Spatial Observation of Giant Panda Habitat: Techniques and Methods 9811987939, 9789811987939

Citation preview

Xinyuan Wang · Jing Zhen · Qingkai Meng

Spatial Observation of Giant Panda Habitat Techniques and Methods

Spatial Observation of Giant Panda Habitat

Xinyuan Wang · Jing Zhen · Qingkai Meng

Spatial Observation of Giant Panda Habitat Techniques and Methods

Xinyuan Wang Chinese Academy of Sciences Aerospace Information Research Institute (AIR) Beijing, China

Jing Zhen Chinese Academy of Sciences Aerospace Information Research Institute (AIR) Beijing, China

Qingkai Meng Chinese Academy of Sciences Institute of Mountain Hazards and Environment (IMHE) Chengdu, China

ISBN 978-981-19-8793-9 ISBN 978-981-19-8794-6 (eBook) https://doi.org/10.1007/978-981-19-8794-6 Jointly published with Science Press The print edition is not for sale in China (Mainland). Customers from China (Mainland) please order the print book from: Science Press. Translation from the Chinese Simplified language edition: “Da Xiong Mao Qi Xi Di Kong Jian Guan Ce Ji Shu Yu Fang Fa” by Xinyuan Wang et al., © Science Press 2020. Published by Science Press. All Rights Reserved. © Science Press 2023 This work is subject to copyright. All rights are reserved by the Publishers, whether the whole or part of the material is concerned, specifically the rights of reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed. The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. The publishers, the authors, and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication. Neither the publishers nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made. The publishers remain neutral with regard to jurisdictional claims in published maps and institutional affiliations. This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd. The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721, Singapore

Foreword

Among the various types of multi-international designated areas, World Heritage sites can be considered the crown jewel. As of July 2019, a total of 1121 sites have been inscribed on the World Heritage List, of which cultural heritage accounts for 869, natural heritage for 213, and cultural and natural heritage for 39. Due to the impact of global changes, the conservation and sustainable development of these precious properties are greatly challenged. Among them, the endangered animal world heritage sites are particularly problematic, especially the habitats of these rare and endangered animals. The imminent or ongoing habitat changes due to climate change and the combined effects of accompanying disasters such as heavy precipitation, drought, and extreme temperatures have led to drastic changes in animal survival conditions, coupled with unreasonable human activities, making these rare and endangered animal habitats. Habitats of these rare and endangered animals face a sudden increase in disaster risk. The Giant Panda, the flagship species of the biodiversity conservation, once a representative of the endangered species, was removed from the “endangered” category in 2015 thanks to the joint conservation efforts of the Chinese government and people from all walks of life around the world. On the one hand, this strongly proves that humans can make a difference in protecting these endangered creatures and can slow down or even change the situation; on the other hand, we must not take it lightly, as temporary removal from “endangered” does not mean that we are free of worries forever, and we still need to even further strengthen the monitoring of Giant Panda habitats and further enhance scientific and effective conservation measures. On the other hand, we can never be complacent, and the fact that we are temporarily out of “endangered” does not mean that we are always worried. Monitoring, conservation, and research of World Heritage sites is a long-term process. Without continuous and long time-series observation, it is impossible to truly understand the impact of global changes on heritage sites. With its advantages of long time-series, macroscopic, timely, holistic, and accurate cognitive objects, space-air-ground integrated Earth observation technology provides a unique perspective and macroscopic view of World Heritage from space. In October 2009, the 35th UNESCO plenary session approved the proposal made by the Chinese Academy of v

vi

Foreword

Sciences in May 2007 to establish an international space technology centre. The aim is to use space technology for work in the areas of natural and cultural heritage, biosphere reserves, climate change, and natural disasters, and to support education for sustainable development. In June 2011, UNESCO and the Chinese government signed an agreement to establish the International Centre on Space Technologies for Natural and Cultural Heritage (HIST), which was officially established in Beijing in July of the same year. HIST is the first space technology-based World Heritage research institute established by UNESCO in the world. The objectives of HIST are to help UNESCO Member States to apply space technology to the study and conservation of cultural and natural heritage, thus enhancing their management, conservation, promotion, and participation in UNESCO’s activities related to world heritage; to strengthen the capacity of Member States to use Earth observation technology to acquire data to support decision-making for sustainable development; and to enable all research findings to become new educational materials, thus supporting the United Nations education for sustainable development activities. thus supporting UN activities on education for sustainable development. Due to the increasing impact of climate change and human activities, research on habitat conservation of rare or endangered species has become more urgent. Currently, the research on ecological health diagnosis methods and techniques of natural heritage sites, especially in the evaluation of biodiversity stability, habitat suitability assessment of important species, evaluation of natural landscape integrity, and the construction of technical standards and platforms for dynamic monitoring of heritage sites, has become a key focus of research, and the methods and models of related research have become a hot spot. In view of the different climatic, hydrological, vegetation, and soil characteristics of the habitats of important species, we study the spatial and temporal characteristics of the elements that characterize the outstanding universal values of natural heritage, combine the characteristics of different observation techniques of spacespace-ground, obtain applicable spatial and temporal resolution remote sensing data, select suitable habitat suitability impact factors of important species with the guidance of dynamic data-driven paradigm, and construct a library of habitat suitability assessment models. The analysis of the factors affecting habitat suitability of important species and their interrelationships, the exploration of their change mechanisms and driving forces, and the search for applicable conservation measures and countermeasures are important elements of habitat-based natural heritage site conservation research, and are urgently needed for in-depth research. However, not much work has been done on this aspect of biological habitats by applying Earth observation techniques, especially in terms of comprehensive research. It is very pleased that the HIST research team has conducted more in-depth research on spatial observation and cognition of natural heritage with the support of the State and Chinese Academy of Sciences. In particular, the successive support from the International Cooperation Special Project of the Ministry of Science and Technology, the Key R&D Project of the Ministry of Science and Technology, and the Strategic Priority Research Program of the Chinese Academy of Sciences has enabled the research team to conduct continuous spatial observation and cognition research on natural heritage, especially the research on the impact of global change on

Foreword

vii

Giant Panda habitat based on space technology, and form some scientific cognition. Under the organization and leadership of researcher Xinyuan Wang, the research team explored the information implied by remote sensing images indoors and the true meaning behind the information in the field despite the high mountains and dangerous roads. As a result, all of us worked together and gathered the strengths of our research to write and assemble a book. The study showed that although the two earthquakes in recent years, Wenchuan and Ya’an, had less impact on the habitat and the Wenchuan earthquake had more impact, the natural recovery of vegetation had an obvious effect in 3–5 years. However, the impact of ongoing climate change requires great attention: in Giant Panda habitat, the vegetation boundaries of the mountainous vertical zone are rising due to warming temperatures, advancing northward in latitude, thus affecting the current suitability of giant panda habitat. In addition, excessive human activities (road construction, reclamation, mining, reservoir construction, etc.) have had different degrees of impact on the habitat suitability of giant pandas, resulting in some habitats even needing to consider relocation for conservation. Looking to the future, we should indeed prepare for the rainy days, especially for the long-term changes in the natural environment and the continuous high-intensity impact of human activities, and pay great attention to the scientific conservation and effective countermeasures for the Giant Panda habitat, and conduct in-depth research. As HIST enters the second phase of construction, I sincerely hope that the HIST research team will continue to practice and innovate in the application of space technology to the conservation and recognition of natural and cultural heritage, and achieve more and greater scientific and technological achievements in the new round of journey. This is the preface.

February 2020

Huadong Guo Academician of the Chinese Academy of Sciences Director of International Centre on Space Technologies for Natural and Cultural Heritage (HIST) under the Auspices of UNESCO Beijing, China

Preface

Space-based Earth observation is a science and technology that detects the Earth from space and conducts scientific analysis and research on the Earth’s targets. By applying this technology, we can help achieve scientific and intelligent decisionmaking and management for the Earth in terms of detection and discovery, monitoring and assessment, and scientific cognition and response strategies. In recent decades, space-based Earth observation has grown with the development of aviation and space platform technologies and optical, microwave, laser, and other payload technologies, as well as the progress of information processing methods. At present, space-based Earth observation systems for various applications have been established worldwide, constituting a multifaceted, three-dimensional observation system for all levels of the land, ocean, and atmosphere, playing an increasingly important role in resource and environmental investigation and monitoring, and promoting economic, social and sustainable development. After nearly three decades of advancement, space technology has also played an increasing role in the monitoring and protection of natural and cultural heritage. For example, in terms of cultural heritage detection, in 1994, Guo Huadong, a Chinese scientist who participated in the Space Shuttle Radar Program, discovered the ancient Great Wall of the Ming and Sui Dynasties buried by dry sand by means of radar remote sensing, which was regarded as one of the “three major discoveries” of the scientific program. In 2013, Chinese scientists identified six ancient town ruins, two settlement sites, and one ancient road section of the Silk Road through optical remote sensing and historical maps. The six ancient towns, two residential areas, and one ancient river in the Guazhou-Shazhou section filled the gap of remote sensing archaeological discoveries of Han and Tang sites in the area west of the ancient city site of Bazhou. In 2018, 10 archaeological remains of the Roman period were discovered in Tunisia at the western end of the Silk Road using spatial archaeological technologies and methods, which revealed the layout of the military defense system and the structure of the agricultural irrigation system on the southern route during the Roman period, which was the first time that Chinese scientists used remote sensing technology to discover archaeological remains outside of China. In terms of using space technology for monitoring and evaluation of natural heritage conservation, ix

x

Preface

Chinese World Natural Heritage sites such as Huangshan and Jiuzhaigou have made notable achievements in digital construction. On a macro level, space information technology is a very effective and objective technical tool in the conservation of world heritage sites, especially for places that are off the beaten track, such as giant panda habitats, with high mountains and overgrown trees and grasses, where human footprints need to make arduous efforts to reach. The giant panda is not only a national treasure of China but also a flagship species for biodiversity conservation globally, and in July 2006, at the 30th UNESCO World Heritage Conference, China’s “Sichuan Giant Panda Habitat” was inscribed on the World Heritage List. The heritage site consists of Baoxing County, where the world’s first giant panda was found, 7 nature reserves including Wolong Nature Reserve and 9 scenic spots including Qingcheng Mountain—Dujiangyan, Jiguan Mountain— Jiulonggou, Xiling Snow Mountain, and Tiantai Mountain in Sichuan Province, covering 12 counties in four cities and states, including Chengdu, Ya’an, Aba, and Ganzi, with an area of 9,245 km2 . Since then, the “Sichuan Giant Panda Habitat” has been included in the world heritage protection. Historically, Giant Panda was widely inhabitated in China, with fossil remains found in the Yellow River, Yangtze River, and Pearl River basins, but with climate change, geological landscape changes, and the continuous expansion of human activities have led to a shrinking range of giant panda habitats. Nowadays, the habitat of giant pandas is limited to six narrow mountain systems in the western region of China at the junction of Sichuan, Shaanxi, and Gansu provinces: Qinling, Minshan, Qionglai, Daxiangling, Xiaoxiangling, and Liangshan. According to the results of the fourth national giant panda survey announced by the State Forestry Administration, the number of wild giant pandas nationwide reached 1864 and the habitat area reached 2.58 million hm2 by the end of 2013. Spatial monitoring and overall conservation of Giant Panda habitats and prediction of future environmental changes will not only help improve the current situation of habitat fragmentation and islandization but also carry out habitat suitability assessment on a fine scale and future habitat adjustment. This will create favorable conditions for the establishment of nature reserves and the release of giant pandas. At the same time, it can also explore research ideas for the monitoring and evaluation of natural heritage sites for rare and endangered species, and how to study the change trends of habitats in the context of future global changes. This book is the culmination of the research team’s joint research, capturing relevant parts of Jing Zhen’s doctoral dissertation. The sections are divided as follows: Outline Development, Preface, Executive Summary, Xinyuan Wang. Chapter 1, Sect. 1.1: Xinyuan Wang and Lei Luo; Sect. 1.2: Lanwei Zhu; Sect. 1.3: Xinyuan Wang and Jing Zhen. Chapter 2, Sect. 2.1: Ruixia Yang, Chuansheng Liu; Sect. 2.2: Fulong Chen, Lanwei Zhu; Sect. 2.3: Cheng Wang, Xiaohuan Xi; Sect. 2.4: Chun Chang, Lei Luo, Xinyuan Wang. Chapter 3, Sect. 3.1: Linhai Jing, Lanwei Zhu; Sects. 3.2, 3.3: Linhai Jing, Yunwei Tang; Sect. 3.4: Jingwei Song, Yunwei Tang, Lei Luo, Xinyuan Wang. Chapter 4, Wanchang Zhang, Ning Nie. Chapter 5, Sect. 5.1: Ruixia Yang, Lei Luo; Sect. 5.2: Fulong Chen; Sect. 5.3: Cheng Wang, Xiaohuan Xi. Section 5.4: Yunwei Tang, Linhai Jing. Chapter 6, Sect. 6.1, 6.4: Qingkai Meng; Sects. 6.2, 6.3: Fulong Chen; Sect. 6.5: Chuansheng Liu, Qingkai Meng. Chapter 7,

Preface

xi

Sect. 7.1: Jing Zhen, Lei Luo; Sects. 7.2, 7.3, 7.4: Jing Zhen. Chapter 8, Sect. 8.1, 8.2, 8.3: Jing Zhen; Sect. 8.4: Xinyuan Wang. The whole book is coordinated by Xinyuan Wang and Jing Zhen. Prof. Siyuan Wang made general revisions to Chaps. 1 and 2. Dr. Qingkai Meng unified the draft of Chaps. 1–6 of the English version. Graduate Students Li Li, Ying Liao, Chun Chang, Jingwei Song, Ning Nie, Shaobo Xia, and Yunqi Zhang participated in the work of data processing and analysis of heritage sites. Chuansheng Liu and Jing Zhen were successively responsible for contacting publication matters. The publication of this book is jointly supported by the Strategic Priority Research Program of the Chinese Academy of Sciences (Grant No. XDA19030500), the National Key Research and Development Program of the Ministry of Science and Technology (2016YFC0503302), and the National International Science and Technology Cooperation Program “Spatial Fine Observation and Cognition of Global Change Impact on World Heritage” (2013DFG21640). This book was prepared under the guidance and preface of Academician Huadong Guo, Associate Professor Haiping Li of Renmin University of China, and Researcher Peng Luo of Chengdu Institute of Biology, Chinese Academy of Sciences have made many constructive modifications and suggestions to the content of this book. Since the spatial observation of natural heritage sites is in a rapid development stage, some principles and technical methods are still being explored, and because of the limited level, omissions and errors are inevitable. In addition, during the preparation of this book, references from the Internet or other sources have been cited, and we would appreciate your correction if there are any references that have been omitted due to the oversight of the preparation.

February 2020

Xinyuan Wang Professor of the Aerospace Information Research Institute, Chinese Academy of Sciences Deputy Director of International Centre on Space Technologies for Natural and Cultural Heritage (HIST) under the Auspices of UNESCO Beijing, China

Contents

1 Natural Heritage Sites and Space Observations . . . . . . . . . . . . . . . . . . . 1.1 World Natural Heritage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1.1 Overview of the World Natural Heritage . . . . . . . . . . . . . . . . . 1.1.2 World Natural Heritage Protection Research . . . . . . . . . . . . . 1.2 Global Change and Space Observation Technology . . . . . . . . . . . . . . 1.2.1 Spatial Observation and Cognition of the Impact of Global Change on World Heritage Sites . . . . . . . . . . . . . . . 1.2.2 Space Observation Technique . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.3 Research Status and Trend of Global Change in Space Observation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3 Spatial Observation of Giant Panda Habitat . . . . . . . . . . . . . . . . . . . . 1.3.1 World Natural Heritage ‘Sichuan Giant Panda Habitat’ . . . . 1.3.2 Spatial Observation of Giant Panda Habitat . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2 Spatial Monitoring Techniques and Methods for Natural Heritage Sites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Optical Remote Sensing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1.1 Characteristics of Optical Remote Sensing . . . . . . . . . . . . . . . 2.1.2 Optical Remote Sensing Monitoring Methods for Natural Heritage Sites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Microwave Remote Sensing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.1 Characteristics of Microwave Remote Sensing . . . . . . . . . . . 2.2.2 Microwave Remote Sensing Applications in Natural Heritage Sites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 LiDAR Remote Sensing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.1 Characteristics of LiDAR Remote Sensing . . . . . . . . . . . . . . . 2.3.2 Application of LiDAR Technology in Natural Heritage Sites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1 1 1 2 4 4 8 18 23 23 24 25 27 27 27 34 39 39 44 55 55 60

xiii

xiv

Contents

2.4 Methods for Extracting Spatial Information on Landscape Change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.1 Ground Survey Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.2 Ecological Modelling Approach . . . . . . . . . . . . . . . . . . . . . . . . 2.4.3 Remote Sensing Interpretation Method . . . . . . . . . . . . . . . . . . 2.4.4 The DEM-NDVI Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5 Quantitative Portrayal of Vegetation Verticality . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Technology and Method of Detailed Information Extraction of Animal Habitat Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1 Multi-source High Resolution Remote Sensing Image Fusion Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.1 Pixel-Level Image Fusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.2 Feature-Level Image Fusion . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.3 Decision-Level Image Fusion . . . . . . . . . . . . . . . . . . . . . . . . . . 3.1.4 A New Method of Remote Sensing Image Fusion—RMI Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 Fine Information Extraction Method of High Resolution Remote Sensing Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.1 Overview of Classification Methods . . . . . . . . . . . . . . . . . . . . 3.2.2 Object-Oriented Classification Methods . . . . . . . . . . . . . . . . . 3.2.3 Fuzzy Logic Classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.4 Analytic Hierarchy Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.5 Decision Tree Classifier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.6 K-Nearest Neighbor Classification Algorithm . . . . . . . . . . . . 3.3 High-Resolution Remote Sensing Data Canopy Delineating Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.1 Traditional Method of Locating the Canopy of an Individual Tree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.2 Object-Oriented Single-Wood Canopy Delineating . . . . . . . . 3.3.3 MSAS Single Wood Canopy Delineating Method . . . . . . . . . 3.3.4 MFS Single Wood Canopy Delineating Method . . . . . . . . . . 3.4 Spatially Refined Observation Methods for Animal Habitats . . . . . . 3.4.1 Overview of Animal Suitable Habitat Research . . . . . . . . . . . 3.4.2 Selection of Habitat Suitability Factors for Animals . . . . . . . 3.4.3 Model Based on Hierarchical Analysis and Expert Weighting Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.4 Improved Method of Integrating Expert System and Neural Network . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.5 Neural Network Method Based on Density Map . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

64 65 65 66 66 72 80 83 83 84 87 87 87 89 90 94 97 98 99 101 102 102 103 104 107 111 112 117 120 123 125 130

Contents

4 Analysis of Changes in Key Environmental Parameters of Land Surface Features in Giant Panda Habitat . . . . . . . . . . . . . . . . . . . . . . . . . 4.1 Basic Geographic Overview of Giant Panda Habitat . . . . . . . . . . . . . 4.2 Analysis of Climate Change in Giant Panda Habitat in Recent Years . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.1 Temperature Change Characteristics . . . . . . . . . . . . . . . . . . . . 4.2.2 Precipitation Variation Characteristics . . . . . . . . . . . . . . . . . . . 4.3 Changes in Water Resources in and Around Giant Panda Habitats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 Spatial Observation and Assessment of Ecological Changes in Giant Panda Habitats . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1 Spatial Observation and Evaluation of Habitat Ecological Landscape Diversity by Optical Remote Sensing . . . . . . . . . . . . . . . . 5.1.1 Study Area Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1.2 Research Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1.3 Analysis of Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 Spatial Observation and Assessment of Habitat Ecological Changes by Microwave Remote Sensing . . . . . . . . . . . . . . . . . . . . . . . 5.2.1 Background of the Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.2 Research Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.3 Research Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.4 Degradation Monitoring . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.5 Validation and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3 Fine Spatial Observation and Assessment of Habitat Ecological Changes by LiDAR Remote Sensing . . . . . . . . . . . . . . . . 5.3.1 Single-Station Ground Point Cloud-Based Trunk Mapping of Bamboo Forests . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.2 Inversion of Average Tree Height of Bamboo Forest Based on Ground-Based Point Cloud . . . . . . . . . . . . . . . . . . . 5.3.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4 Fine Observation of Habitat of Giant Panda Based on High-Resolution Remote Sensing Images . . . . . . . . . . . . . . . . . . . 5.4.1 Study Area Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4.2 Overview of Fine Observation Studies of Giant Panda Habitats from High-Resolution Remote Sensing Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4.3 Data Acquisition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4.4 Research Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4.5 Analysis of Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

xv

135 135 138 140 140 142 145 147 147 148 151 163 166 166 167 170 171 179 182 182 183 197 205 205 206

208 211 212 220 229

xvi

Contents

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat and Assessment Model of Ecological Environment Restoration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.1 Research on Distribution Types and Patterns of Seismic and Geological Disasters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.1.1 Regional Geological Background . . . . . . . . . . . . . . . . . . . . . . . 6.1.2 Earthquake-Induced Secondary Geological Hazards . . . . . . . 6.1.3 Spatial Vulnerability Evaluation of Geological Hazards . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 Post-earthquake Landslide Monitoring and Assessment . . . . . . . . . . 6.2.1 Experimental Area and Data Introduction . . . . . . . . . . . . . . . . 6.2.2 MTInSAR Technical Methodology . . . . . . . . . . . . . . . . . . . . . 6.2.3 Landslide Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3 Microwave Remote Sensing Monitoring Model of Animal Habitat Ecology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3.1 Post-Earthquake Forest Restoration Monitoring and Assessment Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3.2 Forest Restoration Monitoring Results and Interpretation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.4 Habitat Evaluation of Post-earthquake Giant Panda Habitats . . . . . . 6.4.1 Impacts of Different Types of Geological Hazards on the Habitat of Giant Pandas Giant . . . . . . . . . . . . . . . . . . . . 6.4.2 Impacts of Geological Hazards on Giant Panda Habitat . . . . 6.4.3 Evaluation of Habitat Suitability Considering Geological Hazard Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.5 Giant Panda Habitat Recovery After Earthquake . . . . . . . . . . . . . . . . 6.5.1 Natural Ecological Restoration . . . . . . . . . . . . . . . . . . . . . . . . . 6.5.2 Human Intervention . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.5.3 Establishing Ecological Corridors . . . . . . . . . . . . . . . . . . . . . . 6.5.4 Strengthening Post-disaster Reconstruction and Ecological Protection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 Detailed Evaluation of Giant Panda Habitats and Countermeasures Against the Future Impacts of Climate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.1 Impact and Assessment of Climate Change . . . . . . . . . . . . . . . . . . . . . 7.1.1 Impact of Climate Change on Ecosystem . . . . . . . . . . . . . . . . 7.1.2 Impact of Climate Change on Giant Panda Habitat . . . . . . . . 7.1.3 Fine-Scale Evaluation of Climate Change Impact on Giant Panda Habitat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.1.4 Technical Roadmap for Assessing the Impact of Climate Change on Giant Panda Habitat . . . . . . . . . . . . . .

233 234 234 239 254 261 263 265 269 278 278 279 281 282 284 287 294 297 298 298 298 299 299

305 305 305 308 312 315

Contents

7.2 Introduction to Dual Scale Study Area . . . . . . . . . . . . . . . . . . . . . . . . . 7.2.1 Sichuan Giant Panda Habitats . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2.2 Introduction to Ya’an Research Area . . . . . . . . . . . . . . . . . . . . 7.2.3 Landform of the Study Area . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2.4 Soil Vegetation in the Study Area . . . . . . . . . . . . . . . . . . . . . . . 7.2.5 Climatic Characteristics of the Study Area . . . . . . . . . . . . . . . 7.2.6 Interference of Human Activities in the Study Area . . . . . . . 7.3 Research Methods and Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3.1 Maximum Entropy Model (MaxEnt) . . . . . . . . . . . . . . . . . . . . 7.3.2 Maximum Entropy Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3.3 Formula and Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3.4 Model Result Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3.5 Advantages and Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3.6 Evaluation Index System of Giant Panda Habitat Suitability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3.7 Research Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.4 Future Habitat Change Assessment of Giant Pandas Under Climate Change Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.4.1 Habitat Change Assessment of Giant Panda in Sichuan . . . . 7.4.2 Assessment of Changes in Core Areas of Giant Panda Habitat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.4.3 Comparison of Habitat Change Trends at Different Scales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.5 Habitat Assessment of Giant Pandas Under the Combined Impact of Human Disturbance and Climate Change . . . . . . . . . . . . . 7.5.1 Assessment of Giant Panda Habitat Change in Sichuan . . . . 7.5.2 Assessment of Giant Panda Habitat Change in Ya’an Research Area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.5.3 Comparative Study of Habitat Change Trends at Different Scales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 Suggestions on Sustainable Development of Giant Panda Habitat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.1 Conclusion of the Impact of Climate Change on Giant Panda Habitat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.2 Proposed Protection Measures for the Sustainable Development of Giant Panda Habitat . . . . . . . . . . . . . . . . . . . . . . . . . . 8.3 Prospect of Space Observation of Giant Panda Habitat . . . . . . . . . . . 8.4 Recommendations for Achieving the Goals of SDG11.4 for the Protection and Defense of World Heritage Sites . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

xvii

316 317 320 321 328 330 330 334 334 335 336 337 337 339 340 347 348 353 355 357 357 359 360 360 363 363 364 368 373 378

Chapter 1

Natural Heritage Sites and Space Observations

1.1 World Natural Heritage 1.1.1 Overview of the World Natural Heritage World heritage is recognized as a cultural relics and natural landscape with prominent significance and universal value by all mankind. It is a rare and irreplaceable wealth, and is also a “material evidence” to understand the evolution of the earth, the evolution and development of human beings, and the evolution of different ethnic groups and related history. They have the significance and role of natural protection knowledge education, civilization inheritance and spiritual motivation, and can make unique contributions to world peace and security. In November 1972, the United Nations Educational, Scientific and Cultural Organization (UNESCO) passed the Convention Concerning the Protection of the World Cultural and Natural Heritage, stipulating that natural heritage includes the following: from a scientific or conservation perspective, geological and natural geographical structures with outstanding universal value, and survival zones clearly classified as endangered animals and plants. From an aesthetic or scientific perspective, a natural feature consisting of geological and biological structures or groups of such structures of outstanding universal value. From a scientific, protective or natural beauty perspective, natural attractions with prominent universal values or clearly delineated natural areas. There are four criteria for the selection of natural heritages: (i) Outstanding examples that represent important stages in the evolution of the earth; (ii) The composition represents ongoing Important geological processes and Biological evolutionary processes, as well as outstanding examples of the interrelationship between humans and the natural environment; (iii) Unique, rare or wonderful natural phenomena, landforms or areas with rare natural beauty; (iv) Remaining habitats of rare or endangered species of wild fauna and flora species. As of July 2019, the total number of world heritages reached 1121, including 869 world cultural heritages, 213 world natural heritages and 39 world cultural and natural mixed heritages. In terms of quantity, the number of world cultural heritage is far greater than the number of © Science Press 2023 X. Wang et al., Spatial Observation of Giant Panda Habitat, https://doi.org/10.1007/978-981-19-8794-6_1

1

2

1 Natural Heritage Sites and Space Observations

world natural heritage and mixed heritage, but in terms of distribution and protection area, the world natural heritage occupies a large proportion. China has a vast territory, a long history, unique natural environment and rich historical and cultural heritage. In 1985, China acceded to the World Heritage Convention. By July 2019, China had 55 world heritages, including 37 world cultural heritages, 14 world natural heritages and 4 world cultural and natural mixed heritages. The total number of world heritages ranked first in the world and was a veritable world heritage power. China’s 14 world natural heritage sites are Wulingyuan Scenic Area, Huanglong Scenic Area, Jiuzhaigou Scenic Area, Three Parallel Rivers of Yunnan Protected Areas, Giant Panda Habitat in Sichuan, Karst in South China, Sanqingshan National Park, Danxia, Chengjiang Fossil Site in China, Tianshan in Xinjiang, Hubei Shennongjia, Qinghai Hoh Xil, Fanjingshan and Yellow Sea (Bohai) migratory bird habitat in China (Phase I). Four world cultural and natural mixed heritages are Mount Tai, Mount Huangshan, Mount Emei-Leshan Buddha and Mount Wuyi. China’s world natural heritage is mainly distributed in the south of the QinlingHuaihe line, and the characteristics of the distribution around large natural geographical units are very obvious. Sichuan Huanglong, Jiuzhaigou Scenic Area, EmeishanLeshan Giant Buddha Scenic Area, Giant Panda Habitat and Three Parallel Rivers of Yunnan Protected Areas are distributed in the eastern margin of the Qinghai-Tibet Plateau in a strip along the north-south direction. Anhui Huangshan, Jiangxi Sanqingshan and Hunan Wulingyuan are mainly distributed in the hilly areas of Jiangnan. Wuyi Mountain in Fujian is distributed in hilly areas of Fujian and Zhejiang; Karst in South China is distributed on the Yunnan-Guizhou Plateau, across Guangxi, Yunnan and Guizhou Provinces (regions). Fanjing Mountain in Guizhou is also located in the transition slope zone from Yunnan-Guizhou Plateau to Xiangxi Hills. In terms of terrain, it is mainly on the second and third steps of the three major terraces, and each natural geographic unit presents significant differences and geographic gradients in terms of ground elevation, geomorphological combination, and bioclimatic aspects, with obvious edge effects, and non-equilibrium changes such as variation, disturbance, enhancement and weakening make the natural environment in these areas more complex in terms of geographical differentiation, thus becoming an important distribution area for many rare wildlife resources and peculiar natural landscapes.

1.1.2 World Natural Heritage Protection Research The Convention on the Protection of the World Cultural and Natural Heritage sets forth the responsibility of the entire international community to participate in the protection of cultural and natural heritage of outstanding universal value through the provision of collective assistance. To develop scientific and technological research and to adopt appropriate scientific, technological and other measures for the effective protection, preservation and display of cultural and natural heritage.

1.1 World Natural Heritage

3

Space-to-earth observation technology has the characteristics of macroscopic, timely and accurate, which provides a platform for mankind to understand the world heritage from the air. UNESCO list has a wide distribution, large area and many types of heritage, and space-to-ground observation technology plays a key role in this regard. Space-to-earth observation technology plays an important role and significance in monitoring and protecting the world list heritage sites for people to understand the changes of greenhouse gas concentrations, land use and land cover patterns, as well as the expansion of cities and other human settlements. Monitoring, protection and research of world heritage require spatial information technology. In October 2009, the 35th UNESCO assembly approved in May, 2007, Chinese Academy of Sciences, Suggestions, in the territory of China to establish an international space technology center, help it to use the technology to carry out the natural and cultural heritage, biosphere reserve, climate change and natural disasters in the field of education and support sustainable development. In June 2011, UNESCO and the Chinese government signed the establishment of the International Center on Space Technologies for Natural and Cultural Heritage under The Auspices of UNESCO, HIST), HIST was officially established in Beijing in July 2011. HIST is the first space technology-based world heritage research institution established by UNESCO in the world, supported by the Institute of Remote Sensing and Digital Earth of the Chinese Academy of Sciences. HIST is a type II center of UNESCO. Its objective is to help UNESCO member States apply space technology to the research and conservation of cultural and natural heritage, thereby strengthening their management, protection, introduction and promotion of world heritage and participating in UNESCO related activities. Strengthening the capacity of Member States to use earth observation technologies to obtain data to support decision-making on sustainable development; Enable all research findings to become new educational material in support of United Nations education for sustainable development activities. In addition, the spatial dynamic monitoring research on the changes of typical heritage sites is one of the main research directions of HIST. Based on the multi-scale and multisource remote sensing data, the representative world natural and cultural heritage sites are monitored and evaluated on different platforms of “sky-sky-ground”, so as to provide decision-making reference for the management department of heritage sites. Natural heritage is faced with the dual impacts of changes in natural environment (especially catastrophic mutations) and human activities. Natural threats are mainly due to environmental changes caused by floods, diseases, natural extinction of species and global changes. However, most of the threats come from human activities and their effects. At present, illegal hunting and fishing have become the primary threat to natural heritage sites, and the most important and direct human activities of species disappearance. Grazing, agriculture and deforestation affect animal and plant populations and natural landscapes by altering habitats; mining and acquisition change the surface morphology and destroy the ecological balance; alien species invasion directly changed the ecological balance of the original species; the construction of water conservancy facilities has directly changed the water cycle and ecological process in the heritage site, so the threat to the heritage site is fatal. Poor management

4

1 Natural Heritage Sites and Space Observations

cannot control the adverse factors of heritage protection, and further aggravate the destruction of heritage sites. The surrounding development is mainly due to the influence of surrounding urbanization, population growth and industrial development, making the heritage site in the isolated state of peripheral development. Heritage sites are generally more or less affected by tourism development, which is manifested in the influx of tourists and the construction of tourism facilities that destroy the ecological balance and landscape of heritage sites. Fire is due to backward agricultural production methods or natural fire, resulting in ecological imbalance. Roads, airports and engineering pipelines cut the ecological links of the heritage sites. The destruction of heritage sites in troubled areas caused by armed conflicts and military incursions is also a serious concern (Thorsell and Sigaty 1997). Due to the increasing impact of climate change and human activities, especially for rare or endangered animal and plant species habitat protection of natural heritage research, is of great urgency. At present, the research on the methods and techniques of ecological health diagnosis in natural heritage sites, especially in the evaluation of biodiversity stability in natural heritage sites, the evaluation of habitat suitability of important species, the evaluation of the original integrity of natural landscape, and the standard specification and platform construction of dynamic monitoring technology in heritage sites have become the focus of attention. Related research methods and models have also become a research hotspot. Based on the characteristics of different climate, hydrology, vegetation, soil and other characteristics of important species’ habitats, the time-space characteristics of elements representing outstanding universal value (OUV) of natural heritage are studied and combined with the characteristics of different sky-air-ground observation technologies. And choose the right time—spatial resolution remote sensing data, guided by the dynamic data driven model, select the appropriate important species and habitat suitability factor, construct the habitat suitability assessment model base. By analyzing the factors influencing the habitat suitability of important species habitat and its mutual relations, to explore the change reason and driving force, looking for suitable protection measures and countermeasures, is an important part of the habitat class natural heritage sites protection research.

1.2 Global Change and Space Observation Technology 1.2.1 Spatial Observation and Cognition of the Impact of Global Change on World Heritage Sites Global change refers to the global-scale changes in the function of the Earth system caused by natural and human factors, including atmospheric and marine circulation, water cycle, biogeochemical cycle, and changes in resources, land use, urbanization and economic development. Global warming is a prominent symbol of global change (Xu et al. 2013).

1.2 Global Change and Space Observation Technology

5

Global change is increasingly becoming a major issue of concern to all countries in the world, including China. According to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC 2007), the average temperature in the Northern Hemisphere in the second half of the 20th century is likely to be higher than any 50-year period in the past 500 years and may be the highest average temperature in at least 1300 years; atmospheric concentrations of CO2 and CH4 in 2005 far exceeded the natural range of variation for the past 650,000 years. At the same time, a series of major global environmental problems such as environmental pollution, land degradation, species extinction and resource scarcity are becoming more and more serious, threatening human lifestyle and survival (Foley et al. 2005). Global change has the characteristics of large-scale and long-period spatial and temporal evolution. It is a complex system and needs to be studied by various theories and methods. Earth observation technology has a unique advantage in the study of global change due to its characteristics of macroscopic, dynamic, rapid and accurate detection. From the assessment report of the Intergovernmental Panel on Climate Change, the Earth System Science Partnership (ESSP), the World Climate Research Programme (WCRP), From the Climate Change Science Program (CCSP) of the United States, the development and application of space observation technology has been the priority of global change research (Alonso et al. 2008). After more than half a century of development, it has formed a three-dimensional earth observation system covering land, sea and atmosphere. The working band of the sensor covers the band range from visible light, infrared to microwave. The multisystem integrated network observation provides a very effective technical means for the fine observation of global change phenomenon. Among the 50 essential climate variables (ECVs) proposed by Global Climate Observing System (GCOS), 28 depend on satellite observation. Satellite observation has provided rich data for global change research (Xu et al. 2013). Since the 1950s, many observed climate changes have been unprecedented in previous decades to thousands of years. In recent decades, climate change has affected natural and human systems on all continents and in the oceans, indicating that natural and human systems are very sensitive to climate change. Climate change, together with other pressure sources such as habitat loss and fragmentation, will lead to changes in species distribution, population composition, phenological period and ecosystem function. For species with small population size, narrow distribution, weak migration ability, single diet and weak genetic ability, they may face the risk of extinction due to climate change. How to interpret these changes and put forward targeted measures and suggestions is the current concern of domestic and foreign experts and scholars. World heritage is a treasure with outstanding universal value in the process of natural evolution and human development. Its natural, cultural, economic and social value is immeasurable. The protection of world heritage is related to the sustainable development of national culture, ecological environment and social economy. Global change, especially climate warming, has brought great damage to the world heritage sites, and posed serious challenges to heritage protection, ecological system security and global biodiversity. Global climate change affects almost all organisms, and 80%

6

1 Natural Heritage Sites and Space Observations

of species are under the stress of climate warming. The scale and impact of global climate change is already affecting wildlife, from the polar regions to the equatorial tropics, from the oceans to the interior, showing signs of change. At present, a large number of studies and observations have shown that the distribution pattern of species has changed due to climate change, and migration to high latitudes or high altitudes is the adaptation strategy of most species. The reason for this may be that with global warming, wildlife in search of similar cooler environments will move north in latitude as a whole, and the distribution of animals and plants is relatively dense mountains will show the move to a higher place on the elevation direction. On the whole, global warming will make more wildlife at a loss. Climate change will be more pronounced in alpine regions than in low-altitude areas, especially with rising temperatures, changing precipitation patterns and other extreme weather events. Therefore, how to rely on the rapid development of earth observation technology to study and analyze the change trend of species habitat under the background of climate change, whether the existing protected areas can continue to maintain the integrity of species and ecosystem, and how to adjust to mitigate and reduce the impact of climate change on species are the urgent problems to be solved in the field of ecological protection. UNESCO world heritage convention is put forward: world heritage protection aims to protect and sustainable resource utilization, to ensure the integrity and authenticity of the current and future of the world heritage. In order to achieve this goal, all the most effective and appropriate methods must be used to collect, collate, retrieve, maintain and communicate data on the nature and spatial distribution of world heritage sites. Spatial information technology can meet this requirement by describing the status of world heritage sites and obtaining the most up-to-date and accurate data on their nature and location. With the help of spatial information technology, the physical condition, cultural characteristics, social and administrative system environment of world heritage sites can be evaluated to prepare for the formulation of corresponding conservation planning; The effectiveness of resources and their conservation and management measures can be monitored and evaluated. Therefore, spatial information technology has extensive and practical application value for world heritage monitoring and protection. The primary concern of natural heritage monitoring is the prominent universal value of heritage. Natural heritage monitoring is divided into: (i) Systematic monitoring, which includes comprehensive monitoring of the implementation of conservation planning, heritage protection, management, display, publicity, etc.; the key monitoring contents include the monitoring of the solutions and effectiveness of the protection problems; (ii) Reactive monitoring. Reactive monitoring is a special monitoring for the problems in protection management, including the monitoring of abnormal situations or risk factors that threaten heritage protection. Global change, especially climate warming, has brought great damage to world heritage sites and posed serious challenges to heritage protection. By using space-toground observation technology, we will develop algorithms for spatial monitoring, information extraction and multi-source data fusion of ecological and environmental changes of heritage sites based on medium and high resolution remote sensing data combined with ground observation, and the theory and technology system from

1.2 Global Change and Space Observation Technology

7

data to information acquisition and knowledge discovery are gradually established. Focusing on biodiversity in ecologically fragile areas, it is extremely urgent to study the acquisition, analysis and evaluation of time-space fine-change information of biological habitat in the context of global change, which can be explored in the following four aspects. (1) To carry out the research of fine spatial change analysis method of heritage site based on GIS spatial analysis, artificial intelligence and fuzzy mathematics discrimination. Deepen the understanding of the process and mechanism of the impact of global change on world heritage and realize the technological breakthrough of the identification and extraction of detailed spatial change information of specific objects. (2) To develop data-intensive time-varying information analysis techniques for heritage sites. Develop a model for vegetation and ecological information extraction from remote sensing data with medium and low spatial resolution, and use the characteristics of high temporal resolution of remote sensing data with medium and low spatial resolution to study the data fusion algorithm with high spatial resolution data, so as to achieve dynamic information acquisition of ecology and environment of heritage sites with high temporal resolution, as well as breakthroughs in rapid recognition and accurate extraction of habitat change information of endangered species. (3) Study on animal habitat suitability model and suitability mapping. Animal habitat is the place for animal survival and reproduction. The quality of habitat is very important for the continuation and reproduction of species. Habitat suitability is an important indicator to measure the quality of habitats. Habitat suitability mapping can provide the spatial distribution information of wildlife suitable habitats, thus providing decision-making basis for wildlife population management and conservation planning. Combined with ground survey and GPS tracking and positioning technology, and remote sensing on the ecological environment (vegetation-related animal food and living environment) to obtain information, data assimilation to build a more accurate habitat suitability model, habitat suitability mapping fine spatial change. (4) Empirical and comparative study on the impact of global change on heritage sites. Select typical regions to conduct empirical and comparative research on the impact of global change on heritage sites. The research focuses on the process and mechanism of global change and the impact of human activities on heritage sites. This paper studies the natural effects (drought, floods, earthquakes, landslides, warming, etc.), as well as human activities (deforestation, planting, hunting, roads, etc.) to study the influence of rare creatures, revealing the different regional ecosystem safety pattern and the evolution trend of space and time, from different scales to know its landscape pattern change impact on biodiversity, ecosystem stability and the feedback function. The ecological status assessment method of rare and endangered animal habitat and the spatial correlation between the spatial pattern of various vegetation types and environmental factors such as vegetation ecology will be discussed.

8

1 Natural Heritage Sites and Space Observations

1.2.2 Space Observation Technique Before satellite observation is widely used, scientists mainly seek a global perspective through ground observation, which requires international cooperation and large-scale field investigation. Placing data points together requires interpolation and extrapolation to fill the gaps in data, especially those far away. In addition, large-scale sampling requires extensive logistics support and pre-planning to reduce frequent duplication of work. Since in the era before the emergence of artificial satellites, the change rate of many research parameters is faster than that of global map drawing, it is impossible to observe the complete dynamic characteristics of the Earth system. Even if the individual surface observation can be combined into a global picture, a lot of work needs to be done due to the coverage rate and density of the network and the lack of longitudinal resolution. Other geophysical and biological phenomena are less frequently sampled and are often used as part of a dynamic “snapshot” of Earth processes interacting. The advent of artificial satellites triggered a transformation in Earth science, providing the world with the first complete global record of biological, physical, and chemical parameters (e.g., cloudiness, wind fields, ice cover). The synchronicity of observations with larger spatial coverage provided by artificial satellites is not available from ground-based measurements; the time series data provided by artificial satellites reveal large scale processes and features that have not been discovered by other methods. Thus, artificial satellites give scientists quantifiable global images and maps with a frequency and coverage unmatched by any ground-based observation technique. At present, the space observation techniques commonly used in the space monitoring and evaluation of natural heritage sites include optical remote sensing technology, microwave remote sensing technology and LiDAR remote sensing technology. 1. Optical remote sensing technology Optical Remote Sensing is a passive remote sensing technology that uses electromagnetic wave as the transmission medium and the working band of sensor is limited to the range of visible light (0.38–0.76 μm). It is the most commonly used working band in traditional aerial photogrammetry. Since the color range of the photosensitive film is exactly in this wavelength range, optical remote sensing can obtain black-and-white panchromatic or color images with high ground resolution, thereby improving the performance of image interpretation and mapping. Optical remote sensing is mainly limited by the solar illumination conditions. Due to the successive emergence of infrared photography and multiband remote sensing, visible light remote sensing has extended the working band to the near-infrared region (about 0.9 μm). The imaging mode has also developed from a single photographic imaging to a variety of ways including black-and-white photography, infrared photography, color photography, color infrared photography, multi-band photography and multi-band scanning, which greatly improves the detection ability of optical remote sensing. Low-medium resolution optical remote sensing

1.2 Global Change and Space Observation Technology

9

technology mainly refers to the scientific research based on low-medium resolution (less than 5 m) optical remote sensing images. Medium and low resolution remote sensing images have the advantages of wide observation range, short observation period and strong data timeliness (Li 2008). After the 21st century, with the rapid development of space technology, the development of remote sensing observation system has also appeared a new climax. Countries all over the world compete to research, develop and launch high-resolution remote sensing satellites. Since 1998, high-resolution remote sensing satellites have been on the stage of remote sensing application, and their resolution has been greatly improved compared with the previous remote sensing satellites. These highresolution remote sensing satellites mainly include EarlyBird (3 m resolution) and QuickBird (1 m resolution) of Earth-Watch Company. SpaceImaging’s IKONOS (1 m resolution); Orbital Sciences’ OrbVIEW-1 (1, 2, 4 m resolution). In particular, the successful launch of the IKONOS satellite, the world’s first to provide high-resolution satellite imagery at a resolution of more than 1 m, marks the establishment of a new, faster and more economical way to obtain the latest basic geographical information. Over the next decade, the level of satellite resolution developed rapidly. Spatial resolution has reached sub-meter level, even comparable to low-altitude aerial photographs, which provides favorable conditions for the extraction of fine information. For example, the WorldView-1 satellite launched in 2007 can shoot 0.5 m resolution images of up to 500,000 km2 every day. The WorldView-2 satellite launched in 2009 can provide 0.5 m panchromatic images and 1.8 m resolution multispectral images. The Pleiades-1 satellite launched by the SPOT family in 2011 has a resolution of 0.5 m and a width of 20 km × 20 km. In the research of high-resolution remote sensing satellites, China has also made remarkable development. In order to ensure the autonomy of China’s space information resources and promote the development of space information industry, China has incorporated the National Science and Technology Major Project of High Resolution Earth Observation System into the National Medium and Long Term Science and Technology Development Plan (2006–2020), becoming one of the 16 national major science and technology projects. Since the launch of the high-resolution special project in 2010, the first satellite GF-1 was successfully launched on April 26, 2013. GF-1 is equipped with two 2 m resolution panchromatic/8 m resolution multispectral cameras and four 16 m resolution multispectral wide-swath cameras. On August 19, 2014, GF-2 was successfully launched, which can query the resolution better than 1 m in the remote sensing market platform. At the same time, it has the characteristics of high radiation accuracy, high positioning accuracy and fast attitude maneuver ability, which marks that China’s remote sensing satellite has entered the sub-meter “high-resolution era”. Table 1.1 lists the current mainstream high-resolution satellite remote sensing and related technical parameters. It can be seen that compared with traditional medium and low resolution remote sensing images, high-resolution satellite images have many advantages such as more obvious geometric structure of ground objects, clearer position and related layout of ground objects, finer texture and size information, and from two-dimensional

France

India

Japan

Israel

SPOT5

Cartosat-1 (IRS-P5)

ALOS

EROS-B

2006.4

2006.1

2005.5

2002.5

2001.10

2000.12

Israel

U.S

EROS-A1

1999.9

U.S

IKONOS

QuickBird

Launch time

Country

Satellite

0.7

2.5

2.5

2.5/5

0.61

1.8

0.82

PAN/m

N/A

10

N/A

10

2.44

N/A

3.2

Multispectral/m

Yes

Yes

Yes

N/A

N/A

Yes

Yes

Stereoscopic acquisition capability

7

70

26

60

16.8

14

11.3

Width/km

Positioning accuracy (CE90)

–

Panchromatic

–

Panchromatic + 4 1 m (Control multispectral points bands available)

Panchromatic

–

Collection stopped

–

400

panchromatic + 4 30 m multispectral bands

– Collection stopped

–

Collection stopped

Acquisition capability/(ten thousand km2 /d)

panchromatic + 4 14 m multispectral bands

Panchromatic

Panchromatic + 4 9 m multispectral bands

Spectral features

Table 1.1 Parameters of mainstream high-resolution satellite remote sensing technology in China and abroad

(continued)

520

691

618

832

482

520

681

Orbit height/km

10 1 Natural Heritage Sites and Space Observations

U.S

U.S

France

China

GeoEye-1

WorldView-2

Pléiades (1A,1B)

ZY-1 02C

2011.12

2011.12

2009.10

2008.9

2008.8

2007.9

U.S

Germany

WorldView-1

RapidEye

2006.7

Korea

KOMPSAT-2

Launch time

Country

Satellite

Table 1.1 (continued)

2.36 /5

0.5

0.46

0.41

N/A

0.5

1

PAN/m

10

2

1.85

1.65

5

Inapplicability

4

Multispectral/m

N/A

Yes

Yes

Yes

Yes

Yes

Stereoscopic acquisition capability

54

20

16.4

15.2

77

17.7

15

Width/km

Positioning accuracy (CE90)

4m

130

100

100

140

Panchromatic + 8 3.5 m multispectral bands Panchromatic + 4 3 m multispectral bands Panchromatic + 3 – multispectral bands

400 35

–

170

Acquisition capability/(ten thousand km2 /d)

panchromatic + 4 3 m multispectral bands

5 multispectral bands

Panchromatic

Panchromatic + 4 4430 >4400

−0.45–0.7 −0.25 to 0.7

Altitude/m

NDVI

−0.25 to 0

−0.35 to −0.15

−0.3 to 0.05

−0.5 to 0

NDVI

2.5 Quantitative Portrayal of Vegetation Verticality 75

76

2 Spatial Monitoring Techniques and Methods for Natural Heritage Sites

Fig. 2.11 Distribution of sample points

sample points is shown in Fig. 2.11. The variation of NDVI with elevation in 2007 at the sample points is shown in Fig. 2.12. The results of projecting the sample points onto the DEM-NDVI scatter plot are shown in Fig. 2.13. A comparison of Figs. 2.11 and 2.13 gives the vegetation types at each stage in the DEM-NDVI scatter plot, as shown in Table 2.5. The vegetation in the first stage is dominated by coniferous and broad-leaved forests, with a small amount of meadow scrub also present. This is due to the presence of landslides in the area, which caused the destruction of the depressional forest and the secondary vegetation of meadows and scrubs to grow preferentially on the landslides. The second stage is the distribution area of the interlocking forest line. The coniferous forests are characterized by a gradual decrease in the area of distribution with increasing altitude. Within the gradual zone, coniferous forests grow interspersed with alpine scrub and alpine meadows. From the beginning of the gradual zone, the NDVI starts to decline. The main vegetation in the later stages is alpine scrub and alpine meadows, while their area decreases slowly with increasing altitude. The alpine vegetation zone should start at a low altitude in this area. For the southeast slope, the starting point is 3255 m above sea level, and for the northwest slope, 3193 m above sea level.

2.5 Quantitative Portrayal of Vegetation Verticality

77

Fig. 2.12 Variation of NDVI with elevation at sample sites

The third stage is the area of alpine scrub, alpine meadows, and their dramatic decline in the area. The fourth stage is bare rock and bare soil. The upper limit of the distribution of alpine meadows is the boundary between the third stage and the fourth stage. The upper limit of the distribution of alpine meadows and alpine scrub on the southeast slope is about 4415 m above sea level, and on the northwest slope is about 4473 m. From Table 2.5, it can be seen that there is a large difference between the second and third stages of the boundary between the south-eastern and north-western slopes, and by analyzing Figs. 2.11 and 2.12, it can be found that this difference is not caused by coniferous forests. The main reason for this difference is the difference in vegetation growth conditions or the difference in vegetation distribution types between the southeast and northwest slopes, and the reason for this difference needs to be studied in depth through ground survey data. 5. DEM-NDVI sliding mean correlation analysis The sliding mean method was used to find the mean and standard deviation of NDVI at each elevation. Correlation analysis of the sliding averages was carried out to illustrate the reliability and stability of the vertical zonation of alpine vegetation using DEM-NDVI. Taking Guan Gou as an example, the sliding mean and standard deviation of NDVI on its southeast and northwest slopes are shown in Fig. 2.14.

78

2 Spatial Monitoring Techniques and Methods for Natural Heritage Sites

Fig. 2.13 Results of overlaying the sample sites with the 2007 DEM-NDVI scatter plot of the southeast slope

Table 2.5 Vegetation types by stage of DEM-NDVI Stage Stage 1

Elevation range

Vegetation type

SE: 4473 m

Theoretically, the NDVI sliding mean curve is a smooth “knife edge” curve. However, vegetation distribution is not only influenced by elevation but also by local geographical factors such as slope orientation and soil. This results in the NDVI mean curve in Fig. 2.14 showing some fluctuations. The standard deviation of the NDVI mean represents the degree of NDVI concentration within the window. The smaller the standard deviation, the more concentrated the NDVI distribution is and the more representative the NDVI value is of the average NDVI value at that elevation, and vice versa.

2.5 Quantitative Portrayal of Vegetation Verticality

79

Fig. 2.14 NDVI segmented mean and variance

From Fig. 2.14a, c, it can be seen that the correlation coefficients of NDVI in 1994 and 2007 for the same region are high, both reaching above 0.99. This indicates that for the same region, there is consistency in NDVI variation with DEM between different years. This illustrates that the DEM-NDVI scatter maps generated for alpine areas where vertical vegetation zonation exists all show a segmented structure. Therefore, the use of DEM-NDVI scatter maps to quantitatively characterize the vertical zonation of alpine vegetation can be extended to other alpine areas with vertical zonation characteristics. The standard deviation of NDVI in 2007 on the south-eastern slope is closer to that of 1994 and 2007 on the north-western slope, and the standard deviation of NDVI in 1994 on the south-eastern slope being significantly higher. This illustrates that the NDVI distribution on the south-eastern slope in 1994 was more dispersed than in the other three cases. Nevertheless, the correlation coefficient between the 1994 DEM-NDVI sliding mean curve for the southeastern slope and the 2007 DEM-NEVI sliding mean curve for the southeastern slope reached 0.9974. This indicates that the degree of NDVI concentration has little influence on the structural analysis of the DEM-NDVI scatter distribution.

80

2 Spatial Monitoring Techniques and Methods for Natural Heritage Sites

References Askne J, Santoro M (2003) Multitemporal repeat-pass SAR interferometry of boreal forests. IEEE Trans Geosci Remote Sens 41(7):1540–1550 Attema EPW, Ulaby FT (2009) Vegetation modeled as a water cloud. Process Phys Geogr 33(4):457– 473 Chen F, Lin H, Yeung K et al (2010) Detection of slope instability in Hong Kong based on multibaseline differential SAR interferometry using ALOS PALSAR data. Data GIsci Remote Sens 47(2):208–220 Chen F, Lin H, Cheng S (2013) Principles, methods and applications of satellite-based radar interferometry and time series analysis. Science Press, Beijing (in Chinese) Chen S, Tong Q, Guo H (1998) Research on the mechanics of remote sensing information. Science Press, Beijing (in Chinese) Cloude SR, Papathanassiou KP (2003) Three-stage inversion process for polarimetric SAR interferometry. IEEE Proc Radar Sonar Navig 150(3):125–134 Devi N, Hagedorn F, Moiseev P et al (2008) Expanding forests and changing growth forms of Siberian larch at the Polar Urals treeline during the 20th century. Glob Change Biol 14:1581–1591 Dirnböck T, Dullinger S, Grabherr G (2003) A regional impact assessment of climate and landuse change on alpine vegetation. J Biogeogr 30(3):401–417 Dong L, Wu B, Tang S (2011) Combined inversion of forest above-ground biomass by LiDAR GLAS and ETM. J Peking Univ (Nat Sci Edn) 47(4):703–710 (in Chinese) Ferretti A, Fumagalli A, Novali F et al (2011) A new algorithm for processing interferometric data-stacks: squeeSAR. IEEE Trans Geosci Remote Sens 49(9):3460–3470 Fornaro G, Guarnieri AM, Pauciullo A et al (2005) Joint multi-baseline SAR interferometry. EURASIP J Appl Signal Process 20:3194–3205 Fu Y, Han Y, Zhang Y et al (2014) Advances in treeline response to climate change. J Ecol 33(3):799– 805 (in Chinese) Fung AK, Li Z, Chen KS (1992) Backscattering from a randomly rough dielectric surface. IEEE Trans Geosci Remote Sens 30(2):356–369 Gabriel AK, Goldstein RM (1988) Crossed orbit interferometry: theory and experimental results from SIR-B. Int J Remote Sens 9(5):857–872 Gomez-Chova L, Fernandez-Prieto D, Calpe J et al (2006) Urban monitoring using multi-temporal SAR and multi-spectral data. Pattern Recogn Lett 27:234–243 Guo H et al, Radar theory and application to earth observation. Science Press, Beijing (in Chinese) Guo H, Wang W, Zhu B (2004) Simulation and validation of multimodal sensor radiation modal data in SZ-4 spacecraft. National Conference on the Application of Remote Sensing Technology to Land Resources and Environment, Wuhan (in Chinese) Hiemstra CA, Liston GE, Reiners WA (2006) Observing, modelling, and validating snow redistribution by wind in a Wyoming upper treeline landscape. Ecol Model 197(1–2):35–51 Holmgren J, Persson A, Soerman U (2008) Species identification of individual trees by combining high resolution LiDAR data with multispectral images. Int J Remote Sens 29(5):1537–1552 Holtmeier FK, Broll G (2007) Treeline advance-driving processes and asverse factors. Landsc Online 1:1–33 Holtmeier FK (1994) Ecological aspects of climatically caused timberline fluctuations. In: Beniston M (ed) Mountain environment in changing climates. Routledge, London, New York, pp 220–233 Hooper A (2008) A multi-temporal InSAR method incorporating both persistent scatterer and small baseline approaches. Geophys Res Lett 35:L16302 Huang HB, Li Z (2011) Automated methods for measuring DBH and tree heights with a commercial scanning LiDAR. Sensing 77(3):219–227 Jiang L, Liu X, Feng Z (2014) Remote sensing identification of alpine forest lines and its spatial pattern analysis in the Sanjiang River merging area. Resour Sci 36(2):259–266 (in Chinese) Klikoff LG (1965) Photosynthetic response to temperature and moisture stress of three timberline meadow species. Ecology 46(4):516–517

References

81

Krieger G, Papathanassiou KP, Cloude SR (2005) Spaceborne polarimetric SAR interferometry: performance analysis and mission concepts. J Adv Signal Process 20:3272–3292 Le Toan L, Beaudoin A, Riom J et al (1992) Relating forest biomass to SAR data. IEEE Trans Geosci Remote Sens 30(2):403–411 Lefsky MA, Cohen WB, Parker GG et al (2002) Lidar remote sensing for ecosystem studies. Bioscience 52(1):19–30 Liu D, Fan W, Li M (2012) Estimation of stand parameters and biomass using small-spot Lidar. J Northeast Forest Univ 40(1):39–43 (in Chinese) Liu G et al (2012) Interference theory and methods for permanent scatterer radar. Science Press, Beijing (in Chinese) Lucas R, Armston J, Fairfax R et al (2010) An evaluation of the ALOS PALSAR L-band backscatterabove ground biomass relationship Queensland, Australia: impacts of surface moisture condition and vegetation structure. IEEE J Sel Top Appl Earth Observ Remote Sens 3(4):576–593 Luo SZ, Wang C, Pan FF et al (2014) Estimation of wetland vegetation height and leaf area index using airborne laser scanning data. Ecol Ind 48:550–559 Luo SZ, Wang C, Zhang GB et al (2013) Forest leaf area index (LAI) estimation using airborne discrete return LiDAR data. Chinese J Geophys 56(3):233–242 Luo S, Wang C, Xi X et al (2015) Joint inversion of forest leaf area index by satellite-based LIDAR GLAS and TM optical remote sensing. J Infrared Millim Waves 34(2):243–249 (in Chinese) Marlow GP (1998) Palaecology of postglacial treeline shifts in th enorthern Cascade Mountains, Canada. Palaeoecology 141:123–138 Massonnet D, Feigl K (1998) Radar interferometry and its application to changes in the earth’s surface. Rev Geophys 36(4):441–500 McDonald KC, Dobson MC, Ulaby FT (1990) Using mimics to model L-band multiangle and multitemporal backscatter from a walnut orchard. IEEE Trans Geosci Remote Sens 28(4):477–491 Oh Y, Sarabandi K, Ulaby FT (1992) An empirical model and an inversion technique for radar scattering from bare soil surfaces. IEEE Trans Geosci Remote Sens 30(2):370–382 Peake WH, Oliver TL (1971) The response of terrestrial surfaces at microwave frequencies. Columbus Electroscience Lab, Ohia State University, Columbus Price EJ, Sandwell DT (1998) Small scale deformation associated with the 1992 Landers, California Earthquake mapped by synthetic aperture radar. J Geophys Res 103(B11):27001–27016 Qin HM, Wang C, Xi XH et al (2017) Simulating the effects of the airborne Lidar scanning angle, flying altitude, and pulse density for forest foliage. Appl Sci 7(712):1–18 Rauste Y (2005) Multi-temporal jers sar data in boreal forest biomass mapping. Remote Sens Environ 97(2):263–275 Santoro M, Beer C, Cartus O et al (2013) Retrieval of growing stock volume in boreal forest using hyper-temporal series of Envisat ASAR ScanSAR. Remote Sens Environ 115(2):490–507 Shang R, Xi X, Wang C et al (2015) Extraction of single wood structure parameters based on ground laser scanning data. Mapp Sci 40(9):78–81 (in Chinese) Shao Y, Guo H, Fan X et al (2001) Time domain scattering analysis of rice and its application. J Remote Sens 5(5):340–345 (in Chinese) Shi J, Wang J, Hsu A et al (1997) Estimation of bare surface soil moisture and surface roughness parameter using L-band SAR image data. IEEE Trans Geosci Remote Sens 35(5):1254–1266 Shu N (2000) Principles of microwave remote sensing. Wuhan University Press, Wuhan (in Chinese) Ulaby FT, Moore RK, Fung AK (1981) Microwave remote sensing: microwave remote sensing fundamentals and radiometry. Addison Wesley Publishing Company, Advanced Book Program World Science Division, New Jersey Ulaby FT, Van Deventer TE, Haddock TF et al (1988) Millimeter wave bistatic scattering from ground and vegetation targets. IEEE Trans Geosci Remote Sens 26(3):229–243 Wang C, Glenn N (2009) Integrating LiDAR intensity and elevation data for terrain characterization in a forested area. IEEE Geosci Remote Sens Lett 6(3):463–466 Wang C, Xi X, Luo S et al (2015) Data processing and applications of satellite-based Lidar. Science Press, Beijing (in Chinese)

82

2 Spatial Monitoring Techniques and Methods for Natural Heritage Sites

Wang X, Zhou X, Sun Z (2005) Progress on the relationship between alpine forest line and climate change. J Ecol 24(3):301–305 (in Chinese) Wang X, Liu H (2011) Population-soil relationships of the forestline Yuehua birch on the northern slopes of Changbai Mountain. Geogr Stud 30(3):531–5391 (in Chinese) Xi X, Luo S, Wang F et al (2012) Review of the status and development of terrestrial 3D laser scanning systems. Geosp Inf 10(6):13–15 (in Chinese) Yang T, Wang C, Li GC et al (2014) Forest canopy height mapping over China using GLAS and MODIS data. Sci China Earth Sci 57(1):1–11 Zebker HA, Goldstein RM (1986) Topographic mapping from interferometric synthetic aperture radar observations. J Geophys Res Solid Earth 91(B5):4993–4999 Zhang Y, Chen J (2011) Progress and prospects of domestic research on forest line response to climate change. World Forest Res 24(6):18–22 (in Chinese) Zhang Y (2012) Spatial and temporal variation of forest line height in the mountains of Guangdong over the past 50 years. Guangzhou University, Guangzhou (in Chinese) Zhou C, Ouyang, Ma T (2009) Advances in geographic grid models. Adv Geogr Sci 28(5):657–662 (in Chinese) Zhu F, An S, Guan B et al (2007) Ecological staggered zones and their research progress. J Ecol 27(7):3032–3042 (in Chinese)

Chapter 3

Technology and Method of Detailed Information Extraction of Animal Habitat Elements

3.1 Multi-source High Resolution Remote Sensing Image Fusion Technology Remote sensing image fusion refers to the process in which remote sensing image data of different types of sensors are complementary to each other in certain ways to generate new images. It can combine many advantages of multi-source remote sensing data in one, and achieve the purpose of enriching advantages and reducing defects. Users can find a lot of unconspicuous and valuable information in the original image from the fused image to achieve the effect that one plus one can be greater than two. Image fusion can improve the accuracy of subsequent image segmentation and image classification, which is of great help to high-level image analysis and image understanding. Therefore, image fusion is widely used in military target identification and image processing. The synthetic image obtained by using remote sensing image fusion algorithm not only maintains the spectral information of multispectral (MS) image, but also enhances the spatial details of MS image, thus enhancing the analysis and processing ability of remote sensing image, which is conducive to the extraction and classification of targets. Remote sensing image fusion can achieve multiple purposes (Li 2013): (i) It enhances the spatial details of MS images while maintaining their spectral information, and effectively increases their spatial information; (ii) The features that are not obvious in a single image are highlighted; (iii) It improves the accuracy of subsequent image segmentation and classification; (iv) It can be used for change detection, that is, the change information of the region can be obtained by the fusion of remote sensing images of the same region in different phases; data complementarity is realized: the fusion processing of remote sensing data obtained by different types of sensors can make up for the lack of spatial information and spectral information in a single image, overcome the limitations of a single sensor, and achieve the maximum of data utilization efficiency. Remote sensing image fusion can be divided into pixel-level image fusion, featurelevel image fusion and decision-level image fusion according to the different degree © Science Press 2023 X. Wang et al., Spatial Observation of Giant Panda Habitat, https://doi.org/10.1007/978-981-19-8794-6_3

83

84

3 Technology and Method of Detailed Information Extraction of Animal …

of abstraction and application level of information in the fusion process (Zhao and Wang 2013). Pixel-level fusion is performed directly on the original data layer. This level has the highest fusion accuracy and can provide richer, more accurate and more reliable details which other level fusion processing does not have. Pixel-level fusion is conducive to further analysis, processing and understanding of images. It is the basis of feature-level image fusion and decision-level image fusion.

3.1.1 Pixel-Level Image Fusion Pixel-level image fusion is a comprehensive data analysis based on visible light, infrared and SAR data. Existing pixel-level image fusion methods can be divided into three categories (Zhang 2010): component substitution method, image modulation method and multi-resolution method. (1) Component substitution method: First, the MS images with low spatial resolution were spatially transformed and more components were obtained. Then panchromatic (panchromatic, PAN) band images were used to replace one component. Finally, the fusion image was obtained by spatial inverse transformation.Typical algorithms of component substitution include: IHS (intensityhue-saturation) fusion method, PCA (principal component analysis, Principal Component analysis) fusion method (Chavez et al. 1991) and Gram-Schmidt spectral sharpening method (Laben and Brower 1998). This kind of fusion algorithm is fast and easy to implement, but there is obvious spectral distortion in the fusion result, so it is mainly used to assist in the tasks of original image interpretation, ground object classification, ground object recognition and information extraction. In the IHS fusion method, firstly, the MS images of three bands are transformed from RGB space to color space by using the IHS space transformation to obtain three components of image: intensity (I), hue (H) and saturation (S). Then, the intensity component I in color space is replaced by PAN image with high spatial resolution. Finally, the IHS inverse transformation of intensity, hue and saturation was carried out to obtain the spatial enhanced fusion image. This fusion image not only has the high spatial resolution of PAN image, but also maintains the spectral characteristics of MS image, which is beneficial to improve image interpretation, classification and mapping accuracy, and is suitable for visual interpretation and resource survey. Since PAN and MS images cannot be completely correlated, there will be obvious spectral distortion in the fused images, and the IHS fusion method can only fuse MS and PAN images in three bands at the same time. PCA fusion method uses PCA to fuse original MS images. Principal component analysis is a multi-dimensional (multi-band) optimal orthogonal linear transformation of minimum mean square error based on statistics. There is often a high correlation between different bands of remote sensing images. PCA can

3.1 Multi-source High Resolution Remote Sensing Image Fusion Technology

85

remove this correlation and concentrate the useful information in the image into the new principal component, so as to reduce the amount of data and enhance the image information. PCA fusion algorithm mainly includes the following steps: calculate the eigenvalues and eigenvectors of images according to the correlation matrix between MS bands, and sort the eigenvectors according to the size of the eigenvalues from large to small, so as to obtain the transformation matrix. According to the transformation matrix, the MS image is transformed into each principal component of the image. The PAN image was stretched to have the same mean and variance as the first principal component, and then the first principal component was replaced by the PAN image. All principal components are inverted into the fused MS image, PCA fusion method can effectively improve the clarity of MS images, enhance the interpretability and measurability of images, and can process multiple wavebands simultaneously. (2) Image modulation method mainly includes image fusion algorithm based on band ratio. Representative algorithms include: Brovey method, Pradines method (Pradines 1986), SVR (support vector regression, smoothing filter-based intensity modulation (SFIM) and support vector regession (Munechika et al. 1993), HR (high resolution) method (Jing and Cheng 2009). For MS and PAN images from the same sensor, these algorithms can effectively reduce the spectral distortion of the fused image, but when the two images are from different sensors or their spatial resolution is very different, the application effect of this kind of algorithm will be limited. The low resolution PAN images involved in this kind of algorithm are estimated by band arithmetic operation, physical model simulation, or statistical analysis and other methods, rather than the real low resolution PAN images. (3) The multi-resolution method mainly uses the multi-resolution analysis technology to decompose the original image at a multi-scale and deduce the spatial details of PAN images that MS images should have at a high resolution, and then adds these spatial details to MS images to enhance the spatial details of MS images. Multi-resolution analysis technology not only has the ability of local focusing in spatial domain and frequency domain, but also provides sensitive contrast information of human vision. The fusion process of MS and PAN images based on the fusion algorithm is very similar to the recognition process from coarse to fine in human vision system and computer vision. Typical multi-resolution fusion algorithms include atrous wavelet fusion method (Teggi et al. 2003), AWLP (additive wavelet luminance proportional) method (Otazu et al 2005), Laplacian pyramid method (Alparone and Aiazzi 2006), and non-sampled contourlet transform (Non-sampled contourlet transform, NSCT) (Yang and Jiao 2008). Wavelet transform is known as “mathematical microscope” by using gradually fine sampling steps in time domain (or space domain) for high-frequency components, which can “focus” to any details of the object. Wavelet transform can decompose a signal into independent parts in space domain (or time domain) without losing the information contained in the original signal, and can find the orthogonal basis to achieve signal decomposition without redundancy.

86

3 Technology and Method of Detailed Information Extraction of Animal …

As a multi-resolution image analysis tool, wavelet transform has been widely used in the field of remote sensing image fusion, mainly due to its favourable timefrequency analysis characteristics. Wavelet transform is known as “mathematical microscope” by using gradually fine sampling steps in time domain (or space domain) for high-frequency components, which can “focus” to any details of the object. Traditional wavelet image fusion methods mainly include image fusion algorithms based on discrete Wavelet transformation (discrete wavelet transformation, DWT) and Atrous wavelet transform. The fusion rule of this method is to deal with low frequency and high frequency wavelet coefficients respectively. However, this rule ignores the connection between the latter two, and it has a large amount of computation and slow processing speed. Therefore, it has great limitations when applied directly to remote sensing images, especially high-resolution remote sensing images. To deal with this limitation, a variety of new image fusion algorithms based on wavelet transform keep emerging, such as the fast multi-focus image fusion method based on lifting wavelet transform (Shao et al. 2014) and the use of image contrast to link the processing of high frequency coefficient and low frequency coefficient, and image fusion is carried out with contrast as the criterion of metric coefficient selection (Wang 2014). The former method first uses the lifting wavelet algorithm to decompose the original image into four sub-bands, and then uses lifting wavelet inverse transform to obtain high-frequency detail images of sub-bands in various directions, and calculates the non-uniform weighted area energy of the resulting high-frequency detail images, and finally get the final composite image according to the energy-based image fusion rules. This method is much faster than the original wavelet fusion method. The multi-resolution fusion method can effectively maintain the spectral information in MS images, but after the high frequency details of PAN images are added, the resultant images may have spatial distortion, typical phenomena include ringing effect, virtual scene confusion, edge and texture blur (Amolins et al. 2007). In recent years, more and more researchers began to recombine the existing image fusion algorithms to develop new fusion algorithms. For example, in order to combine the advantages of component substitution fusion algorithm in spatial information retention and the advantages of multi-resolution fusion algorithm in spectral information retention, some researchers proposed a fusion algorithm combining wavelet decomposition and component substitution. Although these methods can obtain better fusion results than the component substitution method and the multiresolution method, they increase the computational complexity. When the correlation between MS band and PAN band is low, there will be more spectral distortion in the synthesized image. In order to further reduce spectral distortion and keep the physical meaning of MS image, an algorithm combining the spectral response function of sensor and spectral characteristics of ground object with the above three categories of fusion algorithm is also proposed. In addition, the fusion of optical and microwave images, the fusion of hyperspectral images and LiDAR data, and the fusion of remote sensing data from different platforms have also been developed.

3.1 Multi-source High Resolution Remote Sensing Image Fusion Technology

87

3.1.2 Feature-Level Image Fusion Feature-level remote sensing image fusion belongs to the middle level of image fusion. This kind of fusion can not only fuse the features of different images, but also fuse the different features of an image. Firstly, the processing method of this kind of fusion is to extract the original information from different sensors, and then to process and analyze the multiple feature information obtained from multiple sensors comprehensively, so as to achieve the classification, collection and synthesis of multi-sensor data. Different from pixel-level image fusion, feature-level image fusion emphasizes the spatial one-to-one correspondence of features rather than pixels. Feature-level fusion achieves considerable information compression, is conducive to real-time processing, and provides features directly related to decision analysis, but its fusion accuracy is lower than pixel-level. Feature level image fusion is mainly applied to remote sensing image classification based on multiple information fusion, extracting different types of feature information from remote sensing images and carrying out comprehensive analysis, change detection, quantitative analysis of feature parameters of ground objects, etc.

3.1.3 Decision-Level Image Fusion Decision-level image fusion is the fusion after feature extraction and feature recognition, and it is a high-level information fusion. The results of decision-level fusion will provide the basis for various control or decision-making, so the feature information must be used selectively in combination with specific application objectives. The advantages of decision-level fusion are strong fault tolerance, good openness, short processing time, low data requirements and strong analysis ability. The cost of decision—level fusion is high because of the high requirement for preprocessing and feature extraction. Decision-level remote sensing image fusion is a process of selecting and synthesizing the recognition results of multiple classifiers. Before decision fusion, basic classification and judgment of remote sensing images are needed. The methods of decision fusion include fuzzy set theory, neural network, majority voting, Bayes theory and so on.

3.1.4 A New Method of Remote Sensing Image Fusion—RMI Method MS and PAN images should be strictly aligned in space. The dislocation between MS and PAN images tends to blur the boundary of ground objects in the fused image, which significantly reduces the spectral quality and visual effect of the fused image. If a PAN pixel and MS subpixel M1 correspond to the same object, but there is a

88

3 Technology and Method of Detailed Information Extraction of Animal …

slight spatial dislocation between the two, then PAN pixel corresponds to another subpixel M2 in space. M2 and M1 are adjacent in space, and their pixel vectors have different lengths, but their directions are similar. Therefore, a new subpixel M1 can be simulated to fuse with PAN images by “maintaining the vector direction of M2 and resetting the pixel vector length of M2 with reference to PAN pixels”, so as to reduce the impact of dislocation between MS and PAN pixels on the fusion results and improve the visual effect and spectral quality of fused images. Based on this idea, Jing and Cheng (2011) proposed the RMI (reduce misalignment impact) fusion method. The first step of RMI method is to establish a linear fitting relationship between low resolution PAN image and MS image: PL =

n Σ

ai Mi + b + e

(3.1)

i=1

M i is the I (I = 1, ···, n) band of MS pixel; PL is PAN pixel with low resolution estimated by average method. The coefficients ai and b can be calculated by the least square method; e is the residual; Since there may be dislocation between MS and PAN images, the edge part of MS and PAN images can be excluded and only use the middle part of the image when establishing the linear fitting relationship. Hypothesis Eq. (3.1) is also applicable to PAN image P and MS image M with high resolution, so the formula can be written as P=

n Σ i=1

ai Mi, f + b + e = |Mf |

n Σ ai Mi, f i=1

|Mf |

+b+e

(3.2)

where, M f is the MS image after fusion; M i,f is the ith MS band after fusion; “||” is a vector length. The vector length of MS pixel after fusion is | | |M f | ≈ Σ P − b ai Mi, f n i=1 | M f |

(3.3)

The unit pixel vector M e,f of the fused MS pixel, f can be expressed as Mf M Me, f = | | = |M f | |M|

(3.4)

Based on the above formula, fusion pixel M f can be obtained: | | P −b M M f = | M f | Me, f = Σn i=1 ai Mi

(3.5)

3.2 Fine Information Extraction Method of High Resolution Remote …

89

Fig. 3.1 The original and fused images: a 4 m original MS image, b 16 m MS image upsampled to 4 m, c 4 m PAN image, d RMI fused image, e Brovey fused image, f PANSHAEP fused image, g SDM fused image, and h Gram-Schimidt fused image

Atmospheric effect is an important factor that must be considered in the image fusion process (Jing and Cheng 2009). The main part of atmospheric effect is atmospheric radiation. Atmospheric radiation can be easily determined by dark pixel method. Atmospheric radiation can be removed from the original image before the image fusion, and then re-added to the fusion image after the fusion process ends. As can be seen from the Fig. 3.1, RMI fusion method can process the mixed pixels at the boundary between vegetation and non-vegetation more effectively. The fusion effect of this method is similar to that of PANSHARP, which is superior to traditional fusion methods such as Brovey, SDM and Gram-Schmidt, making the boundary of ground objects in the fusion image clearer.

3.2 Fine Information Extraction Method of High Resolution Remote Sensing Image Remote sensing information extraction is the reverse process of remote sensing imaging, and it is the process of acquiring target information from remote sensing images. The extraction of remote sensing information includes the extraction of specific ground objects and states, the extraction of indicators, the extraction of physical quantities and detection of changes. The remote sensing information extraction can be roughly divided into three categories: quantitative remote sensing, remote sensing classification and target recognition.

90

3 Technology and Method of Detailed Information Extraction of Animal …

At present, the information extraction and target recognition of high-resolution remote sensing images are still in the stage of research and development. The biggest difficulties are large amount of data, complex algorithm, low degree of automation and low recognition rate. With the improvement of spatial resolution. The spatial, geometric structure and texture information of ground objects in the image are becoming more and more abundant. Single ground objects are presented in the form of multiple pixels distributed in contiguity, and each pixel represents only a small part of the ground objects, rather than the whole of the ground objects. Meanwhile, the spectral difference between the pixels within the ground objects is more significant, or even exceeds the spectral difference between the ground objects. In this case, the traditional automatic classification technology based on pixel spectral statistics is difficult to meet the current requirements of remote sensing image information extraction, and it is difficult for users to use remote sensing data to do further research. This feature of high-resolution remote sensing data has become the main bottleneck restricting its application (Xiao and Feng 2012). In recent years, object-oriented image analysis methods have developed gradually. It is different from the traditional processing method based on pixels, firstly, the remote sensing image is segmented by image segmentation technology to obtain a large number of segments (image regions with relatively consistent internal attributes or high homogeneity), then extract the various features of segment, and perform object recognition and identification of segments in the feature space, so as to finally complete the classification and extraction of high-resolution image information. As the processed objects are transferred from pixels to the object level of feature elements, the object-oriented image analysis is closer to the thinking logic of people observing data, and the number of features that can participate in the subsequent analysis is far more abundant than the former, so it also makes the integration of geoscience knowledge easier. Human-computer interaction using multiple interpretation means is still the main way of remote sensing image interpretation at present, but the accuracy of artificial visual interpretation is related to personal experience. Different people may have different or slightly different results for the same image. In addition, the workflow of human-computer interaction interpretation is complicated, with low efficiency, poor objectivity, low intelligence and high cost, which makes it difficult to integrate with existing software systems. With the development of massive high-resolution remote sensing image acquisition and remote sensing image processing technology, intelligent interpretation and quantitative analysis of remote sensing image based on computer technology has gradually become a current research focus, and objectoriented high-precision classification is one of the key links.

3.2.1 Overview of Classification Methods Remote sensing image classification is to use the computers to analyze the spectral information and spatial information of various ground objects in remote sensing

3.2 Fine Information Extraction Method of High Resolution Remote …

91

images. Select features that can be used as classification criteria (spectral feature, spatial feature, time phase feature, etc.) and then each pixel in the image is assigned to each subspace, finally the discriminant criteria are established for classification. The traditional classification method is mainly based on the pixel. This method uses pixels as the basic unit to extract remote sensing information, mainly based on spectral characteristics of ground objects to classify, that is, the feature vectors of the same features are clustered in the same feature spatial, and different features will have different spectral characteristics or spatial characteristics. Traditional classification methods mainly include unsupervised classification and supervised classification. The unsupervised classification methods do not need to acquire a priori knowledge of the image features (i.e., manual selection of training samples), and rely only on the spectral information (or texture information) of different classes of features on the image for feature selection. Differences of the features are then calculated for classification to confirm the properties of the obtained classes. The commonly used unsupervised classification methods are: classification cluster method, wave spectrum feature curve graphic recognition method, parallel pipeline classification method, dynamic clustering method and K-means method. Supervised classification is a technique based on establishing statistical recognition functions and classifying based on training samples. Commonly used supervised classification methods are: minimum distance method, multi-level cut method, feature curve window method, maximum likelihood ratio method, etc. In recent years, many new artificial intelligence methods have been continuously applied to remote sensing image classification, which has improved the classification accuracy to a large extent. Support vector machine (SVM) is a pattern recognition method based on minimal structural risk and statistical learning theory (Vapnic 1995). This method has strict theory, strong adaptability, global optimization, high training efficiency and excellent generalization performance (Fang 2007). Remote sensing image analysis and processing is a popular research direction of SVM application, especially in the fields of classification of land use, mixed pixel decomposition, remote sensing image fusion, multi-spectral/hyperspectral remote sensing classification and so on. For classification problems, SVM is proposed from the perspective of the optimal classification hyperplane in the linearly divisible case. The basic idea is to transform the input space to a high-dimensional feature space by a nonlinear transformation, and then find the optimal classification hyperplane in this new high-dimensional feature space. This classification hyperplane not only classifies all training samples correctly, but also makes the distance from the nearest point in the training samples to the classification plane the largest, i.e., the classification interval is the largest. As shown in Fig. 3.2, the circle and the fork represent two classes of samples, and if these two classes of samples are separable, the result of machine learning is a hyperplane or called the discriminant function, which is the solid line in the Figure. SVM has obvious superiority in hyperspectral remote sensing classification, so the SVM application is considered as one of the most important advances in hyperspectral remote sensing classification. Neural network classification is a remote sensing image classification technology based on artificial neural networks (artificial neural networks, ANN) technology.

92

3 Technology and Method of Detailed Information Extraction of Animal …

Fig. 3.2 The illustration of the support vector machine classification

ANN is a complex network system connected by a large number of simple basic components—neurons and nodes, which simulates the way of human brain nerve processing information to carry out parallel information processing and non-linear transformation (Ge 2007). Artificial neural network has unique functions such as selforganization, self-learning and high fault-tolerance, which make it very effective in solving complex nonlinear problems. Back propagation (BP) learning algorithm is the most commonly used learning algorithm in MLP neural network. The structure of the algorithm is shown in Fig. 3.3. MLP neural network has good nonlinear mapping ability and is often used in remote sensing classification research. It uses BP neural network algorithm to classify remote sensing images, and its basic process is as follows (Wang 1995): Training samples are provided to the network, including the activity model of input unit and the activity model of expected output unit; determine the allowable error between the actual output of the network and the expected output; all link weights in the network are changed by the back propagation algorithm, so that the output generated by the network is closer to the desired output, until the determined allowable error is satisfied. When the sample training process is over, the parameters of the BP neural network model used in remote sensing image classification are determined, and then the regions to be classified in the image can be substituted into the trained network classifier to obtain the classification results and complete the task of remote sensing image classification. Random forests (random forests, RF) is a statistical learning theory. It uses bootstrap method to extract multiple sample sets from the original sample set, and the number of samples extracted is equal to that of the original sample set. Then, it

3.2 Fine Information Extraction Method of High Resolution Remote …

93

Fig. 3.3 BP network structure diagram

conducts decision tree (DT) modeling for each Bootstrap sample set. The final classification result of each sample is determined by voting results of multiple decision trees (Fang et al. 2011), as shown in Fig. 3.4. The specific steps are as follows: (1) K self-help sample sets are generated from the original training data, and each self-help sample set is the entire training data of each classification tree. (2) Each self-help sample set grows into a single classification tree. At each node of the tree, m features are randomly selected from M features (mmM), and a feature from these m features is grown by branching according to the principle of minimum node impurity. Let this classification tree grow fully without the usual pruning operation so that the impurity d of each node is minimized. (3) The new data are predicted based on the generated multiple tree classifiers, and the classification results are determined by the number of votes for each tree classifier. Each sampling generates a set of self-help samples, and the remaining

Fig. 3.4 The illustration of random forest classifier

94

3 Technology and Method of Detailed Information Extraction of Animal …

samples in the whole sample excluding the self-help samples are called out-ofbag (OOB) data. OOB data are used to predict the classification correctness, and the OOB estimates of the error rate are obtained by aggregating the prediction results of each time, which are used to evaluate the correctness of the combined classifier.

3.2.2 Object-Oriented Classification Methods At present, the spatial information of high resolution remote sensing image data is rich, but the spectral resolution is low and the spectral information is relatively insufficient. Therefore, only relying on the spectral information of pixels for classification, focusing on local pixels and ignoring the texture and structure of adjacent entire patches, will inevitably lead to a decrease in classification accuracy. In addition, the spectral statistical characteristics of high resolution remote sensing images are not as stable as those of medium and low resolution images, and the within-class spectral differences are large. The method that only considers the spectral information of a single pixel is prone to a lot of misclassification, resulting in a lot of salt and pepper noise. It is difficult to obtain satisfactory results by using pixel-based remote sensing classification algorithm to classify high resolution remote sensing images. Baatz and Schäpe (1999) and Baatz (2000) proposed an object-oriented remote sensing image classification method according to the characteristics that spatial features of high-resolution remote sensing images are richer than spectral features (Baatz and Schäpe 1999; Baatz 2000). This method breaks through the limitation of traditional classification methods, which take pixels as basic classification and processing unit, and takes objects (including super objects and sub-objects) composed of multiple adjacent pixels with more semantic information as processing unit, which can achieve higher level remote sensing image classification and object extraction. The remote sensing image is segmented by this method. First, homogeneous objects (or segments) are obtained, and then according to the specific requirements of remote sensing image classification or target feature extraction to extract various features of the target feature (such as spectrum, shape, size, structure, texture, shadow, spatial location, related layout, etc.) in order to achieve the purpose of classifying remote sensing images or extracting target features. The essence of object-oriented remote sensing image classification method is to classify remote sensing images from a higher level (object level) by taking the object (or segment) as the smallest unit of classification or detection, so as to reduce the loss rate of semantic information of traditional pixel-level classification method and make the classification results contain more semantic information. Object-oriented information extraction technology mainly includes two key technologies—multi-scale remote sensing image segmentation and object-oriented image classification. Multi-scale remote sensing image segmentation is the basis and key of object oriented information extraction and the segmentation quality is directly related to the accuracy of subsequent information extraction results. Different objects

3.2 Fine Information Extraction Method of High Resolution Remote …

95

Fig. 3.5 Network hierarchy diagram

have different scales in the image, therefore, different analysis purposes are concerned with different scales. The multi-scale analysis method is implemented through object architecture at different levels, as shown in Fig. 3.5. The segmentation of different scales constructs a network hierarchy. Because of the mesh structure, each object can know its own context (adjacent), parent and child objects. This mesh structure is a topological relationship. For example, the boundary of the parent object determines the boundary of the child object, and the area size of the parent object is determined by the sum of the child objects. Each layer is made up of its immediate children, and at the next higher level, the children are merged into the parent, and this merging is constrained by the boundaries of the existing parent. Adjacent objects cannot be merged if they have different parents (Baatz and Schäpe 1999). Image segmentation is a prerequisite for object-oriented remote sensing information extraction, and accurate image segmentation is the basis for subsequent objectoriented image classification and various other image analyses. The concept of image segmentation was first introduced in the field of computer vision in the 1950s–1960s. After decades of development, image segmentation techniques have been greatly developed and have been widely used in medical images, computer vision, remote sensing image processing and other fields. Existing image segmentation algorithms can be broadly classified into four categories: thresholding methods, boundary-based methods, region-based methods and hybrid methods (Adams and Bischof 1994). The first class of methods classifies images into two or more classes by setting thresholds on the grey-scale space (or multi-dimensional feature space). The first class of methods classifies images into two or more classes by setting thresholds on the grey-scale space (or multi-dimensional feature space). The second class of methods delineates groud objects by detecting boundaries, as in the case of graph theory-based segmentation algorithms. The third class of methods identifies regions by determining the difference between the current region and neighbouring regions, such as clustering algorithms, region growing algorithms, seed region growing algorithms and precipitation watershed segmentation algorithms (Zahn 1970; Urquhart 1982; Adams and Bischof 1994; Cramariuc 1995;

96

3 Technology and Method of Detailed Information Extraction of Animal …

Moga and Gabbouj 1995). The fourth class of methods combines discriminatory rules for regions and boundaries to distinguish groud objects in images, such as the immersion watershed method and the subvariant surface matching method in graphic imagery (Beucher and Lantuejoul 1979; Meyer and Beucher 1990; Vincent 1991; Meyer 1994; Moga and Gabbouj 1995). All these algorithms have a certain range of applicability and are not well generalised. Due to the fact that remote sensing images have more bands and more complex information, the segmentation algorithms developed in the field of computer vision for grey-scale images or three-band colour images are not fully applicable to multi-band remote sensing images. In order to segment high resolution remote sensing images with high quality, the idea of multi-scale image segmentation is often used. The basic idea of multiscale image segmentation is region merging, where image elements with similar properties are combined to form small connected regions, resulting in segments with minimal heterogeneity. The principle of image segmentation is “as little variation as possible within segments and as much variation as possible between segments”. In the final segmented image, the image pixels within the segment have similar spectral and textural characteristics, while the inter-segment differences (as a function of spectrum, texture, shape, etc.) are greater than a specific scale threshold. Multiscale image segmentation is usually done in two steps: initial segmentation of the original image and merging of the initial segments. In these two steps, the initial image segmentation is the key, the quality of the initial segmentation directly determines the subsequent segmentation image, the quality of the classified image, and the effect of other subsequent processing. The spatial, spectral and shape features of the image objects are considered in the remote sensing image segmentation process, so the generated image objects include not only spectral homogeneity, but also homogeneity of spatial and shape features. The segmented image segments are merged into regions based on the principle of minimal heterogeneity. The region merging algorithm with the least heterogeneity is a bottom-up, step-by-step region merging process starting from the image pixels. In terms of high-resolution remote sensing information extraction, it is particularly worth mentioning eCognition, the first object-oriented remote sensing information extraction software developed by the German company Definiens Imaging. eCognition uses object-oriented and fuzzy rule-based processing and analysis techniques, pioneering object-based remote sensing information extraction. The main features of eCognition include: (i) Employs an object-oriented and segment-oriented feature extraction and decision-making process that mimics the cognitive processes of the human brain, using features of objects, and relationships with neighbouring objects, to identify targets; (ii) Adopts fuzzy classifie; (iii) The combination of computer automatic classification and manual information extraction can implement necessary manual intervention on the results. The software uses a multi-scale image segmentation method based on the fractal network evolution approach (FNEA). The method consists of two main steps: firstly, the image is initially segmented to produce a large number of non-overlapping initial segments, and then the initial segments are merged to produce the final segmented image. During the merging process, the spectral and geometric information of the segments is taken into account, including the spectral

3.2 Fine Information Extraction Method of High Resolution Remote …

97

variance of the image pixels within the primitives, the compactness of the segments and the smoothness of the segment boundaries, so that the image elements within the merged segments are spectrally homogeneous, tightly aligned and have smooth segment boundaries. During fuzzy classification process of the segment, various statistical values of the segments, such as the spectral and texture statistics within the segments, the geometry of the segments, and the spatial topological relationships between the segments, are used to determine the class of the target segment. The software is also capable of segmenting a very large multi-band remote sensing image at high speed, with the processing taking up less memory. However, as the definition of the scale parameter in this software is not published, it is difficult for the users to effectively relate this function to parameters such as the physical scale, spectrum, tightness and smoothness of the segments. During the image segmentation process, users have to preset the values of the scale parameters and the weights of the spectra, tightness and smoothness, then make a manual judgement on the segmentation results and later adjust the values or weights of each parameter to re-segment until the segmentation results meet expectations of users. This trial-and-error process takes time, and the final result depends on the prior knowledge of the user.

3.2.3 Fuzzy Logic Classification Fuzzy logic has been used in a wide range of fields over the past few decades. In contrast to classical two-valued logic, simply assigning propositions to “yes” and “no” or “true” and “false” is no longer sufficient. In fuzzy logic, a proposition is no longer either true or false, and affiliation becomes a new approach to problem solving. A fuzzy set is a dataset in which all elements have membership degrees. An element’s affiliation can be either fully or partially affiliated. That is to say, the affiliation of an element is no longer just a strict two values of 0 and 1, but can be any value from 0 to 1. The greater the value, the higher the degree of affiliation, the higher the degree to which the element belongs to the fuzzy set. The mathematical function that defines the degree of affiliation of an element is called the affiliation function. Fuzzy classification is an application of fuzzy theory. Fuzzy classification takes into account factors such as: uncertainty in sensor measurements, parameter variations due to sensor calibration, unclear category descriptions, category mixing due to limited resolution (Benz et al. 2004). Fuzzy classification requires a complete fuzzy system, including fuzzification of the input parameters (to obtain a fuzzy dataset), fuzzy logic combination of the fuzzy dataset, and defuzzification of the fuzzy classification. Fuzzy classification is able to describe the real world better than Boolean logic due to its complexity. Fuzzy logic fosters imprecise thinking and is capable of expressing linguistic rules, making fuzzy classification systems well suited to deal with most of the data fuzziness that occurs in remote sensing information extraction. Fuzzy systems consist of three main steps: fuzzification, combination of fuzzy data sets and defuzzification.

98

3 Technology and Method of Detailed Information Extraction of Animal …

Fuzzification describes the transformation of a rigid system into a fuzzy system. It is defined in terms of several fuzzy sets of feature characteristics that represent feature categories such as “low”, “medium” and “high”. These fuzzy feature classes are defined by the affiliation function. The affiliation function needs to define the relationship between feature characteristics and their categories, which requires the introduction of expert knowledge. The more comprehensive the affiliation function is in describing real-world expert knowledge, the better the final classification will be. A fuzzy rule base is a combination of fuzzy rules, combining different fuzzy sets. The simplest fuzzy rules depend on only one fuzzy set. A fuzzy rule is of the form “if …… then ……”, i.e. an action occurs if a condition is satisfied. More advanced fuzzy rules are obtained from different fuzzy sets by different operations, such as “with”, “or”, “not”, etc. A fuzzy rule base implements a set of fuzzy classifications and contains the return values of the output classes, which represent the probability of belonging to that class. In order to produce a standard land cover and use map, the results of the fuzzy classification must then be converted to the results of the clear classification. The maximum affiliation of the fuzzy classification will be used for the assignment of the cl. However, if the maximum affiliation value is less than a certain threshold, the classification will not be performed to ensure accuracy ear classification. This is the most important step in defuzzification.

3.2.4 Analytic Hierarchy Process The analytic hierarchy process (AHP) is a flexible and simple multi-criteria decisionmaking method proposed by the American operations researcher T. L. Saaty in the mid-1970s. The method decomposes a problem into several basic factors, determines the hierarchy of factors according to the dominant relationship, and then compares the factors two by two to obtain the weights of each factor. The method is an effective tool in systems engineering for dealing with complex problems that are difficult to analyse entirely by quantitative methods. It decomposes complex problems into a number of levels and analyses them step by step at a level much simpler than the original problem, thereby achieving the objective of solving complex problems. In other words, the complex problem is decomposed into a series of simple problems layer by layer, and through the analysis and research of the simple problems, the purpose of solving the complex problems is achieved (Liu 2001). A large number of practical problems are often a complex system formed by a multitude of interrelated and inter-constrained factors that cannot be quantified. Analytic hierarchy process provides a simple and easy-to-use solution to this problem. As the core of the method is to quantify the decision maker’s empirical judgement and thus provide the decision maker with a quantitative form of decision making, it is more practical in situations where the objective structure is complex and the necessary data is not available. The basic steps of the method are as follows.

3.2 Fine Information Extraction Method of High Resolution Remote …

99

(1) Construct a hierarchical model. Each factor is broken down into a number of levels according to its basic decomposition, with each factor in the same level being both subordinate to or influencing the factors in the upper level and dominating or influencing the factors in the lower level. The top level is the target level, the bottom level is the programme or object level, and there are one to more criteria or indicator levels in between. (2) Constructing a pairwise comparison matrix. Starting from the second layer, each factor belonging to the upper layer is compared in pairs up to the next layer. (3) Calculate the weight vectors and do a consistency test. For each pairwise comparison array, the maximum eigen roots and the corresponding eigenvectors are calculated, and the consistency test is performed using the consistency index, random consistency index and consistency ratio. If the test is passed, the feature vector (after normalisation) is the weight vector; if not, the pairwise comparison array is reconstructed. (4) Calculate the combination weight vector and do the combination consistency test. Calculate the combined weight vector of the bottom pair of targets and do a combined consistency test according to the formula. If the test is passed, the decision can be made according to the results expressed in the combined weight vector, otherwise the model needs to be reconsidered or the pairwise comparison array with a larger consistency ratio needs to be reconstructed. The analytic hierarchy process easy to calculate, the results are clear, and it is an important tool for system analysis, combining qualitative and quantitative, with a wide range of applications, which can be applied directly and effectively by decision-makers and analysts, and is easy for decision-makers to understand and master directly. However, the hierarchical analysis method also has certain limitations, such as it can only be preferred from the original solution but not innovate to get a better solution; the comparison, judgement and calculation process is also relatively rough and not suitable for problems with high accuracy; the subjective factor has a large influence, including the process of building a hierarchical structure model and giving a comparison matrix.

3.2.5 Decision Tree Classifier The decision tree algorithm originated from a conceptual learning system and is one of the most widely used inductive reasoning algorithms, which is robust to noisy data. The decision tree uses the attribute of the sample as the node and the value of the attribute as the tree structure of the branch. The middle node is the most informative attribute in the subset of samples contained in the subtree rooted at that node, and the leaf nodes are the category values of the samples. In general, the smaller the tree the better the predictive power of the tree. Decision trees contain many

100

3 Technology and Method of Detailed Information Extraction of Animal …

different algorithms, which are divided into three main categories: (i) Statisticalbased methods, represented by CART, in which there are two branches for nonterminal nodes; (ii) Information-theoretic-based methods, represented by the ID3 algorithm, in which the number of branches for non-terminal nodes is determined by the number of sample categories; (iii) AID, CHAIN algorithms, in which the number of branches of non-terminal nodes is distributed within a range of two to the number of sample categories. Based on the rudimentary ID3 algorithm of decision tree, Quinian has made continuous improvements in view of its limitations, and successively formed the C4.5 algorithm and the C5.0 algorithm. The C5.0 algorithm is an improvement on the C4.5 algorithm (Quinlan 1992), which splits the sample data according to the fields that provide the greatest information gain, and ensures a trade-off between accuracy and complexity of the decision tree by pruning or merging the nodes, ultimately constructing the best multi-branched tree structure quickly, capable of extracting classification rules that are both accurate and simple to understand. Decision trees can usually be built without significant training time and the resulting decision trees are easy to decode. C5.0 adds a powerful Boosting algorithm to improve classification accuracy. the Boosting algorithm builds a series of decision trees in sequence, with the later ones focusing on previously misclassified and underclassified data to produce a more accurate tree. Decision tree algorithm can be used to classify remote sensing images, and the process is as follows: (1) Input remote sensing images to determine the main ground object types in the regions to be divided. (2) Statistical characteristics of feature types within the training area, including spectral characteristics, statistical analysis of each band of information for each feature type (mean, variance, covariance, etc.), and separability between bands for each feature type, as well as non-spectral characteristics: geometric information (shape, size, etc.), elevation information, texture information, etc. for each feature type. (3) Based on the statistical analysis in the previous step, the feature and band (or combination of bands) with the greatest differentiability is identified as the root node and the classifier is selected to begin the classification process. (4) For each category obtained in the previous step, select the feature and band (or combination of bands) with relatively greater differentiability to build an internal node of the decision tree, and classify them according to different attribute values. (5) Repeat the process of the previous step, forming a sample sub-decision tree on each division in a recursive manner until one of the following situations occurs and the recursion stops: all the attributes of the image elements on that node have the same value, there are no attributes remaining for the next division, or there are no samples in the branch. (6) Check whether the classification result satisfies the ground object type determined in the first step. Check whether the classification result satisfies the feature

3.2 Fine Information Extraction Method of High Resolution Remote …

101

type determined in the first step. If the requirements cannot be met, the decision tree needs to be adjusted, such as pruning or adding nodes, until a correct decision tree is created. The main advantages of the decision tree algorithm are: (i) Decision trees are parameter free and do not require assumptions about the distribution of the input data; (ii) Decision trees can handle non-linear relationships between ground objects and categories, even if there are missing values; (iii) There is no specific requirement for the type of input variables, both categorical input variables and numeric input variables are allowed, but it is required that their output variables must be categorical. In addition, decision trees have very intuitive schematic diagrams, so the classification structure is very precise and easy to illustrate, which shows great advantages in remote sensing classification problems.

3.2.6 K-Nearest Neighbor Classification Algorithm The main application field of K-NN (K-nearest Neighbor, KNN) nearest neighbor classification is the recognition of unknown things, that is, to judge which category unknown things belong to. The idea is to determine which features of the unknown are closest to the features of the known class based on Euclid’s theorem. The nearest neighbour (NN) method is one of the most important methods in the non-parametric approach to pattern recognition. A major feature of the NN method is that all sample points in each category are used as “representative points” (Vapnic 1995). Therefore, the distance between sample X and all training samples should be calculated during classification, and the result is the category of the training sample closest to X. The K-NN method is a generalisation of the NN method, which is a theoretically mature method and one of the simplest machine learning algorithms. The idea is that a sample belongs to a category if most of the K most similar samples in the feature space belong to that category. In the K-NN method, the selected neighbours are objects that have been correctly classified. The method relies only on the category of the nearest neighbour or samples to determine the category to which the sample to be classified belongs. Although the K-NN method also relies on the limit theorem in principle, it only relates to a very small number of neighbouring samples when it comes to category decisions. Since the K-NN method relies on a limited number of surrounding neighbouring samples rather than on a discriminative method of class domains, it is more suitable than other methods for sets of samples to be divided where there is a large crossover or overlap of class domains.

102

3 Technology and Method of Detailed Information Extraction of Animal …

3.3 High-Resolution Remote Sensing Data Canopy Delineating Technology Forest resources are one of the most important natural resources on earth. Forest resources can be assessed by a variety of indicators, such as tree diameter at breast height, height, canopy width, and depression. The canopy is the main place where trees obtain light energy and perform energy conversion, and it is also the most obvious part of remote sensing images reflecting tree information. Single tree canopy delineating has an important role in fine forestry management. High-resolution remote sensing images contain fine canopy information, and single-wood canopy delineating technology based on high-resolution remote sensing images has made great development in the last 20 years. The first step in crown delineating of single trees is to determine the location of each crown. Canopies usually exhibit a shape with a high center and low surroundings: broadleaf tree canopies are approximately dome-shaped with a high center and low surroundings, while coniferous tree canopies are approximately spire-shaped with a high center and low surroundings. Therefore, both coniferous and broadleaved canopies can find a central highest point to represent the position of the canopy, and this geometric highest point of the canopy is close to the brightest point of the canopy on the remote sensing image. The method of circling a single wood canopy from remote sensing optical images generally makes use of the radiometric feature of the canopy, i.e., the image has local maxima of brightness near the top of the canopy, while the brightness values in the peripheral regions gradually decrease.

3.3.1 Traditional Method of Locating the Canopy of an Individual Tree Traditional methods for locating individual tree canopy are based on optical features of the canopy, such as the ray method, expert classification, colour-texture method and valley tracking methods. (1) The ray method allows for automatic identification and locating of tree canopys. Pouliot et al. (2002) automatically detected the canopy based on the spectral characteristic that the canopy top of secondary forest had a local maximum brightness, and then extracted several rays from the canopy, detected the boundary points of the canopy along each ray, and connected the boundary points to determine the range of the canopy. Xiong and Wu (2006) proposed a semi-automatic tree canopy location method based on ray method. This method allows the user to specify the canopy center, a number of spectral rays were extracted from the center of the tree canopy to obtain the high-order fitting curve of the brightness of the tree canopy. Then, inflection points were found on the curve as the boundary points of the tree canopy, and the inflection points were connected into polygons, and the polygons were smoothed to obtain a

3.3 High-Resolution Remote Sensing Data Canopy Delineating Technology

103

single tree canopy. Finally, the overlap area of the tree canopy was processed to obtain the canopy contour. The shaded areas of the canopy cannot be extracted by the ray method, and the area of the circled canopy is generally smaller than the result of visual interpretation. (2) The expert classification method is to use image classification to circle the tree canopys by adding expert knowledge to build an expert knowledge base and a reasoning machine. When the classification accuracy is not sufficient, analyse the reasons, modify the rules of the inference machine and reclassify until the accuracy of the canopy is reached. (3) The colour-texture feature-based approach is to outline tree canopies of similar colour and texture on remote sensing images, and thus obtain information on the density, size and species of the forest stand. The colour information separates out roads and buildings, etc. in the background, and the texture features separate out grass and shrubs that are spectrally indistinguishable. Individual tree canopy locating in forest area can achieve higher accuracy, on the basis of eliminating the non-forest area. (4) The valley tracking method (Gougeon 1995, 2014) was proposed by Gougeon (1995). The method detects and delineates the canopy profile by tracking local minima of brightness at the canopy boundary. Firstly, the method distinguishes forest region from non-forest region, then searches for local minimum in forest region and valley pixel in surrounding pixel, and finally takes the region surrounded by connected valley pixel as individual tree canopy. High point density airborne LiDAR data can also be used for individua-canopy locating. The canopy locating method based on lidar data usually distinguishes the point cloud of trees according to the distribution characteristics and waveform characteristics of laser point cloud. Calculate the canopy elevation model from the point cloud of the forest tree, and then use algorithms such as marker watershed to determine the top position and canopy edge extraction, and finally delineate individual tree canopys. With the development of high-resolution remote sensing technology, people gradually use high-resolution remote sensing data to delineate and locate individual tree canopys. In the high-resolution remote sensing images, the spatial information of tree canopys is extremely rich, and the geometric structure and texture information of tree canopys are clearly visible, so the conventional image element-based information extraction techniques can hardly meet the requirements of tree canopy information extraction efficiently, object-oriented techniques are often used to implement delineating and locating of individual tree canopy.

3.3.2 Object-Oriented Single-Wood Canopy Delineating Object-oriented technology breaks through the traditional extraction method that takes image element as the basic unit. It takes the object as the basic unit and uses the

104

3 Technology and Method of Detailed Information Extraction of Animal …

spectral, texture, structure, geometry and other information of the feature to achieve the extraction of the target feature on the basis of the object. The idea of extracting tree canopy information by using object-oriented technology is generally: firstly, multi-scale segmentation of the image is performed to establish the image object, and then various image analysis methods are used to classify the image, and finally the size and shape of the tree canopy information is extracted. The maximum-minimum method can be used to extract the boundary of the canopy and delineate the canopy, or fuzzy classification based on samples to extract information such as canopy size, or spatial features can be used to extract the canopy.

3.3.3 MSAS Single Wood Canopy Delineating Method The canopy height mode (CHM) obtained from LiDAR data can be segmented to obtain a single-wood canopy. However, the similarity between branches, crowns and clumps, which have similar shapes and sizes, can significantly reduce the accuracy of single-wood canopy delineating in dense forests. Based on multi-scale analysis and watershed segmentation techniques, Jing et al. (2012a) proposed a new method for canopy delineating based on multi-scale analysis and segmentation (MSAS) of images. The method first determines the scale range of the canopy by mathematical morphology, then filters the canopy elevation model CHM obtained from LiDAR data at multiple scales, uses the local maximum points in the filtered CHM as markers to mark the watershed segmentation of the original CHM, and finally superimposes and optimizes the multi-scale segmentation results to obtain the complete canopy map. The method mainly contains the following three processing steps. 1. scale analysis The main scale of the target canopy is determined by applying the mathematical morphology open operation on CHM. A structuring element (SE) of suitable size and shape is selected, and objects of different sizes on the grayscale image can be separated using the mathematical morphology open operation. The structuring element SE is a matrix containing only 0 and 1 and can be of arbitrary shape. On the image after the opening operation, the objects containing the structuring element can be completely retained, and the others will be eliminated. When a series of disc structure elements of different sizes and the open operation are applied to the CHM of the forest image (Fig. 3.6), the scale distribution of the target canopy can be statistically obtained.

3.3 High-Resolution Remote Sensing Data Canopy Delineating Technology

105

Fig. 3.6 Forest canopy and multiscale distribution

In turn, multiple major scales {k 1 , k 2 ,…,k n } of the canopy are obtained, where k i represents the minimum scale of each major scale distribution. 2. Multi-scale filtering and image segmentation The CHM is filtered at different scales and then segmented using the Marked Watershed algorithm to obtain segmented images at different scales. Design a series of Gaussian filters based on several major scales of the tree canopy. Specifically, for a canopy with a major scale of k i pixels, the window size of the Gaussian filter can be set to k i × k i pixels and the standard variance σ of the filter set to a fraction of the scale k i . Gaussian filters of different scales are used to simulate the geometric three-dimensional shape of the canopy, i.e. a three-dimensional model of the canopy. When the CHM is filtered with this filter, canopies of similar size and shape to the filter are enhanced and smaller canopies are suppressed (Fig. 3.7). The Marked Watershed algorithm is used to segment the CHM at different scales (Fig. 3.8). The watershed approach simulates the submergence process from the marker point to determine the catchment basin. The most critical step in the algorithm is to find the marker points. Here the local maximum of the filtered CHM at different scales is used as the marker point.

Fig. 3.7 The tree crown height models (CHM) filtered at a small, b medium, and c large tree crown levels, respectively

106

3 Technology and Method of Detailed Information Extraction of Animal …

Fig. 3.8 The segmentation maps at the a small, b medium, and c large tree crown levels, respectively

3. Merge the multi-scale segmentation layers to obtain the final canopy map Segments at large canopy scales may cover multiple segments at small scales. In order to combine segments at different scales, it is necessary to know whether the segment at a large scale (referred to as a rough segment) is a single tree or a clump. In CHM, tree canopies are generally circular, whereas a clump is a combination of multiple canopies and is relatively less circular. Based on this phenomenon, a rough segment can be compared to a segment at the corresponding fine scale (referred to as a fine segment) for roundness. The roundness c of the segment can be calculated as follows. c = A/(πr 2 )

(3.6)

where A is the area of the segment; r is the maximum distance from the edge of the segment to the centre point. The closer the roundness value is to 1, the closer the segment is to circular. Rough segments and their corresponding fine segments can be merged in the following steps: (i) Remove fine segments that are not covered by rough segments so that fine and rough segments can be better matched; (ii) If the circularity of a rough segment is less than a predetermined threshold and the fine segment satisfies the following condition (Eq. (3.7)), then the rough segment will be treated as a clump and replaced by all its corresponding fine segments. ( A L > T A A0 ) & (c L > c0 )

(3.7)

where A0 and c0 are the area and roundness of the rough segments. AL and cL are the area and roundness of the fine segments; and T A is a user-defined threshold value. After these steps, the larger scale segments are first combined with the medium scale segments, and the resulting segments are then combined with the smaller scale segments. This merging process results in a layer containing segments at multiple scales. After removing the small segments from the layer and filling in the gaps in the segments, a canopy distribution is created (Fig. 3.9).

3.3 High-Resolution Remote Sensing Data Canopy Delineating Technology

107

Fig. 3.9 Canopy distribution obtained from the CHM

3.3.4 MFS Single Wood Canopy Delineating Method Forests consist of multi-scale branches, canopies, and clumps of trees. When singlewood canopy circling is performed with a watershed segmentation method, branches often cause over-segmentation of the image. To reduce this over-segmentation, Jing et al. (2012b) proposed a new method for single-wood canopy delineating based on image multi-scale filtering and image segmentation, referred to as the MFS (multiscale filtering and segmentation) method. The method first determines the main scales of the canopy, then low-pass filters the grayscale image with Gaussian filters and segments it with a watershed segmentation algorithm, and finally combines the obtained multi-scale segmentation images to obtain the canopy distribution map. The method mainly includes the following five steps. 1. Scale analysis The shape of the canopy can be thought of as a three-dimensional ellipsoid formed by the rotation of a semi-ellipse representing the outermost branch, and the perimeter of the underlying disc is the perimeter of the semi-ellipse, which is also the perimeter of the canopy. Therefore, using the open operation of image morphology and the discshaped structural element SE, the main dimensions of the tree canopy in the image can be determined, as shown in Fig. 3.10. The main purpose of the open operation is to remove ‘foreground’ objects, i.e. those ground objects that are smaller than the

108

3 Technology and Method of Detailed Information Extraction of Animal …

Fig. 3.10 The illustration of tree crowns and their horizontal slices: a tree crowns, b horizontal slices of tree crowns, and c a horizontal slice and the maximum structuring element inside (white circle)

structural elements. The shape of the structural element is determined by the size of the target ground object, with each element being either 0 or 1, as shown in Fig. 3.11. 2. Filetering Multiple Gaussian filters were designed with reference to the main scales of the canopy obtained from the scale analysis, and the Gaussian filters were used to filter the forest grey-scale image to produce a multi-scale image of the forest, as shown in Fig. 3.12. Assuming that the canopy has a diameter of d pixels, creates a Gaussian filter with a size of d × d is and the standard variance is set to 0.3d. When the

Fig. 3.11 Morphological opening operation to separate tree crowns of different sizes: a the CHM of two tree crowns of different sizes, b a structuring element of 7-pixel diameter, c the eroded image, d the opened image, e the difference image between the original and opened CHMs, and f opened image with a structuring element slightly larger than the tree crown

3.3 High-Resolution Remote Sensing Data Canopy Delineating Technology

109

Fig. 3.12 a The original gray forest image and the filtered versions at the b small, c medium, and d large tree crown levels, respectively

Gaussian filter is applied to the forest grey-scale image, canopies with similar shape and size will be enhanced and small features will be filtered out. 3. Watershed image segmentation The watershed segmentation algorithm was initially used to segment a DEM by finding a local minimum in the DEM as a marker and then flooding from the marker, designating the point where two different sources of water meet as the watershed and marking each area surrounded by the watershed as a basin. Using this watershed segmentation algorithm to process the Gaussian filtered multi-scale image, multiple segmentation layers are obtained, as shown in Fig. 3.13. 4. Edge thinning When large scale segmentation layers and small scale segmentation layers are superimposed together, a rough segment does not always overlap well with a fine segment. The boundary of a rough segment can be refined with reference to the boundary of a fine segment as follows: (i) Select all fine segments that are covered by more than half of the rough segment; (ii) Combine the selected fine segments into a new

110

3 Technology and Method of Detailed Information Extraction of Animal …

Fig. 3.13 The segmentation maps of a the original grey image and the filtered versions at the b small, c medium, and d large tree crown levels, respectively

segment; (iii) Replace the original rough segment with the combined new segment. In the process of edge refinement, the large and medium-scale canopy segmentation layers are refined, and the refined segment boundaries can better overlap with the gaps between the canopies. 5. Synthesize multi-scale segmentation layers In order to combine the large-scale and small-scale segmentation layers, a segment can be judged by analyzing its thinning ratio to determine whether it is a canopy or a bush. The flatness k of the segment is calculated by the following formula: k = 4π A/P 2

(3.8)

where A and P are the area and perimeter of the segment. The more complex the shape of the segment, the smaller the k value. When the segment shape is circular, the k value reaches the maximum value of 1. The shape of the canopy is generally round. The bushes are composed of adjacent circles, and the edges of the bushes are often jagged. Based on this phenomenon, canopies are closer to circles than bushes.

3.4 Spatially Refined Observation Methods for Animal Habitats

111

Fig. 3.14 Canopy distribution

With reference to the different thinning ratio of the canopy and bushes, the rough segments and fine segments can be combined according to the following steps: (1) Removing fine segments not covered by rough segments; (2) Finding fine segments that are slightly larger than the area threshold; (3) If the number of fine segments found is less than 2, exit the program, otherwise continue to the next step; (4) If the thinning ratio of the rough segment is less than the threshold, and the fine segment with the largest coverage area meets the following conditions, the rough segment is considered to be a bush, and the fine segment is cropped to replace the rough segment to produce the final tree canopy map (Fig. 3.14): (Am > T A A0 ) & (km > k0 )

(3.9)

Where A0 and k 0 are the area and flatness of the rough segment, respectively; Am and k m are the area and flatness of the fine segment with the largest area, respectively; and T A is the threshold value (here taken as 0.85).

3.4 Spatially Refined Observation Methods for Animal Habitats The concept of habitat was first introduced by Grinnel (1917) and refers to the sum of the living space of an organism and the spatial conditions that regulate its life. Habitat, also known as animal habitat, is the place where animals live, survive and

112

3 Technology and Method of Detailed Information Extraction of Animal …

multiply, and the environment around it. Any animal that lives here has gradually adapted to the environment of the habitat after millions of years of natural evolution. The two interact and depend on each other to form the most basic ecosystem (Hu 2001). Habitat is a complex of biotic and abiotic factors. The primary habitat of a particular species is the resources (e.g. food, shelter, water), environmental conditions (temperature, rainfall, predators and competitors, etc.) occupied by the species or population and the space in which the species can survive and reproduce. Wildlife in habitats are closely related to habitat in terms of population structure, dynamics and behaviour (Yi et al. 2013), and any species has evolved over time to adapt to the environment, while evolution has also had a countervailing effect on habitat. Therefore, the two interact and depend on each other. If the suitable habitat of a species decreases in quantity or quality, the living condition of the species will deteriorate, and the habits of the species will be affected accordingly. Habitat is a necessary condition for wildlife survival, and research into the theory and methodology of habitat suitability can provide a scientific basis for wildlife conservation.

3.4.1 Overview of Animal Suitable Habitat Research The habitat suitability index (HSI) is the most important and versatile method of habitat suitability evaluation system, and its concept was first introduced by the U.S. Fish and Wildlife Service in 1982 in the paper on the in-stream flow augmentation method. Habitat suitability index is a quantitative index widely used for terrestrial and aquatic habitat assessment and monitoring, and it is often used to support wildlife management decisions (Larson et al. 2004), spatial prediction of potential habitat, and habitat emergency assessment (Zhang et al. 2011; Li et al. 2012). With the development of 3S technology, habitat suitability indices have become more accurate and reliable. The earliest research on habitat selection and evaluation of giant pandas began in the 1930s with the publication of “Records on wild giant pandas” by the Western scholar W.G. Sheldon (Zhang 2014). The earliest scientific and systematic research and analysis of giant panda habitat in China began in 1962. The data acquisition process of these early studies mostly used traditional zoological field observations and consultation with local residents, and the types of data acquired were traditional aspects of giant panda habits, population distribution, and reproduction, etc. Before the 1990s, studies on giant panda habitat mainly focused on qualitative descriptions of characteristics of giant panda activities (Zhang 2014; Jin 2012) after the 1990s, with the introduction of statistical techniques, the analysis of giant panda habitat gradually developed towards the direction of quantification. Reid studied the seasonal selection of habitat by giant pandas using a grid statistical analysis. Tang and Ping used Vanderploeg and Scavia indices to analyze the preference of giant pandas for topography, vegetation depression, and staple bamboo. Yang and Yong (1997) and Yang et al. (1998) studied the habitat selection of giant pandas in Foping by using a 20 m × 20 m random sample square and principal component analysis. Ouyang et al.

3.4 Spatially Refined Observation Methods for Animal Habitats

113

studied the distribution and habitat relationships of giant pandas in Wolong Nature Reserve using GIS technology to evaluate the quality of giant panda habitat and its spatial pattern, and discussed the methods and techniques of habitat evaluation (Ouyang et al. 1995). Since the 21st century, as 3S technology continues to develop, its significance in the study of the habitat suitability of giant pandas has gradually come to the fore. At present, the environmental monitoring of giant panda habitat mainly adopts the combination of remote sensing data interpretation and ground data acquisition and analysis. Ouyang and Liu (2001) first studied the parameter system for testing the habitat quality of giant pandas, dividing the factors influencing giant pandas into three categories: physical environment, biological environment and human activity factors and exploring the relationship between factors, using GIS, and spatial simulation technology to implement this quality evaluation system. This method is the analytic hierarchy process, essentially established in the GIS spatial analysis, using the established quality evaluation system to get the giant panda habitat map. In order to enrich the data sources of the analytic hierarchy process, remote sensing data, digital terrain model (DEM) and giant panda GPS collar data began to be widely added to the habitat model, greatly enriching the original data and the reliability of the model. On the basis of the analytic hierarchy process, subsequent improvements have gradually developed and improved towards multivariate statistical methods, expert systems and fuzzy assignments. In view of the shortcomings of the linear analysis of analytic hierarchy process in simulating the interaction between elements, currently there are some methods using neural network and other machine learning algorithms to monitor the habitat suitability of giant pandas (Liu et al. 2005; Liu 2006; Vina et al. 2007; Linderman et al. 2005), these artificial intelligence-class methods can well fit nonlinear systems and realize the simulation of complexity among elements. These methods are based on expert knowledge, through which the input and output values of machine learning fitting are obtained. The current ecological environment evaluation and analysis of giant panda habitat mainly have the following problems: (i) The ground investigation is difficult. Wild giant pandas often live in solitary mountainous areas. When feeding, they especially like the places which are surrounded by trees. It is very difficult to obtain traces. The three giant panda surveys that have been conducted usually lasted 3–4 years, and the search for traces has also shifted from tooth marks to fecal DNA extraction (Zhang and Hu 2002). (ii) The survey model is not reliable, and the existing giant panda ecological environment assessment model is often a combination of expert system and fuzzy mathematics methods, which belong to the analytic hierarchy process in GIS spatial analysis. The core of this method is expert grading, which is greatly affected by subjective factors. Different experts often reach opposite conclusions. Taking the impact assessment of the Wenchuan earthquake on giant pandas as an example, many studies believe that (Yu et al. 2011; Zheng et al. 2012). The wenchuan earthquake had little impact on giant pandas, while other studies have reached the opposite conclusion (Zhang et al. 2011; Xu et al. 2009). These differences largely stem from different experts’ different views on the importance of impact factors. Therefore, a more objective and effective panda habitat evaluation model supported

114

3 Technology and Method of Detailed Information Extraction of Animal …

by expert experience is urgently needed. (iii) Lack of quasi-real-time ground monitoring means. For example, the third panda census began in June 1999 and ended in late 2003, but the report was not published until 2006. The new fourth panda census only began at the end of October 2011. Therefore, the relatively comprehensive spatial distribution of giant pandas and the growth of bamboo, the staple food, often stay in the state of a few years or even more than ten years ago, which is extremely unfavorable for a comprehensive understanding of the growth and life of giant pandas. In view of the above points, the giant panda habitat evaluation model based on remote sensing and GIS should play a more important role. With the development of remote sensing technology, multi-platform (ground, aviation, aerospace) and multi-sensor (radar, optics, laser) observation methods provide more abundant data sources for giant panda habitat monitoring and evaluation. Rich and detailed observation data provide strong support for habitat evaluation and research, but also put forward higher requirements for effective processing and utilization of massive data. In contrast to traditional statistical, geostatistical and hierarchical analysis process, new algorithms for processing non-homogenous, multi-temporal spatial data mining and analysis need to be proposed urgently. In particular, it is a very challenging task to overcome the inefficient and lagging ground survey by using the quasi-real-time remote sensing data. The ultimate goal of the evaluation of giant panda habitats is to use multi-source data to construct a real-time evaluation system for the ecological environment of giant panda habitats, using realtime remote sensing data, ground data and other available data, combined with a large number of historical observations to make spatial predictions and extrapolations of habitat suitability indices, to assess the status of giant panda populations and the environment in which they are located, and to provide a theoretical and technical basis for subsequent real-time monitoring of giant pandas, spatial and temporal change analysis of giant panda habitats, extraction of highly sensitive response factors to global change and corresponding impact and conservation status assessment of rare and endangered animals. As the scope of human activity continues to increase, the destruction of natural animal habitats by humans is becoming more and more serious, resulting in largescale destruction of ecosystems, fragmentation of animal habitats and a reduction in animal diversity. Habitat fragmentation, mainly as a result of human activities, has led to a decline in habitat function. Wildlife habitat consists of three main elements, namely food, shelter and water. Food is the link between animals and their habitat and the basis for interspecific relationships in animal communities. The abundance of food is closely linked to the survival and reproduction of animal populations and is of great importance in habitat selection. A shelter is any structural resource in the environment that provides for wildlife to perform its functions, improve reproduction and survival, and any structural entity in a habitat that provides for wildlife needs. Habitat suitability evaluation methods currently used to describe species-habitat relationships include habitat suitability index, multivariate statistical methods, fuzzy logic methods, artificial neural networks, and statistical analysis of multiple species and communities (Yi et al. 2013).

3.4 Spatially Refined Observation Methods for Animal Habitats

115

1. Habitat suitability index The Habitat Suitability Index is used to quantify the relationship between organisms’ preference for habitat and habitat factors. It is a classical method for habitat quantification, which was first proposed by THE US Fish and Wildlife Service in habitat evaluation procedures (HEP) and has been widely used. Habitat suitability index has binary format, single variable format and multi-variable format. Generally, physical habitats depend on multiple variables, so multiple suitability indexes should be combined into a comprehensive suitability index. In practical applications, multivariable format is widely used. The commonly used methods to calculate the comprehensive suitability index include arithmetic average method, geometric average method, product method, minimum method, weighted sum method and weighted product method. The comprehensive suitability index can also be expressed by multivariate function. This depends on the combination and interrelationship of habitat factors and is usually expressed by exponential functions of polynomial exponents. Multivariable format method is intuitive, easy to obtain the required data, and has strong practical operation. It is a classic method for habitat quantification, and it is also the most widely used method at present. However, it relies on expert experience more and has strong subjectivity. Habitat suitability evaluation of animals mainly considers physical environment factors, biological environment factors and human activities factors (Ouyang and Liu 2001). However, the animal’s choice of habitat will change with time and place, so the habitat evaluation factors may be different. Elevation and its derived factors (slope and aspect), forest type, staple food type, canopy density of trees, road, water system and human influence are some factors with high influence. 1) Physical environmental factors Physical environment factors include elevation, slope aspect, slope, water source, etc. Elevation factor is directly related to ambient temperature and animal food distribution. Especially in mountainous areas, the vertical distribution of temperature makes the habitat selection of animals more selective, and the distribution of food varies with the temperature. As the seasons change, animals move to areas where food resources are abundant and temperatures are moderate. Slope is also an important factor affecting habitat selection. The slope has a direct impact on the physical energy consumption of animals. Different animals’ living habits have different selection conditions. For example, giant pandas tend to move on gentle slopes, which are not too steep, helping them save energy. Slope orientation has an important influence on the type of vegetation growing in the area. Sunny slopes have better vegetation growth and higher temperature due to more sun exposure, while shady slopes have less sun exposure, lower temperature and more shade, and different slope orientations are suitable for animals with different habits.

116

3 Technology and Method of Detailed Information Extraction of Animal …

2) Biological environmental factors Biological environmental factors include vegetation type, vegetation growth status, etc. Vegetation types affect the habitat distribution of animals in two main ways. First, animals that feed on plants mainly rely on plants for energy, so the distribution of plants, directly affects the habitat selection of animals. Second, different types of plants can be used as concealment in different situations. For example, vegetarian animals often choose sheltered grass and shrubs as important conditions for habitat selection. Vegetation growth status also affects habitat selection. The vegetation is growing well, the biodiversity of the area tends to be high and the abundance of food resources is an important factor in attracting animals to gather. 3) Human activity factors Human activity factors include distance from residential areas, distance from roads, etc. With the increasing range of human activities, the living environment of animals is more and more affected by human activities. Residents’ daily activities, geological disasters, road traffic, farming and logging will all affect the range of animals. In particular, the hunting of wildlife and habitat destruction has directly led to the isolation and fragmentation of animal habitats. 2. Other methods The multivariate statistical method is a simultaneous study and analysis of multiple random variables, which considers the interactions and correlations between physical variables by analysing the observed data of multiple random variables to investigate the characteristics of the random variables in general, the patterns, and the interrelationships between the random variables. The use of multivariate statistical methods in habitat suitability evaluation is increasing. The fuzzy logic approach has advantages in dealing with uncertainty in habitat modelling, making better use of expert knowledge and dealing more rationally with measurement inaccuracies and uncertainties in the modelling process. The interactions between multiple variables are also considered, but when the number of variables considered increases, the number of fuzzy rules increases rapidly, causing inconvenience to the calculation. Artificial neural networks are able to implicitly identify complex relationships between response and environmental variables, but their explanatory power is insufficient and they require large amounts of real-world data to train them, limiting their practical application. Multiple species and communities can be statistically analyzed by ordination analysis or gradient analysis.

3.4 Spatially Refined Observation Methods for Animal Habitats

117

3.4.2 Selection of Habitat Suitability Factors for Animals Habitat quality and structure play a decisive role in the abundance and density of animals, and high-quality habitat should be able to meet the long-term survival and reproduction needs of giant pandas. Habitat quality determines the survival and reproduction rates of individual giant pandas, and is closely linked to the maintenance of the viability of their offspring and the length of time they occupy the habitat. However, the giant panda habitat affected and restricted by many factors, its habitat selection is affected by various factors, these factors are not identical and interfere with each other. According to a large number of literature, the influence of related factors as high as more than 80, but in actual habitat suitability assessment, should try to seize the main influence factors, ignoring secondary factors, The following three points should be noted in the selection of habitat elements. (1) Habitat elements are representative and carry a greater weight in the overall habitat assessment. (2) Habitat factors should be measurable or quantifiable. (3) The habitat factor should be reasonable. From the commonly used factors selected in various literatures, elevation and its derived factors (slope and aspect), forest type, staple bamboo type, canopy density of trees, road, water system and human influence were selected with high frequency, which also reflected researchers’ views on the importance of factors affecting the habitat of giant pandas. This book adopts three types of influencing factors: geographical environmental factors, biological factors and interference factors. See Fig. 3.15 for the hierarchical structure. 1. Geographical environment factor Geographical environmental factors include elevation, slope, aspect, topographic index and drainage distribution.

Fig. 3.15 Classification of giant panda habitat factors

118

3 Technology and Method of Detailed Information Extraction of Animal …

Elevation factor plays a decisive role in temperature and food distribution of giant pandas. Changes in temperature and food during different seasons simultaneously drive the vertical migration of giant pandas. Giant pandas move to lower altitudes in autumn and higher altitudes in spring, which can not only ensure sufficient food supply, but also keep them in a more comfortable temperature environment. Giant pandas move to lower altitudes in autumn and higher altitudes in spring, which can not only ensure sufficient food supply, but also keep them in a more comfortable temperature environment. The upper limit of giant panda activity is usually determined by bamboo, while the lower limit is greatly influenced by human activities, with different proportions of human activities in different regions (Guo and Yang 1999). Slope is an important factor influencing the activity of giant pandas (Hu 2001). On gentle slopes, giant pandas use less energy to move around, thus helping them to conserve energy. In addition, giant pandas spend a lot of time foraging for bamboo every day, and if the terrain is too steep, it is difficult for them to forage and their physical energy is too much. Slope orientation has an important influence on the type of vegetation that grows in the area. Slope orientation has an important influence on the type of vegetation that grows in the area. Sunny slopes have better vegetation growth and higher temperatures due to more sun exposure, and giant pandas prefer sunny slopes (Hu and Xia 1985). Statistical studies have also found that pandas prefer areas with more sunny slopes (Zhang et al. 2011). The topographic position index (TPI), proposed by Weiss (2001), measures the position of a study point up and down the longitudinal profile of the terrain. The basic idea is that the difference between the elevation of a point and the average elevation of its surrounding area, combined with the slope of the point, is used to determine its position on the slope. It has a very important significance in the classification of landforms. Table 3.1 shows the geomorphological significance corresponding to the value of slope index. Although giant pandas can obtain water from bamboo, the water in the plants does not fully satisfy the daily needs of giant pandas, which still require direct drinking water. As the range of the giant panda is between 4 and 7 km in diameter, the area Table 3.1 Geomorphic significance of slope position index Serial number

Type

Classification standard

1

Ridge

TPI > 1 SD

2

Uphill

0.5 SD < TPI ≤ 1 SD

3

Mesoslope

−0.5 SD < TPI ≤ 0.5 SD, slope > 5°

4

Flat slope

−0.5 SD < TPI ≤ 0.5 SD, slope ≤ 5°

5

Downhill

−1 SD ≤ TPI ≤ –0.5 SD

6

Valley

TPI < –1 SD

Note SD represents the standard deviation of pixel elevation between the research site and the neighborhood

3.4 Spatially Refined Observation Methods for Animal Habitats

119

within the range that includes the water source should be defined as suitable or more suitable. 2. Biological factors Giant pandas are specialist feeders, with bamboo making up 99% of their diet. The health and distribution of the giant panda’s staple food, bamboo, is directly related to its survival, reproduction, spread and development, and bamboo is a direct and standard indicator for evaluating habitat suitability. The distribution of Phyllostachys lengjianensis, Phyllostachys short-cone and Phyllostachys slingstick in the third giant Panda census were selected as suitable areas, and other bambuss were selected as suitable areas. In the long evolution process, bamboo and its plant community formed a close cooperative relationship, which influenced and promoted each other. In the long evolution process, bamboo and its plant community formed a close cooperative relationship, which influenced and promoted each other. A large number of studies have found that giant pandas prefer the primeval forest with a certain canopy density. Giant pandas are extremely rare in evergreen broad-leaved forest and meadow, and it is difficult to find their traces in artificial forest. A large number of studies have found that giant pandas prefer the primeval forest with a certain canopy density. Giant pandas are extremely rare in evergreen broad-leaved forest and meadow, and it is difficult to find their traces in artificial forest. Coniferous forest and coniferous and broad-leaved mixed forest were defined as suitable habitats for giant pandas, while broad-leaved forest, shrub and meadow were considered as less suitable areasuitable habitats. 3. Interference factors At present, the giant panda is disturbed by many factors, and the role of each factor is complex. Daily activities of residents, geological disasters, timber cutting, transportation, agricultural activities and understory resource collection are the main factors affecting the habitat quality of giant pandas. These factors directly destroyed the giant panda habitat, especially aggravated the giant panda habitat fragmentation and isolation. Due to the lack of data and the uncertainty of the influence of factors, this study only considers the influence of road traffic. As with water distance, given the large range of the giant panda’s movement, a certain range of areas adjacent to disturbance factors are to be classified as unsuitable. 4. Criteria for habitat suitability assessment Based on the previous research results, combined with the information obtained from field visits to some areas of the study area and the characteristics of the influencing factors in the study area, this book classifies the indicators into three categories: suitable, subsuitable and unsuitable, and establishes a table of criteria for evaluating the influence factors on the suitability of giant panda habitat in the study area.

120

3 Technology and Method of Detailed Information Extraction of Animal …

5. Single factor habitat evaluation The basic idea of single-factor habitat evaluation is to grade each single factor according to the single index formulated by habitat suitability evaluation criteria, i.e. using GIS software such as reclassify, DEM spatial information extraction, spatial interpolation and buffer analysis to conduct single-factor Evaluation. The proportion of suitable areas was then analysed using area statistics tools to obtain areas suitable for giant pandas.

3.4.3 Model Based on Hierarchical Analysis and Expert Weighting Method Analytic hierarchy process was traditionally used for the evaluation of giant panda habitat suitability, in which the hierarchical weighting strategy used the expert weighting scoring method. Specifically, weights were estimated for the eight factors used in the study (elevation, slope, slope direction, topographic index, distance to water sources, vegetation distribution, bamboo and road distance), and each factor was dimensionless, and the suitability distribution map was generated using raster calculator in ArcGIS. 1. Construction of evaluation indicators According to the actual situation of the study area, one evaluation objective, three factor layers and eight factor layers were determined, as shown in Table 3.2. Table 3.2 Hierarchical analysis of the types of factors influencing giant panda habitat

Target

Factors layer

Factor level

Habitat suitability index of giant panda

Geographic environmental factors

Slope

Elevation Aspect Topographic index Distance from water source

Biological factors

The vegetation distribution Bamboo

Interference factors

Distance from road

3.4 Spatially Refined Observation Methods for Animal Habitats

121

Table 3.3 Judgment matrix of AHP Suitability

Elevation classification

1

Two elements are equally important to an attribute

3

When two elements are compared, one element is slightly more important than the other

5

When two elements are compared, one element is clearly more important than the other

7

When two elements are compared, one element is much more important than the other

9

When two elements are compared, one element is more important than the other

2, 4, 6, 8

The median of the two adjacent judgments above

1, bij

Reverse comparison of two elements

2. Construct a two-comparison judgment matrix The basic idea of the analytic hierarchy process proposed by Saaty is to assume that the elements of the upper level serve as guidelines and have a dominant relationship to the elements of the lower level, and to determine empirically the importance of two elements to the elements of the upper level. Among them, the 1–9 comparative scaling (Saaty 1988) is used to determine the importance of the two comparative indicators and determine the judgment matrix of each layer, as shown in Table 3.3. 3. Judgment matrix normalization and consistency test The judgment matrix can generate feature vectors, and since there are multiple judgment matrix after normalization, the normalization method is used to normalize multiple feature vector matrices to a unified weight measure, so that the relative weights of all factors are obtained. The basic steps of the process are as follows: (1) (2) (3) (4)

Calculate the product of the weight values of each layer of the matrix. Take the nth root of the weight product. Normalize the vector Wi = (W1 , W2 , . . . , Wn ). The eigenvectors generated by multiple discriminant matrices are unified into one metric according to the overlapping factors, and the result obtained is the weight of each factor.

The calculated judgment matrix should also be checked for consistency to ensure that the created matrix is reasonable. First, calculate the original judgment matrix to obtain the largest eigenvalue and corresponding eigenvector. Then test the consistency of the matrix, including consistency index test, consistency ratio test and random consistency index test. If the judgment matrix can meet the consistency test standard, the weight vector obtained after the normalization process is the required vector; on the contrary, if the consistency test standard cannot be reached,

122

3 Technology and Method of Detailed Information Extraction of Animal …

the comparison matrix needs to be constructed again or the original matrix needs to be adjusted. 4. Dimensionless of indicators The mean values and value ranges of different impact factors often vary widely, and they represent different physical meanings, so that dimensionless indexing is required. The dimensionlessness of indicators plays a key role in habitat quality evaluation, and the reliability of the evaluation depends to some extent on the dimensionlessness strategy. There are two forms of influence factors in the comprehensive multi-factor habitat evaluation: the actual value of each indicator factor; and the evaluation value of each indicator factor. The process of dimensionless evaluation is to unify the actual values of each influence factor to the evaluation value. Considering that AHP itself is based on linearity, and a large number of studies have scored the influence factors based on linear functions or segmented linear functions, the study used the same segmented linear functions to dimensionlessize the giant panda influence factors. The giant panda impact factors were first divided into suitable, less suitable and unsuitable in the manner described previously, and then all values were unified into a linear range according to the maximum and minimum values in the impact factors. 5. Comprehensive evaluation The evaluation model studied is a weighted summation based on hierarchical analysis. Habitat suitability maps can be easily generated by overlaying raster calculator in ArcGIS with raster type layers of equal resolution and size, provided that the weights are set in advance. This tool is used for the mapping. First, all layers are unified into one coordinate system according to the single factor evaluation method described in the previous section, and ensure the same image element size and layer length and width. Then, all layers are dimensionlessized to generate a map of giant panda habitat suitability under the combined influence of multiple factors based on the weight values of each factor affecting the quality of giant panda habitat. In order to count the area of different suitability degrees, and also to display the giant panda habitat suitability map in layers, the generated habitat suitability map must be classified by setting thresholds before outputting. From the output results, the area of suitable, subsuitable and unsuitable areas can be counted and the habitat suitability of the study area can be evaluated.

3.4 Spatially Refined Observation Methods for Animal Habitats

123

3.4.4 Improved Method of Integrating Expert System and Neural Network 1. Fundamental Artificial intelligence (AI) has also been introduced into wildlife habitat assessment on the basis of the traditional hierarchical method, in which the neural network has been widely used for its simplicity. Neural network can be well nested in GIS based habitat suitability mapping because of its good fault tolerance, adaptability and robustness. The traditional hierarchical approach to habitat suitability mapping for giant pandas is based on the assumption that the importance of each influence factor varies linearly, and the mutual influence between the factors also exists in a linear form. However, the actual situation is the importance of the influence factor and the mutual influence between them is complex and non-linear. This is because: (i) The HSI index is complicated. Even in the traditional giant panda habitat evaluation, the HSI index is often used as a piecewise function (Liu et al. 2005; Vina et al. 2007; Zhang et al. 2011; Shalaby and Tateishi 2007), the nonlinear formula is better in actual situations (Song et al. 2014); (ii) The threshold is usually based on Expert knowledge or statistical results, but in actual situations, the thresholds are fuzzy and inaccurate, and changes in some influencing factors will also lead to changes in other influencing factors. For example, the illumination of different slopes varies greatly. Vegetation growth conditions on shaded and sunny slopes are different, and the humidity and temperature are very different. The weight of altitude is different for shaded and sunny slopes; (iii) Even within a fixed area, different expert knowledge and statistical results often lead to different conclusions, and many factors may exert certain imperceptible influences on a small scale. In summary, the large animal habitat mapping based on the analytic hierarchy process cannot make good use of expert knowledge, because its linear characteristics determine that it is difficult to obtain a more ideal effect, while artificial intelligence fully fits the nonlinear characteristics and complexity between influencing factors, to better dig out the knowledge in the expert scoring method. In the research, artificial neural network is used as a transfer function to dig deeper into the relationship between influence factors and suitability. Liu builds an integrated expert system and echo state neural network classifier (ESNNC) to map the giant panda habitat. His core idea is to use the input and results of the analytic hierarchy process as the initial input and output of neural network training. After training the neural network with the sample data, recalculate the habitat suitability map based on the neural network, which makes full use of the nonlinear characteristics of the neural network, and the results obtained are more convincing than the traditional analytic hierarchy process. The basic steps of giant panda habitat assessment integrated with neural network are shown in Fig. 3.16. The steps of generating a habitat suitability index map based on expert scoring method and analytic hierarchy process are the same as those described in Sect. 3.4.3. Overlay the impact factor layer (input) and the habitat suitability layer (output), and use sampling points to extract input and output values. These input

124

3 Technology and Method of Detailed Information Extraction of Animal …

Fig. 3.16 Improved neural network-based habitat suitability map for giant pandas

and output values can dig out more complex potential knowledge between habitat suitability and impact factors. Put the input and output values into the neural network for training, and after getting a more satisfactory neural network, it can be output into a new habitat suitability map based on artificial neural network (ANN). 2. BP (back propagation) neural network structure Neural networks have been designed inspired by human brain activity and are often used to solve complex nonlinear problems such as fitting, pattern recognition, clustering, and time series prediction. The complexity and nonlinearity of the relationship between habitat suitability and impact factors of giant pandas has been well discussed in the previous section, and therefore neural networks can be used to fit this complex relationship. The study used a traditional error back propagation neural network (BP neural network). Figure 3.17 shows the basic structure of the BP neural network. The BP neural network used in the study is composed of an input layer, an output layer, and a hidden layer. The number of nodes in the input layer is determined by the number of influence factors, the output layer has one node, and the hidden layer is adjusted according to the training effect. 3. Evaluation of fitting accuracy Mean absolute percentage error (MAPE), root-mean-square error (RMSE) and correlation Coefficient were used to evaluate the effect of neural network fitting. MAPE is a common tool to measure the deviation between fitting value and true value. RMSE is often used to measure the standard deviation of the fitting value and the true value of a large number of training samples. The correlation coefficient is used to represent

3.4 Spatially Refined Observation Methods for Animal Habitats

125

Fig. 3.17 BP neural network basic structure

the Linear independence between the fitted value and the true value. It is used to indicate the extent to which the fitting value and the true value can be fitted by a line.

3.4.5 Neural Network Method Based on Density Map 1. Two basic assumptions In order to make the improved giant panda habitat suitability evaluation method proposed in the study theoretically feasible, two basic assumptions were first made: (i) the higher the frequency of giant panda traces in a unit area, the more frequent the giant panda activity is, and the higher the habitat suitability index of that unit area; (ii) the probability of giant panda traces being found in any area is equal, ignoring the effects of light, plant shading and artificial proficiency in manual surveys. In actual small-scale surveys, these two basic assumptions do not hold in many habitat suitability studies, due to the different types of data used in different studies and the different ways of examination. In large scale habitat suitability studies of giant pandas, these two assumptions can be guaranteed. This is because giant pandas are located in areas with complex topography, and conducting a comprehensive survey is labor-intensive, so reliable large-scale data are currently available from national

126

3 Technology and Method of Detailed Information Extraction of Animal …

censuses of giant pandas. The most recent census information available is the Third National Giant Panda Census, which started in 2001. This census used the giant panda dentition on bamboo as the survey index, and the dentition is very stable and less susceptible to environmental changes than other indicators, thus ensuring the correctness of the first hypothesis. In addition, similar hypotheses have been adopted in studies. For the second hypothesis, the giant panda field census used the sample line method, i.e., by dividing the study area into regular grids (2km2 ) and studying the giant panda trace points within each grid. To improve the efficiency of the expedition, the grid is divided with due consideration to topographic factors, vegetation distribution, staple bamboo distribution, water systems, roads, cliffs, elevation and regional accessibility. During the expedition, the giant panda traces in the area where the sample lines cross are searched and recorded. Through these measures, the sample line method fully ensures the objectivity of the sampling process and the equal probability of sample points being found. 2. Habitat density map of giant panda based on two hypotheses Density analysis is the generation of continuous surfaces from measured trace points or lines to reflect the concentration and sparse spatial distribution of points or surfaces in a graph. In other words, density analysis is to calculate the spatial distribution of data aggregation in the whole region according to the input factor data. The basic principle of density analysis is the process of interpolating through discrete point data or line data. According to different interpolation principles, density analysis can be divided into nuclear density analysis and ordinary point/line density analysis. In kernel density analysis, the points that fall into the search area have different weights, and the points or lines close to the search center will be given a larger weight. On the contrary, if the weight is smaller, its calculation result distribution is smoother. In ordinary point or line density analysis, the points or lines that fall in the search area have the same weight, and they are summed first, and then divided by the size of the search area to get the density value of each point. Since there is no weight for the giant panda trace points, in order to more accurately express the influence of trace points on a certain pixel, the density analysis method used in the study is nuclear density analysis. The basic principle of kernel density analysis is that each raster image element is used as a search unit, and a neighborhood is defined around the center of a certain raster image element (the neighborhood can be defined by different shapes such as circle, rectangle, ring, wedge, etc.), and all the points in the neighborhood are weighted, and the points closer to the image element are given higher weights according to the kernel density function, and then divided by the area of the neighborhood to get the density of point elements. Assumption (i) relates the habitat suitability to the frequency of the appearance of giant pandas. If this frequency can be obtained quantitatively within a certain error, that is, if the assumption (ii) holds, then the traces of giant pandas in a unit area can be used to measure the suitability of giant pandas. That is, the density map of trace points is equivalent to the habitat suitability map. In fact, previous studies have found a close relationship between the habitat suitability index and density (Shalaby and

3.4 Spatially Refined Observation Methods for Animal Habitats

127

Fig. 3.18 Flow chart of the improved habitat suitability model for giant pandas

Tateishi 2007). Tirpak et al. equated the density map with the habitat suitability index and used it to infer the spatial distribution characteristics of the bird population. It can be seen that it is feasible to use the density map to measure the habitat of the giant panda. There is no universal habitat suitability index method and formula. This is because the actual conditions faced in the application process are very different, and reliable expert knowledge in one area may not be applicable in another area. Therefore, in the absence of reliable expert knowledge, if there are objective traces of giant pandas, density maps can also be used as a choice for habitat suitability mapping. However, the problem with the density map is that not all areas will be inspected for trace points. In fact, many dangerous areas and difficult-to-access areas have not been inspected, but giant pandas may be very active here. Therefore, although the density map is relatively reliable, it cannot replace the analytic hierarchy process based on expert knowledge. 3. Habitat suitability evaluation method based on density map 1) Basic idea of algorithm Figure 3.18 is the basic flow for improving the method. The core of the method is to establish the nonlinear transfer function between the selected influence factor and the density diagram. The data pre-processing of the influence factor of this method is consistent with the analytic hierarchy process and the traditional neural network method. After the influence factor layer is generated, firstly, the panda trace points collected in the field are transformed into a point vector format according to their spatial coordinates, and then these point coordinates are transformed into a density map within a unit area, finally, the relationship between the density diagram and the influencing factors is established. As mentioned above, the density map is inaccurate in unsurveyed areas (such as hazardous or inaccessible areas), whereas assumption

128

3 Technology and Method of Detailed Information Extraction of Animal …

(ii) guarantees that the density map is accurate in other sampled areas, therefore, the purpose of establishing the nonlinear transfer function is to learn the relationship between impact factors and habitat suitability through these exact regions, and then to re-infer the habitat suitability of those unsampled regions. In this paper, neural network is used as a nonlinear transfer function to study the habitat of giant panda. After obtaining the input and output through the sampling points, the neural network is used to train these samples to obtain satisfactory results. In order to avoid over-refinement and to ensure a certain accuracy, the resolution of all raster data is unified to 30 m, but 300 m is used as a window in the neural network processing, and the pixels of each factor are averaged in the window. The BP neural network mentioned above is used here. During the training process, in order to make the results comparable with each other, the absolute values of the study density maps were unified to 1–10. The normalized suitability index can be used to calculate the average absolute error percentage. The trained neural network can not only regenerate the density map (habitat suitability map), but also re-map based on new data. In particular, remote sensing data can obtain real-time ground information, especially vegetation information (Shalaby and Tateishi 2007; Lunetta et al. 2006). Changes in ground objects, including vegetation information, will have a great impact on wildlife. For example, when evaluating the 2008 Wenchuan earthquake, a large number of studies were conducted through patch analysis (Zhang et al. 2011) and fitness index analysis (Zheng et al. 2012) and other statistical analysis methods, found that landslides caused by earthquakes and vegetation changes have varying degrees of impact on giant pandas. These studies have found that the vegetation type is closely related to the staple food bamboo related to the giant panda’s habitat. Although bamboo living below the tree canopy is difficult to be directly observed by remote sensing, its abundance and frequency can be inferred from the canopy vegetation. Therefore, the continuously updated remote sensing data can generate new habitat suitability maps through the trained neural network, and provide first-hand reference materials for decision-making departments. 2) Sampling point selection problem and data resolution problem The selection of sampling points is a key step in the preparation of the actual neural network training data. By superimposing the influence factor and density map, the input and output values corresponding to the sampling points are extracted. Since some regions in the density map are inaccurate, sampling points that are selected completely at random will inevitably introduce some inaccurate training samples, and this error increases as the error of the density map increases. The study considers that the area where the giant panda sample points 1 itself must have been examined, so these points are chosen as sampling points, and in addition, some points 2 are hand-selected in areas far from the activities of giant pandas. These areas must not exist, so the density is 0. Then some points are randomly selected on the graph as supplementary values 3, where some inaccurate points will be included, but they are relatively few, and the error tolerance of the neural network can keep them within a small range. Combining the sampled points 1, 2 and 3 is the process of selecting the pandas’ trace points.

3.4 Spatially Refined Observation Methods for Animal Habitats

129

Fig. 3.19 Fractured terrain in Ya’an area. Image taken in May 2013 shows a fractured landscape in ya’an, a giant panda habitat, with a power tower about 30 m high

Another issue that should be considered in the habitat suitability index is the spatial scale, that is, what resolution should be unified for all layers. There is very little discussion on this issue. Only Vina et al. mentioned “re-sampling high resolution to low resolution”. For technical convenience, they unified the resolution of all impact factors to the lowest impact factor resolution, which is 80 m × 80 m (resolution of MSS sensor). In practical applications, the choice of resolution must not only be technically correct, but also conform to the principles of biology and ecology. In the research, although the remote sensing image and DEM used have a resolution of 30 m × 30 m, a 300 m × 300 m window average and a 30 m × 30 m resolution are used to process all influencing factors. The reasons are as follows: (i) The range of activity of male and female giant pandas is 6–7 and 4–5 km2 , too high resolution will produce meaningless results. As pointed out by Chen et al. (1999) the fragmentation of giant panda habitat is very serious. A small suitable area surrounded by a large range of unsuitable areas cannot provide enough living area for giant panda activities, so it can only be classified as an unsuitable area. (ii) The habitat of giant pandas is located in mountainous areas, and the elevation of these areas changes drastically, and the slope aspect and slopes generated from the elevation are also severely broken. As shown in Fig. 3.19, if a resolution of 30 m × 30 m is used, hundreds of grids will be generated in this area, and drastic changes in topography will lead to suitable and unsuitable ones. If a 300 m × 300 m window is introduced, although the number of grids in the area remains unchanged, the total number of pixels will be very smooth, which is more in line with the actual situation. Therefore, a 300 m × 300 m window can not

130

3 Technology and Method of Detailed Information Extraction of Animal …

only maintain a high spatial resolution, but also take into account the biological and ecological factors of the giant panda’s living area. For details of the habitat suitability evaluation of giant pandas in the Ya’an area using this method, see the literature (Song et al. 2014).

References Adams R, Bischof LB (1994) Seeded region growing. IEEE Trans Pattern Anal Mach Intell 16(6):641–647 Alparone L, Aiazzi B (2006) MTF tailored multiscale fusion of high resolution MS and Pan imagery. Photogramm Eng Remote Sens 72(5):591–596 Amolins K, Zhang Y, Dare P (2007) Wavelet based image fusion techniques—An introduction, review and comparison. ISPRS J Photogramm Remote Sens 62(4):249–263 Baatz A (2000) Multiresolution segmentation an optimization approach for high quality multi-scale image segmentation. In: Strobl J et al (eds) Angewandte Geographische Informationsverarbeitung XII. Heidelberg, Wichmann, pp 12–23 Baatz M, Schäpe A (1999) Object oriented and multi-scale image analysis in semantic networks. In: Proceeding of the 2nd international symposium on operationalization of remote sensing, Netherlands Benz UC, Hofmann P, Willhauck G et al (2004) Multi-resolution, object-oriented fuzzy analysis of remote sensing data for GIS-ready information. ISPRS J Photogramm Remote Sens 58(3– 4):239–258 Beucher S, Lantuejoul C (1979) Use of watersheds in contour detection. In: International workshop on image processing: real-time edge and motion detection/estimation, Rennes, France Chavez P, Sides SC, Anderson JA (1991) Comparison of three different methods to merge multiresolution and multispectral data: landsat TM and Spot panchromatic. Photogramm Eng Remote Sens 57:295–303 Chen L, Liu X, Fu B (1999) Habitat fragmentation of giant pandas in Wolong Nature Reserve. Acta Ecol Sinica 19(3):291–297 (in Chinese) Cramariuc B (1995) Image segmentation by component labelling. In: Proceedings of international conference on digital signal processing, Limassol, Cyprus Fang Q, Wu J, Zhu J et al (2011) A review of random forest methods. J Stat Inf 26(3):32–38 (in Chinese) Fang R (2007) Support vector machine theory and application analysis. China Electric Power Press, Beijing (in Chinese) Ge Z (2007) Neural network theory and MATLAB R2007 implementation. Electronic Industry Press, Beijing (in Chinese) Gougeon FA (1995) A crown-following approach to the automatic delineation of individual tree crowns in high spatial resolution aerial images. Can J Remote Sens 21(3):274–284 Gougeon FA (2014) A crown following approach to the automatic delineation of individual tree crowns in high spatial resolution aerial images. Canad J Remote Sens J Canad De Télédétection 21(3):274–284 Grinnell J (1917) Field tests of theories concerning distributional control. The American Naturalist, 51(602):115-128 Guo J, Yang H (1999) Study on spatial distribution and dynamics of landscape elements in Guandi Mountain forest region. Acta Ecol Sinica 19(4):468–473 (in Chinese)

References

131

Hu J, Xia L (1985) Giant pandas in Wolong. Sichuan Science and Technology Publishing, Sichuan (in Chinese) Hu J (2001) Giant panda research. Science and Technology Education Press, Shanghai (in Chinese) Jin X (2012) Establishment and application of monitoring parameter system for giant panda living status in Qinling mountains. Master’s thesis of Beijing Forestry University, Beijing (in Chinese) Jing L, Cheng Q (2009) Two inprovement schemes of PAN modulation fusion methods for spectral distortion minimization. Int J Remote Sens 30(8):2119–2131 Jing L, Cheng Q (2011) An image fusion method for misaligned panchromatic and multispectral data. Int J Remote Sens 32(4):1125–1137 Jing L, Hu B, Li J, Noland T (2012a) Automated delineation of individual tree crowns from LiDAR data by multiscale analysis and segmentation. Photogramm Eng Remote Sens 78(12):1275–1284 Jing L, Hu B, Noland T, Li J (2012b) An individual tree crown delineation method based on multi-scale segmentation of imagery. ISPRS J Photogramm Remote Sens 70:88–98 Laben CA, Brower BV (1998) Process for enhancing the spatial resolution of multispectral imagery using pan-sharpening. US Patent 6(11):875 Larson MA, Thompson FR, Millspaugh JJ et al (2004) Linking population viability, habitat suitability, and landscape simulation models for conservation planning. Ecol Model 180(1):103–118 Li D, Ren B, Hu J et al (2012) Impact of snow storms on habitat and death of Yunnan snubnosed monkeys in the baimaxueshan nature reserve, Yunnan, China. International Scholarly Research Notices Li H (2013) Research on remote sensing image fusion algorithm based on contourlet transform. Master dissertation of Shandong Normal University, Jinan. (in Chinese) Linderman M, Bearer S, Li A et al (2005) The effects of understory bamboo on broad scale estimates of giant panda habitat. Biol Cons 121(3):383–390 Liu X, Toxopeus AG, Skidmore AK et al (2005) Giant panda habitat selection in foping nature reserve, China. J Wildlife Manag 69(4):1623–1632 Liu XH (2006) Assessment of giant panda habitat based on integration of expert system and neural network. Chinese J Appl Ecol 17(3):438–443 (in Chinese) Liu Z (2001) Application of analytic hierarchy process in aeromagnetic anomaly classification and prospecting prediction in central and western Tibetan Plateau. Geophys Geochem Explor 25(3):161–168 (in Chinese) Lunetta RS, Knight JF, Ediriwickrema J et al (2006) Landcover change detection using multitemporal MODIS NDVI data. Remote Sens Environ 105(2):142–154 Meyer F, Beucher S (1990) Morphological segmentation. J vis Commun Image Represent 1(1):21– 46 Meyer F (1994) Topographic distance and watershed lines. Signal Process 38(94):113–125 Moga AN, Gabbouj M (1995) A parallel watershed algorithm based on the shortest path computation. In: Workshop on parallel programming and applications, Linkoping, Sweden Munechika CK, Warnick JS, Salvaggio C et al (1993) Resolution enhancement of multispectral image data to improve classification accuracy. Photogramm Eng Remote Sens 59(1):67–72 Otazu X, González-Audícana M, Fors O et al (2005) Introduction of sensor spectral response into image fusion methods. Application to wavelet based methods. IEEE Trans Geosci Remote Sens 43(10):2376–2385 Ouyang Z, Liu J (2001) An assessment of giant panda habitat in Wolong Nature Reserve. Acta Ecol Sinica 21(11):1869–1874 (in Chinese) Ouyang Z, Zhang H, Tan Y (1995) Application of GIS in habitat evaluation of giant panda in Wolong Nature Reserve. Man Biosp (3):13–18 (in Chinese) Pouliot DA, King DJ, Bell FW et al (2002) Automated tree crown detection and delineation in high-resolution digital camera imagery of coniferous forest regeneration. Remote Sens Environ 82(2–3):322–334 Pradines D (1986) Improving SPOT image size and multispectral resolution. In: Proceedings of the SPIE. Earth remote sensing using the landsat thematic mapper and SPOT systems, Innsbruck, Austria

132

3 Technology and Method of Detailed Information Extraction of Animal …

Quinlan JR (1992) C4. 5: programs for machine learning. Morgan Kaufmann Publishers Inc., San Francisco Saaty TL (1988) What is the analytic hierarchy process. Springer, Berlin, pp 109–121 Shalaby A, Tateishi R (2007) Remote sensing and GIS for mapping and monitoring land cover and landuse changes in the Northwestern coastal zone of Egypt. Appl Geogr 27(1):28–41 Shao G, Lin J, Zhang W (2014) A fast image fusion algorithm based on lifting wavelet. Electro-optic Technol Appl 29(4):39–44 (in Chinese) Song J, Wang X, Liao Y et al (2014) An improved neural network for regional giant panda habitat suitability mapping: a case study in Ya’an prefecture. Sustainability 6(7):4059–4076 Teggi S, Cecchi R, Serafini F (2003) TM and IRS-1C-PAN data fusion using multiresolution decomposition methods based on the ‘a tròus’ algorithm. Int J Remote Sens 24(6):1287–1301 Urquhart RB (1982) Graph theoretical clustering based on limited neighborhood sets. Pattern Recogn 15(3):173–187 Vapnic V (1995) The nature of statistical learning theory. Springer, Berlin Vina A, Bearer S, Chen X et al (2007) Temporal changes in giant panda habitat connectivity across boundaries of Wolong Nature Reserve, China. Ecol Appl 17(4):1019–1030 Vincent LSP (1991) Watersheds in digital spaces: An efficient algorithm based on immersion simulations. IEEE Trans Pattern Anal Mach Intell 13(6):583–598 Wang W (1995) Principle of artificial neural network. Beihang University Press, Beijing (in Chinese) Weiss A (2001) Topographic position and landforms analysis. In Poster presentation, ESRI user conference, San Diego, CA (Vol. 200) Xiao P, Feng X (2012) High resolution remote sensing image segmentation and information extraction. Science Press, Beijing (in Chinese) Xiong Y, Wu J (2006) Research on object-oriented method of urban green space information extraction. J East China Norm Univ (Nat Sci) (4):84–90 (in Chinese) Xu W, Dong R, Wang X et al (2009) Impact of China’s May 12 earthquake on giant panda habitat in Wenchuan County. J Appl Remote Sens 3(1):031655 Yang XH, Jiao LC (2008) Fusion algorithm for remote sensing images based on nonsubsampled contourlet transform. Acta Autom Sinica 34(3):274–281 Yang X, Meng S, Yong R (1998) Study on the environmental ecology of foping giant panda (II)selection of summer habitat. J Northwest Univ (Nat Sci Edn) 28(4):348–353 (in Chinese) Yang X, Yong R (1997) Study on the environmental ecology of foping giant panda (I)-distribution of habitat plant community in summer. J Northwest Univ (Nat Sci Edn) 27(6):509–514 (in Chinese) Yi Y, Cheng X, Zhou J (2013) Progress in habitat suitability evaluation methods. Ecol Environ Sci 22(5):887–893 (in Chinese) Yu H, Zhao Y, Ma Y et al (2011) A remote sensing based analysis on the impact of Wenchuan Earthquake on the core value of World Nature Heritage Sichuan giant panda sanctuary. J Mt Sci 8(3):458–465 Zahn CT (1970) Graph theoretical methods for detecting and describing gestalt clusters. IEEE Trans Comput 20(1):68–86 Zhang J, Hull V, Xu W et al (2011) Impact of the 2008 Wenchuan earthquake on biodiversity and giant panda habitat in Wolong Nature Reserve, China. Ecol Res 26(3):523–531 Zhang J (2010) Multi-source remote sensing data fusion: status and trends. Int J Image Data Fusion 1(1):5–24 Zhang W (2014) Habitat quality assessment of giant panda in Wanglang Nature Reserve. Master’s thesis of Beijing Forestry University, Beijing (in Chinese) Zhang Z, Hu J (2002) Analysis on the viability of giant panda population in Tangjiahe. Acta Ecol Sinica 22(7):990–998 (in Chinese) Zhao Y, Wang P (2013) A review of remote sensing image fusion methods. Silicon Valley (2):36, 65 (in Chinese)

References

133

Zheng W, Xu Y, Liao L et al (2012) Effect of the Wenchuan earthquake on habitat use patterns of the giant panda in the Minshan Mountains, southwestern China. Biol Cons 145(1):241–245 Zhenglin W (2014) Research on wavelet image fusion algorithm based on contrast. Laser Infrared 9:1042–1044 (in Chinese)

Chapter 4

Analysis of Changes in Key Environmental Parameters of Land Surface Features in Giant Panda Habitat

4.1 Basic Geographic Overview of Giant Panda Habitat The giant panda is known in ancient times as the brave (Shi Jing), in the south as the tapir (Erya), and in local names such as the flower bear, white bear, bamboo bear, and iron-eating beast, and in modern times as the cat bear, later evolving into the giant panda (Wang 2008). The giant panda has existed on earth for a long time and is known as a “living fossil” in the animal kingdom. It is not only a national treasure endangered and rare species in China, but also a natural and historical heritage shared by the whole world, and its reputation, influence, survival and conservation status are of great concern to the international community. The International Union for Conservation of Nature and Natural Resources (IUCN) has listed the giant panda as an endangered species; the Convention on International Trade in Endangered Species of Wild Fauna and Flora (Convention on International Trade in Endangered Species of Wild Fauna and Flora, CITES) prohibits all international trade in giant pandas and their products. During the Pleistocene (From 2,588,000 years ago to 11,700 years ago), the giant panda was once widely distributed in 16 provinces (cities) in eastern China, and stretched southward to Myanmar and northern Vietnam. During the recorded historical period, there are still some remaining distribution sites in Yunnan, Guizhou, Hunan, Hubei and Henan. Hunan, Hubei and Henan still have some remaining distribution sites. More than 1000 years ago, the range of giant pandas was greatly reduced the range of giant pandas has been greatly reduced. In the past one or two hundred years, due to increasingly intense human resource exploitation and economic activities, the range of giant pandas has shrunk even more sharply and formed small, separated habitats, greatly increasing the risk of species extinction. This has greatly increased the risk of species extinction. In the mid-19th century, giant pandas were still found in eastern Sichuan, northern Hubei and western Hunan. In the mid-19th century, giant pandas were still distributed in eastern Sichuan, northern Hubei and western Hunan, but now they are extinct in these areas (Shen 2002). According to the results of the third national giant panda field population survey, the main distribution © Science Press 2023 X. Wang et al., Spatial Observation of Giant Panda Habitat, https://doi.org/10.1007/978-981-19-8794-6_4

135

136

4 Analysis of Changes in Key Environmental Parameters of Land Surface …

Fig. 4.1 Distribution of giant panda habitat

of giant panda habitat is shown in Fig. 4.1. As shown in Fig. 4.1, the present-day giant panda habitat is mainly distributed in the southwestern part of Shaanxi Province, central Sichuan Province and the border between Sichuan and Gansu Provinces. The habitats are mainly located in southwestern Shaanxi Province, central Sichuan Province and the border between Sichuan and Gansu Provinces. These habitats are mainly concentrated in the humid areas of the Minjiang and Jialing River systems (Fig. 4.2). The altitude of the distribution of giant panda habitat is related to the altitude where human activities occur, generally at the altitude of human disturbance activities above the elevation of human disturbance activities; however, because some habitat resources have to be shared with humans, there is a considerable overlap with the elevation of human disturbance activities (Zhou 2005). Figure 4.2 shows the topographic features of the giant panda habitat and its surrounding area as shown in the 30 m spatial resolution digital elevation model. Spatial overlay analysis of DEM and habitat vector boundary data using ArcGIS software showed that giant panda habitats were mainly distributed at 1300–3800 m above sea level, which is consistent with the results of existing surveys (Zhang and Hu 2000; Ouyang et al. 2001) The results show that giant panda habitats are mainly located at 1300–3800 m above sea level, which is consistent with the results of existing surveys (Zhang and Hu 2000; Xu and Ouyang 2006; Jin 2012; Han 2013).

4.1 Basic Geographic Overview of Giant Panda Habitat

137

Fig. 4.2 Digital elevation model of giant panda habitat

Load the 30 m spatial resolution DEM data of giant panda habitat in ArcGIS, and then generate the slope direction map. By superimposing the vector boundary data of the giant panda habitat and the generated slope data, the slope distribution of the giant panda habitat was analyzed. The slope distribution of the giant panda habitat was analyzed by overlaying the vector boundary data and the generated slope data. The main implementation steps are as follows (Nie et al. 2013): (i) Generate orientation data from DEM; (ii) Reclassify the habitat orientation into new raster data (referred to as reaspect), and the corresponding raster values are converted to new raster values, e.g., reaspect values equal to 1 for orientation values of −1 (i.e., slope of 0), reaspect values equal to 1 for orientation values of 0–22.5 and 337.5–360 (i.e., north orientation) are converted to reaspect values equal to 2, and orientation values of 22.5–67.5 (i.e., northeast orientation) are converted to reaspect values equal to 1. (i.e., facing northeast) is converted to a reaspect value equal to 3, a reaspect value of 67.5–112.5 (i.e., facing east) is converted to a reaspect value equal to 4, and so on. Value is equal to 4, and so on. (iii) Convert the output reaspect raster data into vector data, and export the data of different orientations as separate vector layers to

138

4 Analysis of Changes in Key Environmental Parameters of Land Surface …

Fig. 4.3 Statistics on the proportion of slope distribution of giant panda habitat

get the vector map of habitat area of different orientations; (iv) Totalize the area of different orientations to get the vector map. As a separate vector layer, we can get the vector map of habitat area in different orientations. The distribution ratio of habitats with different orientations can be obtained as shown in Fig. 4.3. From Fig. 4.3, we can see that the proportion of habitat area of giant panda The proportion of habitats facing east and southeast is the largest; the proportion of habitats facing west (west, southwest and northwest) is smaller. Figure 4.4 show the land use/land cover (LULC) types in and around the giant panda habitat. The data were obtained from the Global Change Parameter Database (http://globalchange.nsdc.cn), using ESA GlobCover data products with a spatial resolution of 300 m (ESA GlobCover). Soil type data were obtained from the World Soil Information Center (ISRIC, http://www.isric.org) at a scale of 1:10,000,001. Spatial overlay analysis of LULC data and habitat vector boundary data using ArcGIS software showed that the main LULC types in giant panda habitat were closed needle-leaved evergreen forest and closed to open broadleaved evergreen or semi-deciduous forest; the main soil types were haplic luvisols.

4.2 Analysis of Climate Change in Giant Panda Habitat in Recent Years According to the Chinese ground climate data set released by the China Meteorological Data Sharing Service (http://cdc.nmic.cn), there are 10 meteorological stations in the area where the giant panda habitat is located and its surrounding area. By

4.2 Analysis of Climate Change in Giant Panda Habitat in Recent Years

139

Fig. 4.4 Map of land cover/use in and around giant panda habitat

analyzing the ground meteorological observation data of these 10 meteorological stations for the past 60 years (1953–2012), we can study the seasonal and annual characteristics of the long time scale climate (temperature and precipitation) changes in the giant panda habitat and its surrounding areas. The seasons are defined as follows: spring (March–May), summer (June–August), autumn (September–November), and winter (December–February). The climatic rate of change of meteorological elements is generally expressed by a linear equation: Y = a0 + a1 t, where Y is the meteorological element; a0 is the constant term; t is the time; a1 is the linear trend term, and a1 × 10 is the climatic tendency rate of the meteorological element every 10 years (Nie et al. 2012).

140

4 Analysis of Changes in Key Environmental Parameters of Land Surface …

Table 4.1 Distribution information of meteorological stations around the giant panda habitat Zone station number

Station name

Latitude (N)

Longitude (E)

Elevation/m

Start date

56,079

Ruoergai

33°35'

102°58'

3440

1957–01

56,182

Songpan

32°39'

103°34'

2851

1951–01

56,188

Dujiangyan

31°00'

103°40'

699

1954–07

56,193

Pngwu

32°25'

104°31'

893

1951–10

101°58'

56,374

Kangding

30°03'

2616

1951–11

56,376

Hanyuan

29°21'

102°41'

796

1951–01

56,385

Emeishan

29°31'

103°20'

8047

1951–01

102°31'

56,475

Yuexi

28°39'

1660

1953–04

56,485

Leibo

28°16'

103°35'

1256

1952–07

57,134

Foping

33°31'

107°59'

827

1957–01

4.2.1 Temperature Change Characteristics From Table 4.1 and Fig. 4.5, it can be seen that the spatial distribution of meteorological stations is relatively uneven, and the elevation also ranges from a lower 699 m (Dujiangyan) to a higher 8047 m (Emeishan). Therefore, in order to prevent unnecessary errors or even mistakes caused by inappropriate spatial interpolation, the temperature data of each station were analyzed statistically directly in the study area for the study of temperature change. The average value of the data over many years was used to replace the missing data at some stations or at a certain time. The statistical results of the seasonal and annual mean temperature variation of the 10 meteorological stations around the study area are shown in Table 4.2. From the table, it can be seen that most of the stations show a non-significant increase in temperature at all seasonal and inter-annual scales.

4.2.2 Precipitation Variation Characteristics The changes in seasonal and average annual precipitation at each site from 1953 to 2012 are shown in Table 4.3. It can be seen that the annual precipitation and the precipitation in summer, autumn and winter decreased at most stations. The arithmetic averages of precipitation at the 10 weather stations were calculated to represent the precipitation variation in the study area. Figure 4.6 shows the annual precipitation and seasonal precipitation trends in the giant panda habitat and its surrounding areas from 1953 to 2012. According to Fig. 4.6, the annual precipitation in the region has been decreasing slightly at a rate of −17.7 mm/10a (p < 0.01) over the past 60 years, and the interannual variation of the average annual precipitation fluctuates greatly. The maximum value of annual average precipitation is 1283 mm in 1954 and the minimum value

4.2 Analysis of Climate Change in Giant Panda Habitat in Recent Years

141

Fig. 4.5 Distribution of giant pandas and meteorological stations Table 4.2 Statistics of the linear rate of change of the average temperature by season and year at each station (1953–2012) (Unit: °C/a) Site

Inter-annual

Spring

Summer

Autumn

Winter

Ruoergai

0.03

0.02

0.03

0.03

0.05

Songpan

0.02

0.01

0.02

0.02

0.03

Dujiangyan

0.01

0.01

0.01

0.02

0.02

Pingwu

0.01

0.01

0.01

0.01

0.01

Kangding

0.01

−0.01

0.01

0.02

0.02

−0.01

−0.02

−0.01

−0.01

0.01

0.01

0.01

0.02

0.02

Yuexi

−0.01

−0.02

−0.01

0.01

0

Leibo

0.03

0.03

0.03

0.05

0.04

Foping

0.03

0.03

0.02

0.03

0.03

Hanyuan Emeishan

0

142

4 Analysis of Changes in Key Environmental Parameters of Land Surface …

Table 4.3 Statistics of linear variation rate of precipitation by season and year at each station (1953–2012) (unit: mm/a) Site

Inter-annual

Spring

Ruoergai

−0.07

−0.19

Songpan

−0.40

0.22

Dujiangyan

−4.27

−0.40

Pingwu

Summer

Autumn

Winter

0.60

0.52

0.05

−0.33

−0.31

0.02

−2.89

−1.04

0.11

−3.47

−0.15

−2.01

−0.43

0.05

Kangding

1.20

0.65

0.89

−0.17

−0.12

Hanyuan

1.48

1.13

−0.63

−0.22

−0.05

Emeishan

−9.92

−0.74

−5.52

−2.71

−0.51

Yuexi

−1.11

0.73

−0.28

−1.26

−0.11

Leibo

−1.52

−0.15

−0.64

−0.76

0.03

Foping

−0.80

−0.64

−0.11

−0.05

−0.01

is 774 mm in 1965. Precipitation decreases to some extent in all seasons, but the decreasing trend is stronger in summer, which also has the greatest impact on the annual