.1 25.16 29.56

5.11 19.85 25.24 14.5 23.1

0.66 1.23 0.55 0.88 1.63

2.87 3.06 2.59 3.54 3.56

Modeling and simulation of synthetic aperture radar dataset for retrieval of soil surface parameters

Table 10.1 Measured field and image information used in this study.

191

192

Radar Remote Sensing

Figure 10.3 Snaps of sample site location as per satellite schedule.

Figure 10.4 Flowchart for processing Sentinel-1 synthetic aperture radar image.

Modeling and simulation of synthetic aperture radar dataset for retrieval of soil surface parameters

on the same day as satellite data acquisition. First, we processed SAR images of the required field and calculated the essential parameters for input data for each field (sigma naught [s0] and incidence angle). Work was divided into three steps. In the first step, we evaluated the performance of the model. Then, the measured s0 of each sample field was compared with the simulated s0 from the different models for each value of VV and VH polarization. In the next step, soil moisture was calculated from Sentinel 1 images for each type of field data. We checked the accuracy and the estimated values from the figure compared with the ground measurement. Satellite data processing started from the preprocessing steps; for processing, SNAP software was used to perform various preprocessing steps. After that, radiometric correction and geometric correction were performed. Then, the Digital Number (DN) values of row data of Sentinel-1 data were converted into s0 images using radiometric correction. The resulting dataset was georeferenced using the terrain correction algorithm. In the next steps, we separated the field s0, which was exactly extracted for each field sample. First, the field sample was selected using geographic coordinates. The pixel value of the field sample was calculated from the image, and then the s0 values were converted into decibels. After that, SNAP was used to perform postprocessing (speckle noise reduction) using the filter in the SNAP software tool.

4. Soil moisture modeling For soil moisture modeling, we used the backscattering values of Sentinel-1 data and the local incidence angle in the hybrid model, which combines the Oh model for the surface roughness parameter and the MDB model used for soil moisture. The relative soil permittivity was calculated using Topp’s model. 4.1 Evaluation of Oh, calibrated integral equation, and modified dubois models These models are implemented with interactive data language (IDL) simulation tools. All expressions given in the literature for the related models are written in the IDL software module and simulate parametric values (surface roughness s, soil moisture mv, and incidence angle q). The geophysical parameter was obtained from the image and the ground truth was used as a known parameter to simulate s0 in VH and VV polarizations. Then simulated values were compared with the s0 values calculated from the image corresponding to the field area. The accuracy of the simulated and calculated s0 values was computed using root mean square error (RMSE) and the correlation coefficient (R2). After calculating RMSE and the R2 for all models, it was observed that the Oh model is better suited as compare to calibrated integral equation model (CIEM) and MDB model praposed by Baghdadi et al. (Oh, 2004). The MDB underestimate the parameter’s value. However, these models are useful for making new studies and algorithms.

193

194

Radar Remote Sensing

We used VV and VH polarization-related equations from all of these models (for the Oh model, Eqs. (10.2), (10.3), (10.5), (10.6) and (10.8)), where q is the cross-polarization ratio, [0 is the Fresnel reflectivity of the surface at nadir, εr is the dielectric constant, q is the incidence angle, k is the wavenumber, mv is soil moisture, and s is the surface roughness. To generate a simulated dataset, values for the parameter given in the models are based on the field data range (Table 10.1). We used the initial and final edge values of the field data and introduced the step size accordingly for the necessary parameters, such as 11%e31% for soil moisture, with step size of 0.2% (0.1e3.5 cm), surface roughness with a step size of 0.1 cm, and 30e41 degrees for the incidence angle with a step size of 1 degree. A similar range was used to simulate the large dataset generated by the Oh, CIEM, and MDB models. These datasets were used for further analysis and comparison. Surface roughness parameter ks and real and imaginary parts of the dielectric constant were calculated using Eqs. (10.4), (10.5) and (10.7), respectively using the inversion equation. The dielectric constant was calculated from the Hallikainen model in Eq. (10.17). The CIEM (Eqs. 10.15 and 10.16) and MDB (Eqs. 10.13 and 10.14) models were used to produce a wide range of simulated data: q¼

pffiffiffiffiffiffiffi  s0HV ¼ 0:23 [0 1  eks 0 sVV   1  pffiffiffiffi εr 2  [0 ¼  pffiffiffiffi 1 þ εr 

(10.2) (10.3)

Solving Eq. (10.3) gives the relation of the dielectric constant and Fresnel reflectivity of the surface at nadir. We can also calculate the dielectric constant using HH polarization, but in our work, we do not have HH polarization data: pffiffiffiffiffiffiffi  1  [0 2  pffiffiffiffiffiffiffi εr ¼  (10.4) 1 þ [0 We calculated the surface parameter using Eqs. (10.5) and (10.7), which were inverted from two different models (Oh and MDB), by giving them the same input range of ground measured data. Eqs. (10.2) and (10.6) describe the relationship between the cross-polarization of the image with the incidence angle and the surface roughness parameter: " ( )#0:556 s0VH ks ¼  3:125 ln 1  (10.5) 0:11mv0:7 ðcos qÞ2:2

Modeling and simulation of synthetic aperture radar dataset for retrieval of soil surface parameters

q¼

   s0VH ¼ 0:095ð0:13 þ sin 1:5 qÞ1:4 1  exp  1:3ðksÞ0:9 0 sVV

 2 31:111 q ln 1  1:4 0:095ð0:13 þ sin 1:5 qÞ 6 7 ks ¼ 4 5 1:3    s0VH ¼ 0:11mv0:7 ðcos qÞ2:2 1  exp  0:32ðksÞ1:8

(10.6)

(10.7)

(10.8)

For the imaginary part of the dielectric constant, we can use the bulk density equation to calculate it. Furthermore, it can be used to find the imaginary part of the dielectric constant using Eq. (10.11). First, we calculate the loss tangent using Eq. (10.10). r0 ¼ 3:53 log ε0

(10.9)

tan d ¼ 10ð0:44r0 2:943Þ

(10.10)

Loss tangent

Imaginary part of dielectric constant ε00 ¼ ε0 tan d We calculated an important parameter (skin depth) using Eq. (10.12): pffiffiffiffi ε0 Skin depth d ¼ pffiffiffiffiffi l 2p ε00

(10.11)

(10.12)

Furthermore, to simulate the different parameters as described in earlier sections, we implemented the expression of the MDB by Baghdadi: s0VV ¼ 101:138 ðcos qÞ1:528 100:008 cotanðqÞ mv ðksÞ0:71 sinðqÞ

(10.13)

s0VH ¼ 102:325 ðcos qÞ0:01 100:011 cotanðqÞ mv ðksÞ0:44 sinðqÞ

(10.14)

To simulate the CIEM for HV and VV polarization, we implemented equations: Lopt ðs; q; HVÞ ¼ 0:9157 þ 1:2289ðsin 0:1543 qÞ0:3139 s

(10.15)

Lopt ðs; q; VVÞ ¼ 1:281 þ 0:134ðsin 0:19 qÞ1:59 s

(10.16)

Hallikainen model ε ¼ ða0 þ a1S þ a2Si þ a3CÞ þ ðb0 þ b1S þ b2Si þ b3CÞ*m þ ðc0 þ c1S þ c2Si þ c3CÞ*m

(10.17)

195

196

Radar Remote Sensing

Top model mv ¼  5:3*102 þ 2:92*102 ε0  5:5*104 ε0 þ 4:3*106 ε0 2

3

(10.18)

4.5 Vegetation correction Afterward, vegetation correction was done. For this, a semiempirical model called the radiative transfer model was applied to minimize the impact of vegetation on the SAR backscattering coefficient from the soil surface. The backscattering coefficient is the result of the collective surface and vegetation canopy. Here, we applied the water-cloud model as a radiative transfer model: s0canopy ¼ s0veg þ g2 s0soil   s0veg ¼ AV1 cos q 1  g2

(10.20)

g2 ¼ expð  2V2 B = cosqÞ

(10.21)

(10.19)

where g2 is the canopy transmitting factor; V1 and V2 are canopy descriptors; and A and B are coefficients obtained by regression analysis (Susan et al., 1998). We considered the canopy descriptor as Leaf Area Index (LAI) for both V1 and V2. Using Eq. (10.1), we can easily find s0 soil for different crop covers, which can be used for further calculations. After performing vegetation correction, we have to calculate the backscattering coefficient with the Oh empirical model, which is explained in detail in Baghdadi et al. (2016). The basic idea of implementing this model was explained in the subsequent section. Then, we are in the position to retrieve the surface parameters of soil moisture. In the literature, many researchers compared different physical and empirical models and introduced some methods, such as the genetic algorithm (Mirsoleimani et al., 2019) in addition to the empirical model. We observed that the Oh model gives good results in different frequency bands (Baghdadi et al., 2016; Singh and Katapaliya, 2007).

5. Results and discussion Results show that the hybrid model provides good results. The RMSE for VV polarization was calculated at 1.4318 dB. In this model, we used the combined approach of two models (Oh and MDB), to calculate soil surface parameters. Surface roughness was calculated with the Oh model. These values were used to calculate soil moisture using the MDB model. The RMSE and R2 both were good for taking data samples.Table 10.2 shows a detailed analysis of the observed and calculated data. The results are reflected in terms of the regression model, as shown in Figs. 10.4 and 10.5 for VV and VH polarization.

Table 10.2 Measured field data and calculated data. Details of ground samples and soil moisture measured in field/laboratory Sample no.

Longitude

Latitude

Soil moisture (%)

s-1 s-2 s-3 s-4 s-5 s-6 s-7 s-8 s-9 s-10 s-11 s-12 s-13 s-14 s-15 s-16 s-17 s-18 s-19 s-20 s-21 s-22 s-23 s-24 s-25 s-26 s-27 s-28

75.77 75.77 75.76 75.76 75.74 75.76 75.78 75.79 75.81 75.83 75.83 75.92 76.02 76.01 76.00 76.02 75.78 75.55 75.31 75.07 74.84 75.60 75.36 75.12 75.89 75.65 75.41 75.17

18.98 18.97 18.97 18.96 19.00 18.96 18.93 18.89 18.86 18.82 18.84 18.85 18.87 18.88 18.90 18.91 18.85 18.79 18.73 18.67 18.61 18.55 18.49 18.42 18.36 18.30 18.24 18.18

15.20 8.90 9.50 11.70 16.70 7.30 19.70 8.10 9.80 15.00 16.40 21.30 37.20 22.70 25.60 28.30 25.80 49.10 23.10 26.20 30.00 14.60 17.30 18.30 11.20 27.10 15.90 28.10

12.20 8.60 8.80 10.10 10.40 12.10 15.40 11.60 12.20 16.00 22.90 24.30 19.70 22.40 25.50 48.30 20.00 40.40 19.70 7.50 31.20 19.20 13.10 8.70 8.40 30.90 19.60 24.60

9.60 11.00 9.80 10.70 12.20 13.70 18.40 15.20 13.10 16.40 23.10 25.40 25.70 22.50 28.20 29.60 45.70 46.20 15.80 13.80 34.60 13.20 15.90 7.40 11.00 20.80 17.30 32.10

9.30 10.70 7.20 9.70 13.00 11.40 15.20 14.20 11.00 17.40 23.70 19.60 19.30 20.10 30.30 47.10 25.70 24.80 25.60 14.50 30.10 18.60 17.90 12.20 13.20 19.50 13.90 29.30

Soil moisture calculation from synthetic aperture radar data using modified dubois model

8.70 11.30 8.40 13.30 13.50 14.80 18.30 10.70 13.70 14.50 20.10 23.80 25.70 19.40 23.30 25.20 23.80 48.90 29.90 9.70 25.20 15.60 19.80 13.10 11.70 34.50 19.70 25.90

Average soil moisture (%) 11.00 10.10 8.74 11.10 13.16 11.86 17.40 11.96 11.96 15.86 21.24 22.88 25.52 21.42 26.58 35.70 28.20 41.88 22.82 14.34 30.22 16.24 16.80 11.94 11.10 26.56 17.28 28.00

Gravimetric moisture

Local incidence angle

VH

VV

12.30 9.70 11.90 12.20 15.40 13.60 22.10 13.40 14.10 18.10 18.20 23.40 27.70 18.90 26.10 32.40 26.70 38.60 21.50 17.30 27.30 16.20 18.10 14.50 13.30 30.10 18.60 26.40

39.76 42.66 43.37 39.60 41.67 40.96 43.47 41.81 39.56 43.06 40.39 35.43 41.71 39.96 41.62 40.96 43.12 41.20 39.51 42.66 41.23 40.10 39.60 36.30 42.12 42.31 39.20 38.32

20.77 21.20 20.64 21.95 21.76 21.10 20.18 20.76 20.26 19.85 18.79 18.18 19.29 18.64 19.02 17.69 19.42 17.88 20.46 21.44 19.17 20.42 19.81 19.25 21.01 18.74 20.05 18.71

10.58 11.20 10.22 12.00 11.64 11.33 11.17 10.84 11.49 10.35 9.80 8.30 9.16 9.00 9.49 8.28 10.39 8.29 10.61 12.84 9.67 10.28 9.51 10.30 10.77 9.63 10.15 9.09

VH e VV

10.19 10.00 10.43 9.95 10.12 9.77 9.01 9.93 8.77 9.51 8.99 9.88 10.13 9.64 9.53 9.41 9.02 9.58 9.85 8.60 9.50 10.14 10.30 8.95 10.24 9.11 9.89 9.62

Ks

2.43 1.90 3.32 0.86 1.20 1.50 1.40 2.00 1.38 2.10 1.90 2.80 2.80 3.00 2.06 2.10 1.50 2.90 1.30 0.70 1.90 1.80 2.20 1.20 1.90 2.10 1.60 1.80

εr using topp’s model

Mv (VV)

Mv (VH)

εr (VV)

εr (VH)

8.26 10.10 8.74 14.57 14.45 11.87 19.19 11.96 10.15 18.14 22.43 23.13 22.87 20.65 26.00 37.58 26.00 30.75 20.38 15.21 25.20 18.00 21.23 20.00 14.30 25.15 20.41 27.45

10.49 10.10 8.74 11.10 10.05 12.90 22.50 12.94 19.45 20.52 28.28 25.37 21.42 24.79 26.67 36.46 28.00 32.00 18.48 18.90 26.00 16.04 18.59 25.27 11.53 29.29 19.47 27.55

4.75 5.38 4.91 7.25 7.19 6.06 9.65 6.10 5.40 9.07 11.59 12.04 11.87 10.50 13.97 23.07 13.97 17.47 10.34 7.55 13.42 8.99 10.85 10.12 7.12 13.38 10.36 15.00

5.52 5.38 4.91 5.76 5.36 6.50 11.64 6.51 9.80 10.42 15.60 13.53 10.96 13.14 14.44 22.11 15.40 18.44 9.25 9.49 13.97 7.96 9.31 13.46 5.93 16.35 9.81 15.07

Continued

Table 10.2 Measured field data and calculated data.dcont’d s-29 s-30 s-31 s-32 s-33 s-34 s-35 s-36 s-37 s-38 s-39 s-40 s-41

75.94 75.70 75.46 75.23 75.99 75.75 75.51 75.28 75.04 75.80 75.56 75.33 75.09

18.12 18.06 19.00 18.94 18.87 18.81 18.75 18.69 18.63 18.57 18.51 18.45 18.39

21.60 29.60 24.90 15.10 34.60 21.20 11.40 16.50 10.50 26.00 16.50 13.20 32.70

26.40 21.90 34.60 8.70 29.50 21.70 8.60 19.10 10.60 13.90 12.70 15.50 30.10

19.40 34.60 19.30 10.20 28.20 15.90 11.20 12.30 16.10 13.60 9.40 19.50 25.10

16.70 19.70 25.60 10.60 24.40 18.80 8.10 15.50 8.20 16.80 9.30 18.60 29.40

23.10 24.60 27.90 13.20 35.60 16.30 8.90 19.40 9.10 30.60 10.60 21.60 24.60

21.44 26.08 26.46 11.56 30.46 18.78 9.64 16.56 10.90 20.18 11.70 17.68 28.38

24.60 22.10 27.50 13.40 26.70 17.50 12.30 14.20 11.60 21.60 8.90 14.30 30.10

36.50 40.30 41.25 43.62 41.37 39.46 36.24 41.12 43.31 40.58 37.39 39.90 39.56

18.27 19.54 19.61 20.05 19.25 20.16 20.44 20.37 20.70 19.40 21.23 19.68 19.02

8.68 9.51 10.02 9.96 9.74 10.28 10.51 10.38 11.12 9.28 10.81 9.48 9.29

9.59 10.03 9.58 10.08 9.51 9.88 9.93 9.99 9.58 10.12 10.42 10.20 9.74

2.60 1.90 1.50 2.60 1.60 2.10 1.90 1.80 2.10 2.30 1.60 2.20 1.80

22.12 25.32 26.59 18.19 28.45 14.07 10.36 18.45 9.41 24.12 11.64 21.94 27.53

26.10 22.43 24.88 16.70 27.18 16.31 13.84 16.92 13.37 21.84 10.20 19.71 26.30

11.40 13.50 14.38 9.09 15.73 7.01 5.48 9.23 5.13 12.69 5.97 11.29 15.06

14.04 11.59 13.20 8.30 14.80 8.10 6.91 8.41 6.70 11.22 5.42 9.95 14.18

Modeling and simulation of synthetic aperture radar dataset for retrieval of soil surface parameters

Figure 10.5 Regression model of calculated and measured dataset. SAR, synthetic aperture radar; SM, soil moisture.

6. Conclusion Soil moisture is an important parameter that can be used in various applications such as agriculture and hydrology. Retrieval of soil moisture by traditional methods is accurate but tedious and not feasible for large-scale data. Soil moisture parameters are the top priority for many applications such as agricultural and hydrology modeling. SAR remote sensing is a more powerful technique for retrieving large-scale soil moisture data with a high spatial and temporal resolution up to a depth of 0e5 cm. There are many methods for calculating soil moisture using the theoretical model and semiempirical and empirical

199

200

Radar Remote Sensing

methods. A hybrid model technique is explained in this chapter that fuses the Oh and MDB models. Vegetation cover is a major challenge in soil moisture calculation. One can overcome the effect of vegetation using vegetation correction. A different model was explained that helps to calculate soil surface parameters. The Oh and MDB models can be used for different datasets such as L-, S-, and C-bands and for different polarizations (dual and quad polarization) for calculating soil moisture. The hybrid model is also used to calculate soil moisture accurately. A Sentinel-1 dataset were used for this work because it has great potential to downscale soil surface parameters compared with the available SAR product.

References Alexakis, D.D., Mexis, F.D.K., Vozinaki, A.E.K., Daliakopoulos, I.N., Tsanis, I.K., 2017. Soil moisture content estimation based on Sentinel-1 and auxiliary earth observation products. A hydrological approach. Sensors 17, 1455. Altese, E., Bolognani, O., Mancini, M., Troch, P.A., 1996. Retrieving soil moisture over bare soil from ERS 1 synthetic aperture radar data: sensitivity analysis based on a theoretical surface scattering model and field data. Water Resour. Res. 32, 653e661. Attema, E., Ulaby, F.T., 1978. Vegetation model as water cloud. Radio. Sci. 13, 357e364 ([CrossRef]). Aubert, M., Baghdadi, N., Zribi, M., Douaoui, A., Loumagne, C., Baup, F., El Hajj, M., Garrigues, S., 2011. Analysis of TerraSAR-X data sensitivity to bare soil moisture, roughness, composition and soil crust. Remote Sens. Environ. 115, 1801e1810 (CrossRef). Baghdadi, N., Cresson, R., El Hajj, M., Ludwig, R., La Jeunesse, I., 2012. Soil parameters estimation over bare agriculture areas from C-band polarimetric SAR data using neural networks. Hydrol. Earth Syst. Sci. Discuss. 9, 2897e2933. Baghdadi, N., Choker, M., Zribi, M., Hajj, El, Paloscia, S., Verhoest, N., Lievens, H., Baup, F., Mattia, F., 2016. A new empirical model for radar scattering from bare soil surfaces. Remote Sens. 8, 920 (CrossRef). Barrett, B.W., Dwyer, E., Whelan, P., 2009. Soil moisture retrieval from active spaceborne microwave observations: an evaluation of current techniques. Rem. Sens. 1, 210e242. Bauer-Marschallinger, B., Freeman, V., Cao, S., Paulik, C., Schaufler, S., Stachl, T., Modanesi, S., Massari, C., Ciabatta, L., Brocca, L., et al., 2018. Toward global soil moisture monitoring with Sentinel-1: harnessing assets and overcoming obstacles. IEEE Trans. Geosci. Rem. Sens. 57, 520e539. Bousbih, S., Zribi, M., El Hajj, M., Baghdadi, N., Lili-Chabaane, Z., Gao, Q., Fanise, P., 2018. Soil moisture and irrigation mapping in A semi-arid region, based on the synergetic use of Sentinel-1 and Sentinel-2 data. Rem. Sens. 10, 1953. Dabrowska-Zielinska, K., Musial, J., Malinska, A., Budzynska, M., Gurdak, R., Kiryla, W., Bartold, M., Grzybowski, P., 2018. Soil moisture in the Biebrza Wetlands retrieved from sentinel-1 imagery. Rem. Sens. 10, 1979. Dave, R., Kumar, G., Kr Pandey, D., Khan, A., Bhattacharya, B., 2019. Evaluation of modified Dubois model for estimating surface soil moisture using dual polarization RISAT-1 C-band SAR data. Geocarto Int. 1e11. https://doi.org/10.1080/10106049.2019.1655801. Dubois, P.C., Van Zyl, J., Engman, T., 1995. Measuring soil moisture with imaging radars. IEEE Trans. Geosci. Rem. Sens. 33, 915e926. El Hajj, M., Baghdadi, N., Zribi, M., Belaud, G., Cheviron, B., Courault, D., Charron, F., 2016. Soil moisture retrieval over irrigated grassland using X-band SAR data. Remote Sens. Environ. 176, 202e218. El Hajj, M., Baghdadi, N., Zribi, M., Bazzi, H., 2017. Synergic use of Sentinel-1 and Sentinel-2 images for operational soil moisture mapping at high spatial resolution over agricultural areas. Rem. Sens. 9, 1292.

Modeling and simulation of synthetic aperture radar dataset for retrieval of soil surface parameters

Entekhabi, D., 1993. Retrieval of Soil Moisture Profile by Combined Remote Sensing and Modeling. Available online: https://trs.jpl.nasa.gov/handle/2014/34543. (Accessed 11 May 2020). Ezzahar, J., Ouaadi, N., Zribi, M., Elfarkh, J., Aouade, G., Khabba, S., Er-Raki, S., Chehbouni, A., Jarlan, L., 2020. Evaluation of backscattering models and support vector machine for the retrieval of bare soil moisture from sentinel-1 data. Rem. Sens. 12, 72. Field, C.B., Randerson, J.T., Malmström, C.M., 1995. Global net primary production: combining ecology and remote sensing. Remote Sens. Environ. 51, 74e88. Fung, A.K., Cheng, K.S., 1994. Microwave Scattering and Emission Models and Their Applications. Artech House Inc., Boston, MA, USA. Fung, A.K., Li, Z., Chen, K.S., 1992. Backscattering from a randomly rough dielectric surface. IEEE Trans. Geosci. Rem. Sens. 30, 356e369. Hachani, A., Ouessar, M., Paloscia, S., Santi, E., Pettinato, S., 2019. Soil moisture retrieval from Sentinel-1 acquisitions in an arid environment in Tunisia: application of Artificial Neural Networks techniques. Int. J. Rem. Sens. 40, 9159e9180. Hosseini, M., McNairn, H., 2017. Using multi-polarization C-and L-band synthetic aperture radar to estimate biomass and soil moisture of wheat fields. Int. J. Appl. Earth Obs. Geoinf. 58, 50e64. Jackson, T., Hawley, M., O’neill, P., 1987. Preplanting soil moisture using passive microwave sensor. J. Am. Water Resour. Assoc. 23, 11e19. Jackson, T.J., 1993. Measuring surface soil moisture using passive microwave remote sensing. Hydrol. Process. 7, 139e152. Kerr, J.T., Ostrovsky, M., 2003. From space to species: ecological applications for remote sensing. Trends Ecol. Evol. 18, 299e305. Kite, G., Pietroniro, A., 1996. Remote sensing applications in hydrological modelling. Hydrol. Sci. J. 41, 563e591. Lecomte, V., King, C., Cerdan, O., Baghdadi, N., Bourguignon, A., 2001. Use of remote sensing data as alternative inputs in the stream Runoff model. In: Proceedings of the International Symposium on Physical Measurements & Signatures in Remote Sensing, Aussois, France, pp. 699e704, 8e12 January. Martin, M.E., Aber, J.D., 1997. High spectral resolution remote sensing of forest canopy lignin, nitrogen, and ecosystem processes. Ecol. Appl. 7, 431e443. Mirsoleimani, H.R., Sahebi, M.R., Baghdadi, N., El Hajj, M., 2019. Bare soil surface moisture retrieval from sentinel-1 SAR data based on the calibrated IEM and Dubois models using neural networks. Sensors 19, 3209. Mulla, D.J., 2013. Twenty-five years of remote sensing in precision agriculture: key advances and remaining knowledge gaps. Biosyst. Eng. 114, 358e371. Narvekar, P.S., Entekhabi, D., Kim, S., Njoku, E.G., 2015. Soil moisture retrieval using L-band radar observations. IEEE Trans. Geosci. Rem. Sens. 53, 3492e3506. Oh, Y., Sarabandi, K., Ulaby, F.T., 1992. An empirical model and an inversion technique for radar scattering from bare soil surfaces. IEEE Trans. Geosci. Rem. Sens. 30, 370e380 (CrossRef). Oh, Y., Sarabandi, K., Ulaby, F.T., 1994. An inversion algorithm for retrieving soil moisture and surface roughness from polarimetric radar observation. In: Proceedings of the IEEE International Geoscience and Remote Sensing Symposium, Pasadena, CA, USA, vol. 3, pp. 1582e1584, 8e12 August. Oh, Y., Sarabandi, K., Ulaby, F.T., 2002. Semi-empirical model of the ensemble-averaged differential Mueller matrix for microwave backscattering from bare soil surfaces. IEEE Trans. Geosci. Rem. Sens. 40, 1348e1355. Oh, Y., 2004. Quantitative retrieval of soil moisture content and surface roughness from multipolarized radar observations of bare soil surfaces. IEEE Trans. Geosci. Rem. Sens. 42, 596e601 (CrossRef). Qiu, J., Crow, W.T., Wagner, W., Zhao, T., 2019. Effect of vegetation index choice on soil moisture retrievals via the synergistic use of synthetic aperture radar and optical remote sensing. Int. J. Appl. Earth Obs. Geoinf. 80, 47e57. Rao, S.S., Das, S., Nagaraju, M., Venugopal, M., Rajankar, P., Laghate, P., Reddy, M.S., Joshi, A., Sharma, J., 2013. Modified Dubois model for estimating soil moisture with dual polarized SAR data. J. Indian Soc. Remote Sens. 41, 865e872.

201

202

Radar Remote Sensing

Seneviratne, S.I., Corti, T., Davin, E.L., Hirschi, M., Jaeger, E.B., Lehner, I., Orlowsky, B., Teuling, A.J., 2010. Investigating soil moistureeclimate interactions in a changing climate: a review. Earth Sci. Rev. 99, 125e161. Singh, D., Katapaliya, A., 2007. An efficient modeling with ga approach to retrieve soil texture, moisture and roughness from ers-2 sar data. PIER 77, 121e136. Sivapalan, M., Takeuchi, K., Franks, S., Gupta, V., Karambiri, H., Lakshmi, V., Liang, X., McDonnell, J., Mendiondo, E., O’connell, P., 2003. IAHS decade on predictions in ungauged Basins (PUB), 2003e2012: shaping an exciting future for the hydrological sciences. Hydrol. Sci. J. 48, 857e880. Susan, M., Vidal, A., Troufleau, D., Ionue, Y., Mitchell, T.A., 1998. Ku- and C-band SAR for discriminating agricultural crop and soil conditions. IEEE Trans. Geosci. Rem. Sens. 36, 265e272. Ulaby, F.T., 1982. Microwave remote sensing active and passive. In: Rader Remote Sensing and Surface Scattering and Emission Theory; NASA Technical Report. Kansas University, Lawrence, KS, USA, pp. 848e902.  Verhoest, N.E., Lievens, H., Wagner, W., Alvarez-Mozos, J., Moran, M.S., Mattia, F., 2008. On the soil roughness parameterization problem in soil moisture retrieval of bare surfaces from synthetic aperture radar. Sensors 8, 4213e4248. Wagner, W., Pathe, C., Doubkova, M., Sabel, D., Bartsch, A., Hasenauer, S., Blöschl, G., Scipal, K., Martínez-Fernandez, J., Löw, A., 2008. Temporal stability of soil moisture and radar backscatter observed by the Advanced Synthetic Aperture Radar (ASAR). Sensors 8, 1174e1197. Wagner, W., Sabel, D., Doubkova, M., Bartsch, A., Pathe, C., 2009. The potential of Sentinel-1 for monitoring soil moisture with a high spatial resolution at global scale. In: Proceedings of the Earth Observation andWater Cycle Science. Frascati, Italy, 18e20 November. Walker, J.P., 1999. Estimating Soil Moisture Profile Dynamics from Near-Surface Soil Moisture Measurements and Standard Meteorological Data. Ph.D. Thesis. University of Newcastle, Callaghan, Australia. Yang, L., Feng, X., Liu, F., Liu, J., Sun, X., 2019. Potential of soil moisture estimation using C-band polarimetric SAR data in arid regions. Int. J. Rem. Sens. 40, 2138e2150. Zribi, M., Muddu, S., Bousbih, S., Al Bitar, A., Tomer, S.K., Baghdadi, N., Bandyopadhyay, S., 2019. Analysis of L-band SAR data for soil moisture estimations over agricultural areas in the tropics. Rem. Sens. 11, 1122.

CHAPTER 11

Flood inundation mapping from synthetic aperture radar and optical data using support vector machine: a case study from Kopili River basin during Cyclone Amphan Prasad Balasaheb Wale1, Thota Sivasankar1, Varun Narayan Mishra2 and Ratna Sanyal3 1

Geographic Information Systems (GIS) Area, NIIT University, Neemrana, Rajasthan, India; 2Centre for Climate Change and Water Research, Suresh Gyan Vihar University, Jaipur, Rajasthan, India; 3Computer Science and Engineering Area, NIIT University, Neemrana, Rajasthan, India

1. Introduction Floods are one the most frequent natural disasters in the northeastern states of India, including Assam, West Bengal, and Odisha (Jain et al., 2006; Mohapatra, 2003). These states are sensitive to phenomenal changes and damage caused by floods every year because of their richness in the ecological biodiversity of the Sundarbans and Assam Himalaya ranges (Dixit and Bera, 2012). The Ganges and Brahmaputra rivers are the biggest rivers of India in terms of their size as well as the intensity of flood events (Uddin et al., 2019). Most flood events from the Ganges and Brahmaputra rivers occur in these states. Kopili River is one of the most important tributaries of Brahmaputra River; it flows from Sikkim and Assam states. There are various natural and man-made reasons for the increasing flood events in India (Jain et al., 2006). Primary reasons for the floods are excessive rainfall during the monsoon period, river course changes due to erosion, and some anthropogenic reasons such as dam breams (Tripathi, 2015). In addition, glacial lake outburst floods, rapid snow melting in the upper Himalayas, and landslides are important causes of flash floods specifically in northern India (Bhatt et al., 2014). Floods are considered even more dangerous in the coastal states of India. Catastrophic storms originating in the ocean are a primary reason for these floods. Such phenomena are destructive when they enter the land with high-speed winds and rainfall (Hassan et al., 2020). Cyclone Amphan was one of the largest cyclones in history; it originated in the Bay of Bengal in May 2020. Amphan disrupted the living habitats of all of West Bengal, Odisha, and Assam states. The highest impact of Amphan was reported in Bangladesh (Hassan et al., 2020). All of this combined to attract the attention of researchers to study floods across the world. Studies involve continuous monitoring, mapping, and risk analysis of Radar Remote Sensing ISBN 978-0-12-823457-0, https://doi.org/10.1016/B978-0-12-823457-0.00017-3

© 2022 Elsevier Inc. All rights reserved.

203

204

Radar Remote Sensing

floods at different spatial and temporal scales. Such studies will become positive input to minimize risk during emergencies and reduce social, economic, and demographic damage (Schlaffer et al., 2015; Liang and Liu, 2020). Remote sensing (RS) techniques and geographic information systems (GIS) provide a great opportunity to monitor earth surface changes continuously over time (Rahman and Thakur, 2018). RS deals with information about objects or any phenomena, acquired by a sensor on an airborne or spaceborne platform at a certain distance (Sivasankar et al., 2020a,b), whereas GIS refers to a computer-based system that processes, analyzes, modifies, and visualizes spatial data in different formats (Bhatt et al., 2014). Therefore, the research community started using these techniques in various applications including vegetation analysis, change detection studies, and land use land cover studies. Flood monitoring and flood inundation mapping are important areas in which these techniques have attracted much attention. Flood inundation mapping studies could be done using both optical and microwave RS datasets (Jacinth Jennifer et al., 2020). Both optical and microwave RS techniques are able to provide near real-time information about floods throughout various stages during clear weather conditions (Borah et al., 2018). Unfortunately, optical remote sensors have working limitations during cloudy, rainy, and hazy atmospheric conditions (Rahman and Thakur, 2018). On the other hand, synthetic aperture radar (SAR) microwave remote sensors are able to provide continuous earth surface information during all weather conditions (Tiwari & et al., 2020). In addition, SAR sensors can provide daytime as well as nighttime operational modes with fine resolution (Shen et al., 2019). These abilities make SAR useful for different studies such as soil moisture variation studies (Srivastava et al., 2015), plant density calculation, land cover classifications (De et al., 2018; Patel et al., 2006), flood inundation mapping, and water quality analysis (Borah et al., 2018; Sivasankar et al., 2020b). These working principles and abilities make SAR data a unique alternative to optical data. Floods are always associated with rainfall and extreme weather conditions. Therefore, SAR data are the only reliable source for accurate and precise flood monitoring studies. Continuous and reliable information from SAR makes it easy to monitor the preflood and postflood situation to take some precautionary steps to avoid sensitive damage to the ecology and properties all over the world. Apart from SAR, some researchers prefer the multispectral data of Landsat and Sentinel satellite series for flood inundation mapping studies. An important reason is that optical data are comparatively simpler and userfriendly than SAR data. On the other hand, it has become difficult to get cloud-free optical data during the monsoon period, when most floods take place. In addition, some researchers prefer to use optical data as input to validate the results of SAR flood inundation maps. Digital elevation maps can also have a significant role in finding the magnitude and depth of an flood estimation in flood mapping if we are interested in modeling a flood event (Rahman and Thakur, 2018).

Flood inundation mapping

Advances in computer science, statistical analysis techniques, and machine learning algorithms provide a wide range of analysis techniques to extract flooded areas from satellite imagery. Common techniques for separating water and nonwater pixels from images are supervised and unsupervised image classification techniques. In terms of supervised classification, training data are important to train the classifier. However, one unavoidable fact is that during floods it is dangerous to collect field data to validate the results, whereas unsupervised classification techniques use different stretching and clustering techniques to classify the image (Tiwari et al., 2020). The second important and widely used technique for flood inundation mapping is thresholding (Cao et al., 2019): These techniques involve the automatic and manual threshold to separate water and nonwater pixels. These techniques primarily consider the amount of backscatter received from the target to the sensor to separate different objects of the earth’s surface. Every object on the earth’s surface reacts differently to incoming radiation, depending on different factors. For instance, specular reflection from the water body generates very low backscatter toward the SAR sensor; therefore, water appears the black in the image (Borah et al., 2018; Schlaffer et al., 2015). On other hand, vegetation, urban features, soil, and rough surfaces scatter huge backscatter toward the sensor; therefore, they appear white in images (Manjusree et al., 2012). However, backscatter from any ground target can also change with changes in the moisture content of the object, incident angle, or look angle. Changes in the physical property of objects can even change the backscatter quantity of the target. For instance, calm water and wave water have different backscatter values (Matgen et al., 2011; Manjusree et al., 2012). Therefore, the selection of the technique to separate land cover classes becomes important, depending on the surface. This automatic threshold technique uses bimodal histogram distribution to separate water and nonwater pixels (Cao et al., 2019). The result of these thresholding techniques depends on the presence of water and nonwater pixels in the image. However, most researchers mention that automatic thresholding proves better than manual thresholding (Zeng et al., 2020; Cao et al., 2019; Tiwari et al., 2020). Some researchers also use machine learninge based supervised classification techniques to classify both SAR and Sentinel 2 multispectral images to separate different land covers with greater accuracy. Support vector machine (SVM), random forest, and artificial neural networks are the most used machine learning techniques to classify different complex objects on the earth’s surface. These techniques provide precise results, depending on the training samples and application of kernel function (Lardeux et al., 2006). These machine learning techniques are a great alternative to other classification techniques in both multispectral and SAR data classification. Crop classification, flood inundation mapping, land use land cover classification, and change detection studies are areas in which the SVM algorithm is being used at a rapid rate (Yekkehkhany et al., 2014; Verma et al., 2020).

205

206

Radar Remote Sensing

Floods are common phenomena in the Brahmaputra River basin. Various researchers have performed some flood inundation mapping studies in northeastern India. Borah et al. (2018) monitored floods in Kaziranga National Park, because it is associated with the Brahmaputra River and Assam. In this study, we used Sentinel 1 SAR C-band data (VV polarization) for flood inundation mapping of Kopili River in Assam during cyclone Amphan. Quantitative measurements were calculated to define the maximum water extent during the flood. Furthermore, Sentinel 2 multispectral images were considered to validate the results along with the Normalized Difference Water Index (NDWI) and Modified and Normalized Difference Water Index (MNDWI). In this study, the following section describes the details of the study area and the stepwise methodology and results. Fig. 11.2 provides a graphical representation of the methodology. The results of the SAR flood inundation maps were better than the optical flood maps for defining the flood inundation zones.

2. Study area Kopili River is an important tributary Brahmaputra River. Kopili River originates in one of the richest biodiversity states of India, Meghalaya. Then, it flows northward and enters Assam, where it becomes submerged in the Brahmaputra River and finally merges into the Bay of Bengal. The study area map in Fig. 11.1 shows the Kopili River and major tributaries. The total length of the river is nearly 290 km from its origin. Brahmaputra River is well-known for the floods and damage caused by them in Assam and adjoining states of northeast India. In this study, we focused on a portion of Kopili River, where the maximum flood that occurred during cyclone Amphan is identified in Fig. 11.1 with a rectangle. Fig. 11.1 also shows that Nagaon and Kabri Anglong were the worst affected districts. Maximum flooding was located at 25 degrees 400 N to 26 degrees 150 N latitude to 92 degrees 400 to 93 degrees 100 E longitude. The Kopili River valley and adjoining area is rich in terms of its landscape and ecological habitat. The river has created some important geomorphologic features in the channel, including waterfalls, gorges, and some long rapids when it flows in the mountainous area. When Kopili River enters Assam, the river is almost mature and begins to make flood plains. Nagaon District is wellknown for flood plains created by Kopili River. The Shuttle Radar Topography Mission (SRTM) digital elevation map (DEM) in Fig. 11.1 shows the variations in elevation in the study area. The southeast and northwest parts have the maximum elevation, whereas Nagaon District has a comparatively low elevation. Because of this low elevation, the river deposits sediment and creates flood plains in Nagaon District.

Flood inundation mapping

Figure 11.1 Location map of study area along with Kopili River channel and elevation information of Shuttle Radar Topography Mission digital elevation map.

3. Material and methods 3.1 Dataset used Sentinel 1 (SAR) VV and VH polarized data were used in this study. The scene of SAR before and after the flood was downloaded from the Sentinel portal from archive (preflood) images of May 16, 2020 and the crisis image (during and after the flood) of May 28, 2020. Sentinel satellites are well-known for the constellation of two satellites and high spatial resolution datasets. In addition, these satellites provide the SAR C-band dataset at a 12-day interval for the world with a 29- to 46-degree incident angle (Borah et al., 2018). They provide multiple polarization mode data including HH, VV, VV plus VH, and HH plus HV polarization. In this study, VV-polarized level 1 ground range detection images with the descending instrument mode and interferometric wide swath were used. Sentinel 2 optical images were accessed and downloaded through the Google Earth Engine (GEE) platform. Two datasets of Sentinel 2 Multispectral Instrument (MSI) were

207

208

Radar Remote Sensing

accessed for May 16, 2020 (archive and before the flood) and May 28, 2020 (crisis and postflood image). GEE is a cloud-based platform in which Landsat, Sentinel, Moderate Resolution Imaging Spectroradiometer, and some DEM datasets are available ready to use. We can import other required raster and vector data into the GEE platform and perform various spatial analyses. The GEE platform is helpful for researchers because cloud-based platforms consume analysis time and save the hardware storage-related issues. The Sentinel 2 MSI satellite provides data at a 12-day interval with 10 m spatial resolution (visible and near-infrared [NIR]). SRTM DEM were used to analyze the elevation of the study area. DEM data were downloaded from QGIS, open access GIS software with an SRTM Download plug-in and European Space Agency credentials. The overall methodology of this study is shown in Fig. 11.2. It is categorized into two broad sections. The first section shows SAR data processing and the second section shows Sentinel 2 MSI data processing. Both sections followed similar processing procedures, but SAR has some additional preprocessing over optical data. The detailed processing is mentioned in the following subsection. 3.2 Synthetic aperture radar data processing SAR data are technically more complex than multispectral data of Sentinel 2 MSI because of their complex acquisition process and transformations. Therefore, we needed to preprocess SAR data before using them in the study. All of these preprocessing steps were carried out using the SNAP desktop application. The first step involves the subset of data. We used fixed latitude and longitude extend to make a subset of both archive (May 16) and crisis (May 28) image datasets. The subset step has an important role in analyzing the histogram to separate water and nonwater pixels. In addition, it helps to improve the processing speed. The second step deals with calibration. Calibration is an important step to enhance backscatter values from the target. After calibration, the backscatter intensity values get converted into sigma, beta, and gamma naught values, which improves image visualization and helps to identify high scatter and low scatter features (Jacinth Jennifer et al., 2020). The third step involves the terrain correction. Terrain correction creates the shadow, foreshortening, and layover effects, which can distort the target from its actual position. In this study, we SRTM data for range Doppler terrain correction, as did Borah et al. (2018). After terrain correction, we can render the image smoothly. Active remote sensors always generate unnecessary noise while collecting the ground data owing to different sensing modes and angles. Therefore, we used a Lee filter with a 7  7-pixel window to remove all speckles from both images. Furthermore, both archive and crisis images with VV polarization were converted into decibel format to get precise bimodal and nonlinear histograms to separate water and nonwater pixels.

Flood inundation mapping

Figure 11.2 Overall methodology followed in this study. FCC, false color composite; GEE, Google Earth Engine; MNDWI, Modified and Normalized Difference Water Index; MSI, Multispectral Instrument; NDWI, Normalized Difference Water Index; ROI, region of interest; SAR, synthetic aperture radar; SVM, support vector machine.

After preprocessing, both archive and crisis images of May 16 and May 28, 2020 were taken to make RGB color composites. In the final RGB color composite, the archive VV polarization decibel image was assigned to the red channel whereas the crisis VV polarization decibel image was assigned to both the green and blue channels. This RGB color

209

210

Radar Remote Sensing

composite allows us to understand the extent of the flood as well as permanent water bodies. It is possible because of different backscatter responses from the same region of interest over time. 3.3 Flood inundation mapping In this study, SVM supervised image classification techniques were used to classify permanent water, flood water, and nonwater areas using the ENVI Classic desktop application. An accuracy assessment was also done using ENVI Classic, which gives classification validation. The quantitative measurements and postclassification processing were done using the ArcMap 10.3 desktop application to improve the flood inundation mapping accuracy. Finally, flood inundation maps were prepared and compared with Sentinel 2 MSI flood maps. 3.4 Sentinel 2 Multispectral Instrument classification and Normalized Difference Water and Modified and Normalized Difference Water indices The GEE platform was used to access and process Sentinel 2 MSI images of May 28, 2020 (crisis). A false color composite was made for better visualization with NIR, red, and green bands. Further images were exported to perform SVM supervised classification in the ENVI desktop application. SVM classification was used to separate water and nonwater pixels. Optical data are limited by cloud cover; therefore, the cloud masking function of GEE was used to remove cloud cover from images. The primary motive for removing cloud cover was to avoid the mix classification and quality of results. This classification was validated through an accuracy assessment in the ENVI classic desktop application. The classified maps were imported and further processed to make flood maps. NDWI and MNDWI are important indices for water analysis. In this study, both indices were used to check the extent of water. The formula for both indices is: NDWI [ (green e NIR) / (green D NIR) (Yang et al., 2017) MNDWI [ (green e shortwave infrared) / (green D shortwave infrared) (Yang et al., 2017) The basic difference between two indices is that in the NDWI there is a high chance of mixing the built-up area with the water class (Nair and Babu, 2016). Soil pixels and vegetation represent negative values or close to zero, whereas MNDWI is considered more suitable for water body extraction because it can avoid the mixed pixel problem of built-up areas in water (Ali et al., 2019).

4. Result and discussion SAR VV polarized postprocessed images of the study area are shown in Fig. 11.3. One can easily distinguish the difference between the images based on a visual inspection. On May 16, 2020, a normal situation occurred at Kopili River and the adjoining areas, but

Flood inundation mapping

Figure 11.3 Synthetic aperture radar archive (May 16, 2020) and crisis (May 28, 2020) images of study area.

when we observe the crisis image of May 28, 2020, a flood condition can be seen. Based on Fig. 11.3, we can conclude that almost one-quarter of the area was in a flood emergency. We can easily detect water in the river and some water in oxbow lakes. These oxbow lakes are a geomorphologic features created by the river. When the river deposits a huge amount of sediment in the mature stage, it slowly changes its direction of flow and creates oxbow lakes. However, in a crisis image, everything is changed owing to the presence of high flood water. On other hand, only the interpretation of two images cannot arrive at a good estimation. Therefore, quantitative measurements with better visualization were made. The SAR classification map classified the Kopili River basin into three major classes: flood water, permanent water, and nonwater. The flood water pixels show the area currently covered by water, but it basically belongs to other activities and classes such as agriculture or barren land. We estimated that the 145-km2 area had a flood emergency, whereas the permanent water class shows the Kopili River area. This is an area where you have water all of the time. The 18-km2 area was found in the permanent water category. The nonwater class represents agricultural land, urban features, a forest area, and barren

211

212

Radar Remote Sensing

Figure 11.4 Synthetic aperture radar (SAR) image composite classification map showing Kopili River basin with permanent water, flood water, and nonwater classes.

land. We placed all of these features into only one class because our primary object was to estimate flood inundation areas. The nonwater class covered a 655-km2 area during flood emergencies. To validate the accuracy of the map, a classification accuracy assessment was performed. We received 98.23% overall accuracy with a 97.42% kappa coefficient accuracy. Fig. 11.4 shows that permanent water covered only 2.20% of the total area, whereas flood water covered almost 17.73% of the area and the maximum area was covered by a nonwater area that was 80.07%. In addition, the flooded area was almost eight times bigger than the permanent water area. We also considered Sentinel 2 MSI optical data to compare and validate results from SAR-based flood inundation mapping. Images were captured by the Sentinel 2 satellite on the same dates: May 16, and May 28, 2020. As mentioned earlier, optical data are sensitive to cloud cover. We could detect the presence of clouds in the central portion of the archive image and the southwest portion of the crisis image. Fortunately, in Fig. 11.5, the stream is completely visible in both images. Because of cloud cover, we cannot see the ground situation; therefore, we always prefer SAR data for flood inundation mapping. These images were composed with false color

Flood inundation mapping

Figure 11.5 Sentinel 2 Multispectral Instrument (optical) archive and crisis images of Kopili River basin in false color composite (FCC) format. NIR, near-infrared; ROI, region of interest.

composite by assigning NIR, red, and green bands to RGB color channels to improve visualization. The crisis image is also classified to check the accuracy of optical data for flood inundation mapping. The SVM supervised classification technique was used to classify optical data of May 28, 2020. Fig. 11.6 shows that the classified map divided the entire region into two major classes: water and nonwater. The results are satisfactory because the water and nonwater pixels were separated almost accurately, but owing to cloud masking, we lost some information from the image, which is a major drawback of using optical data for flood inundation mapping. We have estimated 94% overall classification accuracy with an 89% kappa coefficient. Classification accuracy with respect to optical data was lower than the SAR classified image (98.23%). This may have been because of cloud cover and mix pixel classification. We have also noticed that in the central portion of the image, some water pixels were transferred to nonwater pixels and some buildup and urban features were mixed with water pixels. We validated the SAR flood inundation map with the Sentinel 2 Optical flood inundation map with the help of regression analysis. Regression techniques have an important

213

214

Radar Remote Sensing

Figure 11.6 Sentinel 2 optical data image composite classification map showing Kopili River basin with water and nonwater classes.

role in finding similarities between classes. To validate the results, 25 random points were used to extract classified information from both classified maps, and the values were compared with each other. Of 25, points, two got a mixed classification in the optical image (Fig. 11.7). This graph shows a high relation between SAR and optical flood maps. The R2 value proves the statistical similarity of the SAR and optical datasets. Based on the 0.74 R2 value, there is a high positive relationship between the datasets. In addition to the optical data, we calculated the NDWI and MNDWI indices of the Kopili River basin to understand the extent of water flow there. Fig. 11.7 shows that MNDWI performs better than NDWI for marking the water extent boundary. In terms of NDWI, the highest value is 0.95 and the lowest value is e0.51, whereas in the case of MNDWI, the maximum value is 0.98 and the minimum value is 0.64. The difference values range is higher in the case of MNDWI than NDWI values. In the case of NDWI, soil and vegetation have negative values that are close to zero. However, urban features and built-up areas are mixed with water and the value can rise to

Flood inundation mapping

Classification Validation Graph

SAR DATA

3

R² = 0.7404 2

1

1

2

3

Optical Data

Figure 11.7 Regression analysis to check similarity between synthetic aperture radar (SAR) and optical data-based flood inundation maps.

0.25 from the built-up area. Therefore, in the NDWI in Fig. 11.8, surrounding urban features are mixed with water, but in the case of MNDWI, we can precisely see the difference between water and nonwater features. Other features including vegetation, the built-up area, and soil have negative values, whereas water gets positive values. Therefore, MNDWI represents the extent of flooding in Fig. 11.8 more precisely than does NDWI.

Figure 11.8 Normalized Difference Water Index (NDWI) and Modified and Normalized Difference Water Index (MNDWI) calculations of Kopili River basin to identify extent of water.

215

216

Radar Remote Sensing

5. Conclusion We have used SAR and optical data of the Sentinel satellite to delineate the extent of water during the cyclone Amphanedriven flood disaster in Kopili River basin, Assam. Optical data are sensitive to extreme weather conditions. Therefore, continuous monitoring of the earth’s surface is impossible. Thus, SAR data are a great alternative to optical datasets in different applications including flood inundation mapping and monitoring. Such accurate and timely information from SAR datasets can become a valuable input to decision-makers and planners. Based on the results we received in this study, SAR data show better results than optical data for delineating the extent of flooding. The presence of clouds in optical images may lead to wrong interpretations. NDWI and MNDWI results are satisfactory, but in this study MNDWI performed better than NDWI. These indices are useful for water analysis and may have different accuracy results in various case studies. Ultimately, the ability of SAR data along with advanced analytical techniques has a lot of potential in flood inundation mapping and monitoring studies.

References Ali, M.I., Dirawan, G.D., Hasim, A.H., Abidin, M.R., 2019. Detection of changes in surface water bodies urban area with NDWI and MNDWI methods. Int. J. Adv. Sci. Eng. Inf. Technol. 9 (3), 946e951. https://doi.org/10.18517/ijaseit.9.3.8692. Bhatt, G.D., Sinha, K., Deka, P.K., Kumar, A., 2014. Flood hazard and risk assessment in chamoli district, Uttarakhand using satellite remote sensing and GIS techniques. Int. J. Innov. Res. Sci. Eng. Technol. 03 (08), 15348e15356. https://doi.org/10.15680/ijirset.2014.0308039. Borah, S.B., Sivasankar, T., Ramya, M.N.S., Raju, P.L.N., 2018. Flood inundation mapping and monitoring in Kaziranga National Park, Assam using Sentinel-1 SAR data. Environ. Monit. Assess. 190 (9), 1e11. https://doi.org/10.1007/s10661-018-6893-y. Cao, H., Zhang, H., Wang, C., Zhang, B., 2019. Operational flood detection using Sentinel-1 SAR data over large areas. Water (Switzerland) 11 (4). https://doi.org/10.3390/w11040786. De, A., Kumar, D., Patel, P., 2018. Analysis of decomposition methods for classification of land-cover targets based on RISAT-1 hybrid SAR images. In: 2018 3rd Int. Conf. Microw. Photonics, ICMAP 2018, 2018. Icmap, pp. 1e2. https://doi.org/10.1109/ICMAP.2018.8354517. Dixit, S., Bera, S.K., 2012. Holocene climatic fluctuations from Lower Brahmaputra flood plain of Assam, northeast India. J. Earth Syst. Sci. 121 (1), 135e147. https://doi.org/10.1007/s12040-012-0150-5. Hassan, M.M., Ash, K., Abedin, J., Paul, B.K., Southworth, J., 2020. A quantitative framework for Analyzing spatial dynamics of flood events: a case study of super cyclone Amphan. Rem. Sens. 12 (20), 26. https://doi.org/10.3390/rs12203454. Jacinth Jennifer, J., Saravanan, S., Abijith, D., 2020. Integration of SAR and multi-spectral imagery in flood inundation mappingea case study on Kerala floods 2018. ISH J. Hydraul. Eng. 00 (00), 1e11. https:// doi.org/10.1080/09715010.2020.1791265. Jain, S.K., Saraf, A.K., Goswami, A., Ahmad, T., 2006. Flood inundation mapping using NOAA AVHRR data. Water Resour. Manag. 20 (6), 949e959. https://doi.org/10.1007/s11269-006-9016-4. Lardeux, C., Frison, P.L., Rudant, J.P., Souyris, J.C., Tison, C., Stoll, B., 2006. Use of the SVM classification with polarimetric SAR data for land use cartography. Int. Geosci. Remote Sens. Symp. 497e500. https://doi.org/10.1109/IGARSS.2006.131. Liang, J., Liu, D., 2020. A local thresholding approach to flood water delineation using Sentinel-1 SAR imagery. ISPRS J. Photogram. Remote Sens. 159, 53e62. https://doi.org/10.1016/ j.isprsjprs.2019.10.017.

Flood inundation mapping

Manjusree, P., Prasanna Kumar, L., Bhatt, C.M., Rao, G.S., Bhanumurthy, V., 2012. Optimization of threshold ranges for rapid flood inundation mapping by evaluating backscatter profiles of high incidence angle SAR images. Int. J. Disaster Risk Sci. 3 (2), 113e122. https://doi.org/10.1007/s13753-0120011-5. Matgen, P., Hostache, R., Schumann, G., Pfister, L., Hoffmann, L., Savenije, H.H.G., 2011. Towards an automated SAR-based flood monitoring system: lessons learned from two case studies. Phys. Chem. Earth 36 (7e8), 241e252. https://doi.org/10.1016/j.pce.2010.12.009. Mohapatra, R.D.S.P.K., 2003. Flood management in India. Curr. Sci. 120 (3), 455e456. https://doi.org/ 10.1177/0019556120120109. Nair, P.K., Babu, S.S., 2016. Spatial shrinkage of vembanad lake, south west India during 1973-2015 using NDWI and MNDWI. Int. J. Sci. Res. 5 (7), 1394e1401. Patel, P., Srivastava, H.S., Panigrahy, S., Parihar, J.S., 2006. Comparative evaluation of the sensitivity of multi-polarized multi-frequency SAR backscatter to plant density. Int. J. Rem. Sens. 27 (2), 293e305. https://doi.org/10.1080/01431160500214050. Rahman, M.R., Thakur, P.K., 2018. Detecting, mapping and analysing of flood water propagation using synthetic aperture radar (SAR) satellite data and GIS: a case study from the Kendrapara District of Orissa State of India. Egypt. J. Remote Sens. Sp. Sci. 21, S37eS41. https://doi.org/10.1016/ j.ejrs.2017.10.002. Schlaffer, S., Matgen, P., Hollaus, M., Wagner, W., 2015. Flood detection from multi-temporal SAR data using harmonic analysis and change detection. Int. J. Appl. Earth Obs. Geoinf. 38, 15e24. https:// doi.org/10.1016/j.jag.2014.12.001. Shen, X., Wang, D., Mao, K., Anagnostou, E., Hong, Y., 2019. Inundation extent mapping by synthetic aperture radar: a review. Rem. Sens. 11 (7), 1e17. https://doi.org/10.3390/RS11070879. Sivasankar, T., Kumar, D., Shanker Srivastava, H., Patel, P., 2020a. Wheat leaf area index retrieval using RISAT-1 hybrid polarized SAR data. Geocarto Int. 35 (8), 905e915. https://doi.org/10.1080/ 10106049.2019.1566404. Sivasankar, T., Borah, S.B., Das, R., Raju, P.L.N., 2020b. An investigation on sudden change in water quality of Brahmaputra river using remote sensing and GIS. Natl. Acad. Sci. Lett. 43 (7), 619e623. https:// doi.org/10.1007/s40009-020-00938-8. Srivastava, H.S., et al., 2015. Soil moisture variation over parts of Saharanpur and Haridwar districts, India. Int. J. Adv. Eng. Res. Sci. 2 (1), 31e39. Tiwari, V., et al., 2020. Flood inundation mapping-Kerala 2018; Harnessing the power of SAR, automatic threshold detection method and Google Earth Engine. PLoS One 15 (8 August), 1e17. https://doi.org/ 10.1371/journal.pone.0237324. Tripathi, P., 2015. “Flood disaster in India: an analysis of trend and preparedness,” interdiscip. J. Contemp. Res. 2 (4), 91e98 [Online]. Available: https://www.researchgate.net/profile/Prakash_Tripathi/ publication/292980782_Flood_Disaster_in_India_An_Analysis_of_trend_and_Preparedness/links/ 56b36ac208ae156bc5fb25bd.pdf. Uddin, K., Matin, M.A., Meyer, F.J., 2019. Operational flood mapping using multi-temporal Sentinel-1 SAR images: a case study from Bangladesh. Rem. Sens. 11 (13). https://doi.org/10.3390/rs11131581. Verma, P., Raghubanshi, A., Srivastava, P.K., Raghubanshi, A.S., 2020. Appraisal of kappa-based metrics and disagreement indices of accuracy assessment for parametric and nonparametric techniques used in LULC classification and change detection. Model. Earth Syst. Environ. 6 (2), 1045e1059. https:// doi.org/10.1007/s40808-020-00740-x. Yang, X., Zhao, S., Qin, X., Zhao, N., Liang, L., 2017. Mapping of urban surface water bodies from sentinel-2 MSI imagery at 10 m resolution via NDWI-based image sharpening. Rem. Sens. 9 (6), 1e19. https://doi.org/10.3390/rs9060596. Yekkehkhany, B., Safari, A., Homayouni, S., Hasanlou, M., 2014. A comparison study of different kernel functions for SVM-based classification of multi-temporal polarimetry SAR data. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. - ISPRS Arch. 40 (2W3), 281e285. https://doi.org/10.5194/ isprsarchives-XL-2-W3-281-2014. Zeng, Z., et al., 2020. Towards high resolution flood monitoring: an integrated methodology using passive microwave brightness temperatures and Sentinel synthetic aperture radar imagery. J. Hydrol. 582, 124377. https://doi.org/10.1016/j.jhydrol.2019.124377.

217

This page intentionally left blank

CHAPTER 12

Performance assessment of phased array type L-band Synthetic Aperture Radar and Landsat-8 used in image classification Swati Suman8, Prashant K. Srivastava1, George P. Petropoulos2, Ram Avtar3, Rajendra Prasad4, Sudhir Kumar Singh5, S.K. Mustak6, Ioannis N. Faraslis7 and Dileep Kumar Gupta4, 8 1

Institute of Environment and Sustainable Development, Banaras Hindu University, Varanasi, Uttar Pradesh, India; Department of Geography, Harokopio University of Athens, Kallithea, Athens, Greece; 3Institute for the Advanced Study of Sustainability, United Nations University (UNU-IAS), Shibuya City, Tokyo, Japan; 4Department of Physics, Indian Institute of Technology, Banaras Hindu University, Varanasi, Uttar Pradesh, India; 5KBCAOS, University of Allahabad, Allahabad, Uttar Pradesh, India; 6Department of Geography, Central University of Punjab, Bathinda, Punjab, India; 7Department of Environmental Sciences, University of Thessaly, Thessaly, Greece; 8Remote Sensing Laboratory, Institute of Environment and Sustainable Development, Banaras Hindu University, Varanasi, Uttar Pradesh, India

2

1. Introduction Land use and land cover have significant impacts on key aspects of the earth’s system functioning (Tsatsaris et al., 2021). They directly alter the availability of different biophysical resources including soil and water (Silva-Fuzzo et al., 2019; Markogianni et al., 2020) and biotic diversity (Singh et al., 2016) and have an impact on local and regional climates (Srivastava et al., 2019; Fragkou et al., 2020) and ecosystem services (Malhi et al., 2020). Land cover refers to the biophysical attributes of the earth’s surface (Pandey et al., 2019), whereas the manner in which humans employ land cover and its resources for their own use and interest is called land use (Srivastava et al., 2012). Information on prevailing land use practices and the prediction of changes in the near future could be crucial in selecting, planning, and implementing land use schemes for the proper and efficient use of natural resources. Remote sensing technologies provide a broad prospect of viewing Earth from space, which is crucial for visualizing and understanding the influence of anthropogenic activities on natural resources over time (Pandey et al., 2019; Anand et al., 2020). Owing to their large spatial and frequent temporal coverage, satellite images serve as a vital source of consistent and continuous data for atmospheric, ocean, and geographic studies on a variety of spatial and temporal scales. These images reduce a lot of fieldwork complexity, provide an easy way to estimate the extent of change with the help of different computer-based classifications, and change approaches to detection (Brown et al., 2018).

Radar Remote Sensing ISBN 978-0-12-823457-0, https://doi.org/10.1016/B978-0-12-823457-0.00002-1

© 2022 Elsevier Inc. All rights reserved.

219

220

Radar Remote Sensing

With advances in earth observation technologies and the launch of a number of satellites there is no scarcity of remotely sensed data; however, the classification of satellite images to obtain meaningful information efficiently remains a problem (Srivastava et al., 2012; Cass et al., 2019). Although satellite images obtained from different sensors have proven to be the most powerful tool for the earth’s surface and atmospheric monitoring, their limitations are one of the most critical drawbacks because no single sensor offers optimal spectral, spatial, and temporal resolutions at the same time (Simone et al., 2002; Tsatsaris et al., 2021). Each sensor has its own applications as well as limitations. The strength of reflection of incident radiation gives information about land surface conditions, such as in the distribution of plants, forests and farm fields, lake, and mountains. Thermal infrared images obtained from optical sensors sense the surface temperature of the object heated by sunlight. Because optical sensors work with shorter-wave radiation, they are susceptible to atmospheric scattering and depend on sunlight for work and can function only during daytime and in a cloud dust-free atmosphere. On the other hand, more advanced arrays of microwave sensors use longer-wavelength microwave radiation, which can penetrate through all types of atmospheric interference such as cloud cover haze, dust, storms, and heavy rainfall but have limited applications compared with visible and infrared satellite images. Useful information from satellite images acquired from these different types of spaceborne sensors are obtained after a series of manual and computer-based operations and applications called image classification (Pandey et al., 2019). Image classification is the process of categorizing the pixels of a raw image into clusters with similar spectral responses. It is broadly categorized as supervised or unsupervised based on the selection of training pixels for image classification (Dawson et al., 2019). Supervised image classification uses samples of known information classes (training sets) to classify pixels of unknown identity and covers techniques such as maximum likelihood classification (Srivastava et al., 2012; Kuching, 2007), support vector machines (SVMs) (Srivastava et al., 2012; Singh et al., 2014; Petropoulos et al., 2015; Lee et al., 2012), artificial neural networks (ANNs) (Srivastava et al., 2012; Petropoulos et al., 2010b; Egmont-Petersen et al., 2002), spectral angle mapper (Petropoulos et al., 2010b, 2015; Jollineau and Howarth, 2008), and decision tree classifier (Kuching, 2007; Otukei and Blaschke, 2010; Jiang et al., 2011). On the other hand, the unsupervised image classifier first examines unknown pixels of an image and provides natural groupings present in the image values. Examples of some famous unsupervised classification techniques include ISODATA (Melesse and Jordan, 2002), fuzzy sets, and K-means clusters. Supervised image classification techniques such as ANNs and SVMs are also termed as nonparametric classifiers because they do not assume that data for individual classes are distributed normally (Gupta and Srivastava, 2010; Pandey et al., 2019). ANNs are computational system based on biological neural networks, designed to emulate intelligent behavior to solve pattern recognition problems, whereas SVMs are supervised

Performance assessment of PALSAR and Landsat-8

training algorithm designed to find a hyperplane to separate datasets into discrete predefined classes consistent with training pixels (Mountrakis et al., 2011; Chatziantoniou et al., 2017). These two classifiers are the most efficient and extensively used machinebased algorithms in land use/land cover studies. The objective of this study is to appraise and compare the performances of supervised image classifier ANNs and all four kernelbased SVM classifiers (i.e., linear, radial basis function (RBF), polynomial, and sigmoid) on the acquired visible and infrared as well as microwave satellite images on Varanasi, recognized as the oldest continuously inhabited holy city of the world.

2. Datasets 2.1 Study area This study assessing the performance of different supervised image classification techniques on multispectral and microwave satellite images was carried out in the Varanasi, India study area. The city is located 83
30 E and 25
180 N in the eastern part of Uttar Pradesh, India, along the left crescent-shaped bank of the Ganges River (Fig. 12.1). Varanasi is regarded as the abode of Lord Shiva. It is considered the holiest spiritual capital of India for Hindus. Varanasi is a semiurban, densely populated (3,682,194, census 2011) region witnessing remarkable population growth and expansion owing to urbanization and increasing political significance, which has caused rapid alterations and modifications in land use/land cover of the region unlike any other developing city in India (Ramachandran, 1992). Agriculture and small-scale silk weaving industries are major sources of income for local residents. Most of the land is used to cultivate crops. Out of the 152.68 ha total

Figure 12.1 Study site: Varanasi, Uttar Pradesh, India.

221

222

Radar Remote Sensing

geographical area in Varanasi city, 95.748 is cultivable and is used for agriculture, 2.93 ha land is under nonagricultural uses, 2.56 ha is cultivable wasteland, and 0.02 ha land is considered permanent pasture. The rest (2.97 ha) is used for miscellaneous plantations for trees, crops, and groves, whereas 2.15 ha is left barren and uncultivable, followed by 46.30 ha left fallow. Varanasi is located in the middle of the Ganges plain region (IV) of the Vindhyan agroclimatic zone. It is of the humid subtropical climate type with dry summers, humid monsoons, and cold winters with a rain-fed agricultural system producing wheat, rice, pearl millet, pigeon peas, and maize as major crops. Varanasi has witnessed remarkable growth expansion and development, and thus a state-of-the-art land use pattern for the city was needed for the systematic development of the province and was the reason for this study. The sets of multispectral and microwave satellite images used in this study to get an overview of current land use practices in Varanasi were acquired from www.usgs.gov.in and www.eorc.jaxa.jp, respectively. 2.2 Datasets: Advanced Land Observing Satellite Phased Array type L-band Synthetic Aperture Radar The microwave imagery used to study the performance of selected image classifiers in this study was obtained from the Advanced Land Observing Satellite (ALOS-2) Phased Array type L-band Synthetic Aperture Radar (PALSAR) sensor. ALOS is an ambitious project of the Japanese national aerospace and space program. It consists of two series of satellites for atmospheric, marine, and land observations. ALOS contains three remote-sensing instrumentation systems: the Panchromatic Remote-sensing Instrument for Stereo Mapping for digital elevation mapping, the Advanced Visible and Near Infrared Radiometer type 2 for wide-range land observations, and PALSAR as an active image sensor for daytime and nighttime and all-weather land observations. Table 12.1 presents the technical details of ALOS while Table 12.2 shows the specification of PALSAR. 2.3 Datasets: Landsat-8 Landsat-8 is the most advanced addition to the 42-year-long Landsat mission started in 1972 with the aim of providing moderate-resolution, multispectral, global data of the earth’s surface. Landsat-8 was launched on February 11, 2013 from Vandenberg Air Force Base, California. Landsat-8 was designed to carry a two-sensor payload: the Operational Land Imager (OLI), built by the Ball Aerospace and Technologies Corporation, and the Thermal Infrared Sensor, built by the National Aeronautics and Space Administration Goddard Space Flight Center. The spacecraft with its two integrated sensors is referred as the Landsat-8 Observatory. Operations of Landsat-8 are similar to Landsat-7 except for the addition of two bands in the latest version. The OLI Coastal/Aerosol band (band 1; 0.435e0.451 mm) was designed principally for ocean color observations and the Cirrus band (band 9; 1.36e1.38 mm) detects thin clouds composed of ice crystals. Multispectral images of September 25, 2014 requiring the comparative study of the performance of a selected image classifier with a microwave image were downloaded from the US Geological Survey Landsat archive.

Performance assessment of PALSAR and Landsat-8

Table 12.1 Advanced Land Observing Satellite characteristics. Item

Characteristics

Orbit Equator pass time Altitude Inclination Recurrence cycle

Sun synchronous, subrecurrent w10.30 (descending.); w22.30 (ascending) 691.65 km 98.16 degrees 46 days 14 þ 27/46 revolutions/day; 671 revolutions/cycle 1 m (offline) 0.0002 deg (offline) (w2.5 m on ground) 0.0004 degrees/5 s Capacity: 96 Gb Data rate (maximum): 360 Mbps (recording) 240 Mbps (playback) 240 Mbps (via Data Relay Test Satellite (DRTS)) 120 Mbps (direct Ground Station (GS) downlink)

GPS orbital position accuracy Star tracker attitude determination accuracy Attitude stability High-speed Solid State Recorder (HSSR)

Data transmission: Ka-band antenna X-band antenna Solar array paddle Generated power Total weight

3 m  22 m, nine segments >7 kW at End of Line (EOL) 4000 kg

Table 12.2 Phased Array type L-band Synthetic Aperture Radar characteristics. Mode

Central frequency Chirp bandwidth Polarization Incident angle Range resolution Observation swath Bit length Data range

Scan synthetic aperture radar

Fine

Polarimetric (experimental mode)*1

1270 MHz (23.6 cm, L-band) 28 MHz

14 MHz

14 MHz, 28 MHz HH/HV

HH/VV 8e60 degrees 7e44 m

HH þ HV/ VV þ VH 8e60 degrees 18e88 m

40e70 km 5 bit 240 Mbps

14 MHz HH þ HV þ VH þ VV

18e43 degrees

8e30 degrees 24e89 m

40e70 km

100 m (multilook) 250e350 km

5 bit 240 Mbps

5 bit 120/240 Mbps

3 to 5 bit 240 Mbps

20e65 km

223

224

Radar Remote Sensing

3. Methodology 3.1 Image classification techniques A pixel-based supervised image classification (i.e., ANNs and four kernel-based SVM classifiers: linear, RBF, polynomial, and sigmoid) was implemented on multispectral and microwave satellite images for land use/land cover classification (Talukdar et al., 2020; Camargo et al., 2019; Zhang et al., 2020). Layer stacking and image subsetting were performed on both sets of images, which helped to get a small portion from a large image file that was convenient for a clearer view and fast processing time. In stage 1, the classification key (i.e., cover classes) was determined. At the second stage, representative training sites for each cover class were selected. Approximately 50e60 pixels per class (300 total pixels) were selected from homogeneous areas. The separability of selected training points for all cover classes was examined in Environment in Visualizing Images (version 5.2), which allows the computation of spectral separability between selected regions of interest for a given input file. At the third stage, the ANNs and all four kernel-based SVM classifiers were developed and implemented using the selected training site (Fig. 12.2). 3.2 Artificial neural networks ANNs are supervised image classifiers that emulate biological neural networks with a large network with a number of weighed connections between elements such as neurons, which are highly functional for solving pattern recognition, forecasting, data analysis, and data mining (Kadavi and Lee, 2018). Because they are a data-driven, self-adaptive technique with good efficiency in handling large datasets and a high computational rate, ANNs are the most widely and frequently used image classifier for land cover change studies (Novelli et al., 2016). Neural networks are typically a multilayered feed-forward model used for linear classification. Srivastava et al. (2012) explained the structural setup of ANNs as a three-layered structure in which each neuron in the input layer represents one input features, such as one satellite image band (Fig. 12.3). A hidden layer delineates functions to the input data whereas each neuron in the output layer corresponds to one of the land cover classes. When the model functions, the input layer fire to the hidden layers via a system of weighted connections and then link to an output layer where the answer is output as one of the classified land cover classes. The efficiency of neural networks is well-documented in the literature for the multispectral classification of Landsat images (Shahriar and Mountrakis, 2018; Lu and Weng, 2007). Because of the wide application of ANNs in remote sensing studies such as image compression, modeling, and pattern recognition, factors affecting the performance ANNs are of major concern for the selection of classifiers in land use change studies at the preliminary stage of development. To solve this problem, Foody et al. (1997) implemented ANNs for agricultural crop classification from an airborne thematic

Performance assessment of PALSAR and Landsat-8

Figure 12.2 Flowchart of methodology adopted in this study. ALOS, Advanced Land Observing Satellite; PALSAR, Phased Array type L-band Synthetic Aperture Radar.

mapper image and found that the accuracy of the classifier was affected by factors related to the dimensionality of remotely sensed data, the architecture of the neural network, and the characteristics of the training data and testing sites. In a similar study, to find best practices to increase the overall performance of ANN classification,

225

226

Radar Remote Sensing

Figure 12.3 Structural setup of artificial neural networks.

Kanellopoulos and Wilkinson (1997) outlined the significance of the nature of the input data and the use of a hybrid classifier in the performance of ANNs in classification. Major disadvantages of using ANNs as a solution include the black-box nature of these systems and the greater computational burden on the infrastructure available for analysis. 3.3 Support vector machines SVMs are statistical theoretical algorithms based on optimization technique widely used to solve data mining problems such as classification, regression, and feature selection (Lee et al., 2012). SVMs uses the kernel function during classification for the optimal separation of hyperplanes (OSH) (Han et al., 2007) (Fig. 12.4). This OSH is generated by solving an underlying optimization equation. SVMs can discriminate between complex data patterns by generating a highly nonlinear separating hyperplane that is implicitly defined by a nonlinear kernel map that has the best generalization ability for unseen data points based on statistical learning theory (Pal and Foody, 2010; Paoletti et al., 2020). Pal and Mather (2004) stated that the main objective of SVMs is to find OSH among all

Performance assessment of PALSAR and Landsat-8

Figure 12.4 Support vector machines operation schematic diagram.

possible hyperplanes. This is accomplished through an optimization problem using Lagrange multipliers and quadratic programming methods (Mathur and Foody, 2008a). Considering a training dataset with k number of samples (Eq. 12.1), a hyperplane can be linearly separated by two inequalities represented by vector w and scalar b (Eqs. 12.2 and 12.3): fxi  yi g; i ¼ 1.k

(12.1)

where x belongs to Rn and is an n-dimensional vector and y belongs to {e1, þ1}, representing the label of each class: w  xi þ b  1 for all y ¼ þ1

(12.2)

w  xi þ b  1 for all y ¼ 1

(12.3)

Also, the equation representing a constraint must be satisfied to achieve a hyperplane separating the two classes completely and linearly. It can be expressed as (Eq. 12.4): yi ðw  xi þ bÞ  1  0

(12.4)

In the case of two nonlinearly separable classes, a set of slack variables {xi ¼ 1} is introduced (Eq. (12.5):

227

228

Radar Remote Sensing

yi ðw  xi þ bÞ  1  xi  0 A function, C

k P i¼1

(12.5)

xi , is added to penalize solutions exhibiting a large value for xi. C is

used to control the degree of penalty in the optimization given in Eq. (12.6): k X   xi Min kwk2 = 2 þ c

(12.6)

i¼1

The linear, RBF, polynomial and sigmoid kernel functions tend to discover a superior separating margin for the OSH. The SVM kernels can be represented by Eqs. (12.7e12.10):   linear : K xi ; xj ¼ xTi xj (12.7) d    (12.8) Polynomial: K xi ; xj ¼ g xTi xj þ r ; g > 0     2  Radial basis function: K xi ; xj ¼ exp  g xi ; xj  ; g > 0 (12.9)     Sigmoid: K xi ; xj ¼ tan H g xTi xj þ r (12.10) where g is the width of the kernel function, d is the polynomial degree term, and r is the bias term in the kernel function. The SVM kernel type and kernel parameter used for classification affect the shape of the decision boundary, which influences the overall performance of the classifier (Huang et al., 2002; Kavzoglu and Colkesen, 2009) SVMs can produce accurate and robust classification results on a sound theoretical basis, even when input data are nonmonotone and nonlinearly separable (Foody and Mathur, 2004). Thus, they can help to evaluate more relevant information conveniently (Mathur and Foody, 2008b). Although SVMs do not deliver a parametric score function, the local linear approximation can offer important support for recognizing mechanisms linking different financial ratios with the final score of a company. For these reasons, SVMs are regarded as a useful tool for effectively complementing information gained from classical linear classification techniques. An appraisal of SVMs for land use/land cover studies was reported in the early work of Gaultieri and Cromp (1999) and Otukei and Blaschke (2010); Petropouolos et al. (2011), Petropouolos et al. (2010a), and Elatawneh et al. (2014). 3.4 Accuracy assessment 3.4.1 Kappa accuracy and coefficients To evaluate the performance of classifiers, overall accuracy (OA), user’s accuracy (UA), and producer’s accuracy (PA) error matrices were calculated and Cohen’s kappa coefficient (Kc) statistics were determined based on the computation of the error matrix

Performance assessment of PALSAR and Landsat-8

statistics (Congalton and Green, 2019). As a result, the OA, UA, PA, and Kc were computed as: OA ¼

r 1 X nii ; N i¼1

(12.11)

nii ; nicol nii ; UA ¼ nirow PA ¼

Kc ¼ N

r X i¼1

nii 

r X nicol nirow i¼1

N2

(12.12) (12.13) 

r X

nicol nirow ;

(12.14)

i¼1

where nii is the number of pixels correctly classified in a category; N is the total number of pixels in the confusion matrix; r is the number of rows; nicol are column reference data; and nirow are the row predicted classes, respectively. OA measures the proportion of the assessed area that is classified (Srivastava et al., 2012) and is expressed as the percent probability of a pixel of thematic map being classified in a set cover class. PA represents the correctly classified group classes by analyst and is expressed as an percentage. UA represents commission errors, whereas PA measurements are related to omission errors (Gupta and Srivastava, 2010). 3.4.2 Quantity disagreement and allocation disagreement Quantity disagreement and allocation disagreement for the comparative analysis of classification approaches were developed by Pontius and Millones (2011). They are was deployed here to compare the accuracies of classifiers implemented on both sets of multispectral and microwave images. Quantity disagreement quantifies variation in the classification of a biophysical unit by the classifier compared with the real land use/land cover distribution owing to misclassification in the proportion of all categories. Allocation disagreement measures inaccuracies of the classifier in the spatial allocation of biophysical units of the categories in the reference map and the comparison map. Total disagreement is the sum total of quantity disagreement and allocation disagreement.

4. Results and discussion Classified maps produced after the application of both ANNs and all four kernel-based SVM classifiers on multispectral and microwave sets of satellite images are listed in their respective sections and accuracies in Table 12.3. All classification methods produced comparable results in terms of describing not only the spatial distribution but also the cover density of each land cover class for the test site. A detailed discussion of the performances of various classifiers on the respective sets of images follows this section.Table 12.3

229

230

Radar Remote Sensing

Table 12.3 Overall Accuracy and Kappa Coefficient for Landsat-8 and ALOS.

Landsat-8 Image classifier

Artificial neural network SVM linear SVM radial basis function SVM polynomial SVM sigmoid

Advanced Land Observing Satellite Phased Array type L-band Synthetic Aperture Radar

Overall accuracy

Kappa coefficient

Overall accuracy

Kappa coefficient

0.973

0.967

0.923

0.904

0.970 0.960

0.963 0.950

0.923 0.943

0.904 0.929

0.960 0.946

0.950 0.933

0.933 0.913

0.917 0.892

SVM, support vector machines.

4.1 Land use/land cover using Landsat-8 satellite imagery A class separability analysis for selected cover classes was carried out using the transformed divergence (TD) technique for different combinations of Landsat-8 bands. TD is a measure of the statistical distance between classes, calculated from the means and covariance matrices of each pair of classes of interest. It provides information about their separability in the feature space, as recommended by Davis et al. (1978). A graph representing class separability between selected training pixels for five different cover classes for the bands 4 and 5 Landsat-8 image is shown in Fig. 12.5. The plot shows that pixels of all five cover classes have distinct spectral responses and are clustered accordingly; however, there is some intermixing of pixels between agriculture and plantation cover classes, which may be due to their similar spectral responses, leading to confusion and misclassification during the process. The near-infrared band (band 5) shows a wider range of distribution compared with the red band (band 4). Variation in reflectance in different cover classes is also represented in Fig. 12.6. Reflectance for water was reported to be least, while for sand it was highest, probably owing to its high albedo (Fig. 12.6). For Landsat-8 image classification, ANNs generally outperformed both OA and individual class accuracies (Fig. 12.7). ANN OA and Kc were recorded at 97.3% and 0.97, respectively, whereas SVM linear performed parallel to ANNs with OA and Kc of 97.0% and 0.96, respectively. SVM sigmoid performed least, with OA and kappa accuracy of 94% and 0.933, respectively. OA analysis indicated the highest accuracy of ANNs compared with all other classification techniques, whereas SVM sigmoid performed least well. For the built-up area class, all classifiers showed the procedure’s accuracy to be in the range 96e100% and

Performance assessment of PALSAR and Landsat-8

Figure 12.5 Class separability plot of training pixels between bands 4 and 5 for selected cover classes.

Figure 12.6 Variation in reflectance with cover class for bands 4 and 5. DN, digital number; NIR, nearinfrared.

the UA for the class was reported to be nearly 93% for all classifiers. For the water cover class, the PA was reported to be 100% by all classifiers and nearly 98% for UA. For the agriculture land cover class, PA was 95% by ANNs and SVM linear and near 91% for all other classifiers. UA for the same class was reported to be around 98% for all classifiers. Classification of plantation cover class ANNs showed higher accuracy in both PA and UA. For sand, 100% PA was noted by all classifiers, whereas UA nearly 94% (Table 12.4).

231

232

Radar Remote Sensing

Figure 12.7 Landsat-8 unclassified and classified images. ANN, artificial neural network; RBF, radial basis function; SVM, support vector machines.

Table 12.4 Overall accuracy for Landsat-8 image.

Classification technique

Artificial neural network SVM linear SVM radial basis function SVM polynomial SVM sigmoid

User’s accuracy (%) Sand

Water body

Agriculture

Plantation

Sand

95.16

100

100

100

95

98.33

93.33

98.28 98.21

95.16 93.65

100 96.55

98.33 96.67

100 100

95 91.67

98.33 98.33

93.33 93.33

98.36

98.21

93.65

96.55

96.67

100

91.67

98.33

93.33

98.36

93.22

93.65

94.92

90

100

91.67

98.33

93.33

Kappa coefficient

Water body

Agriculture

Plantation

97.33

0.967

93.75

100

98.28

97.00 96.00

0.963 0.950

93.65 93.55

98.36 98.36

96.00

0.950

93.55

94.67

0.933

93.1

SVM, support vector machines.

Producer’s accuracy (%)

Builtup area

Builtup area

Overall accuracy (%)

234

Radar Remote Sensing

4.2 Land use/land cover using Advanced Land Observing Satellite Phased Array type L-band Synthetic Aperture Radar image Similar to a Landsat-8 image, a class separability analysis based on backscatter values for combinations of both single and dual polarization of microwave images was performed (Fig. 12.8). Scatterplots of variations in backscatter values for all polarization combinations showed a similar trend of intermixing of agriculture and plantation pixels and a large range of distribution pixels assigned under the built-up cover class. Backscatter values for single polarization modes (HH and HV) showed negative trends, whereas for the dual polarization mode (HH-HV) it was a positive trend. The highest backscatter value for sand and lowest for the built-up class were reported in the analysis (Fig. 12.9).

Figure 12.8 Variations in backscatter with cover class for polarizations: (A) HH and HV; (B) HH and HH-HV; and (C) HV and HH-HV.

Performance assessment of PALSAR and Landsat-8

Figure 12.9 Variations in backscatter with cover classes.

For the ALOS PALSAR image for land cover classification, SVM RBF was the best classifier, with 94% accuracy efficiency. The uncalssified and classified ALOS PALSAR images using selected algorithms for this study over Varanasi is shown in Fig. 12.10. The least good performance was shown by SVM sigmoid. For the built-up area, PA was at 80% and UA was near 96%. For the water cover class, both PA and UA were 100%. For the plantation class, PA was 96%, and the highest performance was by ANNs at about 98%, whereas UA was recorded to be near 84%. For sand cover, PA was 100%, and it was 92% for UA. For the ALOS PALSAR microwave image, SVM RBF performance was best, with an OA of 94.3% and Kc of 0.92. SVM sigmoid performance for the PALSAR image was recorded to be lowest, at 91.3% for OA and 0.89 for Kc. For the ALOS PALSAR microwave image, SVM RBF performance was the best, with an OA of 94.3% and Kc of 0.92. SVM sigmoid performance for the PALSAR image was recorded to be lowest, with 91.3% for OA and 0.89 for Kc (Table 12.5 and Fig. 12.10). 4.3 Diagnostic evaluation of image classification performance and applicability Based on the results of Producer’s Accuracy (PA), the agriculture and sand cover classes produced the lowest percentage for almost all classifiers for Landsat-8 images while agricultural and built-up classes for ALOS PALSAR which can be attributed to be due to similar spectral characteristic of the two cover classes. For the Landsat image, UA for the built-up area was measured to be almost 93% for all classifiers; for the water cover UA it ranged from 98% to 100%. For the agriculture cover class, UA was reported to be 93%e98%, and 93%e95% for the plantation class. For sand, the UA value lay within the 94%e100% range. For the microwave ALOS PALSAR image, the UA for the builtup class was 96%e100%, and for water, it was recorded at 100% for all cover classes. For

235

236

Radar Remote Sensing

Figure 12.10 Advanced Land Observing Satellite (ALOS) Phased Array type L-band Synthetic Aperture Radar (PALSAR) unclassified and classified images. ANNs, artificial neural networks; RBF, radial basis function; SVM, support vector machines.

Table 12.5 Overall accuracy for Advanced Land Observing Satellite Phased Array type L-band Synthetic Aperture Radar image. User’s accuracy (%) Classification technique

artificial neural network SVM linear SVM radial basis function SVM polynomial SVM sigmoid

Sand

Builtup area

Water body

Agriculture

Plantation

Sand

84.29

89.55

80

100

83.33

98.33

100

88.89 94.34

84.06 84.06

93.75 95.24

85 91.67

100 100

80 83.33

96.67 96.67

100 100

100

92.31

84.06

95.24

90

100

80

96.67

100

100

87.27

82.86

92.3

80

100

80

96.67

100

Overall accuracy (%)

Kappa coefficient

Builtup area

Water body

Agriculture

Plantation

92.33

0.904

100

100

90.91

92.33 94.33

0.904 0.929

96.23 100

100 100

93.33

0.917

96.43

91.33

0.892

96

SVM, support vector machines.

Producer’s accuracy (%)

238

Radar Remote Sensing

agriculture, the UA was 87%e94%, and within 82%e84% for the plantation class. For sand, the UA ranged between 89% and 95%. Fig. 12.11 compares the performance of the Landsat-8 and ALOS PALSAR images. Overall classifications for all classifiers used in the study were comparable and better than results obtained in previous studies reported by various authors. Landsat multispectral images have long been used for land use/land cover change studies (Franklin et al., 1996; Kadavi and Lee, 2018). Srivastava et al. (2012) applied ANNs and all four kernel-based SVMs classifiers on optical/infrared imagery of Landsat to compare classifier performance. They reported ANNs (OA 84.9% and Kc 0.74) to be the best classifier and SVM linear to have comparable performance (OA 84.8% and Kc 0.74). On the other hand, SVM sigmoid performed the least well (OA 84.7% and Kc 0.74). Results obtained in this study showed similar accuracy but better performance for the classifiers compared with those reported. In a similar study on ALOS PALSAR microwave images, Zhang et al. (2009) applied SVMs. However, the author stated that despite the availability of microwave images under all environmental and atmospheric conditions, the large commission error in the image classification procedure using microwave images primarily results from similar backscatter amplitudes within selected cover classes. Tsuchida et al. (2010) also outlined single-band radar images of ALOS PALSAR images as the major drawback of the narrow-range application of microwave images for image classification studies. Unlike these, Walker et al. (2010), in a study to evaluate the suitability of ALOS PALSAR images for large area classification in a Brazilian Amazon site involving Landsat-8 datasets, advocated generating accurate map-based estimates of forest cover by ALOS PALSAR sensors to obtain comparable accuracies for both datasets. They endorsed the application

Figure 12.11 Comparative performances of Landsat-8 and Advanced Land Observing Satellite (ALOS) Phased Array type L-band Synthetic Aperture Radar (PALSAR) overall accuracy (OA) and kappa coefficient (Kc). ANN, artificial neural network; RBF, radial basis function; SVM, support vector machines.

Performance assessment of PALSAR and Landsat-8

of microwave images for land use/land cover studies. In an effort to explore the applicability of microwave images to environmental studies, Shimada et al. (2014) conducted a study using ALOS PALSAR microwave images to map landslide activity in a north California region in the United States. They found a unique mechanism for identifying active sites and developed an early warning system for natural hazards such as landslides, demonstrating the effectiveness of microwave images in hazard prediction and management studies. However, few studies report the classification performance of ALOS PALSAR in an Indian context or with respect to comparing sensors. 4.4 Comparative analysis Kappa and total disagreement were calculated. The classifiers showed similar trends in performance, as reported in the previous section with OA and kappa. In this study, for Landsat images, ANNs performed best in mapping five different classes for the selected study site, whereas for the ALOS PALSAR microwave image, SVM RBF performed best. This represents the strongest agreement between the training and test pixels. Table 12.6 lists the results of kappa, kappa total disagreement (%), allocation disagreement (%), and total disagreement (%) scores for Landsat-8 and ALOS PALSAR images. In a previous study examining the suitability of five-band RapidEye satellite data for tree species classification in a mopane woodland of Botswana using machine learning algorithms, Adelabu et al. (2013) used quantity disagreement and allocation disagreement as a measure of accuracy for cover classification by implementing algorithms and reported results comparable to those obtained in this study.

5. Conclusions and future work A comparative performance analysis of different machine-based supervised image classifiers on multispectral and microwave satellite images for the Varanasi, India, study site was the main objective of this study. The study explored the possibilities of using longerwavelength microwave satellite images over easily available and customarily used multispectral Landsat images for land cover classification. Our results showed the better performance of all classifiers compared with previously reported studies for Landsat images and the comparable performance of identical set classifiers when they were applied on microwave ALOS PALSAR images. The comparatively higher classification accuracy of the visible or infrared image was mainly attributed to the multiband Landsat image and single-band image as a major drawback to the lower performance of the ALOS PALSAR image. However, the availability of microwave satellite images under all environmental and atmospheric conditions shows the wide-range implications of microwave images in future studies revealing unspecified earth system phenomena, natural resource management, and land use/land cover classification studies. This study could serve as the basis for future studies exploring the possibility of using long-range microwave imagery to

239

240 Radar Remote Sensing

Table 12.6 Quantity disagreement and allocation disagreement for Landsat-8 and Advanced Land Observing Satellite Phased Array type L-band Synthetic Aperture Radar images. Advanced Land Observing Satellite Phased Array type L-band Synthetic Aperture Radar

Landsat-8

Kappa Kappa total disagreement (%) Quantity disagreement (%) Allocation disagreement (%) Total disagreement

ANN

SVM linear

SVM RBF

SVM polynomial

SVM sigmoid

ANN

SVM linear

SVM RBF

SVM polynomial

SVM sigmoid

0.973 2.739

0.969 3.093

0.958 4.166

0.958 4.166

0.944 5.633

0.917 8.303

0.917 8.303

0.929 7.14

0.919 8.007067

0.909 9.090

1.333

1.5

2

2

2.66

3.833

3.833

3.333

1.833

4.166

0.404

0.135

0.131

0.132

0.07

0.498

0.482

0.382

0.08

0.548

1.737

1.635

2.131

2.132

2.737

4.331

4.315

3.715

1.913

4.714

ANN, artificial neural network; RBF, radial basis function; SVM, support vector machines.

Performance assessment of PALSAR and Landsat-8

answer questions related to extreme weather events, land use change, and many more such studies. Our results confirmed the reliable performance of employed classifiers to extract land use patterns for both sets of images as a vital source of information for managing an unplanned and rapidly expanding classic Indian city such as Varanasi. For the Landsat-8 image, the performance of ANNs was reported to be best, whereas for ALOS PALSAR, SVM RBF attained the highest OA. The findings of this study are believed to provide useful information to planners, managers, and researchers to formulate and implement effective policies and strategies for the better and more efficient use and monitoring of our natural resources for sustainable development.

Acknowledgments The first author is highly thankful to the University Grant Commission for providing the fellowship for the research work. The authors would like to thank Banaras Hindu University for providing the seed grant for this research.

References Adelabu, S., Mutanga, O., Adam, E., Cho, M.A., 2013. Exploiting machine learning algorithms for tree species classification in a semiarid woodland using RapidEye image. J. Appl. Remote Sens. 7, 073480-073480. Anand, A., Pandey, C.P., Petropoulos, P.G., Pavlides, A., Srivastava, K.P., Sharma, K.J., Malhi, K.M.R., 2020. Use of hyperion for mangrove forest carbon stock assessment in Bhitarkanika forest reserve: a contribution towards blue carbon initiative. Remote Sens. MDPI 12, 597. https://doi.org/10.3390/ rs12040597. Brown, R.A., Petropoulos, P.G., Ferentinos, K., 2018. Appraisal of the Sentinel-1 & 2 use in a large-scale wildfire assessment: a case study from Portugal’s fires of 2017. Appl. Geogr. 100, 78e89. https:// doi.org/10.1016/j.apgeog.2018.10.004. Cass, A., Petropoulos, P.G., Ferentinos, P.K., Pavlides, A., Srivastava, K.P., 2019. Exploring the synergy between Landsat and ASAR towards improving thematic mapping accuracy of optical EO data. Appl. Geomat. 1e12. https://doi.org/10.1007/s12518-019-00258-7. Camargo, F.F., Sano, E.E., Almeida, C.M., Mura, J.C., Almeida, T.A., 2019. Comparative assessment of machine-learning techniques for land use and land cover classification of the Brazilian tropical Savanna using ALOS-2/PALSAR-2 polarimetric images. Rem. Sens. 11, 1600. Chatziantoniou, A., Petropoulos, P.G., Psomiadis, E., 2017. Co-orbital Sentinel 1 and 2 for LULC mapping with emphasis on wetlands in a Mediterranean setting based on machine learning. Rem. Sens. 9, 1e18. https://doi.org/10.1080/10106049.2017.1307460. Congalton, R.G., Green, G., 2019. Assessing the Accuracy of Remotely Sensed Data: Principles and Practices, third ed. CRC Press, Boca Raton. Davis, S.M., Landgrebe, D.A., Phillips, T.L., Swain, P.H., Hoffer, R.M., Lindenlaub, J.C., Silva, L.F., 1978. Remote Sensing: The Quantitative Approach. McGraw-Hill International Book Co, New York, p. 405. Dawson, R., Petropoulos, P.G., Toulios, L., Srivastava, K.P., 2019. Mapping and Monitoring of the Land Use/Cover Changes in the Wider Area of Ltanos, Crete, Using Very High Resolution EO Imagery with Specific Interest in Archaeological Sites. Environment, Development and Sustainability. https:// doi.org/10.1007/s10668-019-00353-0 (in press). Egmont-Petersen, M., de Ridder, D., Handels, H., 2002. Image processing with neural networksda review. Pattern Recog. 35, 2279e2301.

241

242

Radar Remote Sensing

Elatawneh, A., Kalaitzidis, C., Petropoulos, P.G., Schneider, T., 2014. Evaluation of diverse classification approaches for land use/cover mapping in a Mediterranean region utilizing Hyperion data. Int. J. Digital Earth 7, 194e216. Foody, G., Arora, M., 1997. An evaluation of some factors affecting the accuracy of classification by an artificial neural network. Int. J. Rem. Sens. 18, 799e810. Foody, G.M., Mathur, A., 2004. Toward intelligent training of supervised image classifications: directing training data acquisition for SVM classification. Remote Sen. Environ. 93, 107e117. Fragou, S., Kalogeropoulos, K., Stathopoulos, N., Louka, P., Srivastava, K.P., Karpouzas, S., Kalivas, P.D., Petropoulos, P.G., 2020. Quantifying land cover changes in a Mediterranean environment using landsat TM and support vector machines. Forests MDPI 11, 750e769. https://doi.org/10.3390/f11070750. Franklin, S., Wulder, M., Lavigne, M., 1996. Automated derivation of geographic window sizes for use in remote sensing digital image texture analysis. Comput. Geosci. 22, 665e673. Gualtieri, J.A., Cromp, R.F., 1999. Support vector machines for hyperspectral remote sensing classification. In: The 27th AIPR Workshop: Advances in Computer-Assisted Recognition, 1999. International Society for Optics and Photonics, pp. 221e232. Gupta, M., Srivastava, K.P., 2010. Integrating GIS and remote sensing for identification of groundwater potential zones in the hilly terrain of Pavagarh, Gujarat, India. Water Int. 35, 233e245. Han, D., Chan, L., Zhu, N., 2007. Flood forecasting using support vector machines. J. Hydroinf. 9, 267e276. Huang, C., Davis, L., Townshend, J., 2002. An assessment of support vector machines for land cover classification. Int. J. Rem. Sens. 23, 725e749. Jiang, X., Lin, M., Zhao, J., 2011. Woodland cover change assessment using decision trees, support vector machines and artificial neural networks classification algorithms. In: Intelligent Computation Technology and Automation (ICICTA), International Conference on, 2011. IEEE, pp. 312e315. Jollineau, M., Howarth, P., 2008. Mapping an inland wetland complex using hyperspectral imagery. Int. J. Rem. Sens. 29, 3609e3631. Kadavi, P.R., Lee, C.W., 2018. Land cover classification analysis of volcanic island in Aleutian Arc using an artificial neural network (ANN) and a support vector machine (SVM) from Landsat imagery. Geosci. J. 22 (4), 653e665. https://doi.org/10.1007/s12303-018-0023-2. Kanellopoulos, I., Wilkinson, G., 1997. Strategies and best practice for neural network image classification. Int. J. Rem. Sens. 18, 711e725. Kavzoglu, T., Colkesen, I., 2009. A kernel functions analysis for support vector machines for land cover classification. Int. J. Appl. Earth Obs. Geoinf. 11, 352e359. Kuching, S., 2007. The performance of maximum likelihood, spectral angle mapper, neural network and decision tree classifiers in hyperspectral image analysis. J. Comput. Sci. 3, 419e423. Lee, Y.-J., Yeh, Y.-R., Pao, H.-K., 2012. Introduction to support vector machines and their applications in bankruptcy prognosis. In: Handbook of Computational Finance. Springer, pp. 731e761. Lu, D., Weng, Q., 2007. A survey of image classification methods and techniques for improving classification performance. Int. J. Rem. Sens. 28 (5), 823e870. https://doi.org/10.1080/01431160600746456. Malhi, R.K., Anand, A.M., Srivastava, K.P., Sandhya Kiran, G., Petropoulos, P.G., Chalkias, C., 2020. An integrated spatiotemporal pattern analysis model to assess and predict the degradation of protected forest areas. ISPRS Int. J MDPI 9, 530e550. https://doi.org/10.3390/ijgi9090530. Markogianni, V., KalivasD, Petropoulos, P.G., Dimitriou, E., 2020. Estimating chlorophyll-a of inland water bodies in Greece based on landsat data. Remote Sens. MDPI 12, 2087e2109. https://doi.org/ 10.3390/rs12132087. Mathur, A., Foody, G., 2008a. Multiclass and binary SVM classification: implications for training and classification users. Geosci. Remote Sens. Lett. IEEE 5, 241e245. Mathur, A., Foody, G.M., 2008b. Crop classification by support vector machine with intelligently selected training data for an operational application. Int. J. Rem. Sens. 29, 2227e2240. Melesse, A.M., Jordan, J.D., 2002. A comparison of fuzzy vs. augmented-ISODATA classification algorithms for cloud-shadow discrimination from Landsat images. Photogramm. Eng. Rem. Sens. 68, 905e912.

Performance assessment of PALSAR and Landsat-8

Mountrakis, G., Im, J., Ogole, C., 2011. Support vector machines in remote sensing: a review ISPRS. J. Potogram. Remote Sens. 66, 247e259. Novelli, A., Tarantino, E., Caradonna, G., Apollonio, C., Balacco, G., Piccinni, F., 2016. Improving the ANN classification accuracy of landsat data through spectral indices and linear transformations (PCA and TCT) aimed at LU/LC monitoring of a River basin. In: Gervasi, O., et al. (Eds.), Computational Science and its Applications e ICCSA 2016. Lecture Notes in Computer Science, vol. 9787. Springer, Cham. https://doi.org/10.1007/978-3-319-42108-7_32. Otukei, J., Blaschke, T., 2010. Land cover change assessment using decision trees, support vector machines and maximum likelihood classification algorithms. Int. J. Appl. Earth Obs. Geoinf. 12, S27eS31. Pal, M., Foody, G.M., 2010. Feature selection for classification of hyperspectral data by SVM geoscience and remote sensing. IEEE Trans. 48, 2297e2307. Pal, M., Mather, P.M., 2004. Assessment of the effectiveness of support vector machines for hyperspectral data Future Generation. Comput. Syst. 20, 1215e1225. Pandey, P.C., Koutsias, N., Petropoulos, P.G., Srivastava, K.P., Dor, B.E., 2019. Land Use/Land Cover in View of Earth Observation: Data Sources, Input Dimensions and Classifiers -a Review of the State of the Art. Geocarto International (in press). Paoletti, M.E., Haut, J.M., Tao, X., Miguel, J.P., Plaza, A., 2020. A new GPU implementation of support vector machines for fast hyperspectral image classification. Rem. Sens. 12 (8), 1257. https://doi.org/ 10.3390/rs12081257. Petropoulos, P.G., Kontoes, C., Keramitsoglou, I., 2010a. Burnt area delineation from a uni-temporal perspective based on Landsat TM and ALI satellite radiometers: results from the analysis of a fire event in Greece. In: EGU General Assembly Conference Abstracts, p. 926. Petropoulos, P.G., Kalivas, P.D., Georgopoulou, I.A., Srivastava, K.P., 2015. Urban vegetation cover extraction from hyperspectral imagery and geographic information system spatial analysis techniques: case of Athens, Greece. J. Appl. Remote Sens. 9, 096088-096088. Petropoulos, P.G., Kontoes, C., Keramitsoglou, I., 2011. Burnt area delineation from a uni-temporal perspective based on Landsat TM imagery classification using support vector machines. Int. J. Appl. Earth Obs. Geoinf. 13, 70e80. Petropoulos, P.G., Vadrevu, P.K., Xanthopoulos, G., Karantounias, G., Scholze, M., 2010b. A comparison of spectral angle mapper and artificial neural network classifiers combined with Landsat TM imagery analysis for obtaining burnt area mapping. Sensors 10, 1967e1985. Pontius Jr., R.G., Millones, M., 2011. Death to Kappa: birth of quantity disagreement and allocation disagreement for accuracy assessment. Int. J. Rem. Sens. 32, 4407e4429. Ramachandran, R., 1992. Urbanization and Urban Systems in India OUP Catalogue. Shahriar, S.H., Mountrakis, G., 2018. Effect of classifier selection, reference sample size, reference class distribution and scene heterogeneity in per-pixel classification accuracy using 26 Landsat sites. Remote Sen. Environ. 204, 648e658. https://doi.org/10.1016/j.rse.2017.09.035. ISSN 0034-4257. Shimada, M., Itoh, T., Motooka, T., Watanabe, M., Shiraishi, T., Thapa, R., Lucas, R., 2014. New global forest/non-forest maps from ALOS PALSAR data (2007e2010). Rem. Sens. Environ. 155, 13e31. Silva-Fuzzo, D., Carlson, N.T., Kourgialas, N., Petropoulos, P.G., 2019. Coupling Remote Sensing with a water balance model for soybean yield predictions over large areas. Earth Sci. India. https://doi.org/ 10.1007/s12145-019-00424-w. Simone, G., Farina, A., Morabito, F.C., Serpico, S.B., Bruzzone, L., 2002. Image fusion techniques for remote sensing applications. Inf. Fusion 3, 3e15. Singh, S.K., Srivastava, P.K., Gupta, M., Thakur, J.K., Mukherjee, S., 2014. Appraisal of land use/land cover of mangrove forest ecosystem using support vector machine. Environ. Earth Sci. 71, 2245e2255. Singh, S.K., Srivastava, K.P., Szabo, S., Petropoulos, P.G., Gupta, M., Islam, T., 2016. Landscape transform and spatial metrics for mapping spatiotemporal land cover dynamics using earth observation datasets. Geocarto Int. pp1e16. https://doi.org/10.1080/10106049.2015.1130084. Srivastava, K.P., Han, D., Rico-Ramirez, M.A., Bray, M., Islam, T., 2012. Selection of classification techniques for land use/land cover change investigation. Adv. Space Res. 50, 1250e1265.

243

244

Radar Remote Sensing

Srivastava, K.P., Petropoulos, P.G., Gupta, M., Singh, S.K., Islam, T., Loka, D., 2019. Deriving Forest Fire Probability Maps from the Fusion of Visible/Infrared Satellite Data and Geospatial Data Mining. Modeling Earth Systems and Environment. https://doi.org/10.1007/s40808-018-0555-5 (in press). Talukdar, S., Singha, P., Mahato, S., Shahfahad, P.S., Liou, Y.-A., Rahman, A., 2020. Land-use land-cover classification by machine learning classifiers for satellite observationsda review. Rem. Sens. 12 (7), 1135. https://doi.org/10.3390/rs12071135. Tsatsaris, A., Kalogeropoulos, K., Stathopoulos, N., Louka, P., Tsanakas, K., Tsesmelis, D.E., Krassanakis, V., Petropoulos, P.G., Pappas, V., Chalkias, C., 2021. Geoinformation technologies in support of environmental hazards monitoring under climate change: an extensive review. ISPRS Int. J. Geo-Inf. 10 (94), 1e32. https://doi.org/10.3390/ijgi10020094. Tsuchida, Y., Omatu, S., Yoshioka, M., 2010. Land cover estimation with ALOS satellite image using a neural-network. Artif. Life Robot. 15, 37e40. Walker, W.S., Stickler, C.M., Kellndorfer, J.M., Kirsch, K.M., Nepstad, D.C., 2010. Large-area classification and mapping of forest and land cover in the Brazilian Amazon: a comparative analysis of ALOS/ PALSAR and Landsat data sources. Selected Topics in Applied Earth Observations and Remote Sensing. IEEE J. 3, 594e604. Zhang, Y., Wang, C., Wu, J., Qi, J., Salas, W.A., 2009. Mapping paddy rice with multitemporal ALOS/ PALSAR imagery in southeast China. Int. J. Rem. Sens. 30, 6301e6315. Zhang, R., Tang, X., You, S., Duan, K., Xiang, H., Luo, H., 2020. A novel feature-level fusion framework using optical and SAR remote sensing images for land use/land cover (LULC) classification in cloudy mountainous area. Appl. Sci. 10 (8), 2928. https://doi.org/10.3390/app10082928.

CHAPTER 13

Evaluation of speckle filtering methods using polarimetric Sentinel-1A data Varun Narayan Mishra1 and Thota Sivasankar2 1

Centre for Climate Change and Water Research, Suresh Gyan Vihar University, Jaipur, Rajasthan, India; 2Geographic Information Systems (GIS) Area, NIIT University, Neemrana, Rajasthan, India

1. Introduction Remotely sensed datasets acquired from different satellite sensors have various potential for mapping projects, scientific applications, and policy decisions (Kumar et al., 2015, 2018; Inglada et al., 2016; Mishra and Rai, 2016; Mishra et al., 2019). Synthetic aperture radar (SAR) is a kind of microwave remote sensing imaging system used for earth observation around the clock in all weather and illumination conditions (Ouchi, 2013; Inglada et al., 2016). Thus, data gathered by SAR is of critical significance in mapping, monitoring, and managing natural resources at different observational scales (Mishra et al. 2017a,b; De Alban et al., 2018; Winsvold et al., 2018; Kumar et al., 2019; Haarpaintner and Hindberg, 2019). The quantity and performance of datasets obtained from the multidimensional SAR systems has greatly improved over the years (Medasani and Reddy, 2018; Wei et al., 2019). Polarimetric SAR (PolSAR) systems with unique features regarding polarization options emerged as an essential system configuration (Lee and Pottier, 2009; Cloude, 2009). PolSAR information is composed of geometries and characteristics of the scatterers of the acquired images (Medasani and Reddy, 2018). PolSAR data have both amplitude and phase information that can be used to extract information and are reported in many studies (Freeman, 2007; Turkar et al., 2012). Furthermore, SAR permits the polarimetric investigation of electromagnetic wave, which in turn helps to distinguish scattering mechanisms. SAR data are generated by a coherent imaging system that is affected by interference to the signal. Thus, despite several merits, SAR images are inherently affected by speckles caused by random interference of the backscattered electromagnetic waves because of the surface roughness in an image (Goodman, 1976). These images incorporate granular noise known as speckle noise. The structure of the imaged surface and several imaging parameters account for the appearance of speckles in an image (Bonny et al., 2019). Speckles are generally regarded to be multiplicative noise, but they have important information about the area under examination (Dellepiane and Angiati, 2014). The appearance of speckle increases with the gray level of a local area owing to its

Radar Remote Sensing ISBN 978-0-12-823457-0, https://doi.org/10.1016/B978-0-12-823457-0.00006-9

© 2022 Elsevier Inc. All rights reserved.

245

246

Radar Remote Sensing

multiplicative nature (Argenti et al., 2013; Bonny et al., 2019). Therefore, within the context of SAR image analysis, the presence of speckle noise not only degrades the quality of an image, it makes it difficult to interpret the image (Liu et al., 2017; Bonny et al., 2019). The presence of speckles also reduces both the spatial and radiometric resolution of the image, and thus the identification of target features (Bonny et al., 2019). Therefore, it is desirable to reduce speckles in PolSAR image processing (Pang et al., 2013; Liu et al., 2017). Because of their complexity speckled data must be treated using statistical analysis (Kupidura, 2016). Several speckle reduction filters have been developed, including a median filter (Pratt, 1975), Lee’s filter (Lee, 1980), Frost’s filter (Frost et al., 1981), Kuan’s filter (Kuan et al., 1987), a diffusion filter (Perona and Malik, 1990), and a wavelet filter (Donoho et al., 1995). These filters were derived by assuming that the speckle is a multiplicative noise random variable with a mean of 1. Subsequently, numerous developments in these filters were performed over the years for different applications (Bonny et al., 2019). All major commercial and open-source image processing systems include many filters for speckle reduction in SAR images. Speckle filtering methods can efficiently reduce speckle noise. However, these methods smear edges and blur images to some extent. It is not easy to achieve smoothing uniform regions while preserving and/or enhancing edges. Because of the availability of numerous filters, it is essential to recognize and choose the best filter for a specific application. This work carried out a performance evaluation of selected speckle filtering methods on dual-polarization SAR images. The study explored available filters in the Sentinels Application Platform (SNAP) toolbox and suggested filters with high efficiency in speckle suppression. The results of this work will be helpful for the quantitative evaluation of despeckled SAR images that may be used for further analysis and terrain classification.

2. Study site and data used The study site for this work was the Varanasi district, which is a part of eastern India. It is one of the oldest inhabited places in the world and is located along the left crescentshaped bank of the holy river Ganges. It is considered to be the cultural and religious capital of India. This region is productive and rich in agriculture because of its place in the middle of the Ganges plain. It has center latitude of 25
170 38.645800 N and longitude of 82
590 57.385300 E. The geographical location of the study site as viewed on a composite image of Vertical-Vertical (VV), Vertical-Horizontal (VH) and data on the difference of the two polarizations (VVeVH) are shown in Fig. 13.1. A Sentinel-1A satellite carrying a PolSAR sensor was launched by the European Space Agency (ESA) in Apr. 2014. It provides C-band images at dual-polarization within

Evaluation of speckle filtering methods using polarimetric Sentinel-1A data

Figure 13.1 Location map of study site as viewed on the Sentinel 1A composite image. Red indicates VV; green, VH; and blue, VVeVH.

12 days of a repeat cycle. It works in different acquisition modes: strip map, interferometric wide swath, extrawide swath, and wave, with different processing levels. In this study, the Sentinel-1A PolSAR dataset at C-band covering the study area was used. Specifications of SAR data used in this study are given in Table 13.1.

Table 13.1 Specifications of synthetic aperture radar data. Satellite/ sensor

Date of acquisition

Sentinel-1A

Apr. 29, 2015

Major characteristics

Data source

Spatial resolution: 5  20 m Band: C (5.40 GHz) Level: L1-GRDH Mode: interferometric wide swath Polarizations: VV and VH Incidence angle: 29.1
e46.0
Unsigned 16-bit Swath: 250 km

European Space Agency

247

248

Radar Remote Sensing

3. Methodology SNAP software (version 3.0.0) with the Sentinel-1 toolbox (S-1TBX) is a user-friendly and GUI-based open-source toolbox for SAR image processing and analysis. It is downloaded from the ESA website (http://step.esa.int/main/download/) and used to preprocess Sentinel-1A images. The radiometric calibration was performed followed by geometric correction using the ellipsoid correction method. 3.1 Speckle filtering methods This study aimed to select the best filtering method available in SNAP software for speckle reduction before calculating backscattering coefficients. In this study, a nonadaptive boxcar filter (Goodman, 1963), a median filter, and four adaptive filters (Frost [at damping factor 2], Gamma map, Lee, and refined Lee) were used on C-band dual-polarization Sentinel 1A data. 3.2 Evaluation of speckle filters Several indices were used to evaluate the speckle filters quantitatively (Argenti et al., 2013; Di Martino et al., 2014). These indices include the standard deviation to mean ratio (SD/M), relative standard deviation (RSD), speckle suppression index (SSI), and equivalent number of looks (ENL). These indices were used to assess the ability to suppress speckles present in PolSAR images. The lower the SD/M values, the more superior the performance is of the speckle filter (Lee and Pottier, 2009). 3.2.1 Relative standard deviation RSD is the measure of competence for a filter when the image has minimal variation in the mean. Reduction of RSD is a good measure of the filter’s efficiency in speckle suppression: RSD ¼

Variance Mean

(13.1)

3.2.2 Speckle suppression index This index tends to be less than 1 if the performance of the filter is efficient in suppressing speckle noise (Sheng and Xia, 1996). A lower value of SSI shows better performance of the speckle filtering algorithm. SSI is based on: qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi  ffi Variance If Mean ðIo Þ    pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ffi SSI ¼ (13.2) Mean If Variance ðIo Þ where If ¼ filtered image and. Io ¼ noisy image.

Evaluation of speckle filtering methods using polarimetric Sentinel-1A data

3.2.3 Equivalent number of looks The ENL is used to scale the level of speckles in SAR data over a uniform area. Its value depends on the size of the tested area. A higher ENL value reflects the better performance of a filter quantitatively in smoothing speckle noise. This index can be calculated using the equation (Gagnon and Jouan, 1997):  2 Mean ENL ¼ (13.3) Standard Deviation

4. Results and discussion The effectiveness of the filters to reduce speckles was assessed quantitatively based on several indices. Different window sizes of 3  3, 5  5, 7  7, 9  9, and 11  11 were tested on each speckle filter to compare the optimal filtering results. The obtained values of SD/M, RSD, SSI, and ENL were used to evaluate the performance of various filters applied on both VV and VH polarization images (Tables 13.2 and 13.3). For a nonadaptive boxcar filter, lower values of 1.39, 0.26, and 0.45 for SD/M, RSD, and SSI, respectively, were obtained at a window size of 11  11 for the VV polarization image. The higher value of 0.51 of ENL shows better performance for a boxcar filter with a window size of 11  11 for a VV polarization image. The same trend was observed for the VH polarization image. For a median filter, lower values of 0.99, 0.09, and 0.32 of SD/M, RSD, and SSI, respectively, were obtained at a window size of 11  11 for a VV polarization image. The higher value of 1.02 of ENL showed better performance of the median filter with a window size of 11  11 for a VV polarization image. The same trend was observed for the VH polarization image because it had the lowest values of SD/M, RSD, and SSI with the highest ENL value. The VV polarization image and the corresponding despeckled image with a median filter applied at an 11  11 window size are shown in Fig. 13.2A and B, respectively. The VH polarization image and corresponding despeckled image with a median filter at an 11  11 window size are shown in Fig. 13.3A and B, respectively. Despeckled images of VV and VH polarization images are assessed qualitatively and can be used for further analysis such as visual interpretaion, information extraction and classification. The Frost filter with a 9  9 window size results in the smallest values of 2.96, 1.15, and 0.96 for SD/M, RSD and SSI, respectively for the VV polarization image. Moreover, the SSI and ENL show similar values for both 9  9 and 11  11 window sizes for VV polarization. The smallest values of 0.91, 0.01, and 0.81 for SD/M, RSD, and SSI, respectively, were obtained for the Frost filter applied on the VH polarization image. The Frost filter with a 5  5 window size had the highest value ENL value of 1.28

249

250

Radar Remote Sensing

Table 13.2 Evaluation of speckle reduction filters applied on VV polarization image. VV polarization

Filter

Boxcar

Median

Frost

Gamma map

Lee

Refined Lee

Window size

33 55 77 99 11  11 33 55 77 99 11  11 33 55 77 99 11  11 33 55 77 99 11  11 33 55 77 99 11  11

Standard deviation to mean ratio

Relative standard deviation

Speckle suppression index

Equivalent number of looks

2.48 2.00 1.70 1.52 1.39 2.36 1.55 1.19 1.06 0.99 2.51 2.54 2.89 2.96 2.98 2.48 2.00 1.70 1.52 1.41 2.48 2.00 1.70 1.52 1.43 2.14

0.81 0.53 0.38 0.30 0.26 0.66 0.26 0.14 0.11 0.09 0.82 0.83 1.08 1.15 1.17 0.81 0.53 0.38 0.30 0.26 0.81 0.53 0.38 0.31 0.27 0.55

0.80 0.65 0.55 0.49 0.45 0.76 0.50 0.38 0.34 0.32 0.81 0.82 0.93 0.96 0.96 0.80 0.65 0.55 0.49 0.46 0.80 0.65 0.55 0.49 0.46 0.69

0.16 0.25 0.34 0.43 0.51 0.18 0.42 0.71 0.89 1.02 0.16 0.16 0.12 0.11 0.11 0.16 0.25 0.35 0.44 0.50 0.16 0.25 0.35 0.43 0.49 0.22

for VH polarization. For the Gamma map filter, lower values of 1.41, 0.26, and 0.46 for SD/M, RSD, and SSI, respectively, were obtained at a window size of 11  11 for the VV polarization image. The higher value of ENL (0.50) also showed better performance for the Gamma map filter with a window size of 11  11 for the VV polarization image. Similar SD/M, RSD, and SSI values were obtained when the Gamma map and boxcar filters were applied on the VV polarization image at window sizes of 3  3, 5  5, and 7  7. The Gamma map filter with a 9  9 window size had the smallest values of 0.76, 0.01, and 0.68 of SD/M, RSD and SSI respectively and highest ENL value (1.74) for VH polarization image. For Lee filter, lower values 1.43, 0.27, and 0.46 of SD/M, RSD, and SSI, respectively, were obtained at a window size of 11  11 for the VV polarization

Evaluation of speckle filtering methods using polarimetric Sentinel-1A data

Table 13.3 Evaluation of speckle reduction filters applied on VH polarization image. VH polarization

Filter

Boxcar

Median

Frost

Gamma map

Lee

Refined Lee

Window size

33 55 77 99 11  11 33 55 77 99 11  11 33 55 77 99 11  11 33 55 77 99 11  11 33 55 77 99 11  11

Standard deviation to mean ratio

Relative standard deviation

Speckle suppression index

Equivalent number of looks

0.94 0.82 0.75 0.70 0.67 0.94 0.78 0.71 0.67 0.64 0.95 0.88 0.90 0.91 0.91 0.94 0.82 0.77 0.76 0.77 0.94 0.82 0.77 0.74 0.73 0.91

0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.01

0.85 0.74 0.67 0.63 0.60 0.84 0.70 0.63 0.60 0.57 0.85 0.79 0.81 0.81 0.82 0.84 0.73 0.69 0.68 0.69 0.84 0.73 0.69 0.66 0.65 0.81

1.12 1.48 1.77 2.01 2.21 1.14 1.63 2.00 2.26 2.47 1.11 1.28 1.24 1.21 1.20 1.13 1.50 1.69 1.74 1.69 1.12 1.48 1.70 1.82 1.90 1.21

Figure 13.2 (A) VV polarization image; (B) Despeckled image.

251

252

Radar Remote Sensing

Figure 13.3 (A) VH polarization image; (B) Despeckled image.

image. The higher value of ENL (0.49) also showed better performance for the Lee filter with a window size of 11  11 for the VV polarization image. Similar SD/M, RSD, and SSI values were obtained when the Lee and Gamma map filters were applied on the VV polarization image at window sizes of 3  3, 5  5, and 7  7. The smallest values of 0.73, 0.01, and 0.65 for SD/M, RSD, and SSI, respectively, were obtained for the Lee filter applied on the VH polarization image. The Lee filter with an 11  11 window size also had the highest ENL value of 1.90 for VH polarization. The refined Lee filter performed well when it was applied on the VH polarization image. It had lower values of 0.91, 0.01, and 0.81 for SD/M, RSD, and SSI, respectively, with a higher ENL value (1.21) for the VH polarization image. When applied on VV polarization, the boxcar, Gamma map, and Lee filters at window sizes of 3  3, 5  5, and 7  7 demonstrated similar performance quantitively in terms of SD/M, RSD, SSI, and ENL. These filters seem similar in reducing speckle noise up to a 7  7 window size for the VV polarization image. Overall, however, the boxcar filter performed well compared with the Gamma map and Lee filters at an 11  11 window size in suppression speckles while using a VV polarization image in the study area. The boxcar filter also performed well at the 11  11 window when it was applied on the VH polarization image. The median filter was observed to be best in speckle reduction among all filters applied on both VV and VH polarization images in the overall comparative evaluation. It had the smallest values for SD/M, RSD, and SSI and the largest value for ENL at the same time, particularly for the cross (VH) polarization image. The backscattering coefficient was calculated for both VV and VH polarization images after we selected the most efficient speckle reduction filter. The sigma naught (s
) images of the VV and VH polarization image are represented in Fig. 13.4A and B, respectively, which can be used for further analysis and classification.

Evaluation of speckle filtering methods using polarimetric Sentinel-1A data

Figure 13.4 Sigma naught (s
) image of (A) VV polarization; (B) VH polarization.

Only the VV and VH polarization images of the Sentinel 1A at C-band were filtered and evaluated in the current study. Thus, more polarizations (HH and HV) and wavelengths (L- and X-band) need to be studied. In future studies, some new indices can be applied and compared with existing indices.

5. Conclusion This work evaluated the speckle suppression efficiency of selected filters applied on a Sentinel 1A SAR image. Several indices were used for the quantitative evaluation of speckle filters. These indices can achieve the best performance at the same time for dual-polarization (VV and VH) images of Sentinel 1A. Based on the SD/M, RSD, SSI, and ENL values, the median filter with a window size of 11  11 gave the best performance in reducing speckle noise for VV and VH polarization, respectively. Moreover, boxcar, Gamma map, and Lee filters performed well in terms of the quantitative assessment of despeckled images. The filtering process, along with radiometric calibration and geometric correction, is a method for calculating the backscattering coefficient accurately over a heterogenous landscape.

Acknowledgment The authors are very grateful to ESA for freely providing Sentinel-1A SAR data and SNAP software with S-1TBX used in this study.

253

254

Radar Remote Sensing

References Argenti, F., Lapini, A., Bianchi, T., Alparone, L., 2013. A tutorial on speckle reduction in synthetic aperture radar images. IEEE Geosci. Remote Sens. Mag. 1 (3), 6e35. Bonny, S., Chanu, Y.J., Singh, K.M., 2019. Despeckling of SAR images by finding the expected values using the probability distribution of speckle. Iran. J. Sci. Technol. Trans. Sci. 43, 1327e1336. Cloude, S.R., 2009. Polarisation Applications in Remote Sensing. Oxford University Press, Oxford, U.K. De Alban, J.D.T., Connette, G.M., Oswald, P., Webb, E.L., 2018. Combined Landsat and L-Band SAR data improves land cover classification and change detection in dynamic tropical landscapes. Remote Sens. 10, 306. Dellepiane, S.G., Angiati, E., 2014. Quality assessment of despeckled SAR images. IEEE J. Sel. Top. Appl. Earth Obs. Rem. Sens. 7 (2), 691e707. Di Martino, G., Poderico, M., Poggi, G., Riccio, D., Verdoliva, L., 2014. Benchmarking framework for SAR despeckling. IEEE Trans. Geosci. Rem. Sens. 52 (3), 1596e1615. Donoho, D.L., 1995. De-noising by soft-thresholding. IEEE Trans. Inf. Theor. 41 (3), 613e627. Freeman, A., 2007. Fitting a two-component scattering model to polarimetric SAR data from forests. IEEE Trans. Geosci. Rem. Sens. 45 (8), 2583e2592. Frost, V.S., Stiles, J.A., Shanmugam, K.S., Holtzman, J.C., Smith, S.A., 1981. An adaptive filter for smoothing noisy radar images. Proc. IEEE 69 (1), 133e135. Gagnon, L., Jouan, A., 1997. Speckle filtering of SAR images: a comparative study between complexwavelet-based and standard filters. SPIE proceedings 3169, 80e91. Goodman, N., 1963. Statistical analysis based on a certain multivariate complex Gaussian distribution (an introduction). Ann. Math. Stat. 34 (1), 152e177. Goodman, J.W., 1976. Some fundamental properties of speckle. J. Opt. Soc. Am. 66 (11), 1145e1150. Haarpaintner, J., Hindberg, H., 2019. Multi-temporal and multi-frequency SAR analysis for forest land cover mapping of the Mai-Ndombe District (Democratic Republic of Congo). Remote Sens. 11, 2999. Inglada, J., Vincent, A., Arias, M., Marais-Sicre, C., 2016. Improved early crop type identification by joint use of high temporal resolution SAR and optical image time series. Remote Sens. 8 (5), 362. Kuan, D.T., Sawchuk, A.A., Strand, T.C., 1987. Adaptive restoration of images with speckle. IEEE Trans. Acoust. Speech Signal Process. 35 (3), 373e383. Kumar, P., Gupta, D.K., Mishra, V.N., Prasad, R., 2015. Comparison of support vector machine, artificial neural network, and spectral angle mapper algorithms for crop classification using LISS IV data. Int. J. Remote Sens. 36 (6), 1604e1617. Kumar, P., Prasad, R., Gupta, D.K., Mishra, V.N., Vishwakarma, A.K., Yadav, V.P., Bala, R., Choudhary, A., Avtar, R., 2018. Estimation of winter wheat crop growth parameters using time series Sentinel-1A SAR data. Geocarto Int. 33 (9), 942e956. Kumar, P., Prasad, R., Choudhary, A., Gupta, D.K., Mishra, V.N., Vishwakarma, A.K., Singh, A.K., Srivastava, P.K., 2019. Comprehensive evaluation of soil moisture retrieval models under different crop cover types using C-band synthetic aperture radar data. Geocarto Int. 34 (9), 1022e1041. Kupidura, P., 2016. Comparison of filters dedicated to speckle suppression in SAR images. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. XLI-B7 269e276. Lee, J.S., 1980. Digital image enhancement and noise filtering by use of local statistics. IEEE Trans. Pattern Anal. Mach. Intell. 2 (2), 165e168. Lee, J.S., Pottier, E., 2009. Polarimetric Radar Imaging: From Basics to Applications. CRC Press, Cleveland, OH, USA. Liu, L., Zhou, F., Chen, J., Yang, X., Jia, L., Dong, Z., Ai, J., 2017. Despeckling PolSAR images with an adaptive bilateral filter. J. Appl. Remote Sens. 11 (2), 020501. Medasani, S., Reddy, G.U., 2018. Speckle filtering and its influence on the decomposition and classification of hybrid polarimetric data of RISAT-1. Remote Sens. Appl. Soc. Environ. 10, 1e6. Mishra, V.N., Prasad, R., Kumar, P., Gupta, D.K., Srivastava, P.K., Rai, P.K., 2017a. Dual-polarimetric C-band SAR data for land use/land cover classification by incorporating textural information. Environ. Earth Sci. 76 (1), 26.

Evaluation of speckle filtering methods using polarimetric Sentinel-1A data

Mishra, V.N., Prasad, R., Kumar, P., Srivastava, P.K., Rai, P.K., 2017b. Knowledge-based decision tree approach for mapping spatial distribution of rice crop using C-band synthetic aperture radar-derived information. J. Appl. Remote Sens. 11 (4), 046003. Mishra, V.N., Prasad, R., Rai, P.K., Vishwakarma, A.K., Arora, A., 2019. Performance evaluation of textural features in improving land use/land cover classification accuracy of heterogeneous landscape using multi-sensor remote sensing data. Earth Sci. Inform. 12 (1), 71e86. Mishra, V.N., Rai, P.K., 2016. A remote sensing aided multi-layer perceptron-Markov chain analysis for land use and land cover change prediction in Patna district (Bihar), India. Arabian J. Geosci. 9 (4), 249. Ouchi, K., 2013. Recent trend and advance of synthetic aperture radar with selected topics. Remote Sens. 5 (2), 716e807. Pang, B., Xing, S., Li, Y., Xuesong, W., 2013. Novel polarimetric SAR speckle filtering algorithm based on mean shift. J. Syst. Eng. Electron. 24 (2), 222e223. Perona, P., Malik, J., 1990. Scale-space and edge detection using anisotropic diffusion. IEEE Trans. Pattern Anal. Mach. Intell. 12 (7), 629e639. Pratt, W.K., 1975. Median Filtering. Image Process Inst Univ Southern California Los Angeles. Sheng, Y., Xia, Z.G., 1996. A comprehensive evaluation of filters for radar speckle suppression. In: IGARSS 1996. IEEE Int. Geosci. Remote Sens. Symp., pp. 1559e1561. Turkar, V., Deo, R., Rao, Y.S., Mohan, S., Das, A., 2012. Classification accuracy of multi-frequency and multi-polarization SAR images for various land covers. IEEE J. Sel. Top. Appl. Earth Obs. Rem. Sens. 5 (3), 936e941. Wei, S., Zhang, H., Wang, C., Wang, Y., Xu, L., 2019. Multi-temporal SAR data large-scale crop mapping based on U-Net model. Remote Sens. 11 (1), 68. Winsvold, S.H., K€a€ab, A., Nuth, C., Andreassen, L.M., Van Pelt, W.J.J., Schellenberger, T., 2018. Using SAR satellite data time series for regional glacier mapping. Cryosphere 12, 867e890.

255

This page intentionally left blank

SECTION 3

Advanced methods for radar remote sensing

257

This page intentionally left blank

CHAPTER 14

Emerging techniques of polarimetric interferometric synthetic aperture radar for scattering-based characterization Bhanu Prakash and Shashi Kumar Photogrammetry and Remote Sensing Department, Indian Institute of Remote Sensing, Indian Space Research Organisation, Dehradun, Uttarakhand, India

1. Introduction Radar remote sensing uses electromagnetic energy backscattered from ground targets to extract physical and dielectric behavior. The advantage of radar imaging lies in its capability of all-hour and all-weather imaging. Radar toward the ground transmits microwave pulses with a predefined pulse repetition frequency, depending on the platform. The intensity and phase of backscattered energy depend on the surface roughness and moisture content of the ground target. The radar antenna receives backscattered energy from ground targets subject to microwave pulses transmitted by it. Because the system has side-looking imaging and geometry (range direction) with forwarding motion capability (azimuth direction), it is known as side-looking aperture radar. Synthetic aperture radar (SAR) is a concept that was developed at the end of the 20th century to reduce the antenna size of the radar satellite by using multiple looks on the same ground target. Fig. 14.1 shows the SAR geometry with different angles and parameters. Polarimetric SAR (PolSAR) decomposition helps to retrieve information about scattering mechanisms in response to the interaction of different ground objects with SAR signals (Freeman and Durden, 1998). This chapter deals with the development of the PolSAR interferometry (PolInSAR) coherence-based decomposition model for the characterization of man-made and natural targets. The fundamental theories and equations of SAR polarimetry are explained in Section 2. A detailed analysis of SAR polarimetry (PolSAR) decomposition with its advantages and limitations is provided in Section 3. Section 4 includes the concept of the polarization orientation angle (POA) and its effects on scattering characterization. Section 5 provides insight into probability distribution functions adopted in polarimetric decomposition. The concepts of PolInSAR coherence and their incorporation into decomposition modeling are described in Section 6. A detailed description of the PolInSAR decorrelation-based decomposition model is presented in Section 7. Radar Remote Sensing ISBN 978-0-12-823457-0, https://doi.org/10.1016/B978-0-12-823457-0.00014-8

© 2022 Elsevier Inc. All rights reserved.

259

260

Radar Remote Sensing

Figure 14.1 Schematic representation of synthetic aperture radar geometry.

As the pulse is transmitted in the range direction and the radar moves in the azimuth direction, SAR imagery will have a range and an azimuth resolution. The azimuth and range resolutions can be calculated from the SAR geometry diagram. The azimuth resolution is a function of the antenna length whereas the range resolution is a function of the pulse repetition frequency. The row image needs to be compressed in the azimuth and range directions to extract useful information from it. The radiation pattern of the antenna can be controlled to operate the SAR system in different modes with various swaths and resolutions (Fig. 14.2). Important classes of modes are strip-map, spotlight, and scan SAR. In the strip-map mode, a single strip is scanned with a fixed pattern. The beam elevation and profile are kept fixed while the data are acquired. The stripmap mode is suitable for medium resolution imaging applications. The spot-light mode is used to acquire images with high resolution in which the antenna is steered in an azimuth direction toward a fixed swath. In the spotlight and strip-map modes, there is less coverage area. The scan SAR mode is used for large-scale area coverage by steering the antenna sequentially in various subswaths (Moreira et al., 2013).

Emerging techniques of polarimetric interferometric synthetic aperture radar for scattering-based characterization

Figure 14.2 Schematic representation of different modes of synthetic aperture radar (SAR): (A) Strip map mode; (B) Scan SAR mode; (C) Spotlight mode.

Different SAR configurations are defined depending on the transmission of radar pulses and the reception of echoes. If backscattered energy is received at the radar using the same transmitted antenna, the SAR systems are known as monostatic, and if it is received using a different antenna, they are regarded as bistatic (multistatic in general). Application of SAR imaging includes two-dimensional (2-D), 3-D, and 4-D mapping, change detection, earth observation, disaster management, urban planning, and planetary studies. From a single SAR product, 2-D maps can be created using the backscattered image, or with the decomposed image using the PolSAR technique. From multiple SAR products (interferometric pair), digital elevation models (DEMs) can be created using SAR interferometry (InSAR). From temporal SAR images, 4-D maps (space and time) can be generated using differential InSAR. PolInSAR is a combination of polarimetric and interferometric techniques that have many applications in the field of vegetation- and snow-related studies (Wangensteen et al., 2005; Neumann et al., 2012; Kumar et al., 2019; Khati, 2014; Chunxia et al., 2012). SAR images are widely used in many other applications such as earth observation, oil spill detection, climatic studies, and subsidence monitoring. They are even used in the lunar and different planetary studies.

2. Synthetic aperture radar polarimetry The scattering matrix (S) establishes a relationship connecting the incident and backscattered energy, which directly depends on the physical and structural properties target. Each pixel of a fully PolSAR image is represented as a 2
2 scattering matrix: " # SHH SHV S¼ (14.1) SVH SVV

261

262

Radar Remote Sensing

Target vectors are derived to simplify the scattering matrix to a convenient target vector that will contain all necessary information. The linear lexicographic feature vector (KC) is given by: 3 2 SHH 7 6 pffiffiffi pffiffiffi T 7 ¼ ½ SHH KC ¼ 6 (14.2) 2 S 2SHV SVV  HV 5 4 SVV Here, T denotes the transpose of a matrix. Covariance matrix (C) is calculated from the lexicographic feature vector as: 2 pffiffiffi 3 * * SHH SVV 2SHH SHV jSHH j2 6 pffiffiffi 7 pffiffiffi 6 y * * 7 C ¼ KC * KC ¼ 6 2SHV SHH (14.3) 2SHV SVV 2jSHV j2 5 4 p ffiffi ffi * * SVV SHH 2SVV SHV jSVV j2 Here ϯ denotes the conjugate-transpose of a matrix. Pauli vector (KL) is another representation in which elements can be physically interpreted as individual scattering mechanisms. It is defined as: 2 3 SHH þ SVV 7 1 6 7 KL ¼ pffiffiffi 6 S  S (14.4) HH VV 5 24 2SHV The first term in the 3
1 matrix denotes surface (odd-bounce) scattering, the second term denotes double-bounce (even-bounce) scattering, and the third term denotes volume (canopy) scattering. Pauli decomposition is derived from the Pauli vector, and the Pauli false-color composite (FCC) is generated by representing the first term as red, the second as blue, and the third as green. A coherency matrix (T) is generated by multiplying the Pauli vector by its conjugate transpose (Richards, 2009): 2

jSHH þ SVV j2 6 16 T ¼ 6 ðSHH  SVV ÞðSHH þ SVV Þ* 24 2SHV ðSHH þ SVV Þ*

T ¼ KL
KyL ðSHH þ SVV ÞðSHH  SVV Þ* jSHH  SVV j2 2SHV ðSHH  SVV Þ*

(14.5) * 2ðSHH þ SVV ÞSHV

3

7 * 7 2ðSHH  SVV ÞSHV 5 4jSHV j2 (14.6)

Covariance and coherency matrices can be calculated with knowledge of any of them by using the unitary transformation matrix. The covariance and coherency matrices are

Emerging techniques of polarimetric interferometric synthetic aperture radar for scattering-based characterization

based on the second-order form of the scattering matrix; hence, they address more attributes about the ground target. This benefit makes them suitable for defining efficient decomposition models to represent target scatterers in SAR imagery.

3. Polarimetric decomposition PolSAR decomposition refers to the process of extracting the individual backscattering response from all targets present in a single resolution cell. Two important classes of target decomposition techniques are coherent and incoherent target decompositions. Coherent target decomposition addresses fully polarized backscattered returns from coherent targets whose polarimetric information can be completely estimated from the scattering matrix (S). Pauli and Krogager decompositions are well-known examples of this class. It can handle only pure scatterers because it uses the scattering matrix (Richards, 2009; Lopez-Martinez and Pottier, 2015; Canada Center for Remote Sensing, 2005). Incoherent decompositions can be eigenvalueeeigen vectorebased or model-based. Incoherent decomposition systems are introduced to address distributed ground targets because the SAR resolution cell contains multiple scatterers. These are based on a 3
3 coherency matrix (T) or covariance matrix (C) rather than using S, and the objective is to express it in terms of a linear combination of second-order scattering mechanism models. Model-based decomposition and eigenvalueeeigenvector decompositions are examples of this class (Moreira et al., 2013). 3.1 Freeman-Durden decomposition The scattering contribution of the SAR image at the pixel level is expressed as a linear combination of three fundamental, physically interpretable scattering components: first-order Bragg’s surface scatters (surface scattering), canopy scatters from randomly oriented dipoles (volume scattering), and even-bounce scatter. This decomposition model was developed using the assumption that there is zero correlation between copolarization and cross-polarization channel returns and the cross-polarization channel return is only due to the presence of canopy scatterer (SHHS*HVSVVS*HV ¼ 0). This assumption simplifies the estimation of volume scattering power from the cross-polarization scattering term (Freeman and Durden, 1998). The remaining two scattering powers are calculated by the Freeman-Durden algorithm by predefining surface and doublebounce dominant scatterers using the real part of the SHHS*VV value. The model equation is given as: 3 3 2 2 2 3 1 0 1=3 1 b* 0 1 a* 0 7 7 6 6 6 7 2 2 7 7 6 6 7 T ¼ FS 6 (14.7) 4 b jbj 0 5 þ FD 4 a jaj 0 5 þ FV 4 0 2=3 0 5 1=3 0 1 0 0 0 0 0 0

263

264

Radar Remote Sensing

where b and a are derived from the reflection coefficients for horizontal and vertical polarization owing to a rough surface (for surface scattering) and groundetrunk interaction (for double-bounce scattering), respectively. These two are unknowns and need to be solved from the decomposition model, but their values are preassumed for ease of scattering power calculation (Freeman and Durden, 1998; Verma, 2012). This equation can be further simplified as: C ¼ FS CS þ FD CD þ FV CV

(14.8)

where F is the coefficient of scattering power and C is the covariance matrix for surface, double-bounce, and volume scattering components. The Freeman-Durden model decomposes total backscatter into canopy scattering (volume scattering), doublebounce (even-bounce) scattering, and surface (odd-bounce) scattering components. The canopy scattering is defined as follows. The scattering matrix for the canopy with standard orientation is represented as: " # SV 0 S¼ (14.9) 0 SH This is derived assuming the canopy is randomly oriented, very thin, and cylinder-like scatterers. From a standard orientation representation, the random orientation case is derived by rotating the scattering matrix using angle F with respect to the radar look direction. A rotation matrix is defined and multiplied with the standard orientation scattering matrix given in Eq. (14.10): " # " # " # SHH SHV cosF sinF cosF sinF ¼ S (14.10) sinF cosF sinF cosF SVH SVV This can be expanded as follows: " # # " SHH SHV Sh sin2 F þ Sv cos2 F ðSv  Sh ÞcosF sinF ¼ SVH SVV ðSv  Sh ÞcosF sinF Sh cos2 F þ Sv sin2 F

(14.11)

Suppose f(F) is any function of F and p(F) is the probability density function corresponding to the target orientation. Then, the expected value for function f(F) can be represented by the integral: Z2p Cf D ¼

dFf ðFÞpðFÞ 0

(14.12)

Emerging techniques of polarimetric interferometric synthetic aperture radar for scattering-based characterization

Thus, elements of the covariance matrix can be represented by the equations:   CjSVV j2 D ¼ a1 jSV j2 þ 2a2 Re SH SV* þ a3 jSH j2 (14.13)   CjSHH j2 D ¼ a1 jSH j2 þ 2a2 Re SH S*V þ a3 jSV j2   CjSHV j2 D ¼ a1 jSV j2  2a2 Re SH SV* þ a3 jSH j2     * CSHH SVV D ¼ ða1 þ a3 ÞRe SH SV* þ a2 jSH j2 þ jSV j2 þ   iða1  a3 ÞIm SH SV* jSH j2     * * CSHH SHV D ¼ a4 SH SV*  jSH j2 þ a5 jSV j2  SV SH     * * D ¼ a4 jSV j2  SH SV* þ a5 SV SH  jSH j2 CSHV SVV Parameters a1, a2, a3, a4, and a5 can be expressed using the integral (Eq. 14.12) as: Z2p a1 ¼

dF cos4 FpðFÞ

(14.14)

0

Z2p a2 ¼

dF cos2 F sin2 FpðFÞ 0

Z2p a3 ¼

dF sin4 FpðFÞ 0

Z2p a4 ¼

dF cos3 F sinFpðFÞ 0

Z2p a5 ¼

dF cosF sin3 FpðFÞ 0

These parameters are simplified. To simplify the set of Eq. (14.14), two more assumptions are made. The scatterers are assumed to be thin cylinders. Hence, SH is zero and SV is unity. The second assumption is the uniform orientation distribution of p(F), which implies values for parameters a1, a2, a3, a4, and a5 as: a1 ¼ a3 ¼ 3p=4; a2 ¼ p=4 and a4 ¼ a5 ¼ 0

(14.15)

265

266

Radar Remote Sensing

These values can be substituted in Eq. (14.13), which implies: CjSHH j2 D ¼ CjSVV j2 D ¼ 1 * D ¼ CjSHV j2 D ¼ 1=3 CSHH SVV

(14.16)

* * SHH SHV ¼ SHV SVV ¼ 0

Double bounce is modeled using the vertical and horizontal reflection coefficients and surface scattering is modeled using the first-order Bragg’s model. The doublebounce scattering covariance matrix is defined as: CjSHH j2 D ¼ jaj2 ; CjSVV j2 D ¼ 1 * D ¼ a; CjSVV j2 D ¼ 0 CSHH SVV

(14.17)

* * CSHH SHV D ¼ CSHV SVV D ¼ 0

where; a ¼ ej2ðgH gV Þ Rgh Rth =Rgv Rtv Similarly, the surface scattering covariance matrix is expressed by the equations: CjSHH j2 D ¼ jbj2 ; CjSVV j2 D ¼ 1 * CSHH SVV D ¼ b; CjSHV j2 D ¼ 0

(14.18)

* * D ¼ CSHV SVV D ¼ 0 CSHH SHV

From these equations, the Freeman-Durden three-component decomposition model in Eq. (14.7) was derived. On equating the left- and right-hand sides in Eq. (14.7), the equations will be formed: CjSHH j2 D ¼ FS jbj2 þ FD jaj2 þ FV

(14.19)

CjSVV j2 D ¼ FS þ FD þ FV * CSHH SVV D ¼ FS b þ FD a þ FV =3

CjSHV j2 D ¼ FV =3 * * CSHH SHV D ¼ CSHV SVV D ¼ 0

It is evident from these equations that FV can be estimated straight from the SHV. This contribution is then subtracted from |SHH|2, |SVV|2, and SHHS*VV. The solution for the remaining unknowns is carried out by the simplified equations: CjSHH j2 D ¼ FS jbj2 þ FD jaj2

(14.20)

Emerging techniques of polarimetric interferometric synthetic aperture radar for scattering-based characterization

CjSVV j2 D ¼ FS þ FD * CSHH SVV D ¼ FS b þ FD a

Five unknowns and three equations are available for the solution. An assumption is made here to assign the values of alpha and beta according to the value of the real part of the SHHS*VV element. If this is less than zero, the double-bounce scattering dominance is assumed and the value of beta is fixed as 1. Similarly, if this is greater than zero, surface scattering dominance is assumed and the value of alpha is assumed to be 1. Using these assumptions, the remaining unknowns are estimated. 3.2 Yamaguchi four-component decomposition The cross-polarization channel return from a randomly oriented complex urban areas is misidentified as volume scattering in Freeman-Durden three-component decomposition. This creates ambiguity in identifying the ground target from the decomposition. Yamaguchi’s four-component decomposition was developed to overcome limitations of the three-component model by introducing helix scattering as a fourth component to address these complex urban scatterers. Mathematically, the helix power represents * sS S * s0. Helix scattering gives zero value over vegetation and the term SHH SHV VV HV water bodies. The helix scattering component model is represented by the matrix: 2 3 pffiffiffi 1 j 2 1 pffiffiffi 7 1 6 pffiffiffi CH ¼ 6 (14.21) Hj 2 2 Hj 2 7 4 5 4 pffiffiffi 1 j 2 1 The volume scattering component is modified in this model by splitting them into three different cases according to the value of the copolarization power ratio: 8 3 2 > > 8 0 2 >   > > 7 6 jS j2  > > 7 6 1   > HH > 6 0 4 0 7; 10 log10  > > 7 6 2  < 2dB >  15 jSVV j > 5 4 > > > > 2 0 3 > > > > > 3 2 > > > 3 0 1 >   > > 7 jS j2 

84 jSVV j  > 5 > > > 1 0 3 > > > > > 3 2 > > > > 3 0 2 >   > 7 6 > jS j2  > 7 6 > 1 > 6 0 4 0 7; 10 log  HH   2dB > > 10  > 7 6 > 15 jSVV j2  > 5 4 > > > 2 0 8 :

267

268

Radar Remote Sensing

The model equation can be written as: C ¼ FS CS þ FD CD þ FV CV þ FH CH

(14.23)

The surface and double-bounce scattering component models are the same as in the Freeman-Durden model (Eq. 14.7). The four-component Yamaguchi decomposition was again modified by rotating the covariance matrix using the POA (Yamaguchi et al., 2011). 3.3 Multiple-component scattering model decomposition The multiple-component scattering model (MCSM) of decomposition is a fivecomponent model (Zhang et al., 2008) introduced by decomposing the total backscattered return from the urban area into double-bounce scattering, helix scattering, and wire scattering. The wire scattering component addresses returns from linear structures present in urban areas. The volume scattering model was expressed in three different  cases according to the jSHH j2 jSVV j2 value similar to Yamaguchi-component decomposition. The wire scattering component model is defined as: 2 3 pffiffiffi * 2gr g jgj2 6 pffiffiffi 7 * (14.24) CW ¼ 6 2r 7 2jrj2 4 2gr 5 p ffiffi ffi g* 2r* 1 g¼

SHH SHV and r ¼ SVV SVV

(14.25)

The covariance matrix corresponds to the four scattering mechanisms that are the same as in Yamaguchi decomposition (Eqs. 14.2e14.8). The model equation for MCSM decomposition is represented as: C ¼ FS CS þ FD CD þ FV CV þ FH CH þ FW CW

(14.26)

The wire scattering highlights linear structures such as roads, building edges, and bridges. An improved multiple component scattering model decomposition was proposed (Zhang et al., 2008) by rotating the covariance matrix. The covariance matrix is rotated using the orientation angle-based rotation matrix before calculating the scattering powers. 3.4 Other works in polarimetric decomposition modeling Based on the Freeman-Durden three-component, Yamaguchi four-component, MCSM five-component, and PolInSAR coherence-based models, many modified polarimetric decomposition models have been developed. An adaptive polarimetric decomposition model was proposed (Arii et al., 2011) without considering the assumption of reflection symmetry. A generalized volume scattering model (Arii et al., 2010) was proposed by

Emerging techniques of polarimetric interferometric synthetic aperture radar for scattering-based characterization

incorporating the mean orientation angle and the degree of randomness for each pixel. The concept of nonnegative eigenvalue decomposition was adopted to extend threecomponent model-based decomposition. An improved four-component decomposition model was introduced (Maurya et al., 2020) to reduce scattering ambiguity and the negative value problem that occurs in model-based decomposition and to address the proper use of covariance matrix elements. The helix scattering angle was calculated and compensated for pixel by pixel to reduce the overestimation of helix scattering. Compensated helix scattering was further used to derive other scattering component powers. Stochastic distance-based POA estimation and decomposition modeling are also available (Bhattacharya et al., 2015) by choosing the Hellinger distance as the separability analysis tool. Yamaguchi’s four-component decomposition model was improved by incorporating cross-scattering as a fifth scattering component (Xiang et al., 2016). The cross-scattering component addresses the horizontal transmit and vertical receive (HV) channel intensities caused by the oriented buildings. A general four-component polarimetric decomposition was proposed (Singh et al., 2013) using a unitary to the rotated coherency matrix. This was done to eliminate the T23 element to reduce the amount of independent information in the coherency matrix, and it was selected because the helix scattering power is calculated from it. The unitary matrix was applied to estimate all  four component powers. Depending on the value of jSHH j2 jSVV j2 , the probability distribution for the volume scattering component definition was chosen among the uniform, sine, and cosine distributions. The six-component polarimetric decomposition model was proposed (Singh and Yamaguchi, 2018) using the real and imaginary parts of the T13 element. Two scattering models in addition to surface, double-bounce, volume, and helix scattering are oriented dipole scattering and compound dipole scattering (oriented quarter-wave scattering). Oriented dipole scattering was developed for the 45 degree-oriented dipole represented by the real part of T13 and the oriented quarter-wave scattering was developed for the compounded scattering submatrix from 45 degreeoriented dipoles to address the imaginary part of T13.

4. Polarization orientation angle The angle between the major axis of the polarization ellipse and the horizontal axis on the incident plane represents the POA (Zhang et al., 2008). The POA becomes shifted by different targets such as sloppy terrain, oriented complex urban structures, and inclined bridges (Sen Lee and Ainsworth, 2010) (Eq. 14.27). Eq. (14.27) shows the relationship of connecting the POA with the azimuth slope (tan u), range slope (tan g), and radar look angle (tan 4): tanq ¼

tan u tan g cos4 þ sin4

(14.27)

269

270

Radar Remote Sensing

The POA shift is considered a characteristic feature of the target because it has a close relationship to the scattering matrix. It is calculated as: 8 if h  p=4 < h; q¼ p : h  ; if h > p=4 2 " !# 1 4RefðS  S ÞS g HH VV HV tan1 (14.28) h¼   4  jSHH j2  jSVV j2 þ 4jSHV j2 A spike in cross-polarization channel intensity is observed because the POA shift causes the symmetry property of C and T matrices to fail. As a result, the assumptions of zero correlation between copolarization and cross-polarization returns will be no longer be valid, and hence will create ambiguity in the identification of the scatterer. To address this, correction of the POA shift was introduced by rotating the C or T matrix using the rotation matrix (Verma, 2012; Yamaguchi et al., 2011; Sen Lee and Ainsworth, 2010). Compensation of the POA has no impact on HH- or VV-dependent terms in C and T, whereas it reduces HV- or VH-dependent terms.

5. Probability distributions When developing the decomposition model, different probability distributions are assumed to account for the random nature of volume scatterers. The uniform distribution and the delta function distribution are two extreme cases that represent pure random scattering orientation and equal scattering orientation, respectively (Arii et al., 2010). Several models were developed using uniform probability distribution because it is simple, even though it is limited to specific types of vegetation. The delta function is unsuitable for volume scatterers because we cannot expect the same orientation from tree canopies and leaves. A cosine squared distribution was proposed (Arii et al., 2010) to replace uniform distribution by considering the orientation angle of the scatterer:  2    cos ðq  q0 Þ n  PV ðq; q0 ; nÞ ¼ R 2p (14.29) 2 ðq  q Þgn jdq jfcos 0 0 The value n ¼ 1 defines the cosine squared probability distribution. q0 denotes the mean orientation angle estimated from Eq. (14.28). The volume scattering component covariance matrix based on the Freeman-Durden three-component decomposition will be expanded using cosine squared distribution as (Arii et al., 2010; Bhanu Prakash and Kumar, 2020, 2021): 3 2 2 3 pffiffiffi 3 0 1 2cos2q0 2sin2q0 0 7 1 6 pffiffiffi 7 pffiffiffi 16 þ 6 CCosine ¼ 6 0 2 07 (14.30) 2sin2q0 0 2sin2q0 7 5 4 4 5 8 8 pffiffiffi 1 0 3 0 2sin2q0 2cos2q0

Emerging techniques of polarimetric interferometric synthetic aperture radar for scattering-based characterization

The resultant covariance matrix elements now depend on the orientation angle of the target scatterer; hence, the volume scattering model adapts according to the offered orientation angle shift by each target.

6. Polarimetric synthetic aperture radar interferometry PolInSAR evolved as a combination of polarimetry and interferometry for fully polarimetric InSAR images (Fig. 14.3) that provide six degrees of freedom per pixel (Flynn et al., 2002). The volumetric information can be retrieved using the PolInSAR technique because the target scatterers offer different responses to different polarizations of radar pulse (Fomena and Cloude, 2005). This advantage of PolInSAR makes it popular even though the data requirement and processing complexity are more than for conventional SAR techniques. Many studies have been performed on forest structural and biophysical parameter retrieval, which is one of the most important applications of this technique (Kumar et al., 2017, 2019; Khati, 2014; Joshi and Kumar, 2017; Kumar, 2009). 6.1 Polarimetric synthetic aperture radar interferometry coherence and optimization PolInSAR generates interferograms and coherence images for each pair of polarization. The interferometric coherence indicates the time variability of terrain features under the

Figure 14.3 Schematic representation of polarimetric synthetic aperture radar interferometry.

271

272

Radar Remote Sensing

area of interest. Each corresponding resolution cell of PolInSAR data can be expressed as master and slave scattering matrices S1 and S2: # " 1 " 1 2 2 SHH SHV SHH SHV and S2 ¼ S1 ¼ (14.31a) 2 2 1 1 SVH SVV SVH SVV SHH, SHV, SVH, and SVV denote the complex scattering coefficients corresponding to the horizontal transmit and horizontal receive (HH), horizontal transmit and vertical receive (HV), vertical transmit and horizontal receive (VH), and vertical transmit and vertical receive (VV) polarization channels. Pauli vectors K1 and K2 corresponding to the individual pixels are expressed as: 2 2 1 3 2 3 SHH þ SVV SHH þ SVV 6 1 1 6 1 7 and K ¼ p 2 5 ffiffiffi 4 SHH  SVV K1 ¼ pffiffiffi 6 (14.31b) SHH  SVV 5 2 4 2 2 2 1 2SHV 2SHV From Eq. (14.31), a single 6
6 coherency matrix can be formed. However, three 3
3 coherency matrices, T11, T22, and U12, can be formed from the master and slave images of PolInSAR data as the product of Pauli vectors described in Eqs. (14.32e14.34): y

T11 ¼ K1
K1 y

T22 ¼ K2
K2 and y

U12 ¼ K1
K2

(14.32) (14.33) (14.34)

When the complex Lagrangian function is applied on T11, T22, and U12, a couple of equations form that yield the eigenvalues and eigenvectors: ½T22 1 ½U12 y ½T11 1 ½U12 u2 ¼ yu2 and

(14.35)

½T11 1 ½U12 ½T22 1 ½U12 y u1 ¼ yu1

(14.36)

Here, u1 and u2 denote unitary complex vector pairs that address the master and slave image polarizations and y represents the eigenvector matrix. From the estimated optimum eigenvalue and eigenvector, optimum coherence can be derived as: Cuopt1;i ½U12 uopt2;i D gopt;i ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Cuopt1;i ½T11 uopt1;i DCuopt2;i ½T22 uopt2;i D

(14.37)

Emerging techniques of polarimetric interferometric synthetic aperture radar for scattering-based characterization

Because Eqs. (14.35) and (14.36) produce three eigenvectors, it is possible to generate three optimum coherences: optimum coherence 1 (goptimum1), optimum coherence 2 (goptimum2), and optimum coherence 3 (goptimum3). The master and slave PolInSAR images are radiometrically calibrated and multilooked to form the squared pixels. Coregistration is done using the ALOS Phased Array L-Band Synthetic Aperture Radar (PALSAR)-1 DEM of 12.5 m resolution in SNAP software. The coregistered data are exported in bin format and different coherence images are generated using PolSARpro 6.0 software. The window size for the coherence estimation is selected as 11. The gray-scale representation of important coherence results for the X-band TerraSAR-X and TanDEM-X data are shown in Fig. 14.4. The HH channel coherence is interferometric coherence between the HH channels of the master and slave images of the PolInSAR data. Similarly, VV channel coherence, HV channels

Figure 14.4 Various coherence images generated using X-band TerraSAR-X and TanDEM-X polarimetric synthetic aperture radar interferometry data of the Forest Research Institute (FRI) area, Dehradun. (A) HH channel coherence, (B) VV channel coherence, (C) HV channel coherence, (D) HH-VV channel coherence, (E) HH þ VV channel coherence, (F) HV þ VH channel coherence, (G) Optimum coherence 1, (H) Optimum coherence 2, and (I) Optimum coherence 3.

273

274

Radar Remote Sensing

coherence, and VH channel coherence are generated using VV, HV, and VH channels of the master and slave images, respectively. The HH, VV, and HV channel coherence images offer discrimination between permanent features such as urban areas and time-varying features such as vegetation. The urban area and dry riverbed are highlighted with white color indicating high correlation between corresponding channels of the master and slave images of the PolInSAR dataset. The forest area is represented with black color indicating low correlation in that area between corresponding channels of the master and slave images of the PolInSAR dataset. The HH þ VV channel for the PolSAR data generally highlights odd bounce scattering from rough surfaces. Similarly, the HH-VV channel highlights even bounce scattering present in the urban area and resulting from groundetree trunk interaction. HV þ VH image highlights a vegetated area because it is the sum of cross-polarization returns. The HH þ VV, HH-VV, and HV þ VH channel coherence images give good discrimination between permanent and time-varying features, as indicated in Fig. 14.5DeF. Fig. 14.5GeI show a gray-scale representation of optimum coherence 1, optimum coherence 2, and optimum coherence 3. Optimum coherence 1 in Fig. 14.4G is a saturated image with maximum values near 1, whereas optimum coherence 3 in Fig. 14.4F gives perfect discrimination between permanent and time-varying features. The histograms (frequency versus coherence values) for the required coherence images were generated and analyzed and are presented in Fig. 14.5. The histograms for the HH and VV channels (Fig. 14.5A and B) show that the distribution of coherence values is toward the higher side (i.e., unity). Histograms of different optimum coherences are shown in Fig. 14.5CeE. The histogram of the optimum coherence 3 is a Gaussian curve as given in Fig. 14.5E, whereas optimum coherence 1 and 2 are not. Optimum coherences 1 and 2 are skewed toward the higher side. This effect is due to the lower temporal baseline of the X-band PolInSAR data used to estimate it. Because optimum coherence 3 is Gaussian distributed, it can better differentiate between permanent and time-varying features. Thus, it is used here to model volume scattering.

7. Polarimetric synthetic aperture radar interferometry coherencebased decomposition PolInSAR-based coherence-based decomposition was proposed (Chen and Sato, 2011) as a possible solution for the presence of ambiguity in scattering retrieval. Interferometric coherence is considered to be a suitability criterion for selecting SAR interferometric data pairs for different applications: 2 3 gHH 0 gOptimum3 6 7 6 7 gOptimum3 6 7 0 0 Cvol ¼ Fv 6 (14.38) 7 1  gOptimum3 6 7 4 5 g*Optimum3 0 gVV

Emerging techniques of polarimetric interferometric synthetic aperture radar for scattering-based characterization

Figure 14.5 Histograms of coherence images: (A) HH coherence channel; (B) VV channel coherence; (C) Optimum coherence 1; (D) Optimum coherence 2; and (E) Optimum coherence 3.

275

276

Radar Remote Sensing

A volume scattering component covariance matrix (CVol) was proposed (Chen and Sato, 2011) using HH and VV channel coherences (gHH and gVV, respectively) and optimum coherence 3 (goptimum3), as described in Eq. (14.38). The surface and doublebounce scattering components are adopted in this model from Freeman-Durden decomposition (Section 3.1). Fig. 14.6 shows the PolInSAR coherence-based decomposition of ALOS-2 PALSAR-2 data for Doon Valley, Uttarakhand. The PolInSAR coherence-based model gives a reliable scattering information for different features within the study area. However, this model provided very good results

Figure 14.6 The false-color composite (red: double bounce scattering, green: volume scattering, blue: surface scattering) of ALOS-2 PALSAR-2 for Doon Valley, Uttarakhand, India.

Emerging techniques of polarimetric interferometric synthetic aperture radar for scattering-based characterization

to distinguish the scattering mechanism from different classes, but an overestimation of volume scattering could easily be seen for the urban settlements of Dehradun city. To overcome this problem, Bhanu and Kumar proposed a decorrelation-based approach (Bhanu Prakash and Kumar, 2021) to retrieve appropriate scattering elements from different classes.

8. Polarimetric synthetic aperture radar interferometry decorrelation-based decomposition model 8.1 Polarimetric synthetic aperture radar interferometry decorrelation Loss of coherence as a result of the variation of scatterers at the subpixel level can be termed decorrelation. Decorrelation is a measure of the variation in the scattering contribution of ground targets with respect to different independent variables. It introduces noise to interferometric phase generation. Table 14.1 shows types of decorrelation in InSAR data. Temporal decorrelation can be inherently high for a highly vegetated area but volume decorrelation depends on the relation between the radar wavelength and the leaf size of vegetation. The temporal decorrelation will be less for interferometric data with a high temporal resolution. Volume decorrelation is widely used to estimate the height of the observed scatterer (Moreira et al., 2013; Algebra and Chandra, 2003). It can be different types according to the independent variable associated with it, such as temporal terrain decorrelation, baseline or geometry decorrelation, volume decorrelation, Doppler centroid decorrelation, thermal decorrelation, or system noise and processing-induced decorrelation. PolInSAR decorrelation is related to volume decorrelation and temporal decorrelation of the interferometric pair dataset. The expression for total decorrelation is given by: gTotal ¼ gSNR :gQuant :gAmb :gGeometry :gAzimuth :gTemp gVol

(14.39)

Volume decorrelation occurs as a result of the penetration ability of the radar wave (Kumar, 2009). Decorrelation present in SAR interferometric pair datasets is closely related to interferometric coherence. The decorrelation value decreases with an increase Table 14.1 Types of decorrelation. Symbol

Quantity

gSNR gQuant gAmb gGeometry gAzimuth gTemp gVol

Signal-to-noise ratio decorrelation Decorrelation owing to quantization errors Decorrelation owing to ambiguities Geometry decorrelation Decorrelation owing to Doppler shift Temporal decorrelation Volume decorrelation

277

278

Radar Remote Sensing

in the coherence value. The low value of coherence indicates high decorrelation. It is possible to minimize terms other than temporal and volume decorrelations by improving various sensors and data acquiring conditions. Variations in the surface properties of target create temporal decorrelation whereas the penetration ability of the radar wave causes volume decorrelation (Kumar, 2009). Temporal decorrelation can be inherently high for highly vegetated areas. Volume decorrelation depends on the scattering properties of multiple targets in a single resolution cell, which cause uncertainty in the interferometric phase (Algebra and Chandra, 2003). Decorrelation is useful for discriminating among temporally varying and invariant features. The decorrelation parameter is generated for X-band TerraSAR-X and TanDEM-X data. The FRI area subset of the decorrelation parameter image is presented in Fig. 14.7. The gray-scale representation highlights the ability of the decorrelation parameter to distinguish between permanent and time-varying scatterers. The forest area is represented in white, indicating maximum decorrelation, whereas urban and surface scatterers are represented with a dark color indicating low decorrelation. The area in Fig. 14.7 indicated by a red box is the urban area represented in black, showing a low decorrelation parameter value. Similarly, the area indicated in blue is a dry riverbed, which is also represented in black showing a low decorrelation. The area indicated in green is a vegetated area represented in white, showing a high decorrelation.

Figure 14.7 Gray-scale representation of decorrelation parameter (X-band TerraSAR-X and TanDEM-X data).

Emerging techniques of polarimetric interferometric synthetic aperture radar for scattering-based characterization

Figure 14.8 Algorithm for polarimetric synthetic aperture radar interferometry decorrelation-based decomposition models.

The algorithm for the PolInSAR decorrelation-based polarimetric decomposition (Bhanu Prakash and Kumar, 2021) is shown in Fig. 14.8. The reference image scattering matrix and HH channel coherence (gHH ), VV channel coherence (gVV ), and optimum coherence 1 (gOptimum ) are given as input to the algorithm. The simplified model equation for this model is: T ¼ PS þ PD þ PV

(14.40)

279

280

Radar Remote Sensing

where PS, PD, and PV are the surface, double-bounce, and volume scattering powers, and T is the total power image. The detailed model equation is given as: 2 2 3 3 1 b* 0 1 a* 0 6 6 7 7 2 7 þ FD 6 a jaj2 0 7 þ FV
Volume scattering component T ¼ FS 6 b 0 jbj 4 4 5 5 0

0

0

0

0

0 (14.41)

b¼

RH  RV e2jgH RTH RGH þ e2jgV RTV RGV and a ¼ 2jg R R RH þ RV e H TH GH  e2jgV RTV RGV jbj < 1; jaj < 1

(14.42) (14.43)

Here, FS, FD, FV are the coefficients of surface, double-bounce, and volume scattering models. RH and RV are the Fresnel reflection coefficients for horizontally and vertically polarized waves (Verma, 2012). The optimum coherence value is compared pixel by pixel with a predefined threshold to decide the volume model. If the optimum coherence value is less than the threshold, the Freeman-Durden volume model is selected; otherwise, the PolInSAR coherence-based volume model is selected. Optimum coherence 1 gives the maximum coherence values compared with the other two optimum coherences because it corresponds to the largest eigenvalue. However, the selection of optimum coherence should be based on its ability to differentiate between permanent and time-varying features. For short-temporal baseline cases, optimum coherence 3 is used because in optimum coherence 1 and 2, coherence values accumulated at the higher side (Chen et al., 2013). Similarly, for longer temporal baseline cases, optimum coherence 2 is preferred because values for optimum coherence 3 accumulate at the lower side. Moreover, an order 3 Hermitian matrix should satisfy the positive semidefinite condition for physical realizability (Cloude, 1989). With this considered, in this work, optimum coherence 3 is taken for all frequencies and all temporal baseline cases to select the appropriate volume component. The Freeman-Durden volume model and the PolInSAR coherencebased volume model are expressed as: 3 2 1 0 1=3 7 6 7 CFreeman ¼ FV1 6 (14.44) 0 2=3 0 5 4 1=3

0

1

Emerging techniques of polarimetric interferometric synthetic aperture radar for scattering-based characterization

2 CCoherence

6 6 6 ¼ FV2 6 6 4

gHH

0

gOptimum3

0

gOptimum3 1  gOptimum3

0

g*Optimum3

0

gVV

3 7 7 7 7 7 5

(14.45)

The decorrelation parameter is defined in Eq. (14.46). The numerator of the decorrelation parameter is the sum of decorrelations present in the HH and VV channel coherences, and optimum coherence 3. The denominator represents the sum of coherences and the ratio proposed in Chen and Sato (2011). The decorrelation parameter is the ratio of these sums; because of the scaling of decorrelation by the coherence, in the resulting decorrelation parameter, the ability to discriminate between permanent and time-varying features is enhanced. This effect is used here to introduce model-based polarimetric decomposition using PolInSAR decorrelation. The vegetated area appears in white, indicating a high decorrelation, and permanent features that appear in dark colors indicate a low decorrelation. The volume power equation is modified by the proposed decorrelation parameter equation:

 ð1  gHH Þ þ ð1  gVV Þ þ 1  gOptimum3 Decorrelation parameter ¼ (14.46) gOptimum3

 gHH þ gVV þ 1  gOptimum3 PV2 ¼ Decorrelation parameter
FV2
Trace of Matrix



 PV2 ¼ ð1  gHH Þ þ ð1  gVV Þ þ 1  gOptimum3
FV2

(14.47) (14.48)

Here Pv2 and Fv2 are the decorrelation-based volume power and the PolInSAR coherence-based volume coefficient. The combined volume power and covariance matrix PV and CV, respectively, are derived as: ( PV1 ; j gOptimum3 < Th PV ¼ (14.49) PV2 ; j gOptimum3  Th ( CFreeman ; j gOptimum3 < Th (14.50) CV ¼ CCoherence ; j gOptimum3  Th

281

282

Radar Remote Sensing

The covariance matrix is then modified using the combined volume covariance matrix to evaluate the surface and double-bounce scattering powers. The covariance matrix is then updated as: C ¼ C  CV

(14.51)

The double-bounce and surface scattering powers can be calculated using the alpha (a) and beta (b) parameters. The values of a and b are assumed by evaluating the dominance of surface and double-bounce scattering components using the Re {C13} value. If Re{C13} is greater than 0, it is double-bounce scattering dominant and the value of a is assumed to be 1; else, the value of b is assumed to be 1 (Freeman and Durden, 1998). The double-bounce and surface scattering powers are calculated using these two assumptions and combined in the final step. This C matrix is used to determine the surface and double-bounce powers using the Freeman-Durden algorithm (Moreira et al., 2013). The results of PolInSAR decorrelation-based decomposition on the X-band TerraSAR-X and TanDEM-X PolInSAR pair data acquired over the Rudrapur area in Uttarakhand, India are displayed (Fig. 14.9A) along with the Google Earth image (Fig. 14.9B). This area contains wide agricultural fields, urban areas, and water bodies. Some important locations are marked using a yellow rectangle in both figures to validate the results. Urban areas are represented by a dark pink color in the FCC of the decomposed image. Two different agricultural areas are identified. In one, green color in the decomposed image represents planted fields indicating the dominance of volume scattering. The other area shown in blue represents plowed land or barren land indicating the dominance of surface scattering. The results of PolInSAR decorrelation-based decomposition on the X-band TerraSAR-X and TanDEM-X PolInSAR pair data acquired over the Haridwar area in Uttarakhand, India are displayed (Fig. 14.9C) along with the Google Earth image (Fig. 14.9D). The city is represented by a dark pink color showing the higher contribution of double-bounce and surface scattering rather than volume-scattering power. The riverbed is represented in blue, showing the dominance of surface-scattering power. The river is represented in black, indicating smooth surface reflection. The bright red spot in the middle of the river points to the possibility of big stones existing in it. As the X-band electromagnetic wave interacts with the stones, there is the possibility of even bounce scattering along with odd-bounce scattering. The temporal baseline of the PolInSAR data is a crucial factor in the amount of coherence and decorrelation obtained over permanent features such as vegetation. Because the coherence value distribution changes according to the temporal baseline, the decision regarding the threshold is a challenging task.

Emerging techniques of polarimetric interferometric synthetic aperture radar for scattering-based characterization

Figure 14.9 (A) Google Earth imagery for Rudrapur area, India. (B) Decomposed X-band TerraSAR-X and Tandem-X polarimetric synthetic aperture radar interferometry (PolInSAR) pair data for Rudrapur area represented in false-color composite. Red indicates double-bounce scattering; green, volume scattering; and blue, surface scattering. (C) Google Earth imagery for Haridwar area, India. (D) Decomposed X-band TerraSAR-X and Tandem-X PolInSAR pair data for Haridwar area represented in falsecolor composite. Red indicates double-bounce scattering; green, volume scattering; and blue, surface scattering.

283

284

Radar Remote Sensing

Acknowledgment The authors would like to express their sincere gratitude to the whole research team of ESA for providing SNAP 6.0 and PolSARPro 4.2 tools for polarimetric processing of the SAR data. The authors are thankful to the German Aerospace Center (DLR) Oberpfaffenhofen for providing TerraSAR-X/TanDEM-X data under the Project Id -NTI_POLI6635 on PolInSAR Tomography for aboveground biomass estimation and the Japan Aerospace Exploration Agency (JAXA) for providing the L-band ALOS-2 PALSAR-2 datasets under the proposal number 1408 of RA4 with the title Hydrological parameter retrieval and glacier dynamics study with L-band SAR data.

References Algebra, V., Chandra, M., 2003. Volume decorrelation resolution in polarimetric SAR interferometry. Electron. Lett. 39 (3), 6e7. Arii, M., Van Zyl, J.J., Kim, Y., 2010. A general characterization for polarimetric scattering from vegetation canopies. IEEE Trans. Geosci. Rem. Sens. 48 (9), 3349e3357. Arii, M., Van Zyl, J.J., Kim, Y., 2011. Adaptive model-based decomposition of polarimetric SAR covariance matrices. IEEE Trans. Geosci. Rem. Sens. 49 (3), 1104e1113. Bhanu Prakash, M.E., Kumar, S., 2021. PolInSAR decorrelation-based decomposition modelling of spaceborne multifrequency SAR data. Int. J. Rem. Sens. 42 (4), 1398e1419. Bhanu Prakash, M.E., Kumar, S., 2020. PolInSAR based polarimetric decomposition using cosine square distribution. In: 2020 7th International Conference on Signal Processing and Integrated Networks, SPIN 2020, pp. 465e470. Issue no: 3. Bhanu Prakash, M.E., Kumar, S., 2021. Multifrequency analysis of PolInSAR-based decomposition using cosine-squared distribution multifrequency analysis of PolInSAR-based decomposition using. IETE Tech. Rev. 0 (0), 1e8. Bhattacharya, A., Muhuri, A., De, S., Manickam, S., Frery, A.C., 2015. Modifying the Yamaguchi fourcomponent decomposition scattering powers using a stochastic distance. IEEE J. Sel. Top. Appl. Earth Obs. Rem. Sens. 8 (7), 3497e3506. Canada Centre for Remote Sensing, 2005. Advanced Radar Polarimetry Tutorial. www.ccrs.nrcan.gc.ca. Chen, S.W., Sato, M., 2011. Model based polarimetric decomposition using PolInSAR coherence. In: IEEE International Geoscience and Remote Sensing Symposium. IGARSS, pp. 1087e1090. Chen, S.W., Wang, X.S., Li, Y.Z., Sato, M., 2013. Adaptive model-based polarimetric decomposition using polinsar coherence. IEEE Trans. Geosci. Rem. Sens. 52 (3), 1705e1718. Chunxia, Z., Yu, Z., Dongchen, E., Zemin, W., Jiabing, S.U.N., 2012. Estimation of ice flow velocity of Calving glaciers SAR interferometry and feature tracking. In: Fringe 2011 Workshop, pp. 19e23 no. September. Cloude, S.R., 1989. Conditions for the physical Conditions for the physical realisability of matrix operators realisability of matrix operators in polarimetry. In: SPIE, pp. 177e185. Flynn, T., Tabb, M., Carande, R., 2002. Coherence region shape extraction for vegetation parameter estimation in polarimetric SAR interferometry. Int. Geosci. Remote Sens. Symp. 5 (C), 2596e2598. Fomena, R., Cloude, S., 2005. On the role of coherence optimization in polarimetric SAR interferometry. In: Pract. Freeman, A., Durden, S.L., 1998. A three-component scattering model for polarimetric SAR data. IEEE Trans. Geosci. Rem. Sens. 36 (3), 963e973. Joshi, S.K., Kumar, S., 2017. Performance of PolSAR backscatter and PolInSAR coherence for scattering characterization of forest vegetation using single pass X-band spaceborne synthetic aperture radar data. J. Appl. Remote Sens. 11 (2), 026022. Khati, U.G., 2014. Polinsar Based Scattering Information and Physical Property Retrieval of Vegetation. Andhra University. Kumar, S., Joshi, S.K., Govil, H., 2017. Spaceborne PolSAR tomography for forest height retrieval. IEEE J. Sel. Top. Appl. Earth Obs. Rem. Sens. 10 (12), 5175e5185.

Emerging techniques of polarimetric interferometric synthetic aperture radar for scattering-based characterization

Kumar, S., Garg, R.D., Govil, H., Kushwaha, S.P.S., 2019. PolSAR-decomposition-based extended water cloud modeling for forest aboveground biomass estimation. Rem. Sens. 11 (19), 1e27. Kumar, S., 2009. Retrieval of forest parameters from Envisat ASAR data for biomass inventory. In: Dudhwa National Park. University of Twente, UP., India. Lopez-Martinez, C., Pottier, E., 2015. Polarimetric Scattering Decomposition Theorems. https://www. univ-rennes1.fr/. Maurya, H., Kumar, A., Mishra, A.K., Panigrahi, R.K., 2020. Improved four-component based polarimetric synthetic aperture radar image decomposition. IET Radar, Sonar Navig. 14 (4), 619e627. Moreira, A., Prats-Iraola, P., Younis, M., Krieger, G., Hajnsek, I., Papathanassiou, K.P., 2013. A tutorial on synthetic aperture radar. IEEE Geosci. Remote Sens. Mag. 1 (1), 6e43. Neumann, M., Saatchi, S.S., Ulander, L.M.H., Fransson, J.E.S., 2012. Assessing performance of L- and P-band polarimetric interferometric SAR data in estimating boreal forest above-ground biomass. IEEE Trans. Geosci. Rem. Sens. 50 (3), 714e726. Richards, J.A., 2009. Remote Sensing with Imaging Radar: A Review. Springer, Verlag-Berlin Heidelberg. Sen Lee, J., Ainsworth, T.L., 2010. The effect of orientation angle compensation on coherency matrix and polarimetric target decompositions. Proc. Eur. Conf. Synth. Aperture Radar EUSAR 49 (1), 499e502. Singh, G., Yamaguchi, Y., 2018. Model-based six-component scattering matrix power decomposition. IEEE Trans. Geosci. Rem. Sens. 56 (10), 5687e5704. Singh, G., Yamaguchi, Y., Park, S.E., 2013. General four-component scattering power decomposition with unitary transformation of coherency matrix. IEEE Trans. Geosci. Rem. Sens. 51 (5), 3014e3022. Verma, R., 2012. Polarimetric Decomposition Based on General Characterisation of Scattering from Urban Areas and Multiple Component Scattering Model. University of Twente. Wangensteen, B., Weydahl, D.A.N.J., Hagen, J.O.N.O.V.E., 2005. Mapping glacier velocities on Svalbard using ERS tandem DInSAR data. Nor. J. Geogr. 59, 276e285. Xiang, D., Ban, Y., Su, Y., 2016. The cross-scattering component of polarimetric SAR in urban areas and its application to model-based scattering decomposition. Int. J. Rem. Sens. 37 (16), 3729e3752. Yamaguchi, Y., Sato, A., Boerner, W.-M., Sato, R., Yamada, H., 2011. Four-component scattering power decomposition with rotation of coherency matrix. IEEE Trans. Geosci. Rem. Sens. 25 (2), 1e8. Zhang, L., Zou, B., Cai, H., Zhang, Y., 2008. Multiple-component scattering model for polarimetric SAR image decomposition. Geosci. Rem. Sens. Lett. IEEE 5 (4), 603e607.

285

This page intentionally left blank

CHAPTER 15

Advanced method for radar remote sensing: circularly polarized synthetic aperture radar Josaphat Tetuko Sri Sumantyo1, 2 1

Center for Environmental Remote Sensing, Chiba University, Chiba, Japan; 2Fakultas Teknik, Universitas Sebelas Maret, Solo, Indonesia

1. Introduction Synthetic aperture radars (SARs) are well-known as multipurpose sensors that can operate in all weather and in the daytime and nighttime (Cumming and Wong, 2005; Soumekh, 1999; Maitre, 2008). Various linear polarized SARs were developed for satellite, airborne, and unmanned aerial vehicles for some applications (Gustavsson et al., 1993; Reigber et al., 2013; Edrich and Weiss, 2008; Essen et al., 2012; Rosen et al., 2007). Compact polarimetric SAR (CP SAR) offers a better option for many polarimetric SAR applications by considering the need for a compromise among swath coverage, hardware requirements, and scattering information of measured scenes (SAR, 2019). CP SAR is a technique that allows the construction of pseudoquadrature polarization (QP) information from dual polarization (DP) SAR systems, in which the CP has the promise of being able to reduce the complexity, cost, mass, and data rate of an SAR system while attempting to maintain many capabilities of a fully polarimetric (FP) or QP system (Nord et al., 2009). As a special DP configuration, CP SAR transmits signals in a special polarization, such as 45-degree linear polarization (LP) or circular polarization, and receives echoes simultaneously in a two orthogonal polarization such as horizontal polarization and vertical polarizations or right-handed circular polarization (RHCP) and left-handed circular polarization (LHCP). Such systems can provide a greater amount of information on polarization than standard DP linear systems while covering twice the swath width of conventional QP systems and maintaining the benefits (e.g., large ranges of feasible incidence angles and a smaller energy budget) that QP systems cannot provide (SAR, 2019; Fernandez et al., 2008). In the application assessment, land use and land cover classifications from hybrid CP SAR or DP data achieve accuracy levels comparable to those from a QP SAR to within a few percent, for which CP SAR has the advantages of large-scale coverage and a compact data volume (Raney, 2016; Ohki and Shimada, 2018; Dasari and Lokam, 2018). CP SAR is considered a suitable option for spaceborne radar such as RISAT-1, Advanced Land Radar Remote Sensing ISBN 978-0-12-823457-0, https://doi.org/10.1016/B978-0-12-823457-0.00016-1

© 2022 Elsevier Inc. All rights reserved.

287

288

Radar Remote Sensing

Observing Satellite (ALOS-2), and the future RADARSAT constellation. This significantly has increased interest in the remote sensing community in the CP SAR domain and posed new scientific and technological problems involving radar systems, data interpretation, performance assessment, and practical applications. As mentioned, CP SAR systems could construct QP information with the advantages of a simpler system, lower data rate, and wider swath, compared with a QP system. Thus, this pseudo-QP information reconstructed from CP measurements under certain conditions can be used similar way to FP data in polarimetric applications such as polarimetric synthesis and polarimetric decomposition. To get undistorted polarimetric information, the calibration technique of CP measurements was proposed by several researchers (Truong-Loï et al., 2010; Chen and Quegan, 2011; Tan and Hong, 2016; Babu et al., 2019; Sabry, 2018). First, Freeman and Chen proposed methods for CP SAR calibration using merely calibrators, which require at least three calibrators in a range line and require many more calibrators when monitoring changes to the system across the swath (Truong-Loï et al., 2010; Chen and Quegan, 2011). FP SAR calibration methods exploring the scattering characteristics of natural distributed targets (DTs) can reduce the number of deployed calibrators needed. However, it is much more complicated in the case of CP SAR calibration, because only two channels of CP measurements are available. Second, a numerical CP calibration method using natural DTs and only one corner reflector (CR) for a reciprocal CP SAR system was proposed by Tan and Hong (2016). The validity of the method for three typical CP modes was confirmed by simulations with ALOS-2 The Phased Array type L-band Synthetic Aperture Radar (PALSAR-2) data. In the sensitivity analysis, a dihedral at 0 degree was selected as the best CR for the calibration algorithm. Third, CP calibration of the RISAT-1 dataset using a combination of trihedral and dihedral CRs with cloud CP decomposition was employed to assess the ground target characterization quality of the dataset before and after polarimetric calibration to improve system-induced polarimetric distortions, which include channel imbalances, phase bias, cross-talk between channels, and ionospheric distortion (Babu et al., 2019). Fourth, transmit wave polarization (i.e., ellipticity) deviation from the ideal or intended transmit mode affects CP products and associated data exploitation. Sabry (2018) evaluated this effect on the CP model and account for associated variations through data exploitations, proposing an approach to model and estimate ellipticity variations for calibration by taking advantage of a variational model and adjoining polarimetric SAR data using RADARSAT-2 data. As discussed earlier, several existing CP systems were developed to generate CP SAR data. Calibration techniques were also proposed and developed using LP and combinations of it (Truong-Loï et al., 2010; Chen and Quegan, 2011; Tan and Hong, 2016; Babu et al., 2019; Sabry, 2018). Therefore, a C-band broadband (maximum of 400 MHz bandwidth) circularly polarized SAR system was proposed to enrich existing CP SAR systems, as well as a preliminary investigation into the calibration technique of CP SAR using a conventional CR for further investigation into calibration techniques of CP SAR.

Advanced method for radar remote sensing: circularly polarized synthetic aperture radar

2. Circularly polarized scattering for remote sensing This section discusses scattering characteristics of general polarization called circular polarization (Fukusako, 2016) in circularly polarized SAR. This type of polarization includes linear, elliptical, and circular polarization. The axial ratio (AR) of linear and circular polarization is infinity and 1, respectively. Fig. 15.1 shows a scattering model using a plane wave that shows the scattering and transverse waves on the dielectric discontinuity surface. qr ¼ qi h sinqt ¼ 1 sinqi h2

(15.1) (15.2)

p and s indexes (subscript) show the parallel and perpendicular components of reflection and transmission coeficient. The Fresnel reflection (Rp and Rs) and transmission (Tp and Ts) coefficient are: Rp ¼

h2 cosqt  h1 cosqi h2 cosqt þ h1 cosqi

(15.3)

Tp ¼

2h2 cosqi h2 cosqt þ h1 cosqi

(15.4)

Rs ¼

h2 cosqi  h1 cosqt h2 cosqi þ h1 cosqt

(15.5)

Ts ¼

2h2 cosqi h2 cosqi þ h1 cosqt

(15.6)

Figure 15.1 Scattering model of circular polarization. CP SAR, compact polarimetric synthetic aperture radar. (From Josaphat Microwave Remote Sensing Laboratory.)

289

290

Radar Remote Sensing

where hn (n ¼ 1, 2) is the characteristic impedance of each medium: sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi jumn hn ¼ sn þ juεo εrn

(15.7)

where n ¼ 1, 2. If incident angle qi increases from 0, Rp will be 0 at angle qB (Brewster’s angle), as shown in Fig. 15.2. When qi ¼ qB , the transmitted wave and reflected wave k t ¼ 0. If Rp ¼ 0, mr1 ¼ mr2 ¼ 1, s1 ¼ s2 ¼ 0, Brewster’s are perpendicular or b k r $b angle qB is: rffiffiffiffiffiffi 1 εr2 qB ¼ tan (15.8) εr1 As one of the applications, the proposed C-Band circularly polarized SAR is considered for monitoring a tropical forest fire in the Indonesian peatland area. A soil sample of peatland was collected in a ground survey at Siak District, Sumatera Island, Indonesia. The dielectric constant or permittivity of the peatland soil was measured using Agilent Vector Network Analyzer 8510C and the open air probe technique of the Agilent 85070E Dielectric Tool Kit in our laboratory. The acquired value was 3.1 at a frequency of 5.3 GHz (C-band). Hence, εr1 and εr2 are considered to be air and peatland permith iT tivity with values of 1 and 3.1, respectively. If incident wave E i ¼ Epi Esi , reflected h h iT iT wave E r ¼ Epr Esr , and transmitted wave Et ¼ Ept Est , the relationship is: "

Esr Epr

#

" ¼

#" Rs 0

0 Rp

Esi Epi

# (15.9)

Figure 15.2 Fresnel coefficient of C-band circular polarization. (From Josaphat Microwave Remote Sensing Laboratory.)

Advanced method for radar remote sensing: circularly polarized synthetic aperture radar

"

Est Ept

#

" ¼

#" Ts 0

0 Tp

Esi

#

Epi

(15.10)

where Ep and Es are two perpendicular components of a circular polarized wave. The relationship of these components and circular polarized components, EL and ER , is: " #" # " # Es EL 1 1 j ¼ pffiffiffi (15.11) ER 2 1 j Ep Then, Ep and Es can be expressed by: #" " # " # Es EL 1 1 1 ¼ pffiffiffi Ep ER 2 j j

(15.12)

Reflection of circular polarization on boundary condition could be derived as " r # " #" #" #"    i  EL ELi 1 1 j Rs 0 1 1 EL R R c x ¼ ¼ R R (15.13) i i r 0 R 1 j j j x c E 2 p ER ER R where  1 Rs  Rp 2  1 Rx ¼ Rs þ Rp 2 Rc ¼

(15.14) (15.15)

Rc is the reflection coefficient of main polarization (co-polarization) of circular polarization, and Rx is the reflection coefficient of cross-polarization. Transmission of circular polarization on boundary condition could be derived as " t # " #"  EL ELi Tc Tx ¼ (15.16) Tx Tc ERi ERt where  1 Ts þ Tp 2  1 Tx ¼ Ts þ Tp 2 Tc ¼

(15.17) (15.18)

Tc is the transmission coefficient of main polarization (co-polarization) of circular polarization, and Tx is the transmission coefficient of cross-polarization.

291

292

Radar Remote Sensing

Fig. 15.2 shows that Rp is zero at qB (Brewster’s angle). Rp and Rs are negative at qi < qB , which shows that the reflected wave changed phase 180 degrees. Rp changed to be positive and Rs remained negative. Tp and Ts are positive, which means that the phase does not change. In the case where qi ¼ qB , as shown on Fig. 15.3, Rc and Rx of Eqs. (15.14) and (15.15) have the same value, in which the scattered wave’s intensity of the main polarized (co-polarization) wave and cross-polarized wave are the same. This combination generates a linear polarized wave, as shown by the infinite axial ratio in Fig. 15.4. In this case, the scattered wave has only Rs , which is shown by Rp with a zero value in Fig. 15.2. This means that only perpendicular electric field component Esr scatters on the surface; parallel component Epr does not scatter.

Figure 15.3 Reflection and transmission coefficient of C-band circular polarization. (From Josaphat Microwave Remote Sensing Laboratory.)

Figure 15.4 Axial ratio of circular polarization. (From Josaphat Microwave Remote Sensing Laboratory.)

Advanced method for radar remote sensing: circularly polarized synthetic aperture radar

For qi < qB , as shown in Fig. 15.3, Rc and Rx have a negative value, but the absolute value of Rx is larger than Rc . This means that the cross-polarized wave is dominant after being scattered. It shows that electric fields perpendicular and parallel components phases are different, and then there is a 180-degree switch after the waves are scattered, as shown in Fig. 15.2, in which Rs and Rp are negative. If we consider the direction of propagation after scattering, it has the opposite direction, and circular polarized wave after scattering has the opposite sense or polarization. If Tc and Tx have positive and negative values, respectively, as shown in Fig. 15.3, the main polarized (co-polarization) wave penetrated in the medium without a phase change, but the cross-polarized wave penetrated in the medium with a 180-degree phase change. The absolute value of Tc is larger than Tx , which means the main polarized wave (co-polarization) is predominant in penetrating the medium. In the case where qi > qB , as shown in Fig. 15.3, the absolute value shows that Rc and Rx have negative values, but the absolute value of Rc is larger than that of Rx . This means that the main polarized (co-polarization) wave is predominant after scattering. The Esr component has the opposite direction after scattering and Epr component keeps the same direction, as shown by the negative Rs and positive Rp in Fig. 15.2. Thus, the sense of scattered circular polarization is the same. The absolute value of Rc is larger than that of Rx , so the main polarized (co-polarization) wave of incident wave is predominant. As shown in Fig. 15.3, the relationship of Tc and Tx under the conditions of qi > qB has same result as for the conditions of qi < qB , which means main that the polarized (co-polarization) wave penetrated in the medium without a phase change, but the cross-polarized wave penetrated in the medium with a 180-degree phase change. The absolute value of Tc is larger than Tx , which means the main polarized (co-polarization) wave is predominant in penetrating the medium. We assume the axial ratio (AR) of the incident wave is 0 dB or perfectly circular polarization, in which the axial ratio of the scattered wave can be calculated as:   Rp  (15.19) ARðdBÞ ¼ 20 log  Rs As incident angle qi gets closer to 0 degrees, the incident wave sense changes to the opposite sense, or the incident wave becomes cross-polarized. The axial ratio of the scattered wave gets closer to 0 dB and the polarization tends to be circular, as shown in Fig. 15.4. However, for an incident angle qi larger than 35 , the axial ratio is larger than 3 dB and the polarization of scattered wave becomes elliptically polarized at an incident angle of 35 , < qi < qB . The axial ratio reaches its largest value at qi ¼ qB , or LP, in which only perpendicular wave scatters and LP with a perpendicular direction remain after scattering. At qi > qB , the scattering wave closes parallel to the scattered plane and the axial ratio decreases to become circular polarization. In this case, the rotational

293

294

Radar Remote Sensing

Figure 15.5 Scattering polarization of circular polarized waves. LHEP, left-handed circular polarization; LP, linear polarization; RHCP, right-handed circular polarization. (From Josaphat Microwave Remote Sensing Laboratory.)

direction or polarization of scattered wave is the same as the polarization of the incident wave or main wave. In Figs. 15.4 and 15.5, the sense of waves changes to the opposite sense or changes to become a cross-polarized wave at incident angle qi close to 0 , and the axial ratio of the scattered wave is close to 0 dB or to circular polarization. However, at an incident angle qi larger than 35 , as shown in Fig. 15.4, the axial ratio is larger than 3 dB and the polarization of the scattered wave is elliptical polarization at an incident angle of 35 qB, the scattering wave is close to parallel to the scattered plane and the axial ratio decreases to become circular polarization. In this case, the rotation direction or polarization of the scattered wave is the same as the polarization of the incident wave or main wave. The transmission or penetrated wave has same polarization as the incident wave, with an axial ratio less than 3 dB at an incident angle of less than 35o , as shown in Fig. 15.4. Based on these results, the design of the proposed circularly polarized SAR must be efficient in power consumption and keep a stable axial ratio for circular polarization, and we must employ a low nadir angle or incidence angle (less than 35o ) to reduce the transmitting power, particularly for the mission of C-band.

Advanced method for radar remote sensing: circularly polarized synthetic aperture radar

3. Specification of circular polarized synthetic aperture radar for microsatellite The circular polarized SAR onboard microsatellite was designed by referring to Bickel et al. (1993), Tomiyasu (1978), Curlander and McDonough (1991), and Sri Sumantyo et al. (2017) with the Stripmap mode. Based on the results of the analysis of circular polarization in the previous section, we consider incidence angle qi to be 20 e35 , with a center frequency fc of 5.3 GHz, orbit altitude h of 500 km, receiver temperature T of 300 K, noise figure (NF) F of 3 dB, ground target normalized radar cross-section so of 25 dB, signal-to-noise ratio (SNR) of 25 dB, diameter of mesh parabolic antenna D of 3.6 m, antenna efficiency h of 80%, system loss Ls of 3 dB, pulse length sp of 50 ms, and bandwidth B of 100e400 MHz. Fig. 15.6 shows the analytical results of the relationship of the incidence angle and pulse repetition frequency (PRF). We chose the mission parameter or the PRF on the white area with a PRF value larger than the minimum PRF and less than the maximum PRF, avoiding a blind range or eclipsing (yellow area), and a nadir return or echo (green area). The results of this analysis show that the isotropic gain (the ideal value was calculated or simulated using a rigid metal reflector) was 45 dBic, the beamwidth at range and azimuth direction is 1 with swath width coverage of 8000e12,500 m on the ground for an incidence angle of 20 e35 with a peak transmit power of 1500 W. Table 15.1 shows the specification parameters of the microsatellite onboard circular polarized SAR. These parameters were employed to develop the mesh parabolic antenna and radio-frequency (RF) system of circular polarized SAR, as discussed subsequently.

Figure 15.6 Relationship of incidence angle and pulse repetition frequency (PRF). (From Josaphat Microwave Remote Sensing Laboratory.)

295

296

Radar Remote Sensing

Table 15.1 Specifications of circular polarized synthetic aperture radar onboard microsatellite. Parameter

Value

Altitude Mode of observation Carrier/center frequency Bandwidth Pulse width Slant range resolution Pulse repetition frequency Receiver gain Receiver noise figure Antenna polarization Antenna diameter Antenna isotropic gain (ideal) Antenna beamwidth

500 km Stripmap 5.3 GHz (C-band) 100e400 MHz 10e100 ms 1.8e3.6 m 400 Hz to 10,000 Hz 50 dB 3 dB Full circular polarization (LL, RR, LR, and RL) 3.6 m 45 dBic 1 (range) 1 (azimuth) 25 dB 25 dB 20 e35 8000e12,500 m 1500 W

Signal-to-noise ratio Normalized radar cross-section Incident angle Ground swath width Peak transmit power

The circular polarized SAR was designed to transmit and receive LHCP and RHCP. The main mission of the circularly polarized SAR is to conduct basic research on multipolarization scattering and its application developments. Fig. 15.7 shows the concept of the circular polarized SAR sensor. This sensor transmits only one polarization (RHCP or LCHP) and receives RHCP and LHCP scattering waves simultaneously. Fig. 15.8A shows a bird’s-eye view of the structure of the circular polarized SAR onboard microsatellite. The cruising direction is the X axis (roll), the earth-pointing direction is the Z axis (yaw), and the pitch direction is the Y axis. The antenna is composed of 24 ribs to fix the mesh sheet with the parabolic shape of the reflector. The gold-coated molybdenum string woven mesh grid is 1 mm. The flatness of the mesh surface is less than 5 mm, and we fabricated it to be 0.5e0.8 mm (rms) based on laser measurements. The support tower of the feeder is made of aluminum. Fig. 15.8BeD shows views of the circularly polarized SAR onboard microsatellite and the size of the microsatellite envelope and modules. Fig. 15.9 shows modules of the circularly polarized SAR onboard microsatellite with the circularly polarized SAR sensor module, attitude control module, communication module, onboard computer module, and power supply module as the five main modules. Because the sensor has no electronic beam steering, the microsatellite must be rolled into the necessary position by controlling four reaction wheels to point to the desired target. The interface of each module employs a 12 V and 24 V power interface. Communication between the SAR mission unit and satellite bus uses RS422 and TCP/IP.

Figure 15.7 Concept of circular polarized synthetic aperture radar onboard microsatellite mission for earth observation. LHCP, left-handed circular polarization; TX, transmitter; RHCP, right-handed circular polarization; RX, receiver. (From Josaphat Microwave Remote Sensing Laboratory.)

Figure 15.8 Structure of circular polarized synthetic aperture radar onboard microsatellite: (A) Bird’s-eye view; (B) Side view; (C) Bottom view; and (D) Top view. (From Josaphat Microwave Remote Sensing Laboratory.)

298

Radar Remote Sensing

Figure 15.9 Modules of circularly polarized SAR onboard microsatellite: (A) Communications (Comm.) and attitude control modules; (B) Power supply and controller modules; and (C) circularly polarized synthetic aperture radar modules. (From Josaphat Microwave Remote Sensing Laboratory.)

Advanced method for radar remote sensing: circularly polarized synthetic aperture radar

The payload of the circularly polarized SAR sensor is 53.9 kg and the satellite bus system is 64.8 kg 118.7 kg (in total). Therefore, the microsatellite is a 150-kg class microsatellite. The satellite has the sun-synchronous orbit (SSO) with an altitude of 500 km with an inclination 97.6 of degrees. Based on a power budget calculation in each season, for every five rotations of the earth, the circularly polarized SAR sensor could be operated for 0.25 min of earth observation mode (800e1500 W) and 10 min of SAR sensor adjustment mode (120 W). For every eight rotations of the earth, the microsatellite could have communications with the ground station for a maximum of 15 min (120 W) and adjustment mode (90 W). In normal mode, the microsatellite consumes 90 W. For this purpose, we employed a nickelemetal hydride (NiMH) battery weighing 8 kg. As shown in Fig. 15.9C, the circularly polarized SAR module is located on the upper part of microsatellite, close to the installed mesh parabolic antenna, to reduce cable loss, noise, and heat that could influence the microsatellite. The rest of the modules are installed in the lower part of microsatellite in space of 0.8  0.8  0.4 m that close to interface of rocket. Several previous spaceborne parabolic antennas employed were of the umbrella type, in which ribs were installed perpendicular to the center of the antenna; they need a hinge or deployment module for each rib (Sharma et al., 2013; Imbriale; Imbriale et al., 2012). This increased the weight of the satellite and caused a serious issue for the microsatellite. Therefore, in this research, the main antenna of the circularly polarized SAR onboard microsatellite is proposed to be the wrap-tib type mesh parabolic antenna, as shown in Fig. 15.8BeD. This antenna has a mesh surface attached to exploded wrap-ribs. The ribs are stored in an envelope of microsatellite during launch, which deploy for full function as reflectors of the parabolic antenna. Hence, flexible and strong ribs and a mesh surface are required for the antenna of a microsatellite onboard circularly polarized SAR. The ribs are deployed with the elasticity of the material and the structure of rib itself. The mechanical and electrical deployment system was not considered in this mission to realize a lightweight and facile deployable antenna for the microsatellite.

4. Radio-frequency system of circular polarized synthetic aperture radar The proposed circularly polarized SAR system operates in a stripmap mode and adopts the pulsed linear frequency modulated modulation radar principle. Table 15.1 lists the specifications of the C-band circularly polarized SAR system. The operating frequency of the system centers on 5.3 GHz (C-band) with an adjustable bandwidth of 100e400 MHz, resulting in 37.5 cm of the finest slant range resolution. The system is intended for a flight mission operating at an altitude of 500e4000 m, with an incident angle of 25e60 degrees. To maximize gain of compression from signal processing, the operating pulse width and PRF are adjustable, from 0.1 to 100 ms and 100 Hz to 10 kHz, respectively. During stripmap SAR scanning, the circularly polarized SAR system can scan continuously for 65.536 s.

299

300

Radar Remote Sensing

The circularly polarized SAR RF transmitter can transmit a peak power of 280 W. The total gain in the receiver is 50 dB with an NF of 2.5 dB. Four circularly polarized antennas, two units with LHCP and two with RHCP, were designed and fabricated in our laboratory. Each fabricated antenna has approximately 22 dBic of isotropic gain, with a beamwidth of approximately 13 degrees in the range direction and 6 degrees in the azimuth direction. Fig. 15.10 illustrates the top-level architecture and interconnection of the circularly polarized SAR system, which is composed of four major subsystems: the RF subsystem,

Figure 15.10 Architecture and interconnection of circular polarized synthetic aperture radar (SAR) onboard microsatellite: (A) Architecture of C-band circularly polarized SAR system; and (B) interconnection diagram of C-band circularly polarized SAR system. AWG, arbitrary waveform generator; BCU, baseband and control unit; EXT CLK, external clock; Ich, in-phase channel; INU, inertial navigation unit; LAN, local area network; LO, local oscillator; PDU, power distribution unit; TC, time controller; TCU, timing and control unit; UART, Universal Asynchronous Receiver/Transmitter; UPS, Uninterruptible Power Supply; Qch, quadrature channel, RX, receiver; STALO, Stable Local Oscillator. (From Josaphat Microwave Remote Sensing Laboratory.)

Advanced method for radar remote sensing: circularly polarized synthetic aperture radar

the baseband and control unit (BCU), the power distribution unit (PDU), and the inertial navigation unit (INU), as shown on Fig. 15.10A. The RF transmitter of the circularly polarized SAR adopts a two-stage superheterodyne architecture for ease of image frequency removal during up-conversion. Fig. 15.10B shows the block diagram of the RF transmitter. The quadrature baseband chirp (I and Q channel) is first up-converted to an intermediate frequency (IF) at 2850 MHz. The IF chirp (centered at 2850 MHz) is then upconverted to the desired carrier frequency at 5.3 GHz in the second up-conversion stage by mixing the IF chirp with a 2450-MHz local oscillator (LO) signal. LO sources used in up-conversion are from the RF receiver, so that coherency between the two RF subsystems can be properly retained. The direct current (DC) offset for each channel of the baseband chirp is adjustable. The purpose is to minimize leakage of the carrier signal to the output of the transmitter during up-conversion. The low-power, up-converted chirp is amplified using a power amplifier (PA) to a moderate power level. Then, the moderate power chirp is gated to a gate control signal (TX trigger gate) generated by the timing and control unit (TCU). The power level of the gated signal can be adjusted for lower transmitting peak power by an adjustable attenuator (0e31 dB). The gated moderate power chirp is finally amplified to the transmitting power level using a cascaded solid-state power amplifier (SSPA). A circulator is placed between the SSPA and PA to minimize reflected power from the SSPA caused by the impedance mismatch between the SSPA and the PA. The transmitter has a total gain of 58 dB and is clean of spurious signals (as low as 65 dB). The RF receiver adopts the two-stage superheterodyne receiver architecture. The clock synthesizer in the receiver generates all necessary clock signals for the entire circularly polarized SAR system. It has three oscillators: (1) a 10-MHz temperature-compensated crystal oscillator (TXCO) to generate the reference clock signal, (2) a 2850-MHz LO to generate the IF LO signal, and (3) a 2450-MHz LO to generate the second LO signal. The output of the TXCO is split into five branches. These split clock signals are used to synchronize the clock source of the individual subcomponents in the circularly polarized SAR system. In the receiver, the RF signals (echoes) collected by the antenna are amplified to an adequate power level using several cascaded low noise amplifiers (LNAs) with a total gain of 58 dB. A circulator is placed between the receiving antenna and the LNA so that impedance mismatch between the LNA module and the receiving antenna is minimized. The 1-dB compression point of the front-end LNA is 50 dBm and can withstand (without a breakdown) a maximum input (continuous) power of þ30 dBm. To avoid the saturation or breakdown of the LNA caused by the coupled transmitting signal, the receiver is switched off by the gating network when the transmitter is transmitting. The amplified echoes are down-converted to the IF band using a single side-band mixer and then to the baseband by a quadrature demodulator. The IF amplifier amplifies

301

302

Radar Remote Sensing

the baseband echoes and finally outputs them for digitization. Before digitization, the DC offset generated during down-conversion is removed. Overall, the NF of the receiver is 4.5 dB, with a spurious signal level of less than 65 dB and image rejection of greater than 30 dB. The BCU handles several critical tasks in the circularly polarized SAR operation. The BCU is built from a workstation consisting of (1) an arbitrary waveform generator (AWG), (2) a high-speed digitizer, and (3) two units of a high-speed solid-state drive and a TCU. All components except the TCU are commercial off-the-shelf products; the TCU is a custom-designed digital system built on a field programmable gate array fabric. As a brief functional summary of the BCU, it can (1) generate two channels (In-phase I and Quadrature Q) with a 1.2-GHz maximum bandwidth chirp signal in which the pulse width of the chirp is 0.5e100 ms, (2) digitize two channels (I and Q) of a 1-GHz maximum bandwidth chirp echoes with onboard 4-G samples of recording buffer, (3) generate accurate and precise timing and control signals with the finest resolution of 10 ns, and (4) synchronize and control the operation of all submodules and maintain the system’s clock coherence among the submodules in the circularly polarized SAR system. A Signatec PXDAC4800D-DP AWG was employed as AWG hardware, and a Gage-Applied Razormax CSE161G2 High-Speed Digitizer as the high-speed digitizer. The PDU was built from several uninterruptible power supplies (UPSs) and AC to DC converters. The PDU incorporates several large-capacity UPSs into a large pool of power source. It is designed to supply a clean and stable AC current source to power up the entire circularly polarized SAR system for more than 3 h of continuous operation without the need to charge the source. The in-system AC to DC converters draw current from this pool of power source and convert the AC source into a few DC output (þ28 Vdc, þ12 Vdc, and þ5 Vdc) to power up the RF transmitter, the RF receiver, the TCU, and the INU. The INU is important for recording the actual flight attitude of the aircraft during the data collection mission. Actual flight information is needed to process recorded SAR echoes into a useful informative SAR image. The INU of the circularly polarized SAR system is composed of a u-blox EVK-5T GPS and a Memsic AHRS440CA-200 inertial measurement unit. The EVK-5T is a cost-effective, high-performance u-blox 5-based LEA-5 series of GPS modules with a compact 17.0  22.4-mm form factor featuring precision timing and an innovative jamming-resistant RF architecture. Moreover, the AHRS440CA-200 is a compact standalone attitude and heading reference system that provides roll, pitch, and yaw measurement data in both static and dynamic environments. Centralized GUI software was developed to control each subsystem (AWG, digitizer, TCU, transmitter, receiver, and INU) and record raw data (SAR echoes and flight information).

Advanced method for radar remote sensing: circularly polarized synthetic aperture radar

5. Flight test and images 5.1 Flight test of circular polarized synthetic aperture radar system The performance of the circularly polarized SAR system was further tested in the Hinotori-C2, mission which was the maiden flight of the circularly polarized SAR system on March 3e16, 2018, in Makassar, Indonesia. The main objectives of the flight test were to verify the functionality of the circularly polarized SAR system and to acquire full polarimetric circularly polarized SAR images covering several types of natural and artificial targets such as CRs for calibration, man-made buildings, water bodies, ships, agricultural areas, forests, and so forth. Fig. 15.11A shows the CASA/IPTN CN235MPA airplane used in the flight test of the Hinotori-C2 mission. Four circularly polarized antennas (two units of LHCP and twos of RHCP) were installed in the nose cone radome of the airplane. The flight altitude was about 1000e1500 m, which yielded 676.2e1014.3 m of swath width (in range direction), finally the incidence angle was fixed at 54 in the mission (Fig. 15.11B). Meanwhile, the circularly polarized SAR system was firmly installed in the cabin area of the airplane (Fig. 15.11C). The antenna or polarization isolation of each antenna was measured to be 40 to 50 dB. We also connected the receiver antenna to 16 dBm of LNA to increase the gain of the receiver and SNR of the circularly polarized SAR system.

5.2 Images of circularly polarized signal-to-noise ratio Fig. 15.12A shows circularly polarized images collected from the Hinotori-C2 mission on March 14, 2018 with various modes (LL, RL, LR, and RR modes) and its analysis of the study area at Hasanuddin International Airport employing a bandwidth of 200 MHz. The figure shows the full polarimetric mode (LL, RL, LR, and RR) of circularly polarized SAR images observed by multipath or multiflight by CN235MPA. The results show that the cross-polarized images have strong scattering compared with the co-polarized images, the same as in our previous investigation in an anechoic chamber using an N219 aircraft model as the target, and as reported in Sri Sumantyo et al. (2017). We investigated the performance of circularly polarized scattering using a bandwidth of 400 MHz (37.5 cm resolution) with a paddy field and settlement as the study area, as shown on Fig. 15.12B. This figure shows a clear image of the study area, where we could detect a pillar inside a building. Circularly polarized SAR can be implemented for building investigations and maintenance. The condition of the paddy field could also be observed by using the sensor. Therefore, circularly polarized SAR could be assessed for estimating paddy grain volume. Details of the forest or vegetated area could be clearly observed; hence, we could implement circularly polarized SAR for biomass estimation.

303

304

Radar Remote Sensing

Figure 15.11 The circularly polarized synthetic aperture radar (SAR) onboard CN235MPA in Hinotori-C2 mission: (A) CN235MPA aircraft for Hinotori-C2 mission; (B) Antenna installation in nose cone radome of CN235MPA for full polarimetric circularly polarized SAR; (C) Circularly polarized SAR onboard CN235MPA in Hinotori-C2 mission. (From Josaphat Microwave Remote Sensing Laboratory.)

Advanced method for radar remote sensing: circularly polarized synthetic aperture radar

(a)

(b)

(c)

(d)

Figure 15.12 Images of circularly polarized synthetic aperture radar: (A) Full circularly polarized images with LL, LR, RL, and RR modes obtained in Hinotori-C2 mission with a bandwidth of 200 MHz; (B) Paddy field (bandwidth of 400 MHz, LR mode); (C) Pangkep City (bandwidth of 200 MHz, RL mode); and (D) Makassar City (bandwidth of 200 MHz, RR mode). (From Josaphat Microwave Remote Sensing Laboratory.)

Fig. 15.12C and D show the coastal area at Pangkep City and Makassar using the RL and RR modes, respectively, with a bandwidth of 200 MHz. This investigation with different polarizations showed good images with 75 cm resolution. The interface of the water body and land is clearly seen, and industrial areas, settlements, forests, and paddy fields can be easily classified.

6. Summary and future research This chapter explained C-band circularly polarized SAR onboard a microsatellite for environment and disaster monitoring. The chapter also describes the characteristics of circularly polarized scattering in circularly polarized SAR to derive the Fresnel coefficient, reflection and transmission coefficient, and axial ratio; the system configuration of the RF system; and the flight test of circularly polarized SAR in the Hinotori-C2

305

306

Radar Remote Sensing

mission onboard CN235MPA on March 14e15, 2018 in Makassar, Indonesia. The results of the flight test were full polarimetric circularly polarized images showing the performance of circularly polarized SAR. In the Hinotori-C2 mission, we also collected images of paddy fields, settlements, industrial areas, and ships. In the future, we plan to implement the results of this circularly polarized SAR system for our microsatellite, extracting information about full polarimetric images of circularly polarized SAR for agriculture, forests, urban areas, infrastructures, disaster monitoring, and more. Several Hinotori missions will be done by improving the antenna, output power for higher altitude operation, bandwidth expanding to improve the resolution, and so forth.

References Babu, A., Kumar, S., Agrawal, S., October, 2019. Polarimetric calibration of RISAT-1 compact-pol data. IEEE J. Sel. Top. Appl. Earth Obs. Rem. Sens. 12 (No.10), 3731e3736. Bickel, D.L., Brock, B.C., Allen, C.T., March, 1993. Spaceborne SAR Study : LDRD ’92 Final Report, Sandia Report SAND93-0731 e UC e 706. Chen, J., Quegan, S., July, 2011. Calibration of spaceborne CTLR compact polarimetric low-frequency SAR using mixed radar calibrators. IEEE Trans. Geosci. Rem. Sens. 49 (No. 7), 2712e2723. Cumming, I.G., Wong, F.H., 2005. Digital Processing of Synthetic Aperture Radar. Artech House, Boston. Curlander, J.C., McDonough, R.N., 1991. Synthetic Aperture Radar e Systems and Signal Processing. Wiley Publisher. Dasari, K., Lokam, A., October, 2018. Exploring the capability of compact polarimetry (hybrid pol) C band RISAT-1 data for land cover classification. IEEE Access 6, 57981e57993. Edrich, M., Weiss, G., December, 2008. Second-generation Ka-band UAV SAR system. In: The 38th European Microwave Conference (EuMC 2008). Essen, H., Johannes, W., Stanko, S., Sommer, R., Wahlen, A., Wilcke, J., July, 2012. High resolution W-band UAV SAR. In: IEEE International Geoscience and Remote Sensing Symposium (IGARSS 2012). Fernandez, P.C.D., Souyris, J.C., Angelliaume, S., Garestier, F., October, 2008. The compact polarimetry alternative for spaceborne SAR at low frequency. IEEE Trans. Geosci. Rem. Sens. 46 (No.10), 3208e3222. Fukusako, T., November, 2016. Basic of circularly polarized antenna. In: The 56th Antenna and Propagation Design and Analysis Method Workshop. The Institute of Electronics, Information, and Communication Engineers (IEICE). Gustavsson, A., Frolind, P.O., Hellsten, H., Jonsson, T., Larsson, B., Stenstrom, G., August, 1993. The airborne VHF SAR system CARABAS. In: Proceedings on IEEE International Geoscience and Remote Sensing Symposium (IGARSS ’93), pp. 18e21. W.A. Imbriale, Spaceborne Antennas for Planetary Exploration, Wiley Inter. Imbriale, W.A., Gao, S., Boccia, L., 2012. Space Antenna Handbook. Wiley. Maitre, H., 2008. Processing of Synthetic Aperture Radar Images. Wiley, London. Nord, M.E., Ainsworth, T.L., Lee, J.S., Stacy, N.J.S., January, 2009. Comparison of compact polarimetric synthetic aperture radar modes. IEEE Trans. Geosci. Rem. Sens. 47 (No.1), 174e188. Ohki, M., Shimada, M., September, 2018. Large-area land use and land cover classification with quad, compact, and dual polarization SAR data by PALSAR-2. IEEE Trans. Geosci. Rem. Sens. 56 (No.9), 5550e5557. Raney, R.K., June, 2016. Comparing compact and quadrature polarimetric SAR performance. Geosci. Rem. Sens. Lett. IEEE 13 (No.6), 861e864.

Advanced method for radar remote sensing: circularly polarized synthetic aperture radar

Reigber, A., Scheiber, R., Jager, M., Iraola, P.P., Hajnsek, I., Jagdhuber, T., Papathanassiou, K.P., Nannini, M., Aguilera, E., Baumgartner, S., Horn, R., Nottensteiner, A., Moreira, A., March, 2013. Very-high-resolution airborne synthetic aperture radar imaging: signal processing and applications. Proc. IEEE 101 (No.3). Rosen, P., Hensley, S., Wheeler, K., Sadowy, G., Miller, T., Shaffer, S., Muellerschoen, R.J., Jones, C., Madsen, S.N., Zebker, H., December, 2007. UAVSAR: new NASA airborne SAR system for research. IEEE Aero. Electron. Syst. Mag. 22 (No.11), 21e28. Sabry, R., March, 2018. A Tool for analysis and calibration of compact polarimetry SAR mode anomaly. Geosci. Rem. Sens. Lett. IEEE 15 (No.3), 424e428. Foreword to the special issue on compact polarimetric SAR. IEEE J. Sel. Top. Appl. Earth Obs. Rem. Sens. 12 (10), October, 2019, 3708e3711. Sharma, S.K., Sharma, S., Rao, S., Shafai, L., July, 2013. Handbook of Reflector Antennas and Feed Systems. Vol.I, Vol.II, and vol. III, Artech House Publisher. Soumekh, M., 1999. Synthetic Aperture Radar Signal Processing. Wiley Interscience, New York. Sri Sumantyo, J.T., Koo, V.C., Lim, T.S., Kawai, T., Ebinuma, T., Izumi, Y., Baharuddin, M.Z., Gao, S., Ito, K., July, 2017. Development of circularly polarized synthetic aperture radar onboard UAV JX-1. Int. J. Rem. Sens. 38 (No.8e10), 2745e2756. Tan, H., Hong, J., August, 2016. Calibration of compact polarimetric SAR images using distributed targets and one corner reflector. IEEE Trans. Geosci. Rem. Sens. 54 (No.8), 4433e4444. Tomiyasu, K., May, 1978. Tutorial reivew of synthetic-aperture radar (SAR) with applications to imaging of ocean surface. Proc. IEEE 66 (No.5), 563e583. Truong-Loï, M.-L., Dubois-Fernandez, P., Pottier, E., Freeman, A., Souyris, J.C., 2010. Potentials of a compact polarimetric SAR system. In: In Proceeding IEEE International Geoscience and Remote Sensing Symposium, Honolulu, HI, USA, pp. 742e745.

307

This page intentionally left blank

CHAPTER 16

A processing chain for estimating crop biophysical parameters using temporal Sentinel-1 synthetic aperture radar data in cloud computing framework Dipankar Mandal1, 2, Vineet Kumar3, Juan M. Lopez-Sanchez4, Y.S. Rao1, Heather McNairn5, Avik Bhattacharya1 and Scott Mitchell6

1 Microwave Remote Sensing Lab, Centre of Studies in Resources Engineering, Indian Institute of Technology Bombay, Mumbai, Maharashtra, India; 2Department of Agronomy, Kansas State University, Manhattan, KS, United States; 3 Department of Water Resources, Delft University of Technology, Delft, the Netherlands; 4Institute for Computer Research, University of Alicante, Alicante, Spain; 5Ottawa Research and Development Centre, Agriculture and Agri-Food Canada, Ottawa, ON, Canada; 6Geomatics and Landscape Ecology Laboratory, Carleton University, Ottawa, ON, Canada

1. Introduction Operational crop growth monitoring is necessary to realize yield forecasts and map inventories at a local and regional level. To achieve such objects, certain countries have engaged with several operational groups to develop monitoring systems including Monitoring Agricultural Resources (MARS) (Baruth et al., 2008), Forecasting Agricultural output using Space (FASAL) (Parihar et al., 2006), CropWatch (Wu et al., 2014), and Integrated Canadian Crop Yield Forecaster (ICCYF) (Chipanshi et al., 2015). These crop monitoring systems function within a growing period through the collection of timely information on plant conditions, meteorologic data, and yield expectations. Complementary synoptic information about spatiotemporal variations in crop growth and phenology stages can be provided using satellite imagery. With decades of investigation and advancement, researchers well-instituted and advocated an operational crop monitoring and yield forecasting framework in the optical regime (Mulla et al., 2013; Khamala et al., 2017; Fritz et al., 2019; deFourny et al., 2019). These frameworks frequently derive either measurable growth indicators (e.g., leaf area index [LAI], plant water content, chlorophyll absorption) or vegetation indices (e.g., normalized difference vegetation index, enhanced vegetation index) as by-products. Operational frameworks that rely on satellite data from optical sensors are restricted to data acquisition under clear sky states. A cloudy scene may keep a continuous time series from being achieved for crop development. On the contrary, synthetic aperture radar (SAR) measurements are seldom affected by sky conditions including cloud cover. Indeed, it has gained attention for agricultural applications because of the sensitivity of the microwave signal to dielectric and geometrical characteristics of the target. However, radar Radar Remote Sensing ISBN 978-0-12-823457-0, https://doi.org/10.1016/B978-0-12-823457-0.00011-2

© 2022 Elsevier Inc. All rights reserved.

309

310

Radar Remote Sensing

backscatter coefficients or other secondary parameters (such as scattering power decomposition parameters) (Cloude et al., 1996) cannot be used directly in existing optical-driven models. A reasonable trail can be followed by obtaining similar vegetation metrics (e.g., LAI or biomass) from SAR measurements. These metrics have been generated from optical sensors as operational products (e.g., Moderate Resolution Imaging Spectroradiometer (MODIS) vegetation products). We can achieve better yield estimates from agricultural monitoring frameworks by assimilating SAR data-derived crop biophysical parameters along with phenological developments. Such intense measurements have become possible with accelerated efforts by several space agencies to expand the constellations of satellites by commissioning the Sentinel-1 series, Satelite Argentino de Observaci on COn Microondas (SAOCOM), Radarsat Constellation Mission (RCM), and the upcoming NASA-ISRO Synthetic Aperture Radar (NISAR) and Radar Observing System for Europe - L-Band (ROSE-L). The individual observation covers a wide swath, which would facilitate the production of within-season crop inventories with reasonable accuracy. In the SAR literature, the semiempirical water cloud model (WCM) (Attema and Ulaby, 1978) is recognized for realizing SAR scattering phenomena within vegetation targets owing to its relative simplicity and inversion for these vegetation descriptors (Graham, 2003; SteeleDunne et al., 2017). Several studies have been undertaken to estimate biophysical parameters from SAR data for different crops (Lievens et al., 2011; Prevot et al., 1993; Chakraborty et al., 2005; Dabrowska et al., 2007; Inoue et al., 2014; Beriaux et al., 2015; Hosseini et al., 2015, 2017; Mattia et al., 2015; Fieuzal et al., 2016; Mandal et al., 2021). Although these experiments are specific to crop and test sites, it recommends feasible schemes to invert the WCM to retrieve crop biophysical variables with adequate accuracy and scalability. Nevertheless, an estimation of biophysical parameters via the WCM inversion scheme may lead to an ill-posed problem. Such cases have been potentially addressed with the iterative optimization (IO) and lookup table (LUT) search techniques (Prevot et al., 1993; Mandal et al., 2021). However, the traditional methods impartially deliver realistic estimations at the expense of high computational support when optimizing such inversion problems (Mandal et al., 2019a). Recognizing the inherent concerns with conventional methods (IO and LUT search) for deployment at larger scales, ill-posed inversion problems are customarily solved by data-driven nonparametric models to produce a stable and optimum solution (Beriaux et al., 2011; Verrelst et al., 2012; Mandal et al., 2019b). In a cross-site experiment setting, Mandal et al. (2019a) reported the superiority of machine learning regression approaches compared with traditional ones for the inversion of WCM in terms of accuracy and efficient computation time. Data-driven nonparametric models in the machine learning regression family can provide a proper method for WCM inversion in operational applications (Chauhan et al., 2018; Mandal et al., 2019a). Operational crop monitoring benefits from Earth observation (EO) data with high temporal revisit and extended spatial coverage. In this context, open data through the C-band Sentinel-1 SAR constellation mission adheres to EO specifications for global

A processing chain for estimating crop biophysical parameters

agricultural monitoring (Whitcraft et al., 2019). It has become important to evaluate model inversion methodologies to test their accuracy for operational use (GEOGLAM, 2012). Nevertheless, the dense acquisition of Sentinel-1 data created several challenges in operational employment in terms of the data volume and the requirements of computational infrastructure (Wagner, 2015; Ali et al., 2017). To build efficient platforms for regional-scale mapping, efforts have been made by different space agencies by inducting the enactment of commercial cloud computing platforms. For instance, Sentinel-1 data are open on Google Earth Engine (GEE) clouds as well as Amazon Web Services. Similarly, for the upcoming NISAR mission, the Alaska Satellite Facility Distributed Active Archive Centers (DAAC) is exploring a preliminary prototype of a cloud-based system (NISAR Science Team, 2018). This approach is currently being tested with Sentinel-1 (approximately 5 GB of data volume per frame) as a surrogate for NISAR data (approximately 25 GB per frame). The GEE enables users to fetch and process Sentinel-1 data instantly on the cloud platform as an alternative to downloading and processing in a local workstation (Gorelick et al., 2017). Several studies showcased the utility of GEE and the efficiency of custombuilt processing pipelines for crop classification and mapping (Torbick et al., 2017; Xiong et al., 2017; Shelestov et al., 2017; Mandal et al., 2018), LAI products from MODIS data (Campos et al., 2018), and crop production evaluation (Lobell et al., 2015). However, Sentinel-1 data have not been fully explored in GEE for the delivery of crop inventories. The overarching goal of the investigation is to appraise the potential and transferability of the model inversion method from a point to a regional scale with Sentinel-1 data in a GEE processing chain. In this chapter, we introduce a comprehensive evolution of a processing chain, GEE4Bio, in a cloud platform to estimate crop biophysical variables using Sentinel-1 products. Apart from the inventory map generation, coupling the Google Colab with the GEE permits us to achieve WCM calibration for individual crops and to assess the retrieval accuracies. The inversion of WCM is implemented into the GEE platform. Instead of a traditional LUT search and iterative approaches, the Random Forest (RF) regression is used as the inversion approach to retrieve the plant area index (PAI) and wet biomass using information from both the co- and cross-pol (VV-VH) channels of Sentinel-1. Furthermore, PAI and wet biomass maps for different crop growth stages are generated in the same processing chain.

2. Methodology 2.1 Study area and dataset We conducted this investigation over the Joint Experiment for Crop Assessment and Monitoring (JECAM) test site for SAR Inter-Comparison Experiment in Carman, Manitoba (Canada), as given in Fig. 16.1. The Carman test site includes z26
48 km2 of the area and is designated by diverse agricultural crop types and soil conditions. Among several crop types, wheat, oats, soybean, canola, and corn are mostly

311

312

Radar Remote Sensing

Figure 16.1 Study area over the Joint Experiment for Crop Assessment and MonitoringeCarman test  site in Canada. The sampling fields (mint green polygons) are overlayed on the sVV Sentinel-1 image acquired on Jul. 19, 2016. A layout of 16 sampling locations within a sampling field is highlighted.

grown during the summer. Ground data were collected nearly simultaneously with satellite overpasses during the SMAPVEX16-MB campaign in 2016 (Bhuiyan et al., 2018). Throughout the SMAPVEX16-MB campaign, crop and soil in situ data from 50 fields were measured in two definite periods (i.e., June 8e22 and July 8e22, 2016). During this campaign window, measurements in most crop fields showed plant growth from an early to a fully vegetative stage. The nominal size of each field was approximately 800 m
800 m. In each sampling field, soil moisture data were gathered from 16 sampling locations, arranged in two parallel transects along the row direction, as shown in Fig. 16.1. Soil moisture measurements were taken at these locations with three replicate measurements using Steven’s Hydra Probes during both periods of the campaign (Bhuiyan et al., 2018). Among these 16 sampling locations, three were selected for vegetation sampling. In each sampling site, plant biomass (dry and wet), PAI

A processing chain for estimating crop biophysical parameters

(m2 m2), and crop phenological stages were recorded by destructive and nondestructive methods. On the contrary, PAI measurements were estimated using digital hemispherical photography and postprocessing images in CANEYE software. A detailed description of the test site and in situ measurement protocols can be found in SMAPVEX16-MB campaign reports (McNairn et al., 2016). We also used the annual crop inventory map developed by Agriculture and Agri-Food Canada (Davidson et al., 2017), which is open in the GEE Data Catalog. We used three Sentinel-1 acquisitions for this research, as listed in Table 16.1. The preprocessing of these Sentinel-1 measurements to produce SAR backscatter intensities for each sampling location is described in detail in Section 2.2. Extracted back  scatter intensities sVV and sVH are enumerated with corresponding in situ measurements for each acquisition date. These tabulated datasets are subsequently employed to calibrate and validate the WCM. From this entire feature set, the calibration data split is performed by selecting approximately half of the data randomly; the remainder of the data are used as an independent validation dataset for each individual crop. The first dataset is used in the WCM calibration, and the other for validation to evaluate the performance of the inversion method. 2.2 GEE4Bio: Sentinel-1 data processing chain in Google Earth Engine for biophysical parameter estimation The processing chain involves two elemental segments considering the cloud computing environments within a unified framework: Earth Engine mode and Google Colab. Processing steps counted on Google Colab (Hoyos et al., 2006) are the calibration of WCM, creation of the LUT, and validation of inversion for each crop type. The inversion of the WCM and the generation of biophysical inventory maps are implemented on the GEE cloud platform. The GEE can seamlessly manage the processing steps from time series of Sentinel-1 data fetching to model inversion. It typically encompasses five actions: (1) Sentinel-1 data fetching, (2) cloud filtering, (3) image preprocessing, (4) vegetation modeling and Table 16.1 Sentinel-1 acquisitions and in situ measurement dates during the campaign. Sentinel-1 image acquisition date

Beam mode

Incidence angle range

Orbit

In situ measurement dates

June 13, 2016 July 7, 2016 July 19, 2016

IW IW IW

31.32e35.24 31.32e35.24 30.23e34.84

Ascending Ascending Ascending

June 13, June 15 Jul. 5, July 6 Jul. 17, July 20

IW, interferometric wide swath mode.

313

314

Radar Remote Sensing

Figure 16.2 Schematic workflow of GEE4Bio processing chain for plant area index (PAI) and wet biomass retrieval. GEE, Google Earth Engine; PAI, plant area index; RFR, Random Forest regression; WCM, water cloud model.

calibration, and (5) model inversion and crop biophysical parameter map generation. The schematic workflow of the GEE4Bio processing chain for biophysical parameter estimation is presented in Fig. 16.2. 2.2.1 Sentinel-1 data fetching We fetched the Sentinel-1A Ground Range Detected (GRD) products in the GEE processing Web application interface directly from the GEE Image collection. The image collection in GEE contains the calibrated and ortho-corrected products, which are preprocessed from the Single Look Complex data using the Sentinel-1 Toolbox (ESA, 2015). 2.2.2 Cloud filtering The GEE image collection of Sentinel-1 data contains multiple metadata attributes regarding imaging properties including orbit pass type (ascending or descending),

A processing chain for estimating crop biophysical parameters

acquisition mode (IW, etc.), and polarization. In the cloud filtering step, we select required images from the GEE image collection using associated attributes by employing the Metadata filtering function. Subsequently, spatial subsetting is performed followed by sorting images by date of acquisition using the filterDate argument. 2.2.3 Image preprocessing The GEE image collection products of Sentinel-1 backscatter intensities are represented in dB scale. Hence, for further processing, we converted the data products to natural po=

wer scale by employing 10Ii;j 10 conversion. Afterward, a 3
3 boxcar averaging sliding window filter is employed to degrade the speckle effect. The type and window size of the filter are determined according to the high field sizes and homogeneous cropping pattern (Robertson et al., 2018). The locations of in situ measurements are overlayed  on these backscatter intensity images and s values are extracted both in VV and VH channels. 2.2.4 Vegetation modeling and calibration The WCM enables the simulation of radar backscatter intensities from the vegetatione soil system using semiempirical models (Attema and Ulaby, 1978). The form of WCM adapted for backscatter calculations is (Eq. 16.1):


   2BV2 2BV2 E s ¼ AV1 cos q 1  exp  þ D expðCMv Þ
exp  cos q cos q cos q (16.1) The vegetation descriptors are presented as V1 ¼ L and V2 ¼ W, where L and W are the PAI and wet biomass, respectively. The WCM parameters (A, B, C, D, and E) are calibrated for each crop type individually in the Google Colab Web platform. The model parameters are determined using the nonlinear least square optimization employing the LevenbergeMarquardt algorithm (More, 1978). The WCM is calibrated for both coand cross-pols for all crops employing the same in situ measurements. The model calibration realizations are evaluated in terms of the correlation coefficient (r) and root mean   square error (RMSE) with the WCM simulated s and observed s . 2.2.5 Model inversion and crop biophysical parameter map generation After calibration of the WCM, we determine the PAI and wet biomass by inverting the WCM. We employ a regression-based method to solve the ill-posed inversion case. The model inversion chain encompasses three principal actions: (1) WCM forward modeling and LUT generation, (2) regression model development, and (3) derivation of PAI and wet biomass maps.

315

316

Radar Remote Sensing

The forward modeling denotes the generation of response values from an augmented dataset employed to train the regression model. We use the aggregates of crop descriptors  from the calibration data in forward WCM to produce the corresponding s to form the LUT. In the Google Colab platform, a LUT is created using these vegetation parameters  and the corresponding s for individual crop types. These LUTs are then put in tables independently for each crop. The LUT elements are then employed as training data to constitute the RF regression  (RFR) model in the GEE cloud. Radar data derived s in both the co-pol and cross-pol channels are introduced as RFR predictors. On the opposite side, the PAI and wet biomass measurements are employed as RFR targets. The formulation of RFR is well-established in machine learning theories, which often indicate RF as an ensemble learning technique (Breiman, 2001). It uses a large set of independently produced decision trees from the given training datasets. To build each tree, a random bootstrap sampling is conducted, which comprises 67% of the training samples. The remaining 33% of the training sample (often defined as out-of-bag [OOB] samples) is employed to get an error estimate for this subset. After building such multiple decision trees, the prediction is made by averaging the results of all trees, which can provide more accurate and stable predictions than individual decision trees. Another key concept of RF is the selection of random subsets of predictor features for node splitting. At each node, the best split is chosen to form the succeeding child nodes. The numerical value of each child node is taken as the mean of samples in that node (Liaw and Wiener, 2002). In this work, based on the OOB error rate, 300 trees are considered for RF regression, and the node impurity is measured with the mean square error. During the inversion step, the PAI and wet biomass values are estimated using the RFR model trained by the LUT elements for each crop. Afterward, PAI and wet biomass inventory maps are generated over the test area in GEE using Sentinel-1 acquisitions and the annual crop class map.

3. Results and discussion 3.1 Water cloud model calibration The calibration of the WCM is performed for VV and VH polarizations as previously discussed in Section 2.2.4 for the wheat, soybean, canola, corn, and oat crops. This step results in 10 different equations. The model parameters (i.e., A, B, C, D, E, and F) for each combination of crop and polarization are estimated along with the significance test analysis (F statistics and P values). P values for all WCMs are 5.3 m2 m2) is likely due to saturation of the C-band (5.6 cm wavelength) wave with a high accumulation of plant biomass throughout the flowering and podding stage of canola. The validation outcomes for oats show that although variations in terms of accuracy (r ¼ 0.90 and RMSE ¼ 1.25 m2 m2) are modest compared with wheat, some differences over the 1:1 line are apparent. These agreements in results are likely due to higher similarities in the structures of these two cereal crop types. As with wheat, oats exhibit an erectophile leaf geometry during the early tillering phase. However, a difference is observed during the period of plant inflorescence when oat heads emerge. Unlike the formation of spikelets, as occurs during the inflorescence of wheat, oat plants form panicles during inflorescence. These panicles are attached to the central axis via branches and mantled by large paper-like covers (glumes), which contain two to three florets (Bleiholder, 2001). Thus, separation between these two crops with second acquisition estimates might be possible, as observed in Fig. 16.3. An underestimation is apparent during the third acquisition period when fruit development starts. During this period, PAI and biomass trends seldom increase proportionately for cereal crops. As observed in Fig. 16.3, the PAI estimates of corn obey the 1:1 line reasonably well for the entire period of corn growth starting from leaf development to tasseling. The in situ measurement of PAI ranges from 0.1 to 5.6 m2 m2 during the field campaign window. A high correlation coefficient (0.94) and low error (RMSE ¼ 0.74 m2 me2) are noted. Similar sensitivity of the estimated LAI (using VV-HV polarizations combined) with in situ measurement was reported for corn in Hosseini et al. (2015), using RADARSAT-2 data. The larger variations in the estimation of PAI above 4.1 m2 m2 (i.e., during the tasseling of corn) may be due to increased randomness in the canopy. 3.2.2 Validation of wet biomass Estimates of wet biomass are shown in Fig. 16.4. The in situ measurement of wet biomass for wheat ranges from 0.4 to 4.8 kg m2 throughout crop development. We obtain a high correlation coefficient (r ¼ 0.92) with an RMSE of 0.52 kg m2, which is lower

319

320

Radar Remote Sensing

than the PAI evaluation errors. However, the boundary of estimation is spread (0.9 kg m2) across the 1:1 line when wet biomass exceeds 2.5 kg m2. Hosseini and McNairn (2017) also summarized such results when estimating wheat total biomass using VV and HV channels. The behavior of wet biomass estimates (Fig. 16.4) of soybean results in a high correlation coefficient (r ¼ 0.88) and low RMSE of 0.29 kg m2. Early in the season, soybean biomass is low. An overestimation of PAI was observed during the leaf development stage. Early in development, the soybean canopy closure is very low (PAI < 1.48 m2 m2) and exposed soil between rows has a greater contribution to backscatter (Wiseman et al., 2014). The correlation coefficient (r) and RMSE are 0.91 and 0.86 kg m2, respectively, for canola. The model estimates diverge after the flowering stage, when the measured biomass is >4.0 kg m2. The sensitivity of radar backscatter to the accumulation of canola biomass throughout leaf development to the flowering stage is likely. At later growth stages, we also observed saturation of the C-band effect on estimates. In contrast to PAI, the overall estimation is marginally better in the case of wet biomass predictions. The in situ measured wet biomass of oats varies from 0.076 to 5.5 kg m2. In contrast, the estimates range from 0.05 to 4.0 kg m2. The r is 0.95 for wet biomass with an RMSE of 0.54 kg m2. The wet biomass estimates follow the 1:1 line better than PAI. The wet biomass estimates of corn follow the 1:1 line (Fig. 16.4) with r ¼ 0.87 and an RMSE of 1.01 kg m2. Despite this, the deviation in wet biomass estimations is comparatively higher as plants advance to the tasseling stage.

3.3 Generation of plant area index and wet biomass maps using GEE4Bio The WCM inversion algorithm in GEE is used to produce PAI and biomass maps of the Carman test site with an area of 26
48 km2 using VV and VH backscatter intensities for Sentinel-1 acquisitions on three dates. High-resolution (20-m) PAI and biomass maps for the three acquisition dates are shown in Figs. 16.5 and 16.6. Both the spatial and temporal variability in plant growth are noted in these inventory maps. The GEE4Bio processing chain takes about 40 s to derive these inventory maps from a single acquisition of Sentinel-1. The ground data indicate that most of the wheat and oat fields were at tillering through the second week of June, with a mean wet biomass of 1.02 kg m2 and PAI of 3.44 m2 m2. Similar results are evident from Figs. 16.5 and 16.6. On the other hand, soybean seeding was completed through this period, and the plants were in their unifoliate to third trifoliate phenology stage (Agriculture, 2016). Therefore, the biophysical parameter maps derived from the Jun. 13 image indicate very low PAI and biomass levels.

A processing chain for estimating crop biophysical parameters

Figure 16.5 Plant area index (PAI) (m2 m2) inventory maps over the test site for different acquisitions of Sentinel-1. Other land cover classes are masked on the inventory maps.

Figure 16.6 Wet biomass (kg m2) inventory maps over the test site for different acquisitions of Sentinel-1. Other land cover classes are masked on the inventory maps.

The PAI and wet biomass estimates in most of the canola fields indicated low values (0.7 m2 m2 and 0.5 kg m2) compared with the cereal crops on June 13. During this period, canola plants were in the emergent to early rosette growth state, which affects the soil contribution, primarily radar backscatter. A similar effect of soil components (surface roughness and moisture content) is well-explained during the early phase of crop development when vegetation cover is low over the soil surface (Baghdadi et al., 2017). However, temporal changes in surface roughness are moderate after the plant starts vegetative growth. Except for specific cultivation practices (harrowing or tillage) or a heavy rainfall event, roughness seldom changes during the early seeding time. Most cornfields were in their primary vegetative stage during the second week of June, and the average PAI and wet biomass was 0.36 m2 m2 and 0.3 kg m2,

321

322

Radar Remote Sensing

respectively. However, the PAI and wet biomass maps overestimate both the PAI and wet biomass by 0.5 m2 m2 and 0.7 kg m2, respectively. These outcomes agree with the validation results presented in Figs. 16.3 and 16.4. This overestimation is likely due to the low cropping density of corn plants and the significant effect of the soil component. The inventory maps for succeeding acquisition dates indicate an increase in both the PAI and wet biomass values of all crops. Rapid growth is evident in canola with an increased PAI and wet biomass up to 5.7 m2 m2 and 4.2 kg m2, respectively. For corn, the average growth was observed with a PAI and wet biomass of 4.1 m2 m2 and 3.5 kg m2, respectively. During the third week of July, we observe pick values of PAI and wet biomass irrespective of the crop types, as they advanced to the end of their vegetative growth (Agriculture, 2016).

4. Conclusion This chapter demonstrates a unified processing chain for end-users to estimate PAI and wet biomass exploiting Sentinel-1 GRD data on the GEE platform. The RFR model is used in this framework for inversion of the WCM. In this cloud computing framework, applicability of the RFR-based inversion method is evaluated for five major crops using Sentinel-1 dual-pol (VV and VH) SAR data. The crop inventory maps showcased the potential by capturing spatial variability between crop fields over the growing period, leveraging production estimates. To go from scientific applications to operational monitoring, the proposed processing chain, named GEE4Bio, needs to be rigorously tested with more crop types for application to a wider range of cropping systems. A goal of the JECAM SAR Inter-Comparison Experiment is to gather data from regional test sites and diverse cropping systems. This experiment would benefit from the efficiency of a GEE cloud computing platform. Moreover, this representation of a cloud-based framework produces insights into possible prototypes for managing high volumes of data, as anticipated from planned operational SAR missions. The results presented here demonstrate the usefulness of GEE for regional biophysical parameter retrieval. Such an approach could significantly advance the operational use of SAR for agricultural monitoring.

Acknowledgments The authors would like to thank the ground team members for data collection through the SMAPVEX16MB campaign, and the European Space Agency for providing Sentinel-1 through the Copernicus Open Access Hub. Also, authors acknowledge the GEO-AWS Earth Observation Cloud Credits Program, which provided a testbed for cloud computation through the project: "AWS4AgriSAR-Crop inventory mapping from SAR data on the cloud computing platform." This research was supported in part by the Shastri Indo-Canadian Institute, India.

A processing chain for estimating crop biophysical parameters

Code availability Extract by points code in GEE. https://code.earthengine.google.com/d767149c290192a0b17538 5e62bea544. GEE code for mapping. https://code.earthengine.google.com/32e06a03325faa2e6720e11af0e58ad2. Google Colab ipynb link. https://colab.research.google.com/drive/1UGQuSZHuZplZfUKJoPVAVvrpleYQcAM?usp¼sharing.

References Agriculture, M.B., 2016. AgriculturedProvince of Manitoba. http://www.gov.mb.ca/agriculture/crops/ seasonal-reports/crop-report-archive/index.html. Ali, I., Naeimi, V., Cao, S., Elefante, S., Bauer-Marschallinger, B., Wagner, W., 2017. Sentinel-1 data cube exploitation: tools, products, services and quality control. Proc. Big Data Space 40e43. Attema, E., Ulaby, F.T., 1978. Vegetation modeled as a water cloud. Radio Sci. 13 (2), 357e364. Baghdadi, N., El Hajj, M., Zribi, M., Bousbih, S., 2017. Calibration of the water cloud model at C-Band for winter crop fields and grasslands. Rem. Sens. 9 (9), 969. Baruth, B., Royer, A., Klisch, A., Genovese, G., 2008. The use of remote sensing within the MARS crop yield monitoring system of the European Commission. Int. Arch. Photogram. Rem. Sens. Spatial Inf. Sci. 37, 935e940. Beriaux, E., Lambot, S., Defourny, P., 2011. Estimating surface-soil moisture for retrieving maize leaf-area index from SAR data. Can. J. Rem. Sens. 37 (1), 136e150. Beriaux, E., Waldner, F., Collienne, F., Bogaert, P., Defourny, P., 2015. Maize leaf area index retrieval from synthetic quad pol SAR time series using the water cloud model. Rem. Sens. 7 (12), 16204e16225. Bhuiyan, H.A., McNairn, H., Powers, J., Friesen, M., Pacheco, A., Jack- son, T.J., Cosh, M.H., Colliander, A., Berg, A., Rowlandson, T., et al., 2018. Assessing SMAP soil moisture scaling and retrieval in the Carman (Canada) study site. Vadose Zone J. 17 (1). https://doi.org/10.2136/ vzj2018.07.0132. Bleiholder, H., Weber, E., Lancashire, P., Feller, C., Buhr, L., Hess, M., Wicke, H., Hack, H., Meier, U., Klose, R., et al., 2001. Growth Stages of Mono-and Dicotyledonous Plants, BBCH Monograph. Federal Biological Research Centre for Agriculture and Forestry, Berlin/Braunschweig, Germany, p. 158. Breiman, L., 2001. Random forests. Mach. Learn. 45 (1), 5e32. Campos-Taberner, M., Moreno-Martınez, A.., Garcıa-Haro, F., Camps-Valls, G., Robinson, N., Kattge, J., Running, S., 2018. Global estimation of biophysical variables from Google earth engine platform. Rem. Sens. 10 (8), 1167. Chakraborty, M., Manjunath, K., Panigrahy, S., Kundu, N., Parihar, J., 2005. Rice crop parameter retrieval using multi-temporal, multi-incidence angle Radarsat SAR data. ISPRS J. Photogrammetry Remote Sens. 59 (5), 310e322. Chauhan, S., Srivastava, H.S., Patel, P., 2018. Wheat crop biophysical parameters retrieval using hybridpolarized RISAT-1 SAR data. Remote Sens. Environ. 216, 28e43. Chipanshi, A., Zhang, Y., Kouadio, L., Newlands, N., Davidson, A., Hill, H., Warren, R., Qian, B., Daneshfar, B., Bedard, F., et al., 2015. Evaluation of the Integrated Canadian Crop Yield Forecaster (ICCYF) model for in- season prediction of crop yield across the Canadian agricultural landscape. Agric. For. Meteorol. 206, 137e150. Cloude, S.R., Pottier, E., 1996. A review of target decomposition theorems in radar polarimetry. IEEE Trans. Geosci. Remote Sens. 34 (2), 498e518. Dabrowska-Zielinska, K., Inoue, Y., Kowalik, W., Gruszczynska, M., 2007. Inferring the effect of plant and soil variables on C-and L-band SAR backscatter over agricultural fields, based on model analysis. Adv. Space Res. 39 (1), 139e148. Davidson, A., Fisette, T., Mcnairn, H., Daneshfar, B., 2017. Detailed crop mapping using remote sensing data (crop data layers). In: Handbook on Remote Sensing for Agricultural Statistics.

323

324

Radar Remote Sensing

Defourny, P., Bontemps, S., Bellemans, N., Cara, C., Dedieu, G., Guzzonato, E., Hagolle, O., Inglada, J., Nicola, L., Rabaute, T., et al., 2019. Near real- time agriculture monitoring at national scale at parcel resolution: performance assessment of the Sen2-Agri automated system in various cropping systems around the world. Remote Sens. Environ. 221, 551e568. ESA, 2015. User Guides - Sentinel-1 SAR. https://sentinel.esa.int/web/sentinel/user-guides/sentinel-1sar/acquisition-modes/interferometric-wide-swath. Fieuzal, R., Baup, F., 2016. Estimation of leaf area index and crop height of sunflowers using multi-temporal optical and SAR satellite data. Int. J. Rem. Sens. 37 (12), 2780e2809. Fritz, S., See, L., Bayas, J.C.L., Waldner, F., Jacques, D., Becker-Reshef, I., Whitcraft, A., Baruth, B., Bonifacio, R., Crutchfield, J., et al., 2019. A comparison of global agricultural monitoring systems and current gaps. Agric. Syst. 168, 258e272. Geoglam, 2012. The Ad Hoc Working Group on GEOGLAM. http://ceos.org/ourwork/ad-hoc-teams/ geoglam/. Gorelick, N., Hancher, M., Dixon, M., Ilyushchenko, S., Thau, D., Moore, R., 2017. Google earth engine: planetary-scale geospatial analysis for everyone. Remote Sens. Environ. 202, 18e27. Graham, A., Harris, R., 2003. Extracting biophysical parameters from remotely sensed radar data: a review of the water cloud model. Prog. Phys. Geogr. 27 (2), 217e229. Hosseini, M., McNairn, H., 2017. Using multi-polarization C-and L-band syn- thetic aperture radar to estimate biomass and soil moisture of wheat fields. Int. J. Appl. Earth Obs. Geoinf. 58, 50e64. Hosseini, M., McNairn, H., Merzouki, A., Pacheco, A., 2015. Estimation of Leaf Area Index (LAI) in corn and soybeans using multi-polarization C-and L-band radar data. Remote Sens. Environ. 170, 77e89. Hoyos-Rivera, G.J., Gomes, R., Willrich, R., Courtiat, J.-P., 2006. Colab: a new paradigm and tool for collaboratively browsing the web. IEEE Trans. Syst. Man Cybern. Syst. Hum. 36 (6), 1074e1085. Inoue, Y., Sakaiya, E., Wang, C., 2014. Capability of C-band backscattering coefficients from highresolution satellite SAR sensors to assess biophysical variables in paddy rice. Remote Sens. Environ. 140, 257e266. Khamala, E., et al., 2017. Review of the Available Remote Sensing Tools, Products, Methodologies and Data to Improve Crop Production Forecasts. http://www.fao.org/3/a-i7569e.pdf. Liaw, A., Wiener, M., 2002. Classification and regression by random forest. R. News 2 (3), 18e22. Lievens, H., Verhoest, N.E., 2011. On the retrieval of soil moisture in wheat fields from L-band SAR based on water cloud modeling, the IEM, and effective roughness parameters. Geosci. Rem. Sens. Lett. IEEE 8 (4), 740e744. Lobell, D.B., Thau, D., Seifert, C., Engle, E., Little, B., 2015. A scalable satellite-based crop yield mapper. Remote Sens. Environ. 164, 324e333. Mandal, D., Hosseini, M., McNairn, H., Kumar, V., Bhattacharya, A., Rao, Y., Mitchell, S., Robertson, L.D., Davidson, A., Dabrowska-Zielinska, K., 2019a. An investigation of inversion methodologies to retrieve the leaf area index of corn from C-band SAR data. Int. J. Appl. Earth Obs. Geoinf. 82, 101893. Mandal, D., Kumar, V., McNairn, H., Bhattacharya, A., Rao, Y., 2019b. Joint estimation of plant area index (PAI) and wet biomass in wheat and soybean from C-band polarimetric SAR data. Int. J. Appl. Earth Observ. Geoinform. 79, 24e34. Mandal, D., Kumar, V., Bhattacharya, A., Rao, Y.S., Siqueira, P., Bera, S., 2018. Sen4Rice: a processing chain for differentiating early and late transplanted rice using time-series sentinel-1 SAR data with Google earth engine. Geosci. Rem. Sens. Lett. IEEE 15 (12), 1947e1951. https://doi.org/10.1109/ LGRS.2018.2865816. Mandal, D., Bhattacharya, A., Rao, Y.S., 2021. Radar Remote Sensing for Crop Biophysical Parameter Estimation. Springer Singapore. Mattia, F., Satalino, G., Balenzano, A., Rinaldi, M., Steduto, P., Moreno, J., 2015. Sentinel-1 for wheat mapping and soil moisture retrieval. In: Geoscience and Remote Sensing Symposium (IGARSS), 2015 IEEE International. IEEE, pp. 2832e2835. McNairn, H., Hochheim, K., Rabe, N., 2004. Applying polarimetric radar imagery for mapping the productivity of wheat crops. Can. J. Rem. Sens. 30 (3), 517e524. McNairn, H., Tom, J.,J., Powers, J., Belair, S., Berg, A., Bullock, P., Colliander, A., Cosh, M.H., Kim, S.B., Ramata, M., Pacheco, A., Merzouki, A., 2016. Experimental Plan SMAP Validation Experiment 2016 in Mani- Toba, Canada (SMAPVEX16-MB). https://smap.jpl.nasa.gov/internal_resources/390.

A processing chain for estimating crop biophysical parameters

More, J.J., 1978. The Levenberg-Marquardt algorithm: implementation and theory. In: Numerical Analysis. Springer, pp. 105e116. Mulla, D.J., 2013. Twenty five years of remote sensing in precision agriculture: key advances and remaining knowledge gaps. Biosyst. Eng. 114 (4), 358e371. NISAR Science Team, 2018. NASA-ISRO SAR (NISAR) Mission Science Users’ Handbook. Tech. rep. Jet Propulsion Laboratory, California Institute of Technology. https://nisar.jpl.nasa.gov/files/nisar/ NISAR_Science_Users_Handbook.pdf. Parihar, J.S., Oza, M.P., 2006. FASAL: an integrated approach for crop assessment and production forecasting. In: Agriculture and Hydrology Applications of Remote Sensing, vol. 6411. International Society for Optics and Photonics, p. 641101. Prevot, L., Champion, I., Guyot, G., 1993. Estimating surface soil moisture and leaf area index of a wheat canopy using a dual-frequency (C and X bands) scatterometer. Remote Sens. Environ. 46 (3), 331e339. Robertson, L.D., Davidson, A., McNairn, H., Hosseini, M., Mitchell, S., de Abelleyra, D., Veron, S., Cosh, M.H., July 2018. Sar speckle filtering and agriculture field size: development of SAR data processing best practices for the JECAM SAR Inter-Comparison Experiment. In: IGARSS 2018 - 2018 IEEE International Geoscience and Remote Sensing Symposium, pp. 3828e3831. Shelestov, A., Lavreniuk, M., Kussul, N., Novikov, A., Skakun, S., 2017. Exploring Google earth engine platform for Big Data Processing: classification of multi-temporal satellite imagery for crop mapping. Front. Earth Sci. 5, 17. Steele-Dunne, S.C., McNairn, H., Monsivais-Huertero, A., Judge, J., Liu, P., Papathanassiou, K., 2017. Radar remote sensing of agricultural canopies: a review. IEEE J. Sel. Top. Appl. Earth Obs. Rem. Sens. 10 (5), 2249e2273. Torbick, N., Chowdhury, D., Salas, W., Qi, J., 2017. Monitoring rice agri- culture across Myanmar using time series Sentinel-1 assisted by Landsat-8 and PALSAR-2. Rem. Sens. 9 (2), 119. Verrelst, J., Muneoz, J., Alonso, L., Delegido, J., Rivera, J.P., Camps-Valls, G., Moreno, J., 2012. Machine learning regression algorithms for biophysical parameter retrieval: opportunities for Sentinel-2 and-3. Remote Sens. Environ. 118, 127e139. Wagner, W., 2015. Big data infrastructures for processing sentinel data. In: Photogrammetric Week, vol. 15, pp. 93e104. Whitcraft, A.K., Becker-Reshef, I., Justice, C.O., Gifford, L., Kavvada, A., Jarvis, I., 2019. No pixel left behind: toward integrating earth observations for agriculture into the United Nations Sustainable development goals framework. Remote Sens. Environ. 235, 111470. Wiseman, G., McNairn, H., Homayouni, S., Shang, J., 2014. RADARSAT- 2 polarimetric SAR response to crop biomass for agricultural production monitoring. IEEE J. Sel. Top. Appl. Earth Obs. Rem. Sens. 7 (11), 4461e4471. Wu, B., Meng, J., Li, Q., Yan, N., Du, X., Zhang, M., 2014. Remote sensing- based global crop monitoring: experiences with China’s CropWatch system. Int. J. Digital Earth 7 (2), 113e137. Xiong, J., Thenkabail, P.S., Gumma, M.K., Teluguntla, P., Poehnelt, J., Congalton, R.G., Yadav, K., Thau, D., 2017. Automated cropland map- ping of continental Africa using Google Earth Engine cloud computing. ISPRS J. Photogrammetry Remote Sens. 126, 225e244.

325

This page intentionally left blank

CHAPTER 17

Fuzzy logic for the retrieval of kidney bean crop growth variables using ground-based scatterometer measurements Dileep Kumar Gupta1, 2, Rajendra Prasad2, Pradeep Kumar3 and Prashant K. Srivastava1

1 Remote Sensing Laboratory, Institute of Environment and Sustainable Development, Banaras Hindu University, Varanasi, Uttar Pradesh, India; 2Department of Physics, Indian Institute of Technology, Banaras Hindu University, Varanasi, Uttar Pradesh, India; 3School of Environmental Sciences, Jawaharlal Nehru University, New Delhi, India

1. Introduction Kidney beans are an worldwide important crop that provide a good source of fat-free, high-quality protein, mineral molybdenum, and cholesterol-lowering fiber (Gupta et al., 2018). Conventional remote-sensing techniques based on visible and infrared radiations have several limitations such as cloud cover, rainfall, and dust. These remote sensing techniques can capture satellite images in daytime because they depend on sunlight. To remove these shortcomings, microwave remote sensing may be a powerful tool to monitor the kidney bean growth cycle by estimating its crop variables at innumerable growth stages (Gupta et al., 2016). Microwave remote sensing offers some advantages over conventional remote sensing owing to its all-weather operations through 24hour operation capabilities with more penetrating power into the vegetation canopies (Kumar et al., 2019). Kim et al. (2013) set up a ground-based fully polarimetric scatterometer to monitor the different soybean crop growth stages over its full growing season at L-, C-, and X-bands. They investigated the relationships between temporal variations in backscattering coefficients and noticed soybean crop biophysical parameters over the entire growth period. A ground-based scatterometer response in the incidence angle of 20e70 degrees at vertical transmitevertical receive (VV) and horizontal transmite horizontal receive (HH) polarizations was shown for the kidney bean crop at innumerable growth stages at the X-band in Varanasi by Prasad (2011). Brown et al. (2003) used ground-based polarimetric synthetic aperture radar to measure wheat crop growth variables at X- and C-bands. Optimum parameters of their radar configuration were a higher angle of incidence and VV polarization at X-band for the affective monitoring of the wheat canopy. However, a medium range of incidence angle is preferred for a higher sensitivity of s
with biomass at the C-band radar system.

Radar Remote Sensing ISBN 978-0-12-823457-0, https://doi.org/10.1016/B978-0-12-823457-0.00015-X

© 2022 Elsevier Inc. All rights reserved.

327

328

Radar Remote Sensing

Numerous experiments have been conducted, including in situ (Inoue et al., 2002; Kim et al., 2012), spaceborne (Kumar et al., 2018; Shao et al., 2001; Thuy Le et al., 1997), and airborne (Taconet et al., 1994) synthetic aperture radar (SAR) to demonstrate the significant correlation of backscattering at different polarizations and frequencies for the crop variables of different crops. The backscattering coefficients are found feasible variable to retrieve the crop growth variables. The reported results have limitations for obtaining quantitative information about crop growth variables. Temporal variations in backscattering coefficients with crop growth variables had a saturation effect on certain growth periods. Radar measurements in a bistatic mode are needed to overcome this drawback (Ferrazzoli et al., 2000). Few researchers worldwide have used bistatic radar geometry to retrieve different crop variables. The German Aerospace Center and EADS Astrium launched a mission for a bistatic radar system with twin satellites (TerraSARX and TanDEM-X) with different polarizations. Many retrieval processes of crop variables using radar data have been carried out implementing a variety of techniques including multiple regressions and theoretical models. Inversion of these models is complex and requires a number of parameters. A parameter-free model is needed to overcome existing problems in the inversion of models. Fuzzy logic may be a good approach for the retrieval of crop growth variables using radar data (Pandey et al., 2013). Gupta et al. (2016) made in situ measurements to retrieve kidney bean crop growth variables using a multiangular, multitemporal, and copolarized bistatic scatterometer system. They prepared two similar kidney bean crop beds for bistatic scatterometer measurements at two frequencies (6 and 10 GHz). Data information from both kidney bean crop beds was used to train and test artificial neural networks, respectively. The researchers found that the 10-GHz frequency and 50 degrees incidence angle were suitable system parameters for kidney bean crop growth variable retrieval at similar polarizations (HH and VV) using bistatic scatterometer data. They found good agreement between observed and estimated values of kidney bean crop variables using artificial neural networks. s
were modeled using the first-order radiative transfer theory by the Michigan Microwave Canopy Scattering (MIMICS) model for wheat and soybean crops at C- and L-bands for different incidence and azimuth angles. Outcomes of this simulation showed that a higher frequency is more sensitive for retrieving crop parameters (Zhang and Wu, 2016). The vegetation water content (VWC) of kidney bean crop retrieval was performed using ground-based multitemporal, multiangular, and copolarized scatterometer data at X-band in Varanasi (Gupta et al., 2018). The researchers performed the investigation to retrieve the VWC of a kidney bean crop by an empirical relation dependent on the least-square optimization method using scatterometer data. Promising results were found for the retrieval of the VWC of the kidney bean crop in this investigation. Limited research studies exist based on ground-based bistatic scatterometer measurements. Most research articles on microwave remote sensing are based on satellite images for the retrieval of crop growth variables (Kumar et al., 2017, 2018). Some studies have

Fuzzy logic for the retrieval of kidney bean crop growth variables using ground-based scatterometer measurements

been made based on a bistatic configuration with a major focus on soil moisture and roughness measurements (Johnson and Ouellette, 2014; Mittal and Singh, 2010). Therefore, it is necessary to investigate these type of studies focused on a bistatic configuration for different types of crop growth variable monitoring. This study describes bistatic scatterometer measurements for kidney bean crops in an angular range of incidence angles of 20e70 degrees for X-band at HH and VV polarization. Measurements of crop growth variables were also carried out simultaneously at each bistatic scatterometer measurement. The fuzzy logic algorithm is used to retrieve crop growth variables for the kidney bean crop.

2. Method and observations 2.1 Bistatic scatterometer setup and measurements An outdoor bistatic scatterometer measurement facility near the physics department was designed by the microwave remote sensing laboratory at the Indian Institute of Technology (BHU) Varanasi campus. The transmitting and receiving antennas were mounted on the specially designed antenna support tower, allowing multiangular and dual-polarized measurements. The height of the antennas aboveground was 3 m. Bistatic scatterometer measurements were made at X-band (10 GHz) for HH and VV polarizations for an angular range of 20e60 degrees incidence in the specular way (f ¼ 0) over the entire life cycle of the kidney bean crop. The bistatic scatterometer transmits a linearly (H or V) polarized electromagnetic wave and the receiver receives a linearly polarized (H or V) wave after scattering from the object or crop. Internal and external calibrations have been performed to calibrate the system. Internal calibration of the bistatic scatterometer system was performed using a power ratio method. The bistatic scattering coefficient (s
) of the distributed target is directly proportional to the ratio of the received power from the target and transmitted power. All gains and losses are considered for the measurement of received and transmitted power except antenna gain of the bistatic scatterometer system. External calibration of the bistatic scatterometer system was performed using a flat and conducting aluminum sheet. Amplitudes of the scattered signal from scatter as well as reflected power from a perfectly smooth and conducting aluminum sheet are observed for the entire angular range. The amplitudes of scattered power from scatter are normalized using reflected power by the aluminum sheet to calibrate the bistatic scatterometer system. Normalized power is used to calculate s
using the radar equation given by Ulaby (1982). In the bistatic configuration, the transmitter and receiver were opposite each other during measurement of the microwave response of the kidney bean crop. The transmitters consisted of a pathway signal generation (PSG) high-power signal generator (E8257D, 10 MHz to 20 GHz), a C- and X-band pyramidal dual-polarized horn antenna, and an antenna support tower. The receivers consisted of an EPM-P series power

329

330

Radar Remote Sensing

meter (E4416A), a peak and average power sensor (E9327A, 50 MHz to 8 GHz), a pyramidal dual-polarized horn antenna, and an antenna support tower. Polarization of the horn antenna was changed using a 90-degree E-H twister. Portable wooden antenna support towers were specially made in the workshop to carry the transmitting and receiving antennas. The antenna support tower has the ability to change the incidence angle and height of the transmitting and receiving antennas. The incidence angle and height of transmitting and receiving antennas can be measured by a pointer provided on the circular scale and linear scale, respectively. The laser pointer was used to match the focus of the transmitting and receiving antennas at the center of the crop bed at all angular measurements. Bistatic measurements were carried at HH, VV, and horizontal transmitevertical receive polarization for incidence angles of 20e60 degrees at steps of 5 degrees with three independent samples at each incidence angle during the wheat growth cycle. A 20 dBm signal was transmitted. Table 17.1 lists detailed specifications of the bistatic scatterometer framework. It consists of a PSG-high power signal generator (E8257D, 10 MHz to 20 GHz), an EPM- P series power meter (E4416A), a peak and average power sensor (E9327A, 50 MHz to 18 GHz), an X-band pyramidal dualpolarized horn antenna, and an antenna support tower. Table 17.1 Specification of scatterometer system.

Radio-frequency generator

Power meter

Power sensor Frequency (GHz) Beam width

E plane (degrees) H plane (degrees)

Band width (GHz) Antenna gain (dB) Cross-polarization isolation (dB) Polarization modes Antenna type Calibration accuracy (dB) Platform height (m) Incidence angle (degrees) Measurement interval

E8257D, PSG high-power signal generator, 10 MHz to 20 GHz (Agilent Technologies) E4416A, EPM-P series power meter, 10 MHz to 20 GHz (Agilent Technologies) Peak and average power sensor (E9327A, 50 MHz to 18 GHz) 10  0.05 (X-band) 17.3118 19.5982 0.8 20 40 Horizontal transmitehorizontal receive Vertical transmitevertical receive Dual-polarized pyramidal horn 1 3 20 degrees (nadir)e70 degrees 20 min

Fuzzy logic for the retrieval of kidney bean crop growth variables using ground-based scatterometer measurements

2.2 Kidney bean crop variables measurements An outdoor crop bed of 4  4 m2 is specially prepared at the Indian Institute of Technology (BHU), India, for bistatic measurements at numerous growth stages of a kidney bean crop. The crop bed is prepared for in situ measurements at the geolocation of 25
13ʹ52ʹʹ N, 82
38ʹ41ʹʹ E. The kidney bean crop is well-known as a broad leaf crop; it becomes mature at 97  5 days after the date of sowing. The kidney bean crop attained a maximum average height of 49  2 cm in our crop bed during the entire observation. The VWC of the plant is the total water content available in the crop constituents. The total number of plants is calculated in the crop bed and divided by 16 to compute the number of kidney bean plants per square meter (plant density) for the VWC measurement. Samples were taken from the outer edge of the crop bed to compute the VWC. The central area of the crop bed is unaffected by the bistatic scatterometer measurement. The leaf area index (LAI) is measured using an ACCUPAR LP-80 instrument. A linear wooden scale 2 m long is used to measure the plant height (PH) in the crop bed at every data collection day. Soil samples are collected manually at a depth of 5 cm for gravimetric soil moisture measurement from three locations in the crop bed.

3. Fuzzy inference system The fuzzy inference system (FIS) algorithm is realized by formulating data clustering to retrieve kidney bean crop variables. To apply fuzzy rules, the clustering of datasets is required. It helps to set fuzzy rules according to the nature of different clusters of datasets. The number of clusters is basically equal to the number of fuzzy rules. The subtractive clustering algorithm to cluster different datasets and the Takagi-Sugeno-Kang type algorithm are used to generate fuzzy rules between s
and crop variables. In the data clustering by subtractive clustering method, the optimum value of cluster radii is required to obtain the desired result by the FIS algorithm (Chiu, 1994). The accuracy of the desired retrieval depends on the selection of the optimum value of radii. The extreme case of zero error occurs if each data point is considered to be its own cluster. The subtractive clustering method is used to estimate the number of clusters and the cluster centers in a set of data. This method assumes each data point to be a potential cluster center. It computes every data information point to characterize the cluster center depending on the density of near data points. First, it considers all of the highest potential data points to be the first cluster. The surrounding area of the cluster center is dictated by the radii of the clusters. The next cluster and its center position can be dictated by the iterative process of data within radii of the cluster center. The FIS algorithm, a Sugeno-type fuzzy model, is used to generate fuzzy rules from a given inputeoutput data set (Takagi and Sugeno, 1985). FIS algorithms are mostly made of five layers: input, membership function or fuzzification, rule, consequent, and output

331

332

Radar Remote Sensing

or defuzzification. The first layer contains input nodes and directly transmit inputs to the next layer. A mathematical representation of the first layer is given in Eq. (17.1): ð1Þ

Oi

¼ xi

(17.1)

At the second layer, the membership value specifies the degree to which an input value belongs to a fuzzy set. The Gaussian membership function is assumed to calculate the membership values for input xi :  ð1Þ 2 ! Oi  mij ð2Þ Oij ¼ exp  (17.2) s2ij where sij and mij are the standard deviation and mean of the Gaussian membership function of the ith term of jth input variables, respectively. In the third layer, the output at each node is computed by the AND operation between the outputs of the second layer. The product rule is applied to determine the firing strength of each rule. The function of each rule is: Y ð2Þ ð3Þ Oj ¼ Oij (17.3) j

In the fourth layer, inputs to the layer four are delivered from the output of layer 3 and other inputs from the input variables of layer 1. Mathematically, the output at the nodes of layer 4 is: ! n X ð4Þ ð3Þ woj þ (17.4) wij xi Oj ¼ Oj i

where wij are the corresponding parameters of the consequent part. The last layer (or output layer or defuzzification layer) has only one node computing the overall output using all incoming signals as:  PR ð3Þ  Pn PR ð4Þ woj þ j¼1 Oj i wij xi j¼1 Oj 5 y¼O ¼ P ¼ (17.5) PR ð3Þ ð3Þ R j¼1 Oj j¼1 Oj where R is the number of fuzzy rules and n is the number of input variables.

4. Results and discussion 4.1 Time series analysis of bistatic scattering coefficients and crop variables The temporal trend of kidney bean crop variables (i.e., VWC, LAI, and PH) at innumerable growth stages is shown in Fig. 17.1. The trend of all of the crop variables increased

Fuzzy logic for the retrieval of kidney bean crop growth variables using ground-based scatterometer measurements

Figure 17.1 Temporal variation of kidney bean crop variables. LAI, leaf area index; VWC, vegetation water content.

with the crop age. The VWC increased until 97 days after sowing and then began to decrease slightly with the beginning of the fruit-filling stage of the crop. The PH increased sharply until 91 days and then became almost constant. The LAI increased sharply after 60 days of sowing and continued until 91 days. The SM content was approximately constant during the entire observation to investigate the response of crop growth variables. The average gravimetric SM was 18.25% during the field observation. The temporal trends of s
at an angular incidence angle of 20e70 degrees for polarizations (HH and VV) at innumerable growth stages of the crop are shown in Fig. 17.2A and B, respectively. The magnitudes of s
decreased with the growth stages of the kidney

Figure 17.2 Temporal variations in bistatic scattering coefficients at horizontal transmitehorizontal receive and vertical transmitevertical receive polarizations.

333

334

Radar Remote Sensing

bean crop for the entire angular range at both polarizations. The difference in the dynamic range of s
at innumerable growth stages of kidney bean crop was adequate for discriminating the effect of soil moisture and crop variables. At the point when the crop variable was small (VWC ¼ 0.17 kg/m2, LAI ¼ 0.15 m2/m2, and PH ¼ 11.5 cm) after 25 days of sowing, the dynamic ranges of s
were 4.3 and 6.74 dB at HH and VV polarizations, respectively, whereas when the crop variables were high (VWC ¼ 1.96 kg/m2, LAI ¼ 2.83 m2/m2, and PH ¼ 48.8 cm) after 98 days of sowing, the dynamic ranges of s
decreased to 1.77 and 2.13 dB at HH and VV polarizations, respectively. The dynamic range of s
was more prominent when values of crop variables were small at an early age of kidney bean crop compared with the dynamic range of s
of an older age of the crop. Subsequently, the effect of soil moisture on s
was more prominent than the crop effect at an early age. The deficit in the dynamic range of s
represents the influence of the crop effect relative to the soil moisture effect at latter growth stages. Thus, the angular trends became flatter with the age of the crop because the effects of soil were blocked by the developing crop growth parameters.

4.2 Retrieval of kidney bean crop variables The observed datasets such as s
and crop growth variables at 50 and 40 degrees incidence angles were interpolated into 80 datasets for each day at 25e104 days after the crop was sown. Two-thirds of the interpolated datasets are used to train the FIS model whereas the remaining third is used to validate the FIS model. The Pearson (r) correlation test is carried out between s
and kidney crop variables to evaluate the sensitivity of s
with the kidney crop variable from multitemporal, multiangular, and dual-polarized bistatic scatterometer data. The values of r are higher at 50 and 40 degrees incidence angles for HH and VV polarizations, respectively. The values of r indicate the degree of linearity between s
and crop variables. Figs. 17.3 and 17.4 show the correlation matrix plot between s
and kidney crop variables at HH and VV polarizations, respectively. In this study, the value of radii is chosen by simulating the FIS algorithm at different radii values between 0 and 1 at 0.05 steps. For entire radii values, performance is continuously monitored to check the performance of the FIS algorithm at each level. The optimum value of the radii value is 0.2 for the retrieval of crop variables at HH and VV polarizations. The optimum number of clusters is 3 to retrieve VWC, LAI, and PH at HH polarizations. However, the optimum number of clusters is 4, 3, and 4 to retrieve VWC, LAI, and PH at VV polarizations, respectively.

Fuzzy logic for the retrieval of kidney bean crop growth variables using ground-based scatterometer measurements

Figure 17.3 Correlation matrix plot between bistatic scattering coefficients and kidney bean crop variable at horizontal transmitehorizontal receive (HH) polarization. LAI, leaf area index; PH, plant height; VWC, vegetation water content.

The performance of different FIS algorithms is evaluated by comparing the values of performance indices (root mean square error and bias [%]) during the training and validation of datasets using the FIS algorithm. The values of performance indices are computed between estimated and observed crop variables. Figs. 17.5 and 17.6 depict scatterplots with a 1:1 line between estimated and observed values of crop variables by FIS using training and validation datasets. The values of estimated and observed crop variables are close during training and validation by FIS algorithms. Performance of the FIS algorithm for the retrieval of the crop variable at HH and VV polarizations is good during training and validation of the datasets. Retrieval of the VWC is better than the LAI and PH at both polarizations in this investigation.

335

Figure 17.4 Correlation matrix plot between bistatic scattering coefficients and kidney bean crop variable at vertical transmitevertical receive (VV) polarization. LAI, leaf area index; PH, plant height; VWC, vegetation water content.

Figure 17.5 Scatterplot with 1:1 line between estimated crop variables by fuzzy logic and experimentally observed crop variables during training of the model. HH, horizontal transmitehorizontal receive; LAI, leaf area index; RMSE, root mean square error; VV, vertical transmitevertical receive; VWC, vegetation water content.

Fuzzy logic for the retrieval of kidney bean crop growth variables using ground-based scatterometer measurements

Figure 17.6 Scatterplot with 1:1 line between estimated crop variables by fuzzy logic and experimentally observed crop variables during validation of the model. HH, horizontal transmitehorizontal receive; LAI, leaf area index; RMSE, root mean square error; VV, vertical transmitevertical receive; VWC, vegetation water content.

5. Conclusion A decreasing trend of s
with the age of the crop is observed in the angular range of the incidence angle at 20e70 degrees at HH and VV polarizations. The s
increased slightly after 97 days of the crop being sown. Pearson and Spearman correlation analysis shows the highest values of r at 50 and 40 degrees incidence angles for HH and VV polarization, respectively. The sensitivity of HH-polarized s
is higher compared with VV-polarized s
with the kidney bean crop variables. The FIS provides better outcomes for the retrieval of crop variables at HH polarization compared with VV polarization. The retrieval of VWC is more accurate than the other kidney bean crop variables by FIS. Nevertheless, the estimated values of LAI and PH are acceptable with the observed values. The challenge is to get the data information in this way and with the appropriate radar parameters to limit the dependence of radar backscatter on all of the parameters except those related to the crop or vegetation. The importance of conducting good field experiments is to generate a huge bank of data to operate and validate these types of models. It becomes fundamentally necessary to improve our understanding of the microwave interaction process to monitor crop growth and use appropriate soft computing techniques and optimization models for the accurate and timely retrieval of crop yields. The bistatic scatterometer system may be a good choice for the accurate monitoring of crop growth variables.

337

338

Radar Remote Sensing

We envisage that this research work may have the option to monitor the growth of agricultural crops by estimating innumerable crop and soil parameters and to predict for spaceborne and airborne sensor parameters more authentically through ground truth observation using innumerable soft computing techniques. Furthermore, it will fortify a clear understanding of remote sensing application for innumerable purposes, including the design of sensors, crop classification, crop yield, soil moisture retrieval, and agricultural planning. The data information accumulated during the project and the results obtained will be arranged in the form of a final report. The scope of the future study will be designed based on the entire study.

References Brown, S., Quegan, S., Morrison, K., Bennett, J.C., Cookmartin, G., 2003. High-resolution measurements of scattering in wheat canopies-Implications for crop parameter retrieval. IEEE Trans. Geosci. Rem. Sens. 41 (7), 1602e1610. Chiu, S.L., 1994. Fuzzy model identification based on cluster estimation. J. Intell. Fuzzy Syst. 2 (3), 267e278. Ferrazzoli, P., Guerriero, L., Solimini, D., 2000. Simulating bistatic scatter from surfaces covered with vegetation. J. Electromagn. Waves Appl. 14 (2), 233e248. Gupta, D.K., Prasad, R., Kumar, P., Mishra, V.N., 2016. Estimation of crop variables using bistatic scatterometer data and artificial neural network trained by empirical models. Comput. Electron. Agric. 123, 64e73. Gupta, D.K., Prasad, R., Kumar, P., Vishwakarma, A.K., Srivastava, P.K., 2018. Vegetation water content retrieval using scatterometer data at X-band. Geocarto Int. 33 (6), 602e611. Inoue, Y., et al., 2002. Season-long daily measurements of multifrequency (Ka, Ku, X, C, and L) and fullpolarization backscatter signatures over paddy rice field and their relationship with biological variables. Remote Sens. Environ. 81 (2), 194e204. Johnson, J.T., Ouellette, J.D., 2014. Polarization features in bistatic scattering from rough surfaces. IEEE Trans. Geosci. Rem. Sens. 52, 1616e1626. Kim, Y., Jackson, T., Bindlish, R., Lee, H., Hong, S., 2012. Radar vegetation index for estimating the vegetation water content of rice and soybean. Geosci. Rem. Sens. Lett. IEEE 9 (4), 564e568. Kim, Y., Jackson, T., Bindlish, R., Lee, H., Hong, S., 2013. Monitoring soybean growth using L-, C-, and X-band scatterometer data. Int. J. Rem. Sens. 34 (11), 4069e4082. Kumar, P., Prasad, R., Gupta, D.K., Vishwakarma, A.K., Choudhary, A., 2017. Retrieval of rice crop growth variables using multi-temporal RISAT-1 remotely sensed data. Russ. Agric. Sci. 43 (6), 461e465. Kumar, P., et al., 2018. Estimation of winter wheat crop growth parameters using time series Sentinel-1A SAR data. Geocarto Int. 33 (9), 942e956. Kumar, P., et al., 2019. Comprehensive evaluation of soil moisture retrieval models under different crop cover types using C-band synthetic aperture radar data. Geocarto Int. 34 (9), 1022e1041. Mittal, G., Singh, D., 2010. Critical analysis of microwave specular scattering response on roughness parameter and moisture content for bare periodic rough surfaces and its retrieval. Progr. Electromag. Res. 100, 129e152. Pandey, A., Prasad, R., Singh, V., Jha, S., Shukla, K., 2013. Crop parameters estimation by fuzzy inference system using X-band scatterometer data. Adv. Space Res. 51 (5), 905e911. Prasad, R., 2011. Estimation of kidney bean crop variables using ground-based scatterometer data at 9.89 GHz. Int. J. Rem. Sens. 32 (1), 31e48. Shao, Y., et al., 2001. Rice monitoring and production estimation using multitemporal RADARSAT. Remote Sens. Environ. 76 (3), 310e325.

Fuzzy logic for the retrieval of kidney bean crop growth variables using ground-based scatterometer measurements

Taconet, O., et al., 1994. Estimation of soil and crop parameters for wheat from airborne radar backscattering data in C and X bands. Remote Sens. Environ. 50 (3), 287e294. Takagi, T., Sugeno, M., 1985. Fuzzy identification of systems and its applications to modeling and control. Syst. Man Cybernet. IEEE Trans. SMC-15 (1), 116e132. Thuy Le, T., et al., 1997. Rice crop mapping and monitoring using ERS-1 data based on experiment and modeling results. IEEE Trans. Geosci. Rem. Sens. 35 (1), 41e56. Ulaby, F.T., Moore, R.K., Fung, A.K., 1982. Microwave Remote Sensing Active and Passive-Volume II: Radar Remote Sensing and Surface Scattering and Emission Theory. Addison-Wesley, Reading, MA. Zhang, Y.Y., Wu, Z.S., 2016. Bistatic scattering characteristics of wheat and soybean by radiative transfer model in L band and C band. Progr. Electromag. Res. 67, 121e136.

339

This page intentionally left blank

CHAPTER 18

Monitoring tropical peatlands subsidence by time-series interferometric synthetic aperture radar (InSAR) technique Deha Agus Umarhadi1, Ram Avtar1, 12, Pankaj Kumar3, Ali P. Yunus4, 5, Tonni Agustiono Kurniawan6, 7, Ali Kharrazi8, 9, 11, Mamoru Ishikawa1, 2 and Wirastuti Widyatmanti10 1

Graduate School of Environmental Science, Faculty of Environmental Earth Science, Hokkaido University, Sapporo, Japan; Faculty of Environmental Earth Science, Hokkaido University, Sapporo, Japan; 3Natural Resources and Ecosystem Services, Institute for Global Environmental Strategies, Hayama, Kanagawa, Japan; 4Department of Earth and Environmental Science, Indian Institute of Science Education and Research Mohali, Mohali, India; 5Center for Climate Change Adaptation, National Institute for Environmental Studies, Tsukuba, Japan; 6Faculty of Social Work, Health, and Nursing, Ravensburg-Weingarten University of Applied Sciences, Weingarten, Germany; 7College of the Environment and Ecology, Xiamen University, Xiamen, PR China; 8Advanced Systems Analysis Group, International Institute for Applied Systems Analysis, Laxenburg, Austria; 9Euro-Mediterranean Center for Climate Change, Ca’ Foscari University of Venice, Venice, Italy; 10Department of Geographic Information Science, Faculty of Geography, Universitas Gadjah Mada, Yogyakarta, Indonesia; 11Faculty of Global Studies, Akita International University, Akita, Japan; 12Institute for the Advanced Study of Sustainability, United Nations University (UNU-IAS), Shibuya City, Tokyo, Japan

2

1. Introduction Peatlands contribute to a quarter of the global carbon stock, accounting for 612 Gt of soil carbon (Yu et al., 2010). These ecosystems are an important carbon sink maintaining abundant decomposed organic matter accumulated over centuries. Peatlands cover about 2.83% (423 Mha) of the earth’s land surface (Xu et al., 2018) and are mainly found in temperate and boreal regions, whereas tropics contribute to approximately 10e16% of the total global peatland area (Page et al., 2011). More than half of tropical peatlands (56%) are located in Southeast Asia, most of which are in Indonesia (i.e., 47% of global tropical peatlands) (Hooijer et al., 2006, 2012; Page et al., 2011). Despite their smaller extent, tropical peatlands store 18%e25% of global peat volume and therefore represent a critical role specifically for the carbon cycle and more broadly greenhouse gas emissions (Leifeld and Menichetti, 2018; Page et al., 2011). Tropical peatlands are at an accelerated rate of degradation owing to the drained condition resulted from drainage canal construction, particularly in Southeast Asian tropical peatlands, which has exponentially increased for plantations and agricultural purposes (Dadap et al., 2021). Peat soils mostly consist of water (90%); therefore, as the organic soils are exposed to air owing to drainage, oxidation occurs, leading to the loss of carbon along with subsidence (Hooijer et al., 2015). Wösten et al. (1997) defined three

Radar Remote Sensing ISBN 978-0-12-823457-0, https://doi.org/10.1016/B978-0-12-823457-0.00013-6

© 2022 Elsevier Inc. All rights reserved.

341

342

Radar Remote Sensing

components of peat subsidence: (1) consolidation: compression of permanently peat layers below groundwater level; (2) oxidation: decomposition of organic matter; and (3) shrinkage: volume reduction caused by desiccation. Not only are they a fount of carbon emission, drained peatlands have a direct impact on their surroundings because they increase vulnerability to floods in the rainy season and fires in the dry season (Lupascu et al., 2020; Vetrita and Cochrane, 2019). Traditionally, peat subsidence monitoring is conducted by observing poles anchored into the underlying mineral subsoil (Evans et al., 2019). More recently, terrestrial methods have been developed to use advanced techniques (e.g., geodetic-based measurements [Dahdal, 2011; Grzywna, 2017; Reeve et al., 2013], the rod surface elevation tableemarker horizon method [Jaya et al., 2016; Webb et al., 2013], and real-time high-resolution cameras [Evans et al., 2021]). These field-based measurements provide high-accuracy results; however, they are weak in spatial coverage. On the other hand, interferometry synthetic aperture radar (InSAR) is renowned for its ability to detect land subsidence with millimeter-scale accuracy. Using this advantage, some studies applied InSAR to detect subsidence in northern and tropical peatlands (Alshammari et al., 2018; Dahdal, 2011; Hoyt et al., 2020; Khakim et al., 2020; Marshall et al., 2018; Susanti and Anjasmara, 2019; Umarhadi et al., 2021; Zhou et al., 2019). Using the case study of a peatland area in Sintang Regency, Indonesia, where the degradation stage varied during the image observation period, this chapter reviews the use of InSAR for tropical peatlands with a time-series approach.

2. Interferometry synthetic aperture radar for tropical peatlands 2.1 Interferometry synthetic aperture radar and time-series interferometry synthetic aperture radar A raw single look complex synthetic aperture radar (SAR) image has two properties (i.e., amplitude and phase) resulting from the return radar signals. Amplitude represents the strength of signals received back to the sensor after interaction with surface objects, which is related to terrain slope, surface roughness, and dielectric constants (Lu et al., 2007), whereas phase is a measurement of the distance between the target object and the radar sensor. In InSAR, phase information in two SAR images is extracted to produce the phase differences of the corresponding points. The result of phase differences between master and slave images is interferometric phase (called an interferogram). Phase differences are composed of contributions from topography, surface displacement, the earth’s curvature, atmospheric phase screening, orbital error, and noise (Ramirez et al., 2020). To generate a digital elevation model (DEM), it is assumed that no surface displacement occurs. A widely known elevation model is Shuttle Radar Topography Mission (SRTM) DEM, the first near-global high-resolution elevation model covering latitude 60
N to 57
S (Rabus et al., 2003). SRTM DEM was generated based on single-pass interferometry with a fixed baseline from two antennas (i.e., the main and outboard

Monitoring tropical peatlands subsidence by time-series interferometric synthetic aperture radar (InSAR) technique

antennas) onboard the Space Shuttle Endeavor (Farr and Kobrick, 2000). Another global DEM mission is TanDEM-X (TerraSAR-X add-on for Digital Elevation Measurement) with the implementation of single-pass interferometry from two identical satellites in close formation (Krieger et al., 2007). On the other hand, a repeat-pass interferometry approach from different acquisition times can measure the displacement of the surface. Because it is intended to derive land motion information, only the deformation phase should remain whereas the contribution of topography is removed (Liu and Mason, 2016). This differential InSAR (DInSAR) method has been widely used to measure displacement driven by earthquakes, volcanic activity, landslides, glacier movement, and other subsidence events caused by mining activity and groundwater extraction (Atzori et al., 2009; Cal
o et al., 2017; Colesanti and Wasowski, 2006; De Novellis et al., 2019; Pawluszek-Filipiak and Borkowski, 2020; Villarroel et al., 2018). To retrieve object information from radar phase data, an object should be wellcorrelated or have enough coherence above 0.2 (Wei and Sandwell, 2010). Coherence refers to phase similarity between two SAR images (scaled in 0e1), presenting a total of geometric, Doppler centroid, volume, thermal, temporal, and processing decorrelation (Zebker and Villasenor, 1992). For repeat-pass InSAR, the temporal aspect is a main source of decorrelation owing to changes in the earth’s surface, because target objects may change between different acquisition times (Li and Lindenschmidt, 2018). For instance, because of wind, the canopy of vegetation always moves. This implies that coherence decreases with denser vegetation coverage. This is a main problem in applying the InSAR method in a densely vegetated area (Pepe and Cal o, 2017). Spaceborne SAR sensors work in microwave frequency bands, including Ka (27e40 GHz; 1.1e0.8 cm), K (18e27 GHz; 1.7e1.1 cm), Ku (12e18 GHz; 2.4e1.7 cm), X (8e12 GHz; 3.8e2.4 cm), C (4e8 GHz; 7.5e3.8 cm), S (2e4 GHz; 15e7.5 cm), L (1e2 GHz; 30e15 cm), and P (0.3e1 GHz; 100e30 cm), whereas most earth observation SAR satellites operate in X-, C-, and L-bands (Meyer, 2019). In general, the length of the SAR wavelength corresponds to its penetration capability. Shorter wavelengths such as the X-band can interact only with the upper part of vegetation, whereas the L-band is able to penetrate vegetation into the ground for coniferous forests (Ji et al., 2020). For vegetated areas, the degree of penetration is equal to the InSAR decorrelation because the longer wavelength pulses interact with more stable objects, minimizing decorrelation. Another way to decrease decorrelation is to apply time-series InSAR with a stack of numerous SAR images acquired in a satellite track. Also, it can diminish artifacts induced by atmospheric delay anomalies and orbit errors, as well as handle covariance characteristics of data distribution by least-square inversion (Lu et al., 2007). Many time-series InSAR methods have been developed referring to the two most prominent approaches. These are based on different approaches: permanent scatterers and distributed scatterers (i.e., persistent scatterer interferometry [PS InSAR] [Ferretti et al., 2001] and small

343

344

Radar Remote Sensing

baseline subset [SBAS] [Berardino et al., 2002], respectively). PS InSAR identifies persistent scatterers of strong and coherent objects over time, which are typical of urban areas. Meanwhile, SBAS InSAR maximizes correlated areas by its redundant networks with short temporal baselines; hence, this method is applied more for rural areas, including peatland areas. 2.2 Peatland interferometry synthetic aperture radar Because degradation in peatlands can be monitored by their subsidence, it extends the range of InSAR applications. Despite the crucial role of peatlands, particularly tropical peatlands, and the potential of InSAR to monitor them, the number of studies is still limited. Table 18.1 lists studies conducting InSAR measurements to estimate subsidence over tropical peatland areas. The use of traditional DInSAR methods has reported uncertainties owing to decorrelation (Dahdal, 2011; Susanti and Anjasmara, 2019). A study conducted by Table 18.1 Studies conducting subsidence estimation in tropical peatlands using interferometry synthetic aperture radar. Synthetic aperture radar data

Reference

Method

Dahdal (2011)

Four-pass and complex interferogram differential InSAR Differential InSAR

ERS-1/2 C-band

Central Kalimantan, Borneo, Indonesia

Sentinel-1 C-band

Riau, Sumatera, Indonesia

SBAS InSAR

Sentinel-1 C-band

Kuala Lumpur International Airport, Malaysia South Sumatera, Indonesia Central Kalimantan, Borneo, Indonesia Sumatera, Borneo, and Peninsular Malaysia Bengkalis Island, Indonesia

Susanti and Anjasmara (2019) Marshall et al. (2018) Khakim et al. (2020) Zhou et al. (2019)

SBAS InSAR

Sentinel-1 C-band

SBAS InSAR

Hoyt et al. (2020)

SBAS InSAR

ALOS PALSAR L-band ALOS PALSAR L-band

Umarhadi et al. (2021)

SBAS InSAR

Sentinel-1 C-band and ALOS PALSAR-2 L-band

Location

ERS, European Remote Sensing satellite; InSAR, interferometry synthetic aperture radar; PALSAR, Phased Array LBand Synthetic Aperture Radar; SBAS, small baseline subset.

Monitoring tropical peatlands subsidence by time-series interferometric synthetic aperture radar (InSAR) technique

Dahdal (2011) using C-band SAR data reported that the long gap of acquisition time (i.e., over 27 months) led to the loss of coherence. Other sources of error are inaccurate baseline estimation and a long perpendicular baseline. For movement detection, a shorter baseline is needed to decrease the decorrelation of the radar signals. Therefore, the baseline is shortened in the successor of ERS (i.e., Sentinel-1 satellite). Nevertheless, with the improved baseline of Sentinel-1, temporal decorrelation remains the main issue, resulting in inconsistencies of subsidence and uplift in the same area with a 1-year temporal baseline (Susanti and Anjasmara, 2019). Time-series SBAS InSAR has proved the potential to estimate subsidence in tropical peatlands with commonly used Sentinel-1 and ALOS Phased Array L-Band Synthetic Aperture Radar (PALSAR) data. Previous studies found that C-band Sentinel-1 could well-characterize peat subsidence owing to high temporal resolution (Khakim et al., 2020; Marshall et al., 2018). In the tropical region, Sentinel-1 acquires data every 12 days in the same satellite track. Marshall et al. (2018) used a newly developed intermittent SBAS (ISBAS), considering the intermittently coherent pixels alongside coherent ones (Cigna and Sowter, 2017), which outperformed conventional SBAS to produce higher coherence in a vegetated area. The number of available ALOS PALSAR (L-band) images is often inadequate for a particular area, even though it is more suitable for use in tropical peatlands. The two overlapped tracks in Central Kalimantan studied by Zhou et al. (2019) provided 12 and 14 images with a large time span from 2006e10, whereas images used by Hoyt et al. (2020) varied from 5 to 15 scenes for each area. The use of PALSAR-2 data (successor of PALSAR) in tropical peatland areas can provide better results because of high penetration that can minimize decorrelation compared with Sentinel-1 (Umarhadi et al., 2021).

3. Case study: Sintang, Indonesia This section demonstrates the application of SBAS InSAR using ALOS PALSAR data to estimate surface displacement in a part of Sintang Regency, Indonesia. Sintang is situated in West Kalimantan, Indonesia, upstream of the Kapuas Watershed, Borneo Island (Fig. 18.1). Peatlands cover 789 km2 of the total of 21,638-km2 administrative area (Anshari et al., 2010). Similar to other peatlands in insular Southeast Asia, peatlands in this regency have been facing degradation owing to deforestation for agricultural and plantations (Rehman et al., 2015; Schoneveld et al., 2019). This leads to continuous subsidence associated with the increase in hotspot occurrence as an indicator of wildfires (Arrizmi, 2015; Yanuarsyah et al., 2016). This study used ALOS PALSAR L-band images acquired from 2007e09. Fourteen images (Table 18.2) were downloaded from the Alaska Satellite Facility Distributed Active Archive Center services in level 1.0.

345

346

Radar Remote Sensing

Figure 18.1 (A) Location of the study area. (B) Landsat 5 image in 2008 retrieved from Google Earth Engine with median filtering.

3.1 Methods PALSAR data were processed using two toolboxes: Interferometric SAR Scientific Computing Environment (ISCE) to produce a stack of differential interferograms with SBAS, and with the Miami InSAR Time-Series Software in Python (MintPy) toolbox for time-series processing (Rosen et al., 2012; Yunjun et al., 2019). Within the ISCE workflow, steps consist of coregistration, interferogram generation, Goldstein adaptive filtering, topographic phase removal, and phase unwrapping. A total of 55 image pairs were constructed with a maximum temporal baseline of 1 year for coregistration. We applied multilooking of 16 (azimuth looks)  4 (range looks), resulting in a pixel size of about 69  62 m. One-arc second SRTM DEM (30-m spatial resolution) was used to remove topographic contribution. The interferograms were then unwrapped by two-dimensional statistical-cost network-flow phase-unwrapping algorithm (Chen and Zebker, 2001). SBAS workflow was carried out in MintPy/smallbaselineApp with the input of a stack of unwrapped interferograms from previous processing (Yunjun et al., 2019). Correlation-based network modification was employed by discarding interferograms below the average spatial correlation of 0.7 (Chaussard et al., 2015). Four interferograms were eliminated in this step, as shown in Fig. 18.2. The network of interferograms was then inverted using weighted least squares inversions based on the weight of the inverse of covariance. Some corrections were applied on raw phase time series, including tropospheric delay, phase deramping, and topographic phase correction. We used ERA-5 data for topographic delay correction (Jolivet et al., 2011). The remaining residual phase was

Image number

Date (dd/mm/yyyy)

Time interval (days)

Perpendicular baseline (m)

Image number

Date (dd/mm/yyyy)

1 2

January 28, 2007 June 15, 2007

e 138

e 216

8 9

3

September 15, 2007 December 16, 2008 January 31, 2008 March 17, 2008 May 2, 2008

92

450

10

June 17, 2008 September 17, 2008 November 2, 2008

458 320 46 46

552 570 146 651

11 12 13 14

December 18, 2008 February 2, 2009 June 20, 2009 August 5, 2009

4 5 6 7

Time interval (days)

Perpendicular baseline (m)

46 92

231 742

46

712

46 46 138 46

461 593 827 111

Monitoring tropical peatlands subsidence by time-series interferometric synthetic aperture radar (InSAR) technique

Table 18.2 Details of ALOS Phased Array L-Band Synthetic Aperture Radar images.

347

348

Radar Remote Sensing

Figure 18.2 Interferogram network visualized with coherence value. Interferograms with spatial coherence below 0.7 were discarded for further processing.

then estimated its root mean square. An image of September 17, 2008 was excluded owing to the high residual phase (Fig. 18.3). Annual line-of-sight (LOS) velocity was calculated based on the corrected phase time series (Fig. 18.4), which was then converted to vertical velocity. In previous studies, carbon loss in tropical peatlands was estimated using subsidence information (Leifeld et al., 2011; van den Akker et al., 2008): Closs ¼ St  DBD  Cdw  area

(18.1)

in which annual carbon loss in an area is a multiplication of subsidence in 1 year (St), DBD is dry bulk density of the peat below the water table, and carbon concentration is Cdw (Couwenberg and Hooijer, 2013). We used the values measured in Sumatra, Indonesia (i.e., 80 kg/m3 and 0.55 for DBD and Cdw, respectively) (Couwenberg and Hooijer, 2013). 3.2 Results and discussion Vertical velocity (cm/year) was estimated from the SBAS InSAR within January 28, 2007 to August 5, 2009. As Fig. 18.5 shows, almost the whole study area is subsiding, including in the nonpeat area. Subsidence in the vicinity of peatlands is the result of a transition area between peat soils and mineral soils. In general, the peat area subsided an average of 1.50 cm/year with a range of 11.58 to 1.82 cm/year. The data

Monitoring tropical peatlands subsidence by time-series interferometric synthetic aperture radar (InSAR) technique

Figure 18.3 Root mean square of residual phase indicates elimination of the date of September 17, 2008 owing to the high residual phase.

349

350

Radar Remote Sensing

Figure 18.4 LOS displacement images. The black point is the location of the reference.

Figure 18.5 Vertical velocity resulted from small baseline subset interferometry synthetic aperture radar.

distribution is depicted in Fig. 18.6, showing that the majority rate is around 0e2 cm/year. The estimated subsidence in this area is lower than the mean regional subsidence (2.24 cm/year) measured by Hoyt et al. (2020) using similar methods. To validate the results, some points are chosen in the most stable area in the center of the urban area (i.e., around the reference point). The mean vertical velocity of the 100

Monitoring tropical peatlands subsidence by time-series interferometric synthetic aperture radar (InSAR) technique

Figure 18.6 Histogram of vertical velocity in peatland area.

extracted pixels is 0.37  0.25 cm/year, which shows that the stable area is also subsiding at a very low rate. The urban area in Sintang is located close to the peat area, which may also consist of peat soil in the transition area. The peat area in Sintang Regency is patchy, consisting of scattered small to large peat domes, classified as basin peat domes separated by rivers and small hills (Anshari et al., 2010). Each area has a different subsidence rate corresponding to how the peatland is treated. For analysis, we focus on three areas: areas A, B, and C. As interpreted from the Landsat image (Fig. 18.1), the three areas have different land cover types. Area A has been partially opened; some parts are covered by pristine forest and some areas are mixed plantations. Almost the whole of area B has been converted to oil palm plantations. Meanwhile, pristine forests cover area C. Boxplots of vertical velocity in these areas are illustrated in Fig. 18.7. The subsidence rate corresponds to the degradation of peatlands, with area B subsided at the highest rate (5.53  1.73 cm/year), whereas the pristine forests of area C have the lowest rate (0.62  0.70 cm/year). The mean vertical velocity in area A is 1.43  0.92 cm/year. Some pixels in area A could not be characterized because of the decorrelation. This is because of the change in land cover that deforestation was ongoing during the observation period. Based on the vertical velocity, carbon loss was calculated using Eq. (18.1). During the observation period of 2007e9, peatlands in the study area emitted a net carbon of 0.354 MtC/year from their 537.39-km2 total area. It is equivalent to 6.59 tC/ha per year. For comparison, the average value is slightly higher than the emission factor of drained shrubs in tropical peatlands (5.3 tC/ha per year) according to the Intergovernmental Panel on Climate Change report (Drösler et al., 2014). This mean rate is much

351

352

Radar Remote Sensing

Figure 18.7 Boxplots of vertical velocity in areas A, B, and C.

lower than the plantation area (acacia and oil palm), reaching 18 tC/ha per year (Couwenberg and Hooijer, 2013). It is because plantations had not been expanded in a large area in Sintang Regency during that time. When we checked recent satellite imagery, most of the area has been converted to oil palm plantations. Consequently, the peat may subside more, accompanied by a the high increase in the rate of carbon loss.

4. Summary In this chapter, we presented the potential of tropical peatland monitoring by employing InSAR. This method has been applied in diverse geographical phenomena associated with land displacement. In an area with a high density of vegetation coverage, such as peatlands, decorrelation is the main problem. Therefore studies applied time-series InSAR based on a small subset baseline to minimize decorrelation. Sentinel-1 C-band

Monitoring tropical peatlands subsidence by time-series interferometric synthetic aperture radar (InSAR) technique

and ALOS PALSAR L-band are commonly used to monitor subsidence by SBAS InSAR, taking advantage of high temporal resolution (Sentinel-1) and high penetration (PALSAR), respectively. This study demonstrated SBAS InSAR in the tropical peatland area in Sintang Regency, Borneo Island, Indonesia using ALOS PALSAR data in 2007e9. During the observation date, 0.354 Mt carbon was estimated to be lost annually, derived from the peat subsidence estimate. Periodic monitoring of Indonesian peatlands is essential to sustainable management, because these peatlands are still degrading and changing to other land use classes. The high temporal resolution and freely available Sentinel-1 have the potential to deal with near real-time monitoring. In the future, the upcoming National Aeronautics and Space AdministrationeIndian Space Research Organization SAR satellite may provide great data combining 12 days’ temporal resolution and L-band sensors (Alvarez-Salazar et al., 2014).

Acknowledgments The authors would like to thank JAXA for providing ALOS PALSAR data through the Alaska Satellite Facility Distributed Active Archive Center. SAR data were processed using InSAR Scientific Computing Environment and MintPy open-source processing tools. We are also thankful to Sousei Grant, Hokkaido University, Japan. We would also like to acknowledge the financial support of MEXTMonbukagakusho, Japan for the scholarship of the first author.

References Alshammari, L., Large, D., Boyd, D., Sowter, A., Anderson, R., Andersen, R., Marsh, S., 2018. Long-term peatland condition assessment via surface motion monitoring using the ISBAS DInSAR technique over the flow country, Scotland. Rem. Sens. 10 (7), 1103. https://doi.org/10.3390/rs10071103. Alvarez-Salazar, O., Hatch, S., Rocca, J., Rosen, P., Shaffer, S., Shen, Y., Sweetser, T., Xaypraseuth, P., 2014. In: Meynart, R., Neeck, S.P., Shimoda, H. (Eds.), Mission Design for NISAR Repeat-Pass Interferometric SAR, p. 92410C. https://doi.org/10.1117/12.2074162. Anshari, G.Z., Afifudin, M., Nuriman, M., Gusmayanti, E., Arianie, L., Susana, R., Nusantara, R.W., Sugardjito, J., Rafiastanto, A., 2010. Drainage and land use impacts on changes in selected peat properties and peat degradation in West Kalimantan Province, Indonesia. Biogeosciences 7 (11), 3403e3419. https://doi.org/10.5194/bg-7-3403-2010. Arrizmi, F., 2015. Pengaruh infiltrasi terhadap penurunan permukaan tanah pada lahan gambut kawasan konservasi taman wisata alam (TWA) baning kabupaten Sintang. Jurnal Teknologi Lingkungan Lahan Basah 3 (1). https://doi.org/10.26418/jtllb.v3i1.9080. Atzori, S., Hunstad, I., Chini, M., Salvi, S., Tolomei, C., Bignami, C., Stramondo, S., Trasatti, E., Antonioli, A., Boschi, E., 2009. Finite fault inversion of DInSAR coseismic displacement of the 2009 L’Aquila earthquake (central Italy). Geophys. Res. Lett. 36 (15). https://doi.org/10.1029/ 2009GL039293. Berardino, P., Fornaro, G., Lanari, R., Sansosti, E., 2002. A new algorithm for surface deformation monitoring based on small baseline differential SAR interferograms. IEEE Trans. Geosci. Rem. Sens. 40 (11), 2375e2383. https://doi.org/10.1109/TGRS.2002.803792. Cal
o, F., Notti, D., Galve, J., Abdikan, S., Gör€ um, T., Pepe, A., Balik Sanli, F., 2017. DInSAR-based detection of land subsidence and correlation with groundwater depletion in konya plain, Turkey. Rem. Sens. 9 (1), 83. https://doi.org/10.3390/rs9010083.

353

354

Radar Remote Sensing

Chaussard, E., B€ urgmann, R., Fattahi, H., Nadeau, R.M., Taira, T., Johnson, C.W., Johanson, I., 2015. Potential for larger earthquakes in the East San Francisco Bay Area due to the direct connection between the hayward and calaveras faults: the Hayward and Calaveras Connection. Geophys. Res. Lett. 42 (8), 2734e2741. https://doi.org/10.1002/2015GL063575. Chen, C.W., Zebker, H.A., 2001. Two-dimensional phase unwrapping with use of statistical models for cost functions in nonlinear optimization. J. Opt. Soc. Am. A 18 (2), 338. https://doi.org/10.1364/ JOSAA.18.000338. Cigna, F., Sowter, A., 2017. The relationship between intermittent coherence and precision of ISBAS InSAR ground motion velocities: ERS-1/2 case studies in the UK. Rem. Sens. Environ. 202, 177e198. https://doi.org/10.1016/j.rse.2017.05.016. Colesanti, C., Wasowski, J., 2006. Investigating landslides with space-borne synthetic aperture radar (SAR) interferometry. Eng. Geol. 88 (3e4), 173e199. https://doi.org/10.1016/j.enggeo.2006.09.013. Couwenberg, J., Hooijer, A., 2013. Towards robust subsidence-based soil carbon emission factors for peat soils in south-east Asia, with special reference to oil palm plantations. Mires Peat 12 (Art. 1). Dadap, N.C., Hoyt, A.M., Cobb, A.R., Oner, D., Kozinski, M., Fua, P.V., Rao, K., Harvey, C.F., Konings, A.G., 2021. Drainage canals in southeast asian peatlands increase carbon emissions. AGU Adv. 2 (1). https://doi.org/10.1029/2020AV000321. Dahdal, B., 2011. The Use of Interferometric Spaceborne Radar and GIS to Measure Peat Subsidence in Indonesia. University of Leicester. De Novellis, V., Atzori, S., De Luca, C., Manzo, M., Valerio, E., Bonano, M., Cardaci, C., Castaldo, R., Di Bucci, D., Manunta, M., Onorato, G., Pepe, S., Solaro, G., Tizzani, P., Zinno, I., Neri, M., Lanari, R., Casu, F., 2019. DInSAR analysis and analytical modeling of Mount Etna displacements: the december 2018 volcano-tectonic crisis. Geophys. Res. Lett. 46 (11), 5817e5827. https://doi.org/10.1029/ 2019GL082467. Drösler, M., Verchot, L.V., Freibauer, A., Pan, G., Evans, C.D., Bourbonniere, R.A., Alm, J.P., Page, S., Agus, F., Hergoualc’h, K., Couwenberg, J., Jauhiainen, J., Sabiham, S., Wang, C., Srivastava, N., Bourgeau-Chavez, L., Hooijer, A., Minkkinen, K., French, N., Larkin, N., 2014. Drained inland organic soils. In: Hiraishi, T., Krug, T., Tanabe, K., Srivastava, N., Jamsranjav, B., Fukuda, M., Troxler, T. (Eds.), 2013 Supplement to the 2006 IPCC Guidelines for National Greenhouse Gas Inventories: Wetlands. Intergovernmental Panel on Climate Change (IPCC). Evans, C.D., Callaghan, N., Jaya, A., Grinham, A., Sjogersten, S., Page, S.E., Harrison, M.E., Kusin, K., Kho, L.K., Ledger, M., Evers, S., Mitchell, Z., Williamson, J., Radbourne, A.D., JovaniSancho, A.J., 2021. A novel low-cost, high-resolution camera system for measuring peat subsidence and water table dynamics. Front. Environ. Sci. 9, 630752. https://doi.org/10.3389/fenvs.2021.630752. Evans, C.D., Williamson, J.M., Kacaribu, F., Irawan, D., Suardiwerianto, Y., Hidayat, M.F., Laur
en, A., Page, S.E., 2019. Rates and spatial variability of peat subsidence in Acacia plantation and forest landscapes in Sumatra, Indonesia. Geoderma 338, 410e421. https://doi.org/10.1016/ j.geoderma.2018.12.028. Farr, T.G., Kobrick, M., 2000. Shuttle radar topography mission produces a wealth of data. EOS 81 (48), 583. https://doi.org/10.1029/EO081i048p00583. Ferretti, A., Prati, C., Rocca, F., 2001. Permanent scatterers in SAR interferometry. IEEE Trans. Geosci. Rem. Sens. 39 (1), 8e20. https://doi.org/10.1109/36.898661. Grzywna, A., 2017. The degree of peatland subsidence resulting from drainage of land. Environ. Earth Sci. 76 (16), 559. https://doi.org/10.1007/s12665-017-6869-1. Hooijer, A., Page, S., Jauhiainen, J., Lee, W.A., Lu, X.X., Idris, A., Anshari, G., 2012. Subsidence and carbon loss in drained tropical peatlands. Biogeosciences 9 (3), 1053e1071. https://doi.org/10.5194/bg-91053-2012. Hooijer, A., Silvius, M., Wösten, H., Page, S., 2006. PEAT-CO2, Assessment of CO2 Emissions from Drained Peatlands in SE Asia. Delft Hydraulics Report Q3943 (2006). Hooijer, A., Vernimmen, R., Mawdsley, N., Page, S., Mulyadi, D., Visser, M., 2015. Assessment of Impacts of Plantation Drainage on the Kampar Peninsula Peatland, Riau. Deltares Report 1207384 to Wetlands International, CLUA and Norad.

Monitoring tropical peatlands subsidence by time-series interferometric synthetic aperture radar (InSAR) technique

Hoyt, A.M., Chaussard, E., Seppalainen, S.S., Harvey, C.F., 2020. Widespread subsidence and carbon emissions across Southeast Asian peatlands. Nat. Geosci. 13 (6), 435e440. https://doi.org/10.1038/ s41561-020-0575-4. Jaya, A., Murdiyarso, D., Sasmito, S.D., Rumbang, N., August 15, 2016. Surface elevation changes of tropical peatland under different land covers: a Preliminary account following 2015 fires in Central Kalimantan, Indonesia. In: 2016 International Peat Society Congress, Serawak. Ji, Y., Huang, J., Ju, Y., Guo, S., Yue, C., 2020. Forest structure dependency analysis of L-band SAR backscatter. PeerJ 8, e10055. https://doi.org/10.7717/peerj.10055. Jolivet, R., Grandin, R., Lasserre, C., Doin, M.-P., Peltzer, G., 2011. Systematic InSAR tropospheric phase delay corrections from global meteorological reanalysis data. Geophys. Res. Lett. 38 (17). https:// doi.org/10.1029/2011GL048757 n/a-n/a. Khakim, M.Y.N., Bama, A.A., Yustian, I., Poerwono, P., Tsuji, T., Matsuoka, T., 2020. Peatland subsidence and vegetation cover degradation as impacts of the 2015 El ni~ no event revealed by Sentinel1A SAR data. Int. J. Appl. Earth Obs. Geoinf. 84, 101953. https://doi.org/10.1016/j.jag.2019.101953. Krieger, G., Moreira, A., Fiedler, H., Hajnsek, I., Werner, M., Younis, M., Zink, M., 2007. TanDEM-X: a satellite formation for high-resolution SAR interferometry. IEEE Trans. Geosci. Rem. Sens. 45 (11), 3317e3341. https://doi.org/10.1109/TGRS.2007.900693. Leifeld, J., Menichetti, L., 2018. The underappreciated potential of peatlands in global climate change mitigation strategies. Nat. Commun. 9 (1), 1071. https://doi.org/10.1038/s41467-018-03406-6. Leifeld, J., M€ uller, M., Fuhrer, J., 2011. Peatland subsidence and carbon loss from drained temperate fens: peatland drainage and carbon loss. Soil Use Manag. 27 (2), 170e176. https://doi.org/10.1111/ j.1475-2743.2011.00327.x. Li, Z., Lindenschmidt, K.-E., 2018. Coherence of Radarsat-2, Sentinel-1, and ALOS-1 PALSAR for monitoring spatiotemporal variations of river ice covers. Can. J. Rem. Sens. 44 (1), 11e25. https://doi.org/ 10.1080/07038992.2018.1419424. Liu, J.G., Mason, P.J., 2016. Introduction to interferometric synthetic aperture radar technique. In: Liu, J.G., Mason, P.J. (Eds.), Image Processing and GIS for Remote Sensing. John Wiley & Sons, Ltd, pp. 113e125. https://doi.org/10.1002/9781118724194.ch10. Lu, Z., Kwoun, O.-I., Rykhus, R.P., 2007. Interferometric synthetic aperture radar (InSAR)dits past, present and future. Photogramm. Eng. Rem. Sens. 73 (3), 217e221. Lupascu, M., Varkkey, H., Tortajada, C., 2020. Is flooding considered a threat in the degraded tropical peatlands? Sci. Total Environ. 723, 137988. https://doi.org/10.1016/j.scitotenv.2020.137988. Marshall, C., Large, D.J., Athab, A., Evers, S.L., Sowter, A., Marsh, S., Sjögersten, S., 2018. Monitoring tropical peat related settlement using ISBAS InSAR, Kuala Lumpur International Airport (KLIA). Eng. Geol. 244, 57e65. https://doi.org/10.1016/j.enggeo.2018.07.015. Meyer, F., 2019. Spaceborne Synthetic Aperture Radar: Principles, Data Access, and Basic Processing Techniques. https://doi.org/10.25966/EZ4F-MG98. Page, S.E., Rieley, J.O., Banks, C.J., 2011. Global and regional importance of the tropical peatland carbon pool. Global Change Biol. 17 (2), 798e818. https://doi.org/10.1111/j.1365-2486.2010.02279.x. Pawluszek-Filipiak, K., Borkowski, A., 2020. Integration of DInSAR and SBAS techniques to determine mining-related deformations using sentinel-1 data: the case study of Rydułtowy mine in Poland. Rem. Sens. 12 (2), 242. https://doi.org/10.3390/rs12020242. Pepe, A., Cal o, F., 2017. A review of interferometric synthetic aperture RADAR (InSAR) multi-track approaches for the retrieval of Earth’s surface displacements. Appl. Sci. 7 (12), 1264. https://doi.org/ 10.3390/app7121264. Rabus, B., Eineder, M., Roth, A., Bamler, R., 2003. The shuttle radar topography missionda new class of digital elevation models acquired by spaceborne radar. ISPRS J. Photogramm. Remote Sens. 57 (4), 241e262. https://doi.org/10.1016/S0924-2716(02)00124-7. Ramirez, R., Lee, S.-R., Kwon, T.-H., 2020. Long-term remote monitoring of ground deformation using sentinel-1 interferometric synthetic aperture radar (InSAR): applications and insights into geotechnical engineering practices. Appl. Sci. 10 (21), 7447. https://doi.org/10.3390/app10217447.

355

356

Radar Remote Sensing

Reeve, A.S., Glaser, P.H., Rosenberry, D.O., 2013. Seasonal changes in peatland surface elevation recorded at GPS stations in the Red Lake Peatlands, northern Minnesota, USA. J. Geophys. Res. Biogeosci. 118 (4), 1616e1626. https://doi.org/10.1002/2013JG002404. Rehman, S.A.U., Sudadi, U., Anwar, S., Sabiham, S., 2015. Land use changes and above-ground biomass estimation in peatlands of Riau and West Kalimantan, Indonesia. J. Int. Soc. Southeast Asian Agric. Sci. 21 (1), 123e136. Rosen, P.A., Gurrola, E., Sacco, G.F., Zebker, H., 2012. The InSAR Scientific Computing Environment, pp. 730e733. Schoneveld, G.C., Ekowati, D., Andrianto, A., van der Haar, S., 2019. Modeling peat- and forestland conversion by oil palm smallholders in Indonesian Borneo. Environ. Res. Lett. 14 (1), 014006. https:// doi.org/10.1088/1748-9326/aaf044. Susanti, R.D., Anjasmara, I.M., 2019. Analysing peatland subsidence in pelalawan regency, riau using dinsar method. IPTEK J. Proc. Series 0 (2), 60. https://doi.org/10.12962/j23546026.y2019i2.5308. Umarhadi, D.A., Avtar, R., Widyatmanti, W., Johnson, B.A., Yunus, A.P., Khedher, K.M., Singh, G., 2021. Use of Multi-frequency (C-band and L-band) SAR Data to Monitor Peat Subsidence Based on Time-series SBAS InSAR Technique. Land Degradation & Development. https://doi.org/ 10.1002/ldr.4061 ldr.4061. van den Akker, J.J.H., Kuikman, P.J., de Vries, F., Hoving, I.E., Pleijter, M., Hendriks, R.F.A., Wolleswinkel, R.J., Sim~ oes, R.T.L., Kwakernaak, C., 2008. Emission of CO2 from agricultural peat soils in The Netherlands and ways to limit this emission. In: Proceedings of the 13th International Peat Congress after Wise Use e the Future of Peatlands, vol. 1, pp. 645e648. Oral Presentations. https://edepot.wur.nl/159747. Vetrita, Y., Cochrane, M.A., 2019. Fire frequency and related land-use and land-cover changes in Indonesia’s peatlands. Rem. Sens. 12 (1), 5. https://doi.org/10.3390/rs12010005. Villarroel, C., Tamburini Beliveau, G., Forte, A., Monserrat, O., Morvillo, M., 2018. DInSAR for a regional inventory of active rock glaciers in the dry andes mountains of Argentina and Chile with sentinel-1 data. Rem. Sens. 10 (10), 1588. https://doi.org/10.3390/rs10101588. Webb, E.L., Friess, D.A., Krauss, K.W., Cahoon, D.R., Guntenspergen, G.R., Phelps, J., 2013. A global standard for monitoring coastal wetland vulnerability to accelerated sea-level rise. Nat. Clim. Change 3 (5), 458e465. https://doi.org/10.1038/nclimate1756. Wei, M., Sandwell, D.T., 2010. Decorrelation of L-band and C-band interferometry over vegetated areas in California. IEEE Trans. Geosci. Rem. Sens. 48 (7), 2942e2952. https://doi.org/10.1109/ TGRS.2010.2043442. Wösten, J.H.M., Ismail, A.B., van Wijk, A.L.M., 1997. Peat subsidence and its practical implications: a case study in Malaysia. Geoderma 78 (1e2), 25e36. https://doi.org/10.1016/S0016-7061(97)00013-X. Xu, J., Morris, P.J., Liu, J., Holden, J., 2018. PEATMAP: refining estimates of global peatland distribution based on a meta-analysis. Catena 160, 134e140. https://doi.org/10.1016/j.catena.2017.09.010. Yanuarsyah, I., Suwarno, Y., Hudjimartsu, S., 2016. The use of hotspot spatial clustering and multitemporal satellite imagery to facilitate peat land degradation in West kalimantan, Indonesia (case study in mensiku miniwatershed of Kapuas river). IOP Conf. Ser. Earth Environ. Sci. 47, 012046. https://doi.org/ 10.1088/1755-1315/47/1/012046. Yu, Z., Loisel, J., Brosseau, D.P., Beilman, D.W., Hunt, S.J., 2010. Global peatland dynamics since the last glacial maximum. Geophys. Res. Lett. 37 (13). https://doi.org/10.1029/2010GL043584. Yunjun, Z., Fattahi, H., Amelung, F., 2019. Small baseline InSAR time series analysis: unwrapping error correction and noise reduction. Comput. Geosci. 133, 104331. https://doi.org/10.1016/ j.cageo.2019.104331. Zebker, H.A., Villasenor, J., 1992. Decorrelation in interferometric radar echoes. IEEE Trans. Geosci. Rem. Sens. 30 (5), 950e959. https://doi.org/10.1109/36.175330. Zhou, Z., Li, Z., Waldron, S., Tanaka, A., 2019. InSAR time series analysis of L-band data for understanding tropical peatland degradation and restoration. Rem. Sens. 11 (21), 2592. https://doi.org/10.3390/ rs11212592.

CHAPTER 19

Toward a North American continental wetland map from space: wetland classification using satellite imagery and machine learning algorithms on Google Earth Engine Masoud Mahdianpari1, 2, Brian Brisco3, Bahram Salehi4, Jean Granger1, Fariba Mohammadimanesh1, Megan Lang5 and Souleymane Toure4 1

C-CORE, St. John’s, NL, Canada; 2Department of Electrical and Computer Engineering, Memorial University of Newfoundland, St. John’s, NL, Canada; 3The Canada Centre for Mapping and Earth Observation, Ottawa, ON, Canada; 4 Environmental Resources Engineering, College of Environmental Science and Forestry, State University of New York, New York, NY, United States; 5U.S. Fish and Wildlife Service, National Wetlands Inventory, Falls Church, VA, United States

1. Introduction With the realities of increasing globalization, population growth, economic shifts, and climate change, the ability to capture large-scale earth-surface information is more important than ever. This is particularly true where natural ecosystems such as wetlands are under a growing threat owing, in part, to the constantly shifting use of the anthropogenic landscape. A thorough understanding of the causes and effects of ecosystem modification and loss requires information about not only the direct and local drivers of change, but also global megatrends (Gardner and Finlayson, 2018) that have the potential to influence ecosystems all over the earth in a myriad of ways. Of these threatened ecosystems, wetlands are among the most vulnerable. Although wetlands can be defined in a number of specific ways depending on the context, the term generally refers to lands that have soils inundated or saturated by water long enough to allow for the establishment of hydrophytic vegetation (Chen et al., 2010). Wetlands take on a variety of forms, dominated by different assemblages of vegetation depending on the location as well as the landscape and sources of water and nutrients (Cowardin et al., 1979; National Wetlands Working Group, 1997). Wetlands provide valuable ecosystems that are services for humanity, including, but not limited to, water filtration, flood prevention, recreational and aesthetic space, and carbon storage (Clarkson et al., 2014; Hanson, 2008). These services are made possible via the natural functions of wetland habitats, which can easily be disturbed and degraded through the

Radar Remote Sensing ISBN 978-0-12-823457-0, https://doi.org/10.1016/B978-0-12-823457-0.00021-5

© 2022 Elsevier Inc. All rights reserved.

357

358

Radar Remote Sensing

influence of direct or indirect anthropogenic activity (Breeuwer et al., 2009; Edvardsson et al., 2015; Pasquet et al., 2015; von Sengbusch, 2015). It is largely the loss of these services that has triggered a desire to protect these habitats in North America. Historically ignored and sometimes reviled (Hook, 1993), efforts to protect and manage wetlands better were not meaningfully manifested in North America until the 1970s. Already, millions of acres of wetland had been lost in Canada and the United States since the time of European settlement (Byun et al., 2018; Davidson, 2014). Although not widely understood at the time, the destruction of these wetlands resulted in the associated loss of wetland services, the impacts of which were reported as early as the 1700s, when there was a noted decrease in ducks and geese available for hunting and food (McAlpine and Smith, 2010; Melinchuk, 1995). It was difficult to quantify, but the loss of these wetlands and the services they provide likely represent a loss of millions of dollars as well as a loss of quality of life for countless humans and nonhuman animals alike (Caramel, 2008; Li et al., 2018). Since then, a multitude of local efforts to protect wetlands have been implemented. These efforts have slowed wetland loss in more recent times (Davidson, 2014), but the impacts of global megatrends, such as globalization and climate change, have made protecting wetlands at the local scale more difficult to implement meaningfully. It requires cross-country initiatives and big data tools to help protect these important habitats. Remote sensing tools and data have long been employed to collect information on wetlands for use in policy and conservation efforts (Cameron, 1950; Lukens, 1968; Shaw and Fredine, 1956). In fact, much of what we know about wetlands (how many wetlands there are and what the major drivers are of changes to wetlands) have been derived from implementations of remote sensing. Unfortunately, data on continental to global scale trends have often been pieced together via a combination of numerous local and regional wetland assessments and maps, the quality of which is often not well-documented (Davidson, 2014; Hu et al., 2017). Until recently, this had been one of the only feasible ways to describe continental to global wetlands spatial information. However, current developments in big data, computational power, data availability, and cross-country collaboration have created a scenario not previously feasible: that is, the ability to capture more consistent categorically detailed (i.e., more than just a couple of classes) wetland information at the scale of a continent or globe. Continental-scale remotely sensed wetland data, in the form of inventories, maps, and databases, have great potential to elucidate large-scale changes to wetland habitats owing to climate change, contribute to the development of more accurate climate change models, and facilitate data exchange across national borders. In addition, such information may help contribute to the creation of cross-country initiatives to protect wetland habitats. This is particularly important because the provision of particular wetland services is relevant at scales far beyond the specific geographic location of the wetland. For example, many species of waterfowl, endangered and otherwise, take refuge, breed,

Toward a North American continental wetland map

feed, and raise young birds in wetlands across national- or continental-scale migration routes. Moreover, large extents of wetland complexes cross these borders, such as those in the Prairie Pothole region and the Great Lakes (Bourgeau-Chavez et al., 2015; van der Kamp et al., 2016). Finally, the destruction of carbon-storing wetlands may influence not only the wetlands geographic area of interest (Bates et al., 2008; Erwin, 2009) but the global climate as a whole. Inventories at such a scale are greatly needed, but many technical and methodological barriers remain that impede their immediate development and widespread accessibility and will likely require much research and collaboration across many political and disciplinary boundaries. Some of this work was performed by Mahdianpari et al. (2020a,b), who produced the first and second Canada-wide wetland inventory using supervised classification of Sentinel-1 and Sentinel-2 imagery at 10-m spatial resolution in the Google Earth Engine (GEE) cloud computing platform. This research has laid the groundwork for a future large-scale wetland map of Canada and the United States, representing much of North America. Major barriers that must now be addressed include the collection of US wetland field data and the merging of each country’s classification systems. This and more details will be outlined in the following chapter.

2. Wetland classification systems Terms and definitions for wetland habitats vary widely across countries. Similarly, they vary across disciplines, dialects, and cultures. As such, a name for a single wetland may include mire, peatland, bog, fen, moor, or muskeg. To address this problem, in addition to communicating the biological and ecological diversities of wetlands, a number of wetland classification systems exist (Brinson, 1993; Cowardin et al., 1979; Ducks Unlimited Canada, 2014; National Wetlands Working Group, 1997). These systems simplify, describe, and classify wetland habitat variability based on a range of shared characteristics, including vegetation patterns, hydrology, the position in the landscape, and the different purposes. Although useful within specific contexts, the large number of classifications contribute to difficulty associated with accurately assessing the large-scale wetland extent via the compilation of numerous regionalized datasets, which are likely to use a range of naming and classification conventions (Davidson, 2014; Hu et al., 2017). Davidson (2014) described the difficulty in extracting large-scale information relating to wetland change from varying quality data collected using a diversity of conventions during different periods of time. Wetland classification systems in Canada and the United States alone are numerous. The most commonly applied are the Cowardin Classification System in the United States and the Canadian Wetland Classification System in Canada (Cowardin et al., 1979; National Wetlands Working Group, 1997). Fig. 19.1 shows examples of different types of wetlands based on the Canadian classification system. These systems or derivations

359

360

Radar Remote Sensing

Figure 19.1 Demonstration of the heterogeneity of wetlands. Bog (top left), fen (top right), swamp (midleft), marsh (midright), and shallow water (bottom) are examples of the five classes defined by the Canadian Wetland Classification System.

have commonly been used in wetland remote sensing research and have been particularly useful owing to their use of vegetation characteristics to define classes, a wetland characteristic most easily captured using fine to medium resolution optical remote sensing data. However, issues arise in applying these systems at a continental scale, because there are terminology- and nomenclature-related issues among these systems. For example, peatland wetlands (wetlands dominated by the buildup of peat), are far more extensive in

Toward a North American continental wetland map

Canada compared with the United States. As such, Canadian classification systems tend to have more extensive classification definitions for peatland wetlands. A peatland that was classified as bog, fen, or swamp (all peatlands according to the Canadian Wetland Classification System) in Canada may simply be referred to as a palustrine wetland with an associated vegetation type (emergent, scrub-shrub, or forest) based on the Cowardin standard. In addition, the Cowardin system puts a greater emphasis on relationships with deepwater habitats (lakes, rivers, and oceans) and substrate than does the Canadian system, in part because the Cowardin system classifies not only wetlands but also hydrologic systems. The selection of a method for defining land cover is an integral step to remote sensing-based land cover classification. This choice will ultimately define and affect the quality of both inputs and outputs from a remote sensing classification methodology and how the product will be used in future applications. Among the biggest challenges to producing a North American continental-scale map of wetlands is creating a standardized and meaningful classification system that will capture extensive wetland heterogeneity across both the United States and North America using remote sensing techniques. A continent-wide classification system must (1) capture a broad range of wetland expression, (2) define wetlands in a way that is meaningful across multiple countries, and (3) define wetlands in a way that is effectively captured using available remote sensing data. A small number of broadscale wetland classification systems exist, including the Ramsar Classification System and the Hydro geomorphic Classification System (Brinson, 1993; Ramsar Convention Secretariat, 2010). The former was designed to identify and classify protected Ramsar sites, and the latter was designed to classify wetlands based on wetland functionality. Some remote sensing-based wetlands research has applied these systems, although much less commonly than the Canadian Wetland Classification and the Cowardin Systems. Global systems have the potential for use in a continent-wide classification system, but the option also exists to create a novel system derived from more commonly used national and regional Canadian and US classification systems. This option may allow for the creation of a classification product that is more practical and meaningful to Canadian- and US-based users and allow for enhanced applicability of the classification system to remote sensing methodology. This work would require the synergistic input of biologists, ecologists, remote sensing scientists, and natural resource managers from both Canada and the United States, and it may not always be a simple task. However, the resulting classification system would perhaps contribute to the production of a more accurate and meaningful remote sensing product, and greatly support the exchange of wetland-related information and collaboration across the CanadianeUS border.

361

362

Radar Remote Sensing

3. Wetland field data The production of a continental-scale wetland map will require the collaboration of a number of individuals across many disciplines, with a variety of goals for the application of the final product. These include various remote sensing scientists and resource managers across Canada and the United States, and biologists, ecologists, and conservationists. These collaborators could contribute to knowledge, advice, field data, satellite data, and other support to help create a continental wetland classification map. Field data provided by these individuals will be particularly important because reference (ground-truth) data are a necessary but often difficult aspect of data compilation for remote sensing, particularly at such a large scale. Field data in general are known to be a bottleneck in many remote sensing efforts. Sourcing field data representing the heterogeneity of wetlands across the entire North American continent will require concessions in terms of data standardization, spatial coverage, quality, and the time period. Most wetland ground-truth data collection will likely have been carried out for smaller, noneremote sensing projects that used a number of different methods and classification terminologies. However, this issue of ground-truth data collection is highly common in remote sensing studies. Thus, extensive time and effort will need to be dedicated to preparing a cohesive and standardized final dataset for use in training and validating remote sensing algorithms. Such data preparation will include checking the accuracy of data, standardizing naming conventions, altering wetland boundaries, and filtering by size. For example, an area where wetlands were previously ground-truthed has been converted to an urban or agricultural landscape in the last year. Perhaps a field program required only general wetland boundary delineation, whereas another study enforced more strict requirements. Similarly, perhaps a wetlands field team was required to classify wetlands only at a broad level (treed wetland vs open wetland), whereas another team classified wetlands at the level of vegetation species composition. All of these issues and more must be addressed before inputting these data into a remote sensing methodology if the product is to be of high quality. Another consideration is the need for nonwetland data, because this will ensure a higher-quality final product. This type of data may largely be extracted from previous land cover classification work and will be made of common nonwetland land cover, including forests, impervious surfaces, and agriculture, among many others. Further consideration must be given to distributing data across the study area and how to split data into testing and training groups. These steps are necessary for implementing a machine learning remote sensing methodology. Millard and Richardson (2015) provide a thorough overview of the considerations that must be given when preparing this type of data, including issues of spatial autocorrelation, the size of datasets, and the proportionality of classes within the dataset. Although these rules may be strictly followed in an optimal situation, this is often not the case given the availability of wetland datasets,

Toward a North American continental wetland map

particularly at a continental scale. For example, there will likely be large swaths of land where no ground-truth data are available. Similarly, most wetland ground-truth data are not collected using a random or semirandom sampling methodology because of budgetary and time constraints. As such, spatial autocorrelation can be an issue. However, efforts can be made to meet these requirements even given these restrictions. Prior work by Mahdianpari et al. (2020a,b) collected and standardized a large wetland field dataset across Canada, establishing a methodology that could be extended across North America. These data were collected across Canada and are located in multiple provinces and ecozones. These data were sourced from a large number of previous wetland-related projects. Nevertheless, there are some gaps in spatial coverage in the dataset. For example, large swaths of northern Canada lack wetland data. This research also collected a large amount of nonwetland land cover data using previous provincial or local land cover datasets, such as the National Land Cover Classification from Natural Resources Canada and a crop inventory map of Canada (Agriculture and Agri-food Canada, 2018). The Canadian Wildlife Service (CWS) at Environment and Climate Change Canada (ECCC) started a working group to continue to develop a national database of wetland ground-truth data to aid in these Canada-wide classifications. Because of this, more effort can be dedicated to collecting data from sources in the United States and ultimately combining cross-continental information into a cross-continental cohesive dataset.

4. Remote sensing data Remote sensing data come in a variety of spatial, spectral, and temporal resolutions from a diverse satellites, sensors, and data providers in a range of prices, capturing various parts of the globe or its entirety. However, only a narrow portion of these data will be useful for mapping wetlands at a continental scale. These data should be free of cost, provide coverage of the entire continent on a regular or semiregular basis, have at least a medium spatial resolution, and be able to capture land cover information useful for differentiating and classifying wetlands. Only a small number of data sources meet these standards, including the American Landsat and European Sentinel missions, which together provide global optical and synthetic aperture radar (SAR) data. These datasets have been used effectively, alone and in combination, to map wetlands across Canada and the United States (Amani et al., 2019; Mahdianpari et al., 2020a,d). Sentinel-1 and Sentinel-2 provide 12- and 10-day repeat coverage of the globe, respectively, and are freely available via several data portal sites. Sentinel-1 collects dual (HH/HV) and single (HH) polarized C-band SAR data over polar regions and dual (VV/VH) or single (VV) polarized data elsewhere. These data must be radiometrically calibrated and terrain corrected before use. After preprocessing, a number of data-containing features can be extracted (e.g., type of backscatter, scattering power).

363

364

Radar Remote Sensing

SAR data are great beneficial for wetland mapping because they enable the collection of structural information that may help differentiate among the composition, height, and amount of vegetation and detect soil saturation and inundation. Many studies demonstrated the usefulness of Sentinel-1 for wetland mapping. Millard et al. (2020), for example, found that Sentinel-1 classified wetlands with greater accuracy than did RADARSAT-2 because of its fine temporal resolution and the greater contribution of Sentinel-1 coherence to overall accuracy. Similarly, Li et al. (2020) explored the potential of Sentinel-1 data to discriminate treed and nontreed wetlands, with high accuracy results depending on the selection of different imagery features. Another benefit of using SAR data is the ability to collect data at any time of day, such as night, and under most weather conditions, including cloud coverage. Cloud cover is a frequent challenge when applying optical imagery, and it is often an issue over large parts of the United States and Canada. Sentinel-2 captures information about the earth’s surface across the visible and infrared portion of the electromagnetic spectrum, specifically collecting data across 12 bands. Of those bands, the use of blue, green, red-edge, near-infrared, and shortwave infrared bands is common in wetland remote sensing. In addition, these bands are available at 10- to 20-m spatial resolutions, which is relatively high for freely available remote sensing data. Preprocessing of this type of data can be difficult because of the impact of cloud cover on Sentinel-2 coverage. Thus, different-date mosaicking techniques and analysis of weather data are sometimes necessary to achieve continuous coverage. Sentinel-2 data have proven to be useful in several wetland remote sensing studies, many of which extracted multiple vegetation indices from the data. For example, Bhatnagar et al. (2020) successfully mapped peatland vegetation using Sentinel-2 data and extracted normalized difference vegetation index, enhanced vegetation index, and normalized difference water index features. Similarly, Kaplan and Avdan (2017) demonstrated the effective application of Sentinel-2 for mapping wetlands in Turkey using raw bands and extracted indices. Moreover, Sentinel-2 generally produced higher accuracy classification products compared with Landsat, another source of free optical satellite imagery, although the file size and processing time are greater (S
anchez-Espinosa and Schröder, 2019). Although both Sentinel-1 and -2 are effective for wetland classification, the synergistic use of both resulted in generally higher-quality remote sensing products (Mahdianpari et al., 2020c). This is because each data type can capture features of wetlands that the other cannot. That is, Sentinel-1 can extract structural and saturation information whereas Sentinel-2 captures optical information. Studies using a synergistic combination of these include work by Kaplan and Avdan (2019), who found that different combinations of Sentinel-1 and Sentinel-2 data are best for classifying different wetlands, and by Slagter et al. (2020), who showed that combined Sentinel-1 and Sentinel-2 data classified wetlands with a greater accuracy than either alone. Fig. 19.2 gives a general outline of classification using Sentinel-1 and Sentinel-2 data fusion.

Toward a North American continental wetland map

Figure 19.2 General workflow of a Sentinel-1 (radar) and Sentinel-2 (optical) data fusion classification. Each data type contributes differently to classification accuracy. NDVI, normalized difference vegetation index.

5. Cloud computing platforms and machine learning algorithms The ability to apply multitemporal, remotely sensed data fusion techniques to classify land cover at a moderate resolution for a continent had been almost impossible given the computational power required, the need to access a large amount of data in the form of satellite imagery, and so on, and software limitations. This was referred to as the geo big data problem (Tamiminia et al., 2020). However, advances in technology allowed for substantial efforts in developing remote-sensing products at scales not possible before. These advances include the increasing availability of open-access earth observation data (from sources such as NASA, the US Geological Survey, the National Oceanic and Atmospheric Administration, and the European Space Agency), increases in computational power, the development of new cloud-based computational framework such as GEE, Amazon’s Web Services, and Microsoft Azure, and improvements in machine learning algorithms and tools (Hird et al., 2017; Liu, 2015). Of the several cloud-computing platforms available, GEE has become a wellestablished, free-to-use, cloud-based geospatial platform that facilitates the large-scale analysis of earth observation data (Gorelick et al., 2017). Within the GEE framework, large amounts of satellite data can be processed using various spatial tools and machine learning algorithms not possible using standard spatial analysis software (Tamiminia et al., 2020). Several studies demonstrated the ability of GEE to produce remotesensing classification products at very large scales (Kumar and Mutanga, 2019; Midekisa et al., 2017; Xiong et al., 2017a,b), including wetland products. For example, Hird et al. (2017) implemented GEE to predict the probability of wetland occurrences in the province of Alberta, Canada, using a boosted regression tree machine learning framework with an overall accuracy of 85%.

365

366

Radar Remote Sensing

Mahdianpari et al. (2018a) also used GEE and the integration of Sentinel -1 and Sentinel-2 to produce the first provincial wetland map of Newfoundland, Canada with an overall accuracy of 88%. In addition, Mahdianpari et al. (2020a,d) produced the first and second generations of a Canadian wetland inventory map using an object-based random forest (RF) algorithm in GEE. DeLancey et al. (2019) used GEE to map peatlands in Alberta at an overall accuracy of 87% by integrating radar, optical, and elevation data. Also, Wang et al. (2020) mapped changes to coastal tidal flats across China’s coastal zones using 44,258 Landsat images. The availability of these cloud-computing platforms is not the only advancement that has allowed for improvements in large-scale remote sensing implementation. A shift toward implementing machine-learning algorithms over more traditional methods (Li et al., 2014) also enabled the cohesive use of large amounts of different types of data. Older methods were limited by an inability to deal with data with many predictor variables, or high dimensionality, which is effectively dealt with by machine learning (Maxwell et al., 2018). Machine learning algorithms are freely available for use in GEE and elsewhere. Some machine learning algorithms include support vector machines, single and boosted decision trees, artificial neural networks, and RF (Maxwell et al., 2018). RF in particular has been shown to be robust for wetland classification (Banks et al., 2019; Mahdianpari et al., 2017; Maxwell et al., 2016; Tian et al., 2016). RF is particularly beneficial to wetland classification studies because the algorithm does not assume the normality of training data. This is important because typical wetland training datasets have nonnormal distributions as a result of the nonrandom methods often used to collect field data because of restricted budgets, time, and accessibility. Another benefit of RF is the ability to estimate the importance of a large number of predictor variables (highdimensionality data). This enables the identification of a large number of variables to contribute the greatest to classification results, allowing for the optimization of feature space, and therefore the reduction of the required computation power (Corcoran et al., 2013; Rodriguez-Galiano et al., 2012). A drawback implementing RF is the large number of tuning parameters available, which may act as a roadblock when used by nonexperts. In addition, many of these parameters require trial-and-error implementation to establish optimal values; as a result, RF may take some time to perfect.

6. Wetland classification results for Canada In a culmination of the geoebig data techniques and data previously discussed, Mahdianpari et al. (2020a) first produced a 10-m resolution wetland map of the entirety of Canada, a total of 9,984,670 km2 in size. This map specifically classified wetlands at the level of class (bog, fen, swamp, and marsh) described by the Canadian Wetland Classification System (National Wetlands Working Group, 1997). Using field-collected

Toward a North American continental wetland map

data from a number of sources, 13,519 Sentinel-1 SAR images, 211,926 Sentinel-2, and RF classification, all within the GEE framework, wetlands across Canada were classified on province-by-province with overall accuracies ranging from 74% to 84%. A major roadblock encountered by Mahdianpari et al. (2020a) was the limited availability of field data at the time, because a number of provinces had no available field data. Moreover, a suggestion was made that implementing classification at the ecozone level may be more meaningful than doing so at the provincial level. In Canada, ecozone boundaries are determined based on similar biological, ecological, and geomorphologic factors such as land cover, human activity, climate, wildlife, soil, and vegetation (Statistics Canada, 2018) (Fig. 19.3). As such, classification within ecozone boundaries may contribute to the production of more ecologically meaningful and accurate results. Thus, after the production of the first Canada remote sensing-based wetland map at 10-m resolution, Mahdianpari et al. (2020b) produced the second generation of Canada’s wetland map (Fig. 19.4), but with the addition of a large amount of new field data and processing at the level of the ecozone rather than at the province. Fig. 19.3 shows the results of this work. Although a large amount of new training data contributed to the research, a small number of ecozones lacked training data, particularly in the north. As such, Mahdianpari et al. (2020b) used high-resolution WorldView-2 and Pleiades imagery to classify wetlands visually and added this information to the available training data. Accuracy results ranged from 76% to 91% across ecozones, an increase over the firstgeneration product. These accuracies may be better than actually realized because they

Figure 19.3 Fifteen ecozones of Canada.

367

368

Radar Remote Sensing

Figure 19.4 Second generation of Canada’s wetland inventory map.

are based on available testing samples. Accordingly, the lack of samples or availability of only a few in some ecozones created a degree of uncertainty in the reported accuracies. It is expected that incorporating well-distributed training samples in the future will decrease the degree of uncertainty in the results. Ecozones with the lowest overall accuracies tended to have the smallest amount of field data available for training, demonstrating the bottleneck caused by training data availability. Another issue involved discriminating the swamp class. Many ecozones had an overclassification of swamps, likely owing to the difficulty associated with separating swamps from upland forests using optical and C-band SAR imagery, a common problem (Jahncke et al., 2018). This highlights the need for an improved national database, which the CWS is addressing. This database includes the best available wetlands datasets across Canada, which have been collected from various sources, including federal, provinces, territories, municipalities, nongovernmental organizations, academia, and the private sector. The third generation of this work is being carried out integrating the L-band of Advanced Land Observing Satellite (ALOS) data. The inclusion of structural data captured by ALOS will improve wetland class discrimination across Canada, particularly in the case of swamp wetlands. As shown in the literature, wetland classification accuracy tends to increase when both optical and radar data are applied (Mahdianparai et al.,

Toward a North American continental wetland map

2020c). Therefore, it is anticipated that the third-generation Canada wetland map will have increased accuracy compared with that of the first and second versions. The purpose of this work is to contribute to an progressing understanding of Canadian wetlands while improving management of this expansive natural resource. In addition, this work demonstrates the possibility of expanding this methodology to other countries and entire continents.

7. Conclusion1 The ability to apply big data remote sensing methods at scales not previously possible has resulted in the production of classification products for entire provinces, states, countries, and even continents. This is a result of the increasing availability of cloud computing, large amounts of open source data, and advances in machine learning. However, despite such potential, there has been no effort dedicated to the production of a North Americanescale wetland map. This type of work would be particularly useful given the shared borders of the United States and Canada, which would enable improvements in cross-country policies and initiatives and contribute to better modeling of climate and landscape changes in megatrends including globalization, population growth, and climate change. The groundwork for this effort was laid by Mahdianpari et al. (2020a,b), who mapped wetlands across Canada at high accuracy. However, much work must be completed before such a project can be implemented, including the development of a cohesive and meaningful wetland classification system that effectively captures heterogeneous wetlands across both Canada and the United States, and the collection and preparation of wetland field data from across states as well as provinces and territories of both countries. This work would require the collaboration of biologists, ecologists, remote sensing scientists, and natural resource managers across the continent. The results of this effort would be the first wetland map of the entirety of North America (Canada and the United States) and a step toward better managing and understanding wetlands in an ever-changing world.

References Agriculture and Agri-food Canada, 2018. ISO 19131 Annual Crop Inventory e Data Product Specifications. Agriculture and Agri-food Canada, p. 27. Amani, M., Mahdavi, S., Afshar, M., Brisco, B., Huang, W., Mohammad Javad Mirzadeh, S., White, L., Banks, S., Montgomery, J., Hopkinson, C., 2019. Canadian wetland inventory using Google Earth engine: the first map and preliminary results. Rem. Sens. 11, 842.

1

The findings and conclusions in this article are those of the author(s) and do not necessarily represent the views of the US Fish and Wildlife Service.

369

370

Radar Remote Sensing

Banks, S., White, L., Behnamian, A., Chen, Z., Montpetit, B., Brisco, B., Pasher, J., Duffe, J., 2019. Wetland classification with multi-angle/temporal SAR using random forests. Rem. Sens. 11, 670. https:// doi.org/10.3390/rs11060670. Bates, B., Kundzewicz, Z.W., IPCC (Eds.), 2008. Climate Change and Water. IPCC Technical Paper, p. 6. Bhatnagar, S., Gill, L., Regan, S., Naughton, O., Johnston, P., Waldren, S., Ghosh, B., 2020. Mapping vegetation communities inside wetlands using SENTINEL-2 imagery IN Ireland. Int. J. Appl. Earth Obs. Geoinf. 88, 102083. https://doi.org/10.1016/j.jag.2020.102083. Bourgeau-Chavez, L., Endres, S., Battaglia, M., Miller, M., Banda, E., Laubach, Z., Higman, P., ChowFraser, P., Marcaccio, J., 2015. Development of a Bi-national great lakes coastal wetland and land use map using three-season PALSAR and landsat imagery. Rem. Sens. 7, 8655e8682. https://doi.org/ 10.3390/rs70708655. Breeuwer, A., Robroek, B.J.M., Limpens, J., Heijmans, M.M.P.D., Schouten, M.G.C., Berendse, F., 2009. Decreased summer water table depth affects peatland vegetation. Basic Appl. Ecol. 10, 330e339. https://doi.org/10.1016/j.baae.2008.05.005. Brinson, M.M., 1993. A Hydrogeomorphic Classification for Wetlands. East Carolina Univ Greenville NC. Byun, E., Finkelstein, S.A., Cowling, S.A., Badiou, P., 2018. Potential carbon loss associated with postsettlement wetland conversion in southern Ontario, Canada. Carbon Bal. Manag. 13, 6. https:// doi.org/10.1186/s13021-018-0094-4. Cameron, H.L., 1950. The use of aerial photography in seaweed surveys. Photogramm. Eng. 9. Caramel, E., 2008. The economic cost of wetland destruction. In: 2008 Conference (52nd), February 5-8, 2008, Canberra, Australia 5985, vol. 19. Australian Agricultural and Resource Economics Society. Chen, H., Lu, X., Wang, G., 2010. Wetland definitions: creation, evolution and application. Wetland Sci. 8, 299e304. Clarkson, B.R., Ausseil, A.-G.E., Gerbeaux, P., 2014. Wetland Ecosystem Services. Ecosystem Services in New Zealand e Conditions and Trends, vol. 12. Corcoran, J., Knight, J., Gallant, A., 2013. Influence of multi-source and multi-temporal remotely sensed and ancillary data on the accuracy of random forest classification of wetlands in Northern Minnesota. Rem. Sens. 5, 3212e3238. https://doi.org/10.3390/rs5073212. Cowardin, L.M., Carter, V., Golet, F.C., Laroe, E.T., 1979. Classification of wetlands and deepwater habitats of the United States. In: Lehr, J.H., Keeley, J. (Eds.), Water Encyclopedia. John Wiley & Sons, Inc., Hoboken, NJ, USA, p. sw2162. https://doi.org/10.1002/047147844X.sw2162. Davidson, N.C., 2014. How much wetland has the world lost? Long-term and recent trends in global wetland area. Mar. Freshw. Res. 65, 934. https://doi.org/10.1071/MF14173. DeLancey, E.R., Kariyeva, J., Bried, J.T., Hird, J.N., 2019. Large-scale probabilistic identification of boreal peatlands using Google Earth Engine, open-access satellite data, and machine learning. PLoS One 14, e0218165. https://doi.org/10.1371/journal.pone.0218165. Ducks Unlimited Canada, 2014. Boreal Wetland Classes in the Boreal Plains Ecozone of Canada: Field Guide. Ducks Unlimited Canada, Stonewall, Man. Edvardsson, J., Simanauskien_e, R., Taminskas, J., Bauzien_e, I., Stoffel, M., 2015. Increased tree establishment in Lithuanian peat bogs d insights from field and remotely sensed approaches. Sci. Total Environ. 505, 113e120. https://doi.org/10.1016/j.scitotenv.2014.09.078. Erwin, K.L., 2009. Wetlands and global climate change: the role of wetland restoration in a changing world. Wetl. Ecol. Manag. 17, 71e84. https://doi.org/10.1007/s11273-008-9119-1. Gardner, R., Finlayson, M., 2018. Global Wetland Outlook: State of the World’s Wetlands and Their Services to People, Ramsar Convention on Wetlands. Ramsar Convention Secretariat, Gland, Switzerland. Gorelick, N., Hancher, M., Dixon, M., Ilyushchenko, S., Thau, D., Moore, R., 2017. Google earth engine: planetary-scale geospatial analysis for everyone. Rem. Sens. Environ. 202, 18e27. https://doi.org/ 10.1016/j.rse.2017.06.031. Hanson, A.R., 2008. Wetland Ecological Functions Assessment: An Overview of Approaches. Canadian Wildlife Service, Ottawa. Hird, J., DeLancey, E., McDermid, G., Kariyeva, J., 2017. Google earth engine, open-access satellite data, and machine learning in support of large-area probabilistic wetland mapping. Rem. Sens. 9, 1315. https://doi.org/10.3390/rs9121315.

Toward a North American continental wetland map

Hook, D.D., 1993. Wetlands: history, current status, and future. Environ. Toxicol. Chem. 12, 2157e2166. https://doi.org/10.1002/etc.5620121202. Hu, S., Niu, Z., Chen, Y., 2017. Global wetland datasets: a review. Wetlands 37, 807e817. https://doi.org/ 10.1007/s13157-017-0927-z. Jahncke, R., Leblon, B., Bush, P., LaRocque, A., 2018. Mapping wetlands in Nova Scotia with multi-beam RADARSAT-2 Polarimetric SAR, optical satellite imagery, and Lidar data. Int. J. Appl. Earth Obs. Geoinf. 68, 139e156. https://doi.org/10.1016/j.jag.2018.01.012. Kaplan, G., Avdan, U., 2019. Evaluating the utilization of the red edge and radar bands from sentinel sensors for wetland classification. Catena 178, 109e119. https://doi.org/10.1016/j.catena.2019.03.011. Kaplan, G., Avdan, U., 2017. Mapping and monitoring wetlands using SENTINEL-2 satellite imagery. ISPRS Ann. Photogramm. Rem. Sens. Spatial Inf. Sci. IV-4/W4, 271e277. https://doi.org/ 10.5194/isprs-annals-IV-4-W4-271-2017. Kumar, L., Mutanga, O., 2019. Google Earth Engine Applications. MDPI - Multidisciplinary Digital Publishing Institute. Li, M., Zang, S., Zhang, B., Li, S., Wu, C., 2014. A review of remote sensing image classification techniques: the role of spatio-contextual information. Eur. J. Rem. Sens. 47, 389e411. https://doi.org/10.5721/ EuJRS20144723. Li, X., Bellerby, R., Craft, C., Widney, S.E., 2018. Coastal wetland loss, consequences, and challenges for restoration. Anthropocene Coasts 1e15. https://doi.org/10.1139/anc-2017-0001. Li, Z., Chen, H., White, J.C., Wulder, M.A., Hermosilla, T., 2020. Discriminating treed and non-treed wetlands in boreal ecosystems using time series Sentinel-1 data. Int. J. Appl. Earth Obs. Geoinf. 85, 102007. https://doi.org/10.1016/j.jag.2019.102007. Liu, P., 2015. A survey of remote-sensing big data. Front. Environ. Sci. 3. https://doi.org/10.3389/ fenvs.2015.00045. Lukens, J.E., 1968. Color aerial photography for aquatic vegetation surveys. Rem. Sens. Environ. 5, 441e446. Mahdianpari, M., Brisco, B., Granger, J.E., Mohammadimanesh, F., Salehi, B., Banks, S., Homayouni, S., Bourgeau-Chavez, L., Weng, Q., 2020a. The second generation Canadian wetland inventory map at 10 meters resolution using Google earth engine. Can. J. Rem. Sens. 1e16. https://doi.org/10.1080/ 07038992.2020.1802584. Mahdianpari, M., Brisco, B., Granger, J.E., Mohammadimanesh, F., Salehi, B., Banks, S., Homayouni, S., Bourgeau-Chavez, L., Weng, Q., 2020b. The second generation Canadian wetland inventory map at 10 meters resolution using Google earth engine. Can. J. Rem. Sens. 46, 360e375. https://doi.org/ 10.1080/07038992.2020.1802584. Mahdianpari, M., Granger, J.E., Mohammadimanesh, F., Salehi, B., Brisco, B., Homayouni, S., Gill, E., Huberty, B., Lang, M., 2020c. Meta-analysis of wetland classification using remote sensing: a systematic review of a 40-year trend in North America. Rem. Sens. 12, 1882. https://doi.org/10.3390/ rs12111882. Mahdianpari, M., Salehi, B., Mohammadimanesh, F., Brisco, B., Homayouni, S., Gill, E., DeLancey, E.R., Bourgeau-Chavez, L., 2020d. Big data for a big country: the first generation of Canadian wetland inventory map at a spatial resolution of 10-m using sentinel-1 and sentinel-2 data on the Google earth engine cloud computing platform. Can. J. Rem. Sens. 46, 15e33. https://doi.org/10.1080/ 07038992.2019.1711366. Mahdianpari, M., Salehi, B., Mohammadimanesh, F., Homayouni, S., Gill, E., 2018a. The first wetland inventory map of Newfoundland at a spatial resolution of 10 m using sentinel-1 and sentinel-2 data on the Google earth engine cloud computing platform. Rem. Sens. 11, 43. https://doi.org/10.3390/ rs11010043. Mahdianpari, M., Salehi, B., Mohammadimanesh, F., Motagh, M., 2017. Random forest wetland classification using ALOS-2 L-band, RADARSAT-2 C-band, and TerraSAR-X imagery. ISPRS J. Photogrammetry Rem. Sens. 130, 13e31. https://doi.org/10.1016/j.isprsjprs.2017.05.010. Maxwell, A.E., Warner, T.A., Fang, F., 2018. Implementation of machine-learning classification in remote sensing: an applied review. Int. J. Rem. Sens. 39, 2784e2817. https://doi.org/10.1080/01431161. 2018.1433343.

371

372

Radar Remote Sensing

Maxwell, A.E., Warner, T.A., Strager, M.P., 2016. Predicting palustrine wetland probability using random forest machine learning and digital elevation data-derived terrain variables. Photogramm. Eng. Rem. Sens. 82, 437e447. https://doi.org/10.14358/PERS.82.6.437. McAlpine, D.F., Smith, I.M., 2010. Assessment of Species Diversity in the Atlantic Maritime Ecozone. NRC Research Press. Melinchuk, R., 1995. Ducks Unlimited’s landscape approach to habitat conservation. Landsc. Urban Plann. 32, 211e217. https://doi.org/10.1016/0169-2046(95)07002-C. Midekisa, A., Holl, F., Savory, D.J., Andrade-Pacheco, R., Gething, P.W., Bennett, A., Sturrock, H.J.W., 2017. Mapping land cover change over continental Africa using Landsat and Google Earth Engine cloud computing. PLoS One 12, e0184926. https://doi.org/10.1371/journal.pone.0184926. Millard, K., Kirby, P., Nandlall, S., Behnamian, A., Banks, S., Pacini, F., 2020. Using growing-season time series coherence for improved peatland mapping: comparing the contributions of sentinel-1 and RADARSAT-2 coherence in full and partial time series. Rem. Sens. 12, 2465. https://doi.org/ 10.3390/rs12152465. Millard, K., Richardson, M., 2015. On the importance of training data sample selection in random forest image classification: a case study in peatland ecosystem mapping. Rem. Sens. 7, 8489e8515. https:// doi.org/10.3390/rs70708489. National Wetlands Working Group, 1997. The Canadian Wetland Classification System. Wetlands Research Branch, University of Waterloo, Waterloo, Ont. Pasquet, S., Pellerin, S., Poulin, M., 2015. Three decades of vegetation changes in peatlands isolated in an agricultural landscape. Appl. Veg. Sci. 18, 220e229. https://doi.org/10.1111/avsc.12142. Ramsar Convention Secretariat, 2010. Wetland inventory: a Ramsar framework for wetland inventory and ecological character description. In: Ramsar Handbooks for the Wise Use of Wetlands, fourth ed. Ramsar Convention Secretariat, Gland, Switzerland. Rodriguez-Galiano, V.F., Ghimire, B., Rogan, J., Chica-Olmo, M., Rigol-Sanchez, J.P., 2012. An assessment of the effectiveness of a random forest classifier for land-cover classification. ISPRS J. Photogram. Rem. Sens. 67, 93e104. https://doi.org/10.1016/j.isprsjprs.2011.11.002. S
anchez-Espinosa, A., Schröder, C., 2019. Land use and land cover mapping in wetlands one step closer to the ground: sentinel-2 versus landsat 8. J. Environ. Manag. 247, 484e498. https://doi.org/10.1016/ j.jenvman.2019.06.084. Shaw, S.P., Fredine, C.G., 1956. Wetlands of the United States: their extent and their value to waterfowl and other wildlife. U S Fish Wildl. Serv. Circ. 39, 1e67. Slagter, B., Tsendbazar, N.-E., Vollrath, A., Reiche, J., 2020. Mapping wetland characteristics using temporally dense Sentinel-1 and Sentinel-2 data: a case study in the St. Lucia wetlands, South Africa. Int. J. Appl. Earth Obs. Geoinf. 86, 102009. https://doi.org/10.1016/j.jag.2019.102009. Statistics Canada, 2018. Ecological Land Classification, 2017. Statistics Canada, Ottawa. Tamiminia, H., Salehi, B., Mahdianpari, M., Quackenbush, L., Adeli, S., Brisco, B., 2020. Google Earth Engine for geo-big data applications: a meta-analysis and systematic review. ISPRS J. Photogram. Rem. Sens. 164, 152e170. https://doi.org/10.1016/j.isprsjprs.2020.04.001. Tian, S., Zhang, X., Tian, J., Sun, Q., 2016. Random forest classification of wetland landcovers from multisensor data in the arid region of Xinjiang, China. Rem. Sens. 8, 954. https://doi.org/10.3390/ rs8110954. van der Kamp, G., Hayashi, M., Bedard-Haughn, A., Pennock, D., 2016. Prairie Pothole wetlands e suggestions for practical and objective definitions and terminology. Wetlands 36, 229e235. https:// doi.org/10.1007/s13157-016-0809-9. von Sengbusch, P., 2015. Enhanced sensitivity of a mountain bog to climate change as a delayed effect of road construction. Mires Peat 15, 1e18. Wang, X., Xiao, X., Zou, Z., Chen, B., Ma, J., Dong, J., Doughty, R.B., Zhong, Q., Qin, Y., Dai, S., Li, X., Zhao, B., Li, B., 2020. Tracking annual changes of coastal tidal flats in China during 1986e2016 through analyses of Landsat images with Google Earth Engine. Rem. Sens. Environ. 238, 110987. https://doi.org/10.1016/j.rse.2018.11.030. Xiong, J., Thenkabail, P., Tilton, J., Gumma, M., Teluguntla, P., Oliphant, A., Congalton, R., Yadav, K., Gorelick, N., 2017a. Nominal 30-m cropland extent map of continental Africa by integrating pixel-based

Toward a North American continental wetland map

and object-based algorithms using sentinel-2 and landsat-8 data on Google earth engine. Rem. Sens. 9, 1065. https://doi.org/10.3390/rs9101065. Xiong, J., Thenkabail, P.S., Gumma, M.K., Teluguntla, P., Poehnelt, J., Congalton, R.G., Yadav, K., Thau, D., 2017b. Automated cropland mapping of continental Africa using Google Earth Engine cloud computing. ISPRS J. Photogrammetry Rem. Sens. 126, 225e244. https://doi.org/10.1016/ j.isprsjprs.2017.01.019.

Further reading Mahdianpari, M., Salehi, B., Rezaee, M., Mohammadimanesh, F., Zhang, Y., 2018b. Very deep convolutional neural networks for complex land cover mapping using multispectral remote sensing imagery. Rem. Sens. 10, 1119. https://doi.org/10.3390/rs10071119. Mahdianpari, M., Brisco, B., Granger, J., Mohammadimanesh, F., Salehi, B., Homayouni, S., BourgeauChavez, L., 2021. The third generation of pan-canadian wetland map at 10 m resolution using multisource earth observation data on cloud computing platform. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 14, 8789e8803. Mahdianpari, M., Salehi, B., Mohammadimanesh, F., Brisco, B., Homayouni, S., Gill, E., DeLancey, E.R., Bourgeau-Chavez, L., 2020e. Big data for a big country: the first generation of Canadian wetland inventory map at a spatial resolution of 10-m using sentinel-1 and sentinel-2 data on the Google earth engine cloud computing platform. Can. J. Rem. Sens. 1e19. https://doi.org/10.1080/ 07038992.2019.1711366.

373

This page intentionally left blank

SECTION 4

Future challenges in radar remote sensing

375

This page intentionally left blank

CHAPTER 20

Challenges in Radar remote sensing Prashant K. Srivastava1, Rajendra Prasad2, Sumit Chaudhary Kumar3, Suraj A. Yadav2, Jyoti Sharma2, Swati Suman1, Varsha Pandey1, Rishabh Singh1 and Dileep Kumar Gupta1, 2

1 Remote Sensing Laboratory, Institute of Environment and Sustainable Development, Banaras Hindu University, Varanasi, Uttar Pradesh, India; 2Department of Physics, Indian Institute of Technology, Banaras Hindu University, Varanasi, Uttar Pradesh, India; 3Indian Institute of Remote Sensing, Dehradun, Uttarakhand, India

1. Introduction In microwaves, compared with radiometers, a radar is an active device that uses an artificial source of energy in which radio signals are transmitted and then received in the form of backscatter to detect targets. These backscatters are further used to develop algorithms for different applications related to biochemical and biophysical parameter retrieval, flood detection and monitoring, soil moisture, snow, droughts, and so forth. Synthetic aperture radar (SAR) can be used to develop high-resolution images at various spatial and temporal resolutions of the earth’s surface. However, backscattered energy received from an area depends on the properties of the sensors (frequency, angle, height, etc.) and surface, such as the slope, roughness, humidity, textural inhomogeneities, and dielectric constant. The different environmental conditions and sensor characteristics provide many challenges, which are briefly presented in the following subsections. Section 1.1 provides details about challenges in radar remote sensing for biophysical parameter retrieval; Section 1.2 provides brief information about different challenges in flood monitoring using radar; Section 1.3 describes challenges associated with soil moisture retrieval using radar; Section 1.4 provides challenges in radar remote sensing for drought; whereas Sections 1.5 and 1.6 provide challenges in radar remote sensing for snow and issues that exist for sensor calibration, validation, and development. 1.1 Challenges in radar remote sensing for biophysical parameters Biophysical parameters are critical factors for determining the health of vegetation, estimating crop growth and yields in agriculture, and estimating carbon stocks. Because of the complementary nature of SAR remote sensing, it is the most suited for biophysical parameter retrieval (Lu, 2006; Sinha et al., 2015). SAR response is affected by the shape, size, and orientation of vegetation, landscape, and underlying soil conditions, in addition to the acquisition geometry of SAR systems. Radar backscatter in high-wavelength bands, such as the P- and L-bands, is highly correlated with major forest parameters

Radar Remote Sensing ISBN 978-0-12-823457-0, https://doi.org/10.1016/B978-0-12-823457-0.00020-3

© 2022 Elsevier Inc. All rights reserved.

377

378

Radar Remote Sensing

(Kurvonen et al., 1999; Sader, 1987). Low-wavelength SAR C-band backscatter, on the other hand, is less suitable for forest parameters (Le Toan et al., 1992); however, it is suitable for crop parameter retrieval. Beaudoin et al. (1994) discovered that radar returns in HH polarization were related to both trunk and crown biomass, whereas radar returns in VV and HV polarization were related only to crown biomass. Aside from these sensor characteristics, the spatial and radiometric resolutions of the SAR system have a significant impact on estimating biophysical parameters. Coarse spatial resolution remote sensors primarily capture canopy information rather than individual tree information. However, after a certain stage, the SAR signal becomes saturated, making parameter estimation difficult. Saturation levels differ depending on wavelengths, polarization, and the vegetation stand structure and ground conditions. According to one study, the L-band SAR image becomes saturated once the biomass density reaches 40 tons per hectare (Luckman et al., 1997). As a result, extra caution should be taken when dealing with signal saturation during parameter retrieval; however, some studies proposed the synergic use of multifrequency SAR images to overcome the problem of saturation (Hoekman and Quinones, 1997; Santos et al., 2004). The accuracy of parameter retrieval is also influenced by the underlying heterogeneous soil conditions. For example, rabi crops or dry land crops have more complex soil conditions than do rice or wetland crops, which makes the retrieval of biophysical parameters for dry land crops more difficult. Lucas et al. (2010) also reported that SAR data (L-band) acquired during periods of low surface moisture and rainfall are better suited to estimating woody vegetation above ground biomass over Queensland, Australia. Topographic factors such as slope and aspect, particularly in mountainous areas, cause chaos in estimating biophysical parameters using SAR data (Austin et al., 2003). Estimating biophysical parameters using a statistical correlation of biophysical parameters with radar backscattering coefficients is a simple method. This may range from basic linear regression to advanced machine learning (ML) techniques of establishing the relation between them. However, the spatial transferability of these methods is problematic owing to various uncertainties owing to acquisition and processing of remotely sensed data, field measurements and in situ parameter calculation, the disparity between remote sensing acquisition date and field data collection, and the size of the sample plot compared with the spatial resolution of remotely sensed data (Foody et al., 2003). To overcome the limitation of statistical methods, various microwave backscattering models have been developed, such as Michigan microwave canopy scattering (MIMICS) (McDonald and Ulaby, 1993; Ulaby et al., 1990) and the water cloud model (WCM) (Attema and Ulaby, 1978) for biophysical parameter retrieval. MIMICS is a complicated model that seems impractical for the inversion of biophysical parameters; however, WCM is a semiempirical model that has gained enormous popularity for parameter

Challenges in Radar remote sensing

retrieval owing to its simplicity (Graham and Harris, 2003; McNairn et al., 2012). These are noncoherent scattering models that lack phase information and rely solely on radar backscattering. The retrieval of biophysical parameters from backscattering models is an ill-posed inversion problem. Various mathematical schemes have been developed to solve ill-posed inverse problems, such as iterative optimization (IO), lookup table (LUT) search, and ML. IO runs the model iteratively to find the best fit between measured and simulated backscatter intensities. The IO approach is simple to implement, but it needs huge computational resources. When it comes to optimizing inversion problems for nonlinear and multivariate models, the IO method may produce inaccurate estimates because it settles into local rather than global minima (Perez et al., 2012). Another widely used inversion technique for retrieving biophysical parameters is the LUT search (McNairn et al., 2012). All possible combinations of model input variables are generated by varying each variable at regular intervals. The calibrated model is then used to simulate the model output for each combination. The LUT is made of each combination and the simulated output values that can be used to retrieve the input parameters corresponding to the given output value. The search for an appropriate combination is primarily based on minimizing the root mean square error as a cost function, but the choice of cost function in an operational context is still debatable (Verrelst et al., 2014). The size of the LUT is determined by the intervals used for each input variable to generate the combinations. Small intervals increase the number of possible combinations and thus the computational cost, but they improve inversion results by lowering the chance of saturation. In the case of a large area to be studied, the computational cost becomes the most expensive. It is most likely because the search method has to go through the entire LUT to find the best solution for each resolution cell. Furthermore, LUT approaches have shown a limited ability to generalize (Rivera et al., 2013). ML methods have been demonstrated to be useful in the inversion of biophysical parameters from these backscattering models. The model has to be trained on simulated radar backscatter (i.e., LUT) and the required parameter will be predicted using the trained model. When the intervals are reduced to generalize the LUT, the training of ML models on the entire LUT becomes computationally intensive; however, once the model is trained, prediction can be performed easily even for a large area. A subset of LUT, created by random sampling from the entire LUT, can also be used to reduce training computation costs. The inversion performance of this method depends entirely on how well the model was trained on the LUT dataset. In addition to backscatter intensity, to exploit phase information in the biophysical parameter retrieval process, various polarimetric SAR-based coherent scattering models were devised (Chiu and Sarabandi, 2000; Cloude and Williams, 2003; Lin and Sarabandi, 1999). The coherence scattering model assumes that the vegetation layer is made of

379

380

Radar Remote Sensing

uniform particles that follow a specific orientation distribution function. In practice, these models are complicated and computationally expensive to implement, and inversion of these complex models is more difficult and requires a more extensive approach (Yuzugullu et al., 2017a,b). Aside from the coherent and noncoherent backscattering model, the SAR interferometry, polarimetry SAR interferometry, and SAR tomography techniques have been used to retrieve tree and crop heights (Mette et al., 2004; Pulliainen et al., 2003; Santos et al., 2004). Although the feasibility of these techniques is limited, they have opened up enormous possibilities in the field of biophysical parameter retrieval research. However, owing to the complexity of data acquisition and processing, the growth rate has slowed, but it is still increasing (Erten et al., 2016; Pichierri, 2016). 1.2 Challenges in radar remote sensing for flood monitoring SAR, interferometric SAR, and optical sensors are the most commonly used satellite products for flood mapping (Mohammadimanesh et al., 2018). However, owing to their different acquisition technique and image properties, each of these satellite products has distinct applications and limitations (Ozdogan et al., 2010). Radar satellite, with all-weather availability, constant frequent revisit time, active imaging mechanisms, and global product availability, has improved the spatiotemporal coverage of flood and postflood assessment substantially (Scotti et al., 2020). The radar backscatter strength is determined by a variety of factors, including surface roughness, dielectric characteristics, and local topography in relation to the radar look angle (Schuler et al., 2002). Water bodies are a specular reflector of the radar pulse. The signal returning from these surfaces to the sensor is low. Therefore, a variety of methods have been proposed in the literature to delineate water bodies using these signal properties using either a single process or a combination. Some of these techniques include histogram thresholding (Brivio et al., 2002; Henry et al., 2006), region growing (Matgen et al., 2011), texture analysis (Pradhan et al., 2014), and fuzzy classification (Martinis et al., 2015). Among these techniques, change detection (CD) is a widely used approach that employs multitemporal SAR images to compare preflood and postflood images of the area to demonstrate temporal changes in land cover. CD is often combined with various other image segmentation techniques to identify the low backscatter region in the image to improve the reliability of the flood delineation using radar images. This technique is described as being more advantageous than others using single-image methodologies (Matgen et al., 2011). Despite these operational advantages of SAR products, many challenges are involved in flood identification and assessment. Major issues involving the roughening of the water surface owing to other environmental extremes such as heavy rainfall and wind, radar shadowing; overlaying or foreshadowing owing to the terrain, the dependence of sensor functioning on the incidence angle, and double-bounce effects of radar signals off water bodies

Challenges in Radar remote sensing

situated between urban structures constrain the extent of flood delineation (Clement et al., 2018). However, these uncertainties can be minimized by using ancillary datasets and complementary approaches such as interferometric coherence techniques (Dasgupta et al., 2018). 1.3 Challenges in radar remote sensing of soil moisture Soil moisture has a vital role in environmental processes such as water and energy transfer between the ground and the atmosphere through evaporation and plant transpiration (Srivastava et al., 2014a; Singh et al., 2019). Soil moisture provides principal constituents and food to the plants and controls the soil temperature (Srivastava et al., 2014b). Therefore, knowledge of soil moisture content is essential for such applications as agriculture, ground water prediction, and land and soil management (Srivastava, 2017). The lowfrequency region of the electromagnetic spectrum is appropriate for soil remote sensing because of high penetration through the vegetation layer (Njoku et al., 2002). Radar systems are advantageous when it is necessary to attain soil moisture information at a high spatial resolution (Ulaby et al., 1996). Various spaceborne microwave radar sensors are operated to monitor soil moisture, such as the Ku band (Quicksat and Scatsat-1), X-band (TerraSAR-X and COSMO SkyMed), C-band (ESA ERS-1/2, Sentinel-1, ESA ENVISAT, and Radarsat-2), and L-band (ALOS PALSAR and SAOCOM) (Barrett et al., 2009). Microwave active sensors onboard satellite or airborne platforms are a good alternative for optical and thermal remote sensing to monitor the earth’s surface (Petropoulos et al., 2015). Several forward or inversion scattering algorithms such as the WCM, radiative transfer model, and integral equation model are used to estimate soil moisture from SAR sensors (Singh et al., 2019). All of these approaches are accompanied by the radar backscattering coefficient, which is more sensitive to surface roughness than surface soil moisture (Singh et al., 2020). Surface roughness is the primary cause of variation the radar backscatter, whereas soil moisture is secondary. Therefore, estimation of soil moisture from radar systems is affected owing to soil surface roughness. This is why it is very necessary to account for surface roughness properly in scattering processes when estimating soil moisture. There are many challenges to measure soil moisture in radar remote sensing. Soil moisture information are generally needed at high spatial and temporal resolutions (Srivastava et al., 2013b). Radar observations are beneficial when needed for a particular region, and an airborne sensor is used for this purpose, so that the measurements can be done at the high spatial and temporal resolutions, but if it comes to a broad region or on a global basis, it is challenging to attain high spatial resolution using Radar. Therefore, it is still a requirement in radar remote sensing to achieve soil moisture on a daily basis, or at high spatio-temporal resolutions (Srivastava et al., 2013c).

381

382

Radar Remote Sensing

Second, SAR sensors are primarily used in the C- or X-band frequency region, which cannot penetrate vegetation cover fully. It is crucial for estimating soil moisture. Instead, the L-band is the most suitable frequency band for soil moisture estimation (Srivastava et al., 2013a). Few radar sensors are available that operate in the L-band, and these radar data are not freely available to students. Apart from that, an L-band radar was launched with the Soil Moisture Active Passive mission; it was freely available, but it stopped operating because of technical issues. Hence, L-band radar observations are required for an accurate estimation of soil moisture, because it fully penetrates the vegetation layer (Shimada et al., 2008). A lot of scattering models have been developed to convert radar backscatter into soil moisture, which requires surface roughness, vegetation biomass, and soil texture as the input parameters (Chaudhary et al., 2021). The accurate estimation of these input parameters is still a great challenge for the accurate assessment of soil moisture (Chaudhary et al., 2022). 1.4 Challenges in radar remote sensing of drought The satellite remote sensing platform provides effective and high potential for near real-time drought identification by the repetitive coverage of spatial and temporal continuous measurements (West et al., 2019). The evolution of different remote sensing applications, particularly high-resolution optical and SAR remote sensing, has been used to characterize drought phenomena and impact assessments. Optical data are preferably used for drought measurements because of their wide availability and simplistic data processing methodology. However, optical data have limitations in acquisition under cloudy conditions. Moreover, optical data can capture surface conditions (e.g., vegetation greenness, phenology, leaf water content, leaf area index etc). On the contrary, the longer operating wavelength of SAR remote sensing can penetrate clouds and provide information even in monsoons (Das et al., 2021). SAR remote sensing relies on backscattering impulses from surface and subsurface features and is a promising alternative to optical data. SAR remote sensing also relies on the dielectric electric constant of the features and surface roughness. Because of the working principle, SAR has several limitations and complexity in precise data capturing and processing. SAR depends on sidelooking geometry and scans the surface at an oblique angle along the range direction perpendicular to the flight direction. This induces geometrical distortions such as foreshortening, layover, and shadow in rugged terrain. Similar errors are observed in urban areas, where features beyond tall buildings are not captured in SAR data. Although foreshortening and layover are corrected employing the digital elevation model, shadows remain uncorrected. A major limitation in SAR data is tedious preprocessing, such as applying the orbit file, radiometric calibration, debursting, multilooking, speckle filtering, and terrain correction, which requires high-end computation facilities.

Challenges in Radar remote sensing

Moreover, SAR data were publicly unavailable until the availability of Sentinel-1 Cband SAR data by the European Space Agency in 2014. The inaccessibility of SAR data has greatly hindered its wide applications. Because of its limited availability, Sentinel-1 data can be used. Moreover, Sentinel-1 data have limited capacity or applications owing to its lower operating wavelength (C-bands of w5 cm). Several studies used SAR data for drought analysis. For example, Abdel-Hamid et al. (2020) employed Sentinel-1 SAR data to assess the impact of drought on different grassland types in Eastern Cape Province. Ghazaryan et al. (2020) evaluated agricultural drought and developed an effective approach to crop condition estimation using optical and SAR data. Datta et al. (2021) employed Sentinel-1 data to estimate the surface soil moisture content in agricultural fallow land during the Rabi cropping season employing various ML models. The lower wavelength of C-band Sentinel-1 SAR has a lower penetration ability. Thus, it can provide only surface soil moisture conditions and is unsuitable for assessing subsurface soil moisture conditions. Speckles in SAR images mostly depend on local parameters; thus, accuracy in estimating surface properties (e.g., soil moisture) greatly varies with the scale. 1.5 Challenges in radar remote sensing of snow Passive remote sensing techniques based on microwaves are not feasible for application in the presence of precipitation or thin snow. The grain size and effects of metamorphism complicate the mapping of snow water equivalents (Kongoli et al., 2004). It is difficult to distinguish dry snow cover from a surface without snow unless there are wavelengths of 2 cm or more (Bernier and Fortin, 1998). The backscattering signal from a volume of dry snow cover is 0 at the L-band (l ¼ 26 cm), very low at the C-band (l ¼ 5 cm), existent at the X-band (l ¼ 3 cm), and significant at the Kuband (l ¼ 1.5 cm) (Baghdadi and Zribi, 2016). Passive microwave methods have proved their worth in determining the spatial extent of sea ice accurately, but they lack the potential to determine ice concentration. To monitor ice concentrations, different algorithms are suggested, but there is no universally accepted algorithm and most fail to perform in summer. Icebergs are relatively easy to detect owing to the presence of water around them. However, most icebergs are smaller than 100 m and require high spatial resolution, which makes it difficult to monitor their movement. Distinguishing iceberg from sea ice is challenging and high-resolution imagery is required. In contrast to sea ice, freshwater ice measurement is not well-defined. This is because of the finer spatial scale of the phenomenon, which is restricted by the spatial resolution of the observing satellites. Detection of freshwater ice is done using both Visible/InfraRed and SAR imagery, although black ice remains a problem, as does determining the thickness of ice from space.

383

384

Radar Remote Sensing

The identification of glaciers and large terrestrial ice is well-developed. However, surface faces, equilibrium line identification, and the determination of accumulated rates using radar images remain ambiguous. Determining glacier thickness from space is difficult, but changes in ice volume are monitored using changes in surface topography. However, owing to the spatial constraints of radar altimeters, topographic measurements are suitable for ice caps and sheets but not for smaller glaciers. 1.6 Challenges in radar remote sensing for sensor development and implementation The extraction of target characteristics using electromagnetic radiation reflectivity from the target has created considerable interest for researchers (Farina et al., 1992). The measurement of target reflectivity led to the development and implementation of the radar system (Farina et al., 1992). Because of advances in radar sensors and data processing technology, the quality of target reflectivity measurement is a major concern for radar missions (Rodriguez-Cassola et al., 2011). The need to develop radar sensors includes real-time data analysis in all weather condition and recognizing and classifying specific targets (Mohammed et al., 2020). Radar systems used onboard have basically airborne or mobile platforms. Therefore, the physical size and weight of the radar sensors have been taken care of in the implementation of radar systems (Koo et al., 2012). The physics behind reflectivity measurements and techniques, such as high frequency, complex data processing, and a wide bandwidth, is used in modern radar systems. However, challenges include defining a calibration system for a specific target, analysis procedures, data acquisitions, storage medium, cost, measurement system stability, range requirements, radar angle measurements, and environmental characterization (Sahebi et al., 2002). The quality of target reflectivity measurement from the onboard sensors of the radar system is directly proportional to the sensor’s calibration (Huang et al., 2015). Measurement systems such as sensors, data acquisition equipment, and playback types of equipment are precisely calibrated to get qualitative and quantitative target information (El-Darymli et al., 2014). Bias error in reflectivity measurements could be eliminated by properly defining the calibration system for a specific target. In addition, implementation of the radar system works almost on the data acquisition and its subsequent analysis. The data acquisition system (DAS) is a sophisticated interface to the analogedigital world that records and stores the data and could also be capable of further data analysis procedures (Dixit and Agrawal, 2012). The nature of data, real-time data processing, storage medium, and the cost are useful in the DAS. The analysis procedure techniques of the observed reflectivity for complex targets from a well-executed measurement platform are also challenging in the proper presentation of results. The measurement system stability platform is also an important aspect for implementing the radar system. The basic system requirement is to measure target returns and extract the desired target information from the returns. To identify the radar returns accurately or build stable radar systems,

Challenges in Radar remote sensing

synchronization of the transmitter and receiver site, calibration, clutter characterization, and frequency referencing is needed (Malanowski and Roszkowski, 2011). Accurate measurements of target radar returns require illumination by a uniform plane wave with constant amplitude and phase. The most common method to deal with the range requirement is to place the target at a sufficient distance from the source such that a spherical incident wave can be considered a plane wave over a target’s dimensions. A common Fraunhofer distance criterion is used to achieve the range requirements conditions (Ezuma et al., 2021). Measurement of the different angular positions of the complex scatterer is especially important in identifying complex targets and is called the glint. Environmental characterization includes the errors in reflectivity owing to the poor implementation of environmental parameters such as the surface roughness, the moisture content of complex targets, wind-generated droplets and small facets on the ocean surface, and drop-size distributions as the rate of rain (Chumchean et al., 2004). This is one of the most significant challenges that is overlooked in implementing the radar system.

2. Conclusion Radar backscattering depends on numerous factors such as the sensor frequency, incidence angle, and terrain features, which include slope, roughness, humidity, textural inhomogeneities, and the dielectric constant. Radar remote sensing of biophysical parameters from backscattering models is considered an ill-posed inversion problem, whereas flood monitoring depends on the incidence angle and double-bounce effects of radar signals off water bodies situated between urban structures. Challenges faced by radar remote sensing of soil moisture are basically got affected due to the soil surface roughness, angle of measurements, spatial and temporal mismatch, radio-frequency interference problems, and so forth. In drought monitoring, the major problem is the lower penetration capability. Thus, it can provide only surface conditions and is unsuitable for assessing subsurface conditions especially during drought. For radar, the determination of glacier thickness from space is difficult owing to low penetration and coarse resolution datasets. For radar sensor development, improper calibration causes errors in radar reflectivity measurements and hence biases in the datasets.

References Abdel-Hamid, A., Dubovyk, O., Graw, V., Greve, K., 2020. Assessing the impact of drought stress on grasslands using multi-temporal SAR data of Sentinel-1: a case study in Eastern Cape, South Africa. Eur. J. Remote Sens. 53, 3e16. Baghdadi, N., Zribi, M., 2016. Land Surface Remote Sensing in Continental Hydrology. Elsevier. Barrett, B.W., Dwyer, E., Whelan, P., 2009. Soil moisture retrieval from active spaceborne microwave observations: an evaluation of current techniques. Rem. Sens. 1, 210e242. Bernier, M., Fortin, J.-P., 1998. The potential of times series of C-band SAR data to monitor dry and shallow snow cover. IEEE Trans. Geosci. Rem. Sens. 36, 226e243.

385

386

Radar Remote Sensing

Brivio, P., Colombo, R., Maggi, M., Tomasoni, R., 2002. Integration of remote sensing data and GIS for accurate mapping of flooded areas. Int. J. Rem. Sens. 23, 429e441. Chaudhary, S.K., Gupta, D.K., Srivastava, P.K., Pandey, D.K., Das, A.K., Prasad, R., 2021. Evaluation of radar/optical based vegetation descriptors in water cloud model for soil moisture retrieval. IEEE Sensor. J. 21, 21030e21037. Chaudhary, S.K., Srivastava, P.K., Gupta, D.K., Kumar, P., Prasad, R., Pandey, D.K., Das, A.K., Gupta, M., 2022. Machine learning algorithms for soil moisture estimation using Sentinel-1: model development and implementation. Adv. Space Res. 69, 1799e1812. Chumchean, S., Seed, A., Sharma, A., 2004. Application of scaling in radar reflectivity for correcting rangedependent bias in climatological radar rainfall estimates. J. Atmos. Ocean. Technol. 21, 1545e1556. Clement, M.A., Kilsby, C., Moore, P., 2018. Multi-temporal synthetic aperture radar flood mapping using change detection. J. Flood Risk Manag. 11, 152e168. Das, P., Pandey, V., Dutta, D., 2021. Land surface water resource monitoring and climate change. In: Mapping, Monitoring, and Modeling Land and Water Resources. CRC Press, pp. 311e326. Dasgupta, A., Grimaldi, S., Ramsankaran, R., Pauwels, V.R., Walker, J.P., Chini, M., Hostache, R., Matgen, P., 2018. Flood mapping using synthetic aperture radar sensors from local to global scales. In: Global Flood Hazard: Applications in Modeling, Mapping, and Forecasting, pp. 55e77. Datta, S., Das, P., Dutta, D., Giri, R.K., 2021. Estimation of surface moisture content using Sentinel-1 C-band SAR data through machine learning models. J. Indian Soc. Remote Sens. 49, 887e896. Dixit, S., Agrawal, R., 2012. Development of data acquisition system using VLSI. IJECCE 3, 1091e1095. El-Darymli, K., McGuire, P., Gill, E., Power, D., Moloney, C., 2014. Understanding the significance of radiometric calibration for synthetic aperture radar imagery. In: 2014 IEEE 27th Canadian Conference on Electrical and Computer Engineering (CCECE). IEEE, pp. 1e6. Ezuma, M., Anjinappa, C.K., Funderburk, M., Guvenc, I., 2021. Radar Cross Section Based Statistical Recognition of UAVs at Microwave Frequencies arXiv preprint arXiv:2102.11954. Farina, A., Studer, F., Vitiello, R., 1992. High resolution radar for enhanced target detection. In: 92 International Conference on Radar. IET, pp. 163e166. Ghazaryan, G., Dubovyk, O., Graw, V., Kussul, N., Schellberg, J., 2020. Local-scale agricultural drought monitoring with satellite-based multi-sensor time-series. GIScience Remote Sens. 57, 704e718. Henry, J.B., Chastanet, P., Fellah, K., Desnos, Y.L., 2006. Envisat multi-polarized ASAR data for flood mapping. Int. J. Rem. Sens. 27, 1921e1929. Huang, M.-C., Liu, J.J., Xu, W., Gu, C., Li, C., Sarrafzadeh, M., 2015. A self-calibrating radar sensor system for measuring vital signs. IEEE Transac. Biomed. Circ. & Syst. 10, 352e363. Kongoli, C., Grody, N.C., Ferraro, R.R., 2004. Interpretation of AMSU microwave measurements for the retrievals of snow water equivalent and snow depth. J. Geophys. Res. Atmos. 109. Koo, V., Chan, Y.K., Vetharatnam, G., Chua, M.Y., Lim, C.H., Lim, C.S., Thum, C., Lim, T.S., bin Ahmad, Z., Mahmood, K.A., 2012. A new unmanned aerial vehicle synthetic aperture radar for environmental monitoring. Progr. Electromagn. Res. 122, 245e268. Malanowski, M., Roszkowski, P., 2011. Bistatic noise radar using locally generated reference signal. In: 2011 12th International Radar Symposium (IRS). IEEE, pp. 544e549. Martinis, S., Kersten, J., Twele, A., 2015. A fully automated TerraSAR-X based flood service. ISPRS J. Photogramm. Remote Sens. 104, 203e212. Matgen, P., Hostache, R., Schumann, G., Pfister, L., Hoffmann, L., Savenije, H., 2011. Towards an automated SAR-based flood monitoring system: lessons learned from two case studies. Phys. Chem. Earth 36, 241e252. Parts A/B/C. Mohammadimanesh, F., Salehi, B., Mahdianpari, M., Brisco, B., Motagh, M., 2018. Wetland water level monitoring using interferometric synthetic aperture radar (InSAR): a review. Can. J. Rem. Sens. 44, 247e262. Mohammed, A.S., Amamou, A., Ayevide, F.K., Kelouwani, S., Agbossou, K., Zioui, N., 2020. The perception system of intelligent ground vehicles in all weather conditions: a systematic literature review. Sensors 20, 6532.

Challenges in Radar remote sensing

Njoku, E.G., Wilson, W.J., Yueh, S.H., Dinardo, S.J., Li, F.K., Jackson, T.J., Lakshmi, V., Bolten, J., 2002. Observations of soil moisture using a passive and active low-frequency microwave airborne sensor during SGP99. IEEE Trans. Geosci. Rem. Sens. 40, 2659e2673. Ozdogan, M., Yang, Y., Allez, G., Cervantes, C., 2010. Remote sensing of irrigated agriculture: opportunities and challenges. Rem. Sens. 2, 2274e2304. Petropoulos, G.P., Ireland, G., Srivastava, P.K., 2015. Evaluation of the soil moisture operational estimates from SMOS in Europe: results over diverse ecosystems. IEEE Sensor. J. 15, 5243e5251. Pradhan, B., Hagemann, U., Tehrany, M.S., Prechtel, N., 2014. An easy to use ArcMap based texture analysis program for extraction of flooded areas from TerraSAR-X satellite image. Comput. Geosci. 63, 34e43. Rodriguez-Cassola, M., Prats, P., Schulze, D., Tous-Ramon, N., Steinbrecher, U., Marotti, L., Nannini, M., Younis, M., L opez-Dekker, P., Zink, M., 2011. First bistatic spaceborne SAR experiments with TanDEM-X. Geosci. Rem. Sens. Lett. IEEE 9, 33e37. Sahebi, M.R., Angles, J., Bonn, F., 2002. A comparison of multi-polarization and multi-angular approaches for estimating bare soil surface roughness from spaceborne radar data. Can. J. Rem. Sens. 28, 641e652. Schuler, D.L., Lee, J.-S., Kasilingam, D., Nesti, G., 2002. Surface roughness and slope measurements using polarimetric SAR data. IEEE Trans. Geosci. Rem. Sens. 40, 687e698. Scotti, V., Giannini, M., Cioffi, F., 2020. Enhanced flood mapping using synthetic aperture radar (SAR) images, hydraulic modelling, and social media: a case study of Hurricane Harvey (Houston, TX). J. Flood Risk Manag. 13, e12647. Shimada, M., Ozawa, T., Fukushima, Y., Furuya, M., Rosenqvist, A., 2008. Japanese L-band radar improves surface deformation monitoring. Eos Transac. Amer. Geophys. Union 89, 277e278. Singh, U., Srivastava, P.K., Pandey, D.K., Chaurasia, S., 2020. Assessment of SCATSAT-1 backscattering by using the state-of-the-art water cloud model. In: Applications of Geomatics in Civil Engineering. Springer, pp. 511e516. Singh, U., Srivastava, P.K., Pandey, D.K., Chaurasia, S., Gupta, D.K., Chaudhary, S.K., Prasad, R., Raghubanshi, A., 2019. ScatSat-1 leaf area index product: models comparison, development, and validation over cropland. Geosci. Rem. Sens. Lett. IEEE. Srivastava, P., O’Neill, P., Cosh, M., Kurum, M., Lang, R., Joseph, A., 2014a. Evaluation of dielectric mixing models for passive microwave soil moisture retrieval using data from ComRAD ground-based SMAP simulator. IEEE J. Sel. Top. Appl. Earth Obs. Rem. Sens. 8 (9), 4345e4354. Srivastava, P.K., 2017. Satellite soil moisture: review of theory and applications in water resources. Water Resour. Manag. 31, 3161e3176. Srivastava, P.K., Han, D., Ramirez, M.A., O’Neill, P., Islam, T., Gupta, M., 2014b. Assessment of SMOS soil moisture retrieval parameters using tau-omega algorithms for soil moisture deficit estimation. J. Hydrol. Srivastava, P.K., Han, D., Ramirez, M.R., Islam, T., 2013a. Appraisal of SMOS soil moisture at a catchment scale in a temperate maritime climate. J. Hydrol. 498, 292e304. Srivastava, P.K., Han, D., Ramirez, M.R., Islam, T., 2013b. Machine learning techniques for downscaling SMOS satellite soil moisture using MODIS land surface temperature for hydrological application. Water Resour. Manag. 27, 3127e3144. Srivastava, P.K., Han, D., Rico-Ramirez, M.A., Al-Shrafany, D., Islam, T., 2013c. Data fusion techniques for improving soil moisture deficit using SMOS satellite and WRF-NOAH land surface model. Water Resour. Manag. 27, 5069e5087. Ulaby, F.T., Dubois, P.C., Van Zyl, J., 1996. Radar mapping of surface soil moisture. J. Hydrol. 184, 57e84. West, H., Quinn, N., Horswell, M., 2019. Remote sensing for drought monitoring & impact assessment: progress, past challenges and future opportunities. Rem. Sens. Environ. 232, 111291.

387

This page intentionally left blank

CHAPTER 21

The study of Indian Space Research Organization’s Ku-band based scatterometer satellite (SCATSAT-1) in agriculture: applications and challenges Ravneet Kaur1, 2, Reet Kamal Tiwari3, Raman Maini2, Sartajvir Singh3, 4 and Vishakha Sood5

1 Apex Institute of Technology, Department of Computer Science and Engineering, Chandigarh University, Gharuan, Punjab, India; 2Department of Computer Science and Engineering, Punjabi University, Patiala, Punjab, India; 3Department of Civil Engineering, Indian Institute of Technology, Ropar, Punjab, India; 4Chitkara University School of Engineering and Technology, Chitkara University, Solan, Himachal Pradesh, India; 5Aiotronics Automation, Palampur, Himachal Pradesh, India

1. Introduction As an active microwave sensor, a scatterometer sends electromagnetic microwave pulses toward the land’s surface, and after the object reflects from the earth’s surface, it measures the backscattered energy with respect to the roughness of the object (Liu, 2002). Major advantages of active microwave sensors are that they provide illumination and are independent of ambient radiation such as in passive microwave sensors (Naeimi and Wagner, 2010). Furthermore, compared with airborne or ground observations, scatterometers offer more than 90% of global coverage daily, which is impossible with other measurement methods (Frison et al., 2016). Moreover, a scatterometer measures the cross-section or sigma-nought (s
) over the earth’s surface in daytime and nighttime with the ability to penetrate clouds (Spencer et al., 2000). A scatterometer is primarily designed to measure wind-vectors (direction and speed) over ocean surfaces owing to sensitivity to the roughness of the oceans (Kumar et al., 2012). However, application areas are emerging because of numerous advantages such as global coverage, daily data delivery, penetration through clouds, large-scale data availability through various scatterometers, and free availability to the research community (Ballantyne and Long, 2002; Panday et al., 2011). The first operational wind scatterometer, known as the Seasat-A Satellite Scatterometer onboard Seasat-A, was launched by the National Aeronautics and Space Administration (NASA) in 1978. Afterward, various organizations launched a series of scatterometers; a detailed description can be found in multiple studies (Singh et al., 2020). Scatterometer sensors are broadly categorized based on the frequency band

Radar Remote Sensing ISBN 978-0-12-823457-0, https://doi.org/10.1016/B978-0-12-823457-0.00003-3

© 2022 Elsevier Inc. All rights reserved.

389

390

Radar Remote Sensing

used: Ku-band (12e18 GHz or wavelength of 16.7e25 mm) and C-band (4e8 GHz or wavelength of 3.75e7.5 cm) (Singh et al., 2006). NASA, India Space Research Organization (ISRO), and China National Space Administration scatterometers operate at the Ku-band frequency range. In contrast, European Space Agency (ESA) scatterometers work in the C-band frequency range. Both frequency bands have their own characteristics, but their applicability is nearly the same. C bandebased scatterometers, including the Earth Remote Sensing (ERS) Scatterometer onboard ERS-1 and Advanced Scatterometer (ASCAT) were launched by the ESA in 1991 and 1995, respectively. These were followed by the NASA Scatterometer (NSCAT) (Kimball et al., 2001), QuikSCAT (Mladenova et al., 2009), SeaWinds (Bartsch, 2010), Ocean Scatterometer (OSCAT) (Chakraborty et al., 2013), HY-2 (Haiyang means ocean) series scatterometer (Li and Shen, 2015), Rapid Scatterometer (Lin and Portabella, 2017), and Scatterometer Satellite (SCATSAT-1) (Singh et al., 2020). In this study, we focused on the Ku band-based SCATSAT-1 and explored its applicability in agriculture. The SCATSAT-1, launched in 2016 by ISRO, is an active microwave sensor initially designed to measure ocean surface wind velocity. It has earth scanning capability in two different polarizations: horizontal-transmit and horizontal-receive (HH) and verticaltransmit and vertical-receive (VV). With time, it has been explored in numerous applications such as soil moisture estimations using vegetation cover (Gaur et al., 2019), snow monitoring and mapping (Nikam et al., 2017; Oza et al., 2019; Singh et al., 2021; Sood et al., 2020), sea-ice monitoring (Singh et al., 2018; Singh and Tiwari, 2021), rice crop monitoring (Gaur et al., 2019), crop phenology (Chaube et al., 2019), the identification of paddy crops (Palakuru et al., 2019) and many more (Singh et al., 2021a). Since its inception, SCATSAT-1 has been widely used for ocean studies, but the data are less explored in the agricultural field. It is becoming more challenging to work with coarse-resolution satellite datasets to identify or quantify different types of vegetation class categories (Maurya et al., 2021). However, with advanced models and methods, satellite data could be improved to extract critical class categories on a large scale. SCATSAT-1 offers an enhanced resolution dataset (up to 2 km) processed through a scatterometer image reconstruction algorithm, which improves the applicability of SCATSAT-1 data over the land surface (Long and Drinkwater, 1999). Soil moisture is one of the most critical parameters in hydrologic studies for estimating partitioning water and energy fluxes (Brocca et al., 2017). Therefore, monitoring and mapping soil moisture are necessary to estimate the amount of water within soil (Moran et al., 2015). To estimate soil moisture over the land surface, remote sensing is a costeffective solution compared with in situ observations (Tripathi et al., 2020). However, optical-based remote sensing is generally affected by cloud cover, whereas microwavebased remote sensing offers cloud-free images during the daytime and nighttime (Singh et al., 2021b). Moreover, the scatterometer is more suitable for soil moisture estimation because of daily global coverage. Therefore, soil moisture monitoring on a global scale will have significant societal benefits (Seneviratne et al., 2010). In the literature, various

The study of Indian Space Research Organization’s Ku-band

attempts were made to estimate soil conditions and vegetation analysis using Ku-band and C-band microwave sensors such as synthetic aperture radar (SAR) (Li et al., 2016, Tripathy and Bhattacharya, 2021), QuikSCAT, and Special Sensor Microwave Imager data (Oza and Parihar, 2007). In addition to these microwave sensors, SCATSAT-1 has been applied in the field of agriculture, such as to estimate paddy crops (Palakuru et al., 2019), rice grain yields (Tripathy et al., 2019), crop phenology (Chaube et al., 2019), soil moisture extraction (Murugan et al., 2019), and the fraction of vegetation cover (Maurya et al., 2021). Here, the main focus is on exploring applications of SCATSAT-1 in the agricultural domain. The study also involves the advantages as well as disadvantages of SCATSAT1 in agriculture. This introductory section is followed by a background on SCATSAT-1. The third and fourth sections consist of applications of SCATSAT-1 and a discussion. A conclusion and future scope are presented.

2. Background of SCATSAT-1 SCATSAT-1, a successor OSCAT, was launched on Sep. 26, 2016, by ISRO to monitor the direction and speed of wind. However, with the availability of various products, applications over the land surface have continuously increased. The technical specifications are highlighted in Table 21.1. Different levels of SCATSAT-1 data products are Table 21.1 Technical specifications of SCATSAT-1. S. no.

Parameters

Value

1 2 3 4

Altitude of spacecraft Orbit Operating frequency Polarization modes

5 6 7 8 9 10 11 12 13 14 15 16

Swath width Scanning radius Inclination angle Scanning rate Gain Wind speed range Antenna diameter Pulse repetition frequency Transmit pulse width Wind direction Orbital period No. of orbits per day

723 km Polar (sun-synchronous) 13.5 GHz (at Ku-band) Horizontal-transmit and horizontal-receive; and vertical-transmit and verticalreceive 14001800 km 700920 km 98 degrees 20.5 rpm 39 dBi 3e30 m/s with accuracy of 10% 1m 193 Hz 1.35 ms 0e360 degrees 99.19 min 14 þ 1/2

391

392

Radar Remote Sensing

L3W

L4BH-South L4BH-North

L3SH

L4AW-625

L3BH-DES

SCATSAT-1 Products

L3IC

L3BH-ASC

L4BH (Global)

L4BH-India L4SV-India

L4AW (Global)

Figure 21.1 Overview of various SCATSAT-1 data products.

available (Fig. 21.1): (1) Level-1B (L1B) (s
coefficient at w6  30 km resolution); (2) L2A (s
coefficient at 25  25 km resolution); (3) L2B (wind velocity at 50  50 km resolution); (4) L3S (s
at 50  50 km resolution); (5) L3W (wind velocity at 25  25 km resolution); (6) L3IC (ice cover at 25  25 km resolution); (7) L3BT (brightness temperature at 25  25 km resolution); (8) L4_India (s
, g
, and Tb at 2  2/6.25  6.25 km resolution); (9) L4_North_Polar (s
, g
, and Tb at 2  2/6.25  6.25 km resolution); (10) L4_South_Polar (s
, g
, and Tb at 2  2/6.25  6.25 km resolution); (11) L4_Full_Globe (s
, g
, and Tb at 2  2/6.25  6.25 km resolution); (12) L4AW (analyzed winds at 25  25 km resolution); (13) L4AW_625 (analyzed winds at 6.25 km resolution); (14) L4HA (high-resolution (HR) AW at 6.25  6.25 km resolution); (15) L4HW (HR winds at 6.25  6.25 km resolution); and (16) L4UI (upwelling index at 25  25 km resolution). In the literature, SCATSAT-1 level 1 products are used for nominal wind estimation (Mankad et al., 2019). Level 2 products are employed in cyclone and weather monitoring (Bhowmick et al., 2019; Jaiswal et al., 2019). Level 3 products are extensively used to forecast weather and natural hazards services (Johny et al., 2019; Kumar and Gairola, 2019; Kumar et al., 2019; Mankad et al., 2019; Misra et al., 2019). Finally, Level 4 products have numerous applications such as land cover, cryospheric studies, paddy crop estimation, retrieval of wind vectors, weather forecasting and monitoring, and cyclone

The study of Indian Space Research Organization’s Ku-band

prediction (Chaube et al., 2019; Gaur et al., 2019; Gupta et al., 2019; Mandal et al., 2018; Maurya et al., 2021; Murugan et al., 2019; Palakuru et al., 2019; Ratheesh et al., 2019; Singh et al., 2018, 2019; Singh and Singh, 2020; Tripathy et al., 2019).

3. Applications in agriculture Since the launch of SCATSAT-1, various studies have been conducted in the field of agriculture. From an agricultural point of view, the scatterometer backscattered signal is generally influenced by multiple surface properties such as roughness, the structure of vegetation, water content within foliage, and soil moisture (Naeimi and Wagner, 2010). The estimation of vegetation surface is challenging using a scatterometer dataset owing to computation of the scattering volume contributing to the entire backscatter signal (Mladenova et al., 2009). The incidence angle also affects the backscattered signal, especially for vegetation studies, because the backscattering signal strength increases with decreases in the incidence signal, and it depends on the surface roughness (Naeimi and Wagner, 2010). The roughness over the soil surface does not vary much compared with vegetation, which highly varies with the season. Various pre-processing and advanced models have been proposed to overcome ambiguities in the SCATSAT-1 dataset (Mankad et al., 2019). In this section, we will explore the applications of SCATSAT-1 in the field of agriculture. This section is divided into subsections concerning the applications, including soil moisture, paddy crop phenology, and the leaf area index (LAI). 3.1 Soil moisture Ku-band SCATSAT-1 has the unique potential to measure soil moisture at a global level, with more than 90% of coverage daily without being intercepted by various atmospheric effects such as haze and clouds. Other than scatterometers, multiple options are available to deliver soil moisture products: (1) ASCAT onboard Meteorological Operational (METOP-A and METOP-B) satellites (C-band scatterometer) with 25 km spatial resolution and 1-day temporal resolution (Wagner et al., 2013); (2) Soil Moisture and Ocean Salinity (SMOS) mission product (L-band radiometer) with 50 km spatial resolution and 2-day temporal resolution (Kerr et al., 2012); (3) Advanced Microwave Scanning Radiometer-2 (AMSR-2) onboard the Global Change Observation Mission for Water satellite (C-band and X-band radiometers) with w25 km spatial resolution and 1-day temporal resolution (Kim et al., 2015); and (4) soil moisture active and passive (SMAP) mission with w36 km spatial resolution and 2-day temporal resolution (Enenkel et al., 2016). The availability of soil moisture products started in (1) Jan. 2007 with ASCAT, (2) Jan. 2010 with SMOS, (3) Jul. 2012 with AMSR-2, and (d) Apr. 2015 with SMAP. The major problem associated with these microwave sensors is that the applications are limited owing to low spatial resolution. A detailed description of soil moisture products can be found in different studies (Brocca et al., 2016, 2017). On the other

393

394

Radar Remote Sensing

Figure 21.2 Framework of soil moisture product generation using SCATSAT-1. NDVI, normalized difference vegetation index; VTCI, vegetation temperature condition index.

hand, SCATSAT-1 offers land surface interface information with w2 km enhanced resolution and 1-day temporal resolution, and these data are available from Sep. 2016 onward. Therefore, SCATSAT-1 is satellite data that attract many researchers in the field of agriculture. There are numerous applications of soil moisture products in various applications such as the forecasting of floods and monitoring of drought and soil erosion, and forecasting of crops and vegetation. However, the main interest is in analyzing the dynamics of vegetation. Soil moisture is an essential factor involved in plant growth and climate, and can be considered a vital indicator of agriculture yield and forecasting analysis. Various studies were conducted with microwave sensors to retrieve and analyze soil moisture using theoretical, empirical, and semiempirical methods (Dubois et al., 1995; Fung et al., 1992; Maurya et al., 2021; Oh et al., 1992; Ulaby et al., 1982). Murugan et al. (2019) reported a framework to generate high-quality soil moisture information using SCATSAT-1 (Fig. 21.2).

3.1.1 Framework steps Step 1: Computation of normalized difference vegetation index (NDVI) via red and near-infrared bands using an optical sensor such as Moderate Resolution Imaging Spectroradiometer (MODIS) using the equation: NDVI ¼

BandNIR  BandRed BandNIR þ BandRed

(21.1)

Step 2: The different class categories can be identified from NDVI data. For each pixel of NDVI, a scatterometer cross-section (s
) is calculated. Step 3: The scatterometer backscattered coefficient is divided into two parts: translucent and nontransparent vegetation.

The study of Indian Space Research Organization’s Ku-band

Step 4: Afterward, backscattering from the soil surface is computed and backscattering
from translucent st is calculated as: 

wt Cosq st ¼ (21.2) 1  e2sr=Cosq þ s0soil e2sr=Cosq þ 2wdwt sr e2sr=Cosq 2 Terms wt and sr represent the scattering albedo and optical depth of translucent vegetation, respectively; w and d represent the empirical coefficient and Fresnel power reflectivity, respectively. Step 5: Compute the soil dielectric constant using the Dubois method. Step 6: Estimate the volumetric soil moisture constant. Step 7: Finally, compute the vegetation temperature condition index (VTCI) via MODIS and average it with soil moisture products as: VTCI ¼

Tmax  Ts Tmax þ Ts

(21.3)

Terms Tmax and Ts correspond to the highest and smallest land surface temperatures.

3.2 Paddy crop The paddy crop is an important food crop in South Asian countries such as India, Bangladesh, Bhutan, Nepal, Sri Lanka, and Pakistan. It also contributes to improvements in the national economy. A significant portion of the rice crop is cultivated during the Kharif season. According to Indian weather, rice production depends on monsoon rains; w60% of rice receives irrigation. Therefore, monitoring and forecasting this crop are essential for food safety. The Mahalanobis National Crop Forecasting Center (MNCFC) provides rice yield information with the help of C bandebased Radar Satellite-1 and Sentinel-1 with Ceband SAR (Tripathy et al., 2019). It also uses optical remote sensing data and weather-based models to estimate the paddy crop. SCATSAT-1 can interact with the surface and deliver critical information regarding canopy, grain, and rice crops. Various attempts have been made in the literature to estimate the harvest using active and passive microwave satellite datasets (Oza et al., 2008; Oza and Parihar, 2007). Here, we considered only studies that used SCATSAT-1 data to monitor and map the paddy crop. Various authors proposed different models for estimating paddy crops (Chaube et al., 2019; Gaur et al., 2019; Palakuru et al., 2019; Tripathy et al., 2019). Here, we have summarized three models developed by Chaube et al. (2019), Gaur et al. (2019), and Tripathy et al. (2019). Fig. 21.3 represents model 1 for estimating rice phenology, in which three types of information (NDVI derived from the optical dataset, water index [WI], and SCATSAT-1) (Chaube et al., 2019; Palakuru et al., 2019) were used.

395

396

Radar Remote Sensing

Figure 21.3 A framework of model 1 for estimating rice phenology. NDVI, normalized difference vegetation index; WI, water index.

3.2.1 Methodology Step 1: Compute NDVI from optical data and preprocessing the backscattered coefficient (sigma-nought) for both HH and VV polarizations and the first derivative of WI. Step 2: Perform the statistical analysis with NDVI from optical data and the backscattered coefficient (sigma-nought) and compute the heading stage and spatial distribution. Step 3: Compute the transplanting and harvesting stages using NDVI from optical data and the SCATSAT-1 backscattered coefficient (sigma-nought). Step 4: The pudding stage is computed using WI and the first derivative of WI. Fig. 21.4. represents model 2 for identifying rice crop phenology, in which two types of data are used (i.e., SCATSAT-1 and Sentinel-1) (Gaur et al., 2019).

Figure 21.4 Framework of model 2 for identifying rice crop phenology. Poly., polynomial.

The study of Indian Space Research Organization’s Ku-band

3.2.2 Methodology Step 1: Initially, the digital number values of SCATSAT-1 are converted into the actual backscattered (dB) value to covert sigma-nought into dB (Singh et al., 2021b): s0 ðdBÞ ¼ ðS  DNÞ þ Ov

(21.4)

Parameter S is the slope of product parameters. That is, s0 and Ov represent the offset value normally considered to be 0.001. Step 2: A rice crop map is generated using Sentinel-1 with the help of an unsupervised classification method and decision rule-based classification (Gaur et al., 2019). Step 3: The rice crop is regridded at 2 km resolution. The rice crop maps generated via Sentinel-1 are laid on the SCATSAT-1 grid using GIS software. Step 4: Afterward, the sixth-order polynomial defines the best fit for the SCATSAT1 dataset to identify rice crops in three different phenological stages. Step 5: Finally, each pixel value of SCATSAT-1 is assigned to different phenological stages of the rice crop. Fig. 21.5. represents model 3 for identifying rice yield prediction with the SCATSAT-1 dataset. 3.2.3 Methodology Step 1: Initially, the SCATSAT-1 sigma-nought backscattered coefficient from the level 4 India product (at HH and VV polarization) is used. Step 2: State-wise data are collected using a crop cutting experiment (CCE) corresponding to the HH/VV ratio. Step 3: A rice mask is applied to the ratio (for selected studies), which can be collected from MNCFC, India.

Figure 21.5 Framework for identifying rice yield prediction with SCATSAT-1 dataset. CCE, crop cutting experiment; HH, horizontal-transmit and horizontal-receive; Pol., polymerization; VV, vertical-transmit and vertical-receive.

397

Radar Remote Sensing

Step 4: Afterward, the model (optimum bio-window) is implemented to compute the rice yield estimation. Step 5: Finally, the outcomes are compared and validated with the national- or statelevel rice yield dataset. 3.3 Leaf area index Singh et al. (2019) presented a methodology for generating LAI using SCATSAT-1. The LAI is a critical parameter for understanding biophysical processes in plants. It also acts as a vital component for soil moisture modeling (Singh et al., 2019). Various datasets, such as ERS-1 (in 1991e2000), ERS-2 (in 1995e2011), RADARSAT-1 (in 1995e2013), ENVISATeAdvanced Synthetic Aperture Radar (in 2002e12), RADARSAT-2 (2007), and Sentinel-1 (2014) were used for vegetation analysis. However, these datasets are limited by spatial resolution. To estimate or retrieve the LAI using a SCATSAT-1 cross-section (s
), two different models were developed or improved (Singh et al., 2019): the water cloud model (WCM) (Attema and Ulaby, 1978) and the Oveisgharan et al. (2018) model. The WCM s
ðdBÞ model is computed as:  

 
s
ðdBÞ ¼ 10log10 AV1 Cos q 1  exp 2BV1=Cos q þ exp 2BV2=Cos q ss (21.5)


The term ss represents the contributing factor under the soil; terms A and Brepresent vegetation parameters; and V1 and V2 are canopy descriptors. According to the 
Oveisgharan et al. (2018) model, backscattered power spp; i is given as:


spp; i ¼ fbare  ðc þ d  mV Þ þ ð1  fbare Þ 8 9 < = a Bpp ð LAI 2 Þ Bpp ð LAI 2 Þ þ Cpp ðLAI 2 Þ e A ð1  e : pp ; =

=

=

398

(21.6)

Term, fbare is a vegetation fraction; App , Bpp , Cpp , a, c, and d are training coefficients and are estimated using the nonlinear least square fitting model; and mV represents volumetric soil moisture.

4. Summary and conclusions In the previous section, we reviewed the contribution of Ku bandebased SCATSAT-1 data in the agricultural domain. Three different areas were explored related to agriculture: (1) soil moisture, (2) paddy crop, and (3) LAI. In some of these domains, multiple models were explained concerning the data requirements (other than SCATSAT-1), the

The study of Indian Space Research Organization’s Ku-band

complexity of the model, and their incorporation with existing techniques. Murugan et al. (2019) proposed an approach to generate HR soil moisture products using SCATSAT-1 and MODIS. They validated outcomes with soil moisture data retrieved from SMAP and SMOS sensors. For validation, the root means square error (RMSE) was computed. The RMSE lay between 0.05 and 0.15 m3/m3 for retrieved data. The outcomes concluded that the computed soil moisture from SCATSAT-1 wellmatched with SMAP soil moisture, with an average of 0.1 m3/m3. Moreover, 0.1 RMSE varied for soil moisture generated from SCATSAT-1 and SMAP, whereas 0.15 RMSE varied for SMOS. This model is effective in terms of the error rate, but the soil moisture quality still needs to improve in the extraction of different land use and land cover class categories. To estimate the paddy crop yield, three models were developed by various authors including Chaube et al. (2019), Palakuru et al. (2019), Gaur et al. (2019), and Tripathy et al. (2019). Chaube et al. (2019) reported two different approaches: a statistical model and a polynomial fitebased model. The statistical model uses the MODIS NDVI, WI, and first derivative of WI, whereas the polynomial fit-based model describes the fit of time series backscatter values. From the outcomes, 5 dB of variation exists in crosssection (s
) observed in dry/wet desert and 2 dB of variation in cross-section (s
) on barren land. The backscattered values varied between 10.6 and 12.5 dB with HH polarization and 11 to 12.5 dB with VV polarization during the transplanting to maturity stage. Gaur et al. (2019) used Sentinel-1 with SCATSAT-1 to estimate the paddy crop and analyze temporal behavior. They reported that during the transplanting stage, backscattered values were observed between 13 and 14 dB in the selected study area, and during the transplanting to maturity stage, backscattered values were observed between 9 and 14 dB. The maximum backscattered value was observed during the tillering phase, and the minimum backscattered value was observed during the heading stage, as also reported by various authors (Oza and Parihar, 2007). These outcomes showed a good correlation with respect to field observations as well as MODIS NDVI. Palakuru et al. (2019) reported a 0.069 standard error mean of Gaussian distribution and 0.687 standard error mean of the logistic distribution. Tripathy et al. (2019) estimated rice grain over South Asian countries with the help of SCATSAT-1 L4 data, MNCFC (India) for crop mask, CCE and reported yield data. They reported that dual maxima were observed during the period of tillering and filling, and dual minima were observed before transplanting. Moreover, the coefficient of determination was observed to be 95% for rice yield prediction for all selected states. The correlation coefficient was observed to be 0.95, with an RMSE of 0.28. In the LAI estimation, Singh et al. (2019b) reported two improved models for LAI estimation: the WCM (Attema and Ulaby, 1978) and Oveisgharan et al. (2018) models. In Oveisgharan et al. (2018), the R2 , RMSE, and bias were 0.99, 0.02 m2me2, and 0.01 m2me2,

399

400

Radar Remote Sensing

respectively, whereas in the case of the WCM model, the R2 , RMSE, and bias were 0.88, 0.45 m2me2, and 0.23 m2me2, respectively. Moreover, the Oveisgharan et al. model performed marginally better (1 m2me2) as compared with the WCM model, but performed equally well when the LAI was less than 5 m2me2. These observations show that SCATSAT-1 is a good alternative to existing microwave models in terms of an improvement in resolution, global coverage, and product availability. SCATSAT-1 also encourages researchers owing to its numerous advantages, such as cloud penetration, daily data delivery services, daytime and nighttime measurements, continuous global coverage, and nominal resolution. However, Ku-band SCATSAT-1 applicability is limited because of the inability to identify the wheat heading stage and leaves; as a result, there is a downfall in saturation (Moran et al., 1998). Nevertheless, there is still a need to enhance the scope of SCATSAT-1 in many remote sensing applications. To accomplish this, research activities must be performed developing blended products with the help of daily-based MODIS datasets or products that can solve multispectral information problems. However, it is expected that the future mission of scatterometer will also consider the demand of SCATSAT-1 land surface applications and improve spatial resolution and noise reduction, and add more features. Moreover, it is also expected that the incorporation of advanced or robust algorithms such as deep learning may be helpful to improve information or object extraction.

Acknowledgments The authors would like to thank the Meteorological and Oceanographic Satellite Data Archival Center (MOSDAC), Indian Space Research Organization (ISRO), for providing the scatterometer satellite (SCATSAT-1) for research purposes.

Funding This research work is financially supported by the Science and Engineering Research Board, Department of Science and Technology, India, under a Teachers Associateship for Research Excellence (Grant no. TAR/ 2019/000354).

References Attema, E.P.W., Ulaby, F.T., 1978. Vegetation modeled as a water cloud. Radio Sci. 13, 357e364. https:// doi.org/10.1029/RS013i002p00357. Ballantyne, J., Long, D.G., 2002. A multidecadal study of the number of Antarctic icebergs using scatterometer data. Int. Geosci. Remote Sens. Symp. 5, 3029e3031. https://doi.org/10.1109/ igarss.2002.1026859. Bartsch, A., 2010. Ten years of sea winds on QuikSCAT for snow applications. Rem. Sens. 2, 1142e1156. https://doi.org/10.3390/rs2041142. Bhowmick, S.A., Cotton, J., Fore, A., Kumar, R., Payan, C., Rodríguez, E., Sharma, A., Stiles, B., Stoffelen, A., Verhoef, A., 2019. An assessment of the performance of ISRO’s SCATSAT-1 Scatterometer. Curr. Sci. 117, 959e972. https://doi.org/10.18520/cs/v117/i6/959-972.

The study of Indian Space Research Organization’s Ku-band

Brocca, L., Ciabatta, L., Moramarco, T., Ponziani, F., Berni, N., Wagner, W., 2016. Use of satellite soil moisture products for the operational mitigation of landslides risk in Central Italy. In: Satellite Soil Moisture Retrieval. Elsevier, pp. 231e247. https://doi.org/10.1016/B978-0-12-803388-3.00012-7. Brocca, L., Crow, W.T., Ciabatta, L., Massari, C., de Rosnay, P., Enenkel, M., Hahn, S., Amarnath, G., Camici, S., Tarpanelli, A., Wagner, W., 2017. A review of the applications of ASCAT soil moisture products. IEEE J. Sel. Top. Appl. Earth Obs. Rem. Sens. 10, 2285e2306. https://doi.org/10.1109/ JSTARS.2017.2651140. Chakraborty, A., Deb, S.K., Shikakolli, R., Gohil, B.S., Kumar, R., 2013. Intercomparison of OSCAT winds with numerical-model-generated winds. Geosci. Rem. Sens. Lett. IEEE 10, 260e262. https:// doi.org/10.1109/LGRS.2012.2200094. Chaube, N.R., Chaurasia, S., Tripathy, R., Pandey, D.K., Misra, A., Bhattacharya, B.K., Chauhan, P., Yarakulla, K., Bairagi, G.D., Srivastava, P.K., Teheliani, P., Ray, S.S., 2019. Crop phenology and soil moisture applications of SCATSAT-1. Curr. Sci. 117, 1022e1031. https://doi.org/10.18520/cs/ v117/i6/1022-1031. Dubois, P.C., van Zyl, J., Engman, T., 1995. Measuring soil moisture with imaging radars. IEEE Trans. Geosci. Rem. Sens. 33, 915e926. https://doi.org/10.1109/36.406677. Enenkel, M., Steiner, C., Mistelbauer, T., Dorigo, W., Wagner, W., See, L., Atzberger, C., Schneider, S., Rogenhofer, E., 2016. A combined satellite-derived drought indicator to support humanitarian aid organizations. Rem. Sens. 8, 340. https://doi.org/10.3390/rs8040340. Frison, P.-L., Jarlan, L., Mougin, E., 2016. Using satellite scatterometers to monitor continental surfaces. In: Land Surface Remote Sensing in Continental Hydrology. Elsevier, pp. 79e113. https://doi.org/ 10.1016/B978-1-78548-104-8.50003-6. Fung, A.K., Li, Z., Chen, K.S., 1992. Backscattering from a randomly rough dielectric surface. IEEE Trans. Geosci. Rem. Sens. 30, 356e369. https://doi.org/10.1109/36.134085. Gaur, P., Tahlani, P., Tripathy, R., Bhattacharya, B.K., Ray, S.S., 2019. Identification of rice crop phenology using Scatsat-1 Ku-band scatterometer in Punjab and Haryana. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. - ISPRS Arch. 42, 549e555. https://doi.org/10.5194/isprs-archivesXLII-3-W6-549-2019. Gupta, P.K., Pradhan, R., Singh, R.P., Misra, A., 2019. Scatterometry for land hydrology science and its applications. Curr. Sci. 117, 1014e1021. https://doi.org/10.18520/cs/v117/i6/1014-1021. Jaiswal, N., Kumar, P., Kishtawal, C.M., 2019. SCATSAT-1 wind products for tropical cyclone monitoring, prediction and surface wind structure analysis. Curr. Sci. 117, 983e992. https://doi.org/10.18520/cs/ v117/i6/983-992. Johny, C.J., Singh, S.K., Prasad, V.S., 2019. Validation and impact of SCATSAT-1 scatterometer winds. Pure Appl. Geophys. 176, 2659e2678. https://doi.org/10.1007/s00024-019-02096-5. Kerr, Y.H., Waldteufel, P., Richaume, P., Wigneron, J.P., Ferrazzoli, P., Mahmoodi, A., Al Bitar, A., Cabot, F., Gruhier, C., Juglea, S.E., Leroux, D., Mialon, A., Delwart, S., 2012. The SMOS soil moisture retrieval algorithm. IEEE Trans. Geosci. Rem. Sens. 50, 1384e1403. https://doi.org/10.1109/ TGRS.2012.2184548. Kim, S., Liu, Y.Y., Johnson, F.M., Parinussa, R.M., Sharma, A., 2015. A global comparison of alternate AMSR2 soil moisture products: why do they differ? Remote Sens. Environ. 161, 43e62. https:// doi.org/10.1016/j.rse.2015.02.002. Kimball, J.S., McDonald, K.C., Keyser, A.R., Frolking, S., Running, S.W., 2001. Application of the NASA scatterometer (NSCAT) for determining the daily frozen and nonfrozen landscape of Alaska. Remote Sens. Environ. 75, 113e126. Kumar, P., Gairola, R.M., 2019. Impact of SCATSat-1 retrieved wind vectors on short range WRF model predictions over the south-asia region key points. J. Geophys. Res. Atmos. 124 (22), 11994e12009. https://doi.org/10.1029/2019JD030642. Kumar, P., Kumar, K.P.H., Pal, P.K., 2012. Impact of Oceansat-2 scatterometer winds and TMI observations on Phet cyclone simulation. IEEE Trans. Geosci. Rem. Sens. 51, 3774e3779. https://doi.org/ 10.1109/TGRS.2012.2221720.

401

402

Radar Remote Sensing

Kumar, R., Bhowmick, S.A., Chakraborty, A., Sharma, A., Sharma, S., Seemanth, M., Gupta, M., Chakraborty, P., Modi, J., Misra, T., 2019. Post-launch calibration-validation and data quality evaluation of SCATSAT-1. Curr. Sci. 117, 973e982. https://doi.org/10.18520/cs/v117/i6/973-982. Li, D., Shen, H., 2015. Evaluation of wind vectors observed by HY-2A scatterometer using ocean buoy observations, ASCAT measurements, and numerical model data. Chin. J. Oceanol. Limnol. 33, 1191e1200. https://doi.org/10.1007/s00343-015-4136-4. Li, K., Yang, Z., Shao, Y., Liu, L., Zhang, F., 2016. Rice phenology retrieval automatically using polarimetric SAR. In: 2016 IEEE International Geoscience and Remote Sensing Symposium (IGARSS). IEEE, pp. 5674e5677. https://doi.org/10.1109/IGARSS.2016.7730482. Lin, W., Portabella, M., 2017. Toward an improved wind quality control for RapidScat. IEEE Trans. Geosci. Rem. Sens. 55, 3922e3930. Liu, W.T., 2002. Progress in scatterometer application. J. Oceanogr. 58, 121e136. https://doi.org/10.1023/ A:1015832919110. Long, D.G., Drinkwater, M.R., 1999. Cryosphere applications of NSCAT data. IEEE Trans. Geosci. Rem. Sens. 37, 1662e1670. https://doi.org/10.1109/36.763285. Mandal, S., Sil, S., Shee, A., Swain, D., Pandey, P.C., 2018. Comparative analysis of SCATSat-1 gridded winds with Buoys, ASCAT, and ECMWF winds in the Bay of Bengal. IEEE J. Sel. Top. Appl. Earth Obs. Rem. Sens. 11, 845e851. https://doi.org/10.1109/JSTARS.2018.2798621. Mankad, D., Sikhakolli, R., Kakkar, P., Saquib, Q., Agrawal, K.M., Gurjar, S., Jain, D.K., Ramanujam, V.M., Thapliyal, P., 2019. SCATSAT-1 Scatterometer data processing. Curr. Sci. 117, 950e958. https://doi.org/10.18520/cs/v117/i6/950-958. Maurya, A.K., Murugan, D., Singh, D., 2021. An approach for soil moisture estimation using urban and vegetation fraction cover from coarse resolution Scatsat-1 data. Adv. Space Res. 68, 1329e1340. https://doi.org/10.1016/j.asr.2021.03.022. Misra, T., Chakraborty, P., Lad, C., Gupta, P., Rao, J., Upadhyay, G., Vinay Kumar, S., Saravana Kumar, B., Gangele, S., Sinha, S., Tolani, H., Vithani, V.K., Raman, B.S., Rao, C.V.N., Dave, D.B., Jyoti, R., Desai, N.M., 2019. SCATSAT-1 Scatterometer: an improved successor of OSCAT. Curr. Sci. 117, 941e949. https://doi.org/10.18520/cs/v117/i6/941-949. Mladenova, I., Lakshmi, V., Walker, J.P., Long, D.G., De Jeu, R., 2009. An assessment of QuikSCAT Kuband scatterometer data for soil moisture sensitivity. Geosci. Rem. Sens. Lett. IEEE 6, 640e643. Moran, M.S., Doorn, B., Escobar, V., Brown, M.E., 2015. Connecting NASA science and engineering with earth science applications. J. Hydrometeorol. 16, 473e483. https://doi.org/10.1175/JHM-D-140093.1. Moran, M.S., Vidal, A., Troufleau, D., Inoue, Y., Mitchell, T.A., 1998. Ku-and C-band SAR for discriminating agricultural crop and soil conditions. IEEE Trans. Geosci. Rem. Sens. 36, 265e272. https:// doi.org/10.1109/36.655335. Murugan, D., Maurya, A.K., Garg, A., Singh, D., 2019. A framework for high-resolution soil moisture extraction using SCATSAT-1 scatterometer data. IETE Tech. Rev. 37 (2), 147e156. https:// doi.org/10.1080/02564602.2019.1575293. Naeimi, V., Wagner, W., 2010. C-band scatterometers and their applications. In: Geoscience and Remote Sensing New Achievements. InTech, pp. 229e246. https://doi.org/10.5772/9102. Nikam, B.R., Garg, V., Gupta, P.K., Thakur, P.K., Senthil Kumar, A., Chouksey, A., Aggarwal, S.P., Dhote, P., Purohit, S., 2017. Satellite-based mapping and monitoring of heavy snowfall in North Western Himalaya and its hydrologic consequences. Curr. Sci. 113, 2328e2334. https://doi.org/10.18520/ cs/v113/i12/2328-2334. Oh, Y., Sarabandi, K., Ulaby, F.T., 1992. An empirical model and an inversion technique for radar scattering from bare soil surfaces. IEEE Trans. Geosci. Rem. Sens. 30, 370e381. https://doi.org/10.1109/ 36.134086. Oveisgharan, S., Haddad, Z., Turk, J., Rodriguez, E., Li, L., 2018. Soil moisture and vegetation water content retrieval using QuikSCAT data. Rem. Sens. 10, 636. https://doi.org/10.3390/rs10040636. Oza, S.R., Bothale, R.V., Ram Rajak, D., Jayaprasad, P., Maity, S., Thakur, P.K., Tripathi, N., Chouksey, A., Bahuguna, I.M., 2019. Assessment of cryospheric parameters over the Himalaya and Antarctic regions using SCATSAT-1 enhanced resolution data. Curr. Sci. 117, 1002e1013. https:// doi.org/10.18520/cs/v117/i6/1002-1013.

The study of Indian Space Research Organization’s Ku-band

Oza, S.R., Panigrahy, S., Parihar, J.S., 2008. Concurrent use of active and passive microwave remote sensing data for monitoring of rice crop. Int. J. Appl. Earth Obs. Geoinf. 10, 296e304. https://doi.org/ 10.1016/j.jag.2007.12.002. Oza, S.R., Parihar, J.S., 2007. Evaluation of Ku-band QuikSCAT scatterometer data for rice crop growth stage assessment. Int. J. Rem. Sens. 28, 3447e3456. https://doi.org/10.1080/01431160601034860. Palakuru, M., Yarrakula, K., Chaube, N.R., Sk, K.B., Satyaji Rao, Y.R., 2019. Identification of paddy crop phenological parameters using dual polarized SCATSAT-1 (ISRO, India) scatterometer data. Environ. Sci. Pollut. Res. 26, 1565e1575. https://doi.org/10.1007/s11356-018-3692-5. Panday, P.K., Frey, K.E., Ghimire, B., 2011. Detection of the timing and duration of snowmelt in the Hindu Kush-Himalaya using QuikSCAT, 2000-2008. Environ. Res. Lett. 6, 2000e2008. https://doi.org/ 10.1088/1748-9326/6/2/024007. Ratheesh, S., Chaudhary, A., Agarwal, N., Sharma, R., 2019. Role of ocean dynamics on mesoscale and sub-mesoscale variability of Ekman pumping for the Bay of Bengal using SCATSAT-1 forced ocean model simulations. Curr. Sci. 117, 993e1001. https://doi.org/10.18520/cs/v117/i6/993-1001. Seneviratne, S.I., Corti, T., Davin, E.L., Hirschi, M., Jaeger, E.B., Lehner, I., Orlowsky, B., Teuling, A.J., 2010. Investigating soil moistureeclimate interactions in a changing climate: a review. Earth Sci. Rev. 99, 125e161. https://doi.org/10.1016/j.earscirev.2010.02.004. Singh, C., Trivedi, D., Mohan, S., Ajai, 2006. Comparison of Ku- and C-band radar backscatter signatures over Indian region. In: Geomatics-2006 (National Conference on Geomatics for Infrastructure Development). Chennai, India, pp. 1e12. Singh, R.K., Singh, K.N., Maisnam, M., P., J., Maity, S., 2018. Antarctic sea ice extent from ISRO’s SCATSAT-1 using PCA and an unsupervised classification. Proceedings 2, 340. https://doi.org/ 10.3390/ecrs-2-05153. Singh, R.K., Singh, K.N., Maisnam, M., P, J., Maity, S., 2019. Observing Larsen C ice-shelf using ISRO’s SCATSAT-1 data. Pol. Sci. 19, 57e68. https://doi.org/10.1016/j.polar.2018.12.007. Singh, S., Tiwari, R.K., 2021. Detection of cryospheric parameters with artificial neural network over antarctic region using ku-band based ISRO’s SCATSAT-1 data. In: 2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS. IEEE, pp. 435e438. https://doi.org/10.1109/ IGARSS47720.2021.9555088. Singh, S., Tiwari, R.K., Gusain, H.S., Sood, V., 2020. Potential applications of SCATSAT-1 satellite sensor: a systematic review. IEEE Sensor. J. 20, 12459e12471. https://doi.org/10.1109/JSEN.2020.3002720. Singh, S., Tiwari, R.K., Sood, V., Gusain, H.S., 2021. Detection and validation of spatiotemporal snow cover variability in the Himalayas using Ku-band (13.5 GHz) SCATSAT-1 data. Int. J. Rem. Sens. 42, 805e815. https://doi.org/10.1080/2150704X.2020.1825866. Singh, S., Tiwari, R.K., Sood, V., Gusain, H.S., Prashar, S., 2021a. Image-fusion of ku-band based SCATSAT-1 and MODIS data for cloud-free Change detection over western himalayas. IEEE Trans. Geosci. Rem. Sens. 1e13. https://doi.org/10.1109/TGRS.2021.3123392. Singh, S., Tiwari, R.K., Sood, V., Prashar, S., 2021b. In: Reddy, V.S., Prasad, V.K., Wang, J., Reddy, K.T.V. (Eds.), Unsupervised Snow Cover Classification Using Dual-Polarized SCATSAT-1 Satellite Data BT - Soft Computing and Signal Processing. Springer Singapore, Singapore, pp. 627e635. https://doi.org/10.1007/978-981-33-6912-2_57. Singh, U., Srivastava, P.K., Pandey, D.K., Chaurasia, S., Gupta, D.K., Chaudhary, S.K., Prasad, R., Raghubanshi, A.S., 2019. ScatSat-1 leaf area index product: models comparison, development, and validation over cropland. Geosci. Rem. Sens. Lett. IEEE 1e5. https://doi.org/10.1109/lgrs.2019.2927468. Singh, U.S., Singh, R.K., 2020. Application of maximum-likelihood classification for segregation between Arctic multi-year ice and first-year ice using SCATSAT-1 data. Remote Sens. Appl. Soc. Environ. 18, 100310. https://doi.org/10.1016/j.rsase.2020.100310. Sood, V., Gusain, H.S., Gupta, S., Singh, S., Kaur, S., 2020. Evaluation of SCATSAT-1 data for snow cover area mapping over a part of Western Himalayas. Adv. Space Res. 66, 2556e2567. https://doi.org/ 10.1016/j.asr.2020.08.017. Spencer, M.W., Wu, C., Long, D.G., 2000. Improved resolution backscatter measurements with the SeaWinds pencil-beam scatterometer. IEEE Trans. Geosci. Rem. Sens. 38, 89e104.

403

404

Radar Remote Sensing

Tripathy, R., Bhattacharya, B.K., Tahlani, P., Gaur, P., Ray, S.S., 2019. Rice grain yield estimation over some Asian countries using ISRO’s SCATSAT-1 Ku-band scatterometer data. Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci. - ISPRS Arch. 42, 257e262. https://doi.org/10.5194/isprs-archives-XLII3-W6-257-2019. Tripathy, R., Bhattacharya, B.K., 2021, July. Exploring Use of KU-Band Scatterometer Data from SCATSAT-1 for Crop Monitoring in India, a Case Study for Jute Crop. In 2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS. IEEE, pp. 431e434. https://doi.org/10.1109/ IGARSS47720.2021.9554449. Ulaby, F.T., Moore, R.K., Fung, A.K., 1982. Microwave Remote Sensing: Active and Passive. Volume 2Radar Remote Sensing and Surface Scattering and Emission Theory. Wagner, W., Hahn, S., Kidd, R., Melzer, T., Bartalis, Z., Hasenauer, S., Figa-Salda~ na, J., de Rosnay, P., Jann, A., Schneider, S., Komma, J., Kubu, G., Brugger, K., Aubrecht, C., Z€ uger, J., Gangkofner, U., Kienberger, S., Brocca, L., Wang, Y., Blöschl, G., Eitzinger, J., Steinnocher, K., 2013. The ASCAT soil moisture product: a review of its specifications, validation results, and emerging applications. Meteorol. Z. 22, 5e33. https://doi.org/10.1127/0941-2948/2013/0399.

CHAPTER 22

Radar remote sensing of soil moisture: fundamentals, challenges & way-out Hari Shanker Srivastava1 and Parul Patel2 1

Indian Institute of Remote Sensing (IIRS), Indian Space Research Organisation (ISRO), Dehradun, Uttarakhand, India; Space Applications Centre (SAC), ISRO, Ahmedabad, Gujarat, India

2

1. Introduction Soil moisture is the temporary storage of water within the shallow layer of the earth’s upper surface. Compared to the total amount of water available throughout the globe, soil moisture in this layer seems insignificant, but it is this thin layer that controls all the agricultural activities. Soil moisture is not only important for vegetation, it also significantly affects the proportion of rainfall that percolates, runs off, or evaporates from land. Thus, information on soil moisture status is a crucial parameter in crop-yield prediction, irrigation scheduling, hydrological, agricultural, and meteorological applications. For example, Beljaars et al. (1996) and Paegle et al. (1996) consistently show that operational highresolution numerical weather prediction and regional atmospheric model forecasts of the 1993 Upper Midwest United States flooding event was improved with realistic soil moisture initial conditions. Moreover, the measurement of soil moisture aids in predicting plant stress, desertification, and deforestation. Conventional methods for measuring soil moisture are location specific, hence provide point estimates. Since soil moisture is highly dynamic both spatially and temporally, point estimates cannot be extended over a large area with high accuracy. Hence, for estimating spatial distribution of soil moisture over a large agricultural area, remote sensing methods are best suited as they offer a feasible, practical, timely, and cost-effective means. Furthermore, among the various electromagnetic bands, the microwave bands have the highest potential for remotely sensing soil moisture. The key factor behind soil moisture estimation using microwaves is the large difference between the dielectric constant of water (w80) and that of dry soil (3e4) at microwave frequencies. The radar backscattering coefficient (s
) is strongly related to soil moisture due to high dielectric constant of mixture of soil and water (Wang, 1980). This fact has been experimentally verified using many ground-based experiments (Ulaby et al., 1986). With the verification of these theoretical concepts, lot of expectations were generated for getting soil moisture maps on a routine basis after the launch of the first operational synthetic aperture radar (SAR) sensor onboard ERS-1 in 1991. ERS-1 SAR was followed by large number of operational SAR sensors like JERS-1, ERS-2, Radarsat-1, Envisat-1, Radarsat-2, ALOS Palsar, Risat-1, Risat-2B,

Radar Remote Sensing ISBN 978-0-12-823457-0, https://doi.org/10.1016/B978-0-12-823457-0.00022-7

© 2022 Elsevier Inc. All rights reserved.

405

406

Radar Remote Sensing

Risat-2BR1, Risat-1A, Sentinel-1, etc. Operational SAR sensors along with shuttle imaging radars (like SIR-A, SIR-B, SIR-C/X-SAR) and airborne radars (like ISRO DMSAR, ISRO ASAR, DLR E-SAR, F-SAR, AirSAR, etc.) have provided a variety of SAR data for soil moisture retrieval. The last 3 decades have provided the opportunity to exploit multiincidence angle SAR data, multipolarized SAR data, multifrequency SAR data, interferometric SAR (InSAR) data, polarimetric SAR (PolSAR) data, polarimetric interferometric (PolInSAR) data, and hybrid polarimetric SAR data for soil moisture retrieval. However, based upon the global experience and results achieved during last 3 decades, it has been established beyond doubt that radar remote sensing of soil moisture is a challenging task, and without addressing various challenges involved in radar remote sensing of soil moisture, it is difficult to retrieve soil moisture over a large agricultural area with high accuracy. In this chapter, an attempt has been made to highlight the major issues involved in radar address of soil moisture. Moreover, all these issues are also soundly addressed to enable the researchers to use these approaches for radar remote sensing of soil moisture on an operational basis. To enable readers to easily comprehend the chapter, all the major issues or challenges involved in radar remote sensing of soil moisture are addressed under following seven groups: 1. to address the effect of target parameters on SAR sensitivity toward soil moisture 2. to address the effect of sensor parameters on SAR sensitivity toward soil moisture 3. to identify sensitive polarimetric parameters derived from fully and hybrid polarimetric SAR for soil moisture retrieval 4. to address various issues involved in ground truth data collection 5. to address challenges involved in development of a soil moisture retrieval model 6. to address challenges involved in SAR data processing due to the huge data volume 7. to determine interval and scale of a soil moisture map The problem starts from the basic fact that along with its high sensitivity toward soil moisture, SAR is also sensitive to other target parameters of an agricultural field like surface roughness, vegetation/crop cover, and soil texture (Henderson and Lewis, 1998; Ulaby et al., 1978; Ulaby et al., 1979; Dobson and Ulaby, 1981). Similarly, it is also essential to understand the influence of sensor parameters on SAR sensitivity toward soil moisture. A clear understanding of the influence of SAR sensor parameters on SAR sensitivity toward soil moisture enables us to acquire SAR data with optimum sensor parameters that are best suited for soil moisture retrieval (Srivastava et al., 2007). Availability of fully polarimetric and hybrid polarimetric SAR data from a space platform have provided the opportunity to resolve the surface roughness and soil moisture ambiguity and also allows to retrieve soil moisture underneath crop cover (Pal et al., 2016; Patel et al., 2013a; Sharma et al., 2019). However, out of a large number of polarimetric parameters, it is difficult for a researcher to select only those polarimetric parameters that are more sensitive toward soil moisture and surface roughness for bare soil and for soil moisture in the presence of a crop cover. Therefore, in this chapter, a detailed list of

Radar remote sensing of soil moisture: fundamentals, challenges & way-out

polarimetric parameters is provided for bare soil condition as well as under wheat and mustard crop conditions (Patel and Srivastava, 2013a). Planning of a precise and effective field campaign to collect ground truth data for soil moisture retrieval using radar remote sensing involves two important issues. The first issue is to decide the minimum size of sampling fields with reference to the SAR sensor parameters, and the second issue is related to the scientific basis to decide the minimum sample size required, i.e., number of fields from where soil samples need to be collected, for model development and model validation (Patel and Srivastava, 2013b). The next issue is a general difficulty with SAR data processing irrespective of the type of radar remote sensing application. As the SAR backscatter data is 32-bit real data, it requires huge disk space and high-end computing facilities for various radar remote sensing applications. In this chapter, a simple approach is proposed that can significantly reduce the disk space and computing requirements for soil moisture retrieval using SAR data. The chapter also addresses how to determine the soil moisture interval of a soil moisture map and also the impact of scale of a soil moisture map on its accuracy. The following sections are devoted to discuss in detail not only the various issues or challenges involved in radar remote sensing of soil moisture but also addresses how to handle these issues and challenges.

2. Effect of target parameters on SAR sensitivity toward soil moisture To understand the sensitivity of SAR toward soil moisture, firstly we should look at the parameters that affect the SAR return signal from an agricultural land. SAR return signal is affected by the target dielectric and geometrical properties in general (Srivastava, 2010). SAR backscatter from an agricultural terrain is strongly influenced by the moisture content, soil texture, and surface roughness conditions of the soil along with dielectric and geometric properties of the vegetation prevailing in the agricultural fields at the time of SAR data acquisition along with moisture content of the soil; all other target parameters like soil texture, soil surface roughness, and vegetation/crop cover significantly affect the SAR sensitivity toward soil moisture. Therefore it is absolutely necessary to incorporate the effect of these noise parameters in the soil moisture retrieval model. 2.1 Target parameters that affect SAR backscatter from an agricultural land 2.1.1 Soil moisture At microwave frequencies, the dielectric constant of dry soil is around 3 and that of water is around 80. Hence, the dielectric constant for a moist soil, which is a mixture of the two, ranges between 3 and 30 for dry soil and saturated soil respectively. As the dielectric constant of a material increases, the Fresnel reflectivity also increases, resulting in an increased backscatter. Thus SAR backscatter is directly related to moisture content of the target under consideration. That is, a dry field would yield low backscatter, so it would appear darker in tone, and a moist field would appear brighter in tone due to high backscatter.

407

408

Radar Remote Sensing

2.1.2 Soil texture Wet soil is a heterogeneous mixture of soil, water, and air pockets. Water in wet soil is in the form of bound water and free water. The percentage of free water and bound water present in a soil medium largely determines the dielectric constant of a soil medium (Ulaby et al., 1990). Moreover, the percentage of bound water and free water depends upon the surface area of soil particles present in the soil medium. The surface area of soil particles in a soil medium depends upon the particle size and the relative proportions of various-sized particles in a given soil, which is governed by the soil texture. Thus, soil texture controls the proportion of bound water and free water in a soil medium, so the dielectric constant of soil also depends upon soil texture (Srivastava et al., 2006). For example, due to larger size of sand particles, the surface area of particles of sandy soil is much larger than that of clay soil, as particle size of clay is the smallest among all the major constituents of soil, e.g., sand, silt, and clay. Therefore, due to smaller surface area, the proportion of bound water in the case of sandy soil is much lower than clay soil, so for similar soil moisture condition, the dielectric constant of sandy soil is higher than clay soil. It has been observed that at same soil moisture value of say 40% gm/gm, the dielectric constant of sandy soil may be as high as 30 compared with only about 22 for clay soil. This is because at a given soil moisture, the amount of free water in sandy soil is higher than clay soil. Hence, it is evident that without considering the effect of soil texture in the soil moisture retrieval model, it is difficult to retrieve soil moisture with high accuracy. 2.1.3 Soil surface roughness Surface roughness is another important parameter that significantly affects the SAR backscatter from soil. A field that is smooth would appear dark due to low backscatter, as a smooth surface gives rise to specular reflection, whereas a rough field would appear brighter due to higher noncoherent scattering component, resulting in an increased backscatter toward the SAR antenna, as shown in Fig. 22.1. Here, it is interesting to mention that magnitude of surface roughness itself is a function of frequency and incidence angle at which the surface is being observed. It indicates that the characterization of a soil surface into smooth and/or rough class changes with the SAR sensor parameters. According to Fraunhofer criterion, a surface will appear smooth if the surface root mean square (RMS) height (h) satisfies the following condition given by Eq. (22.1). h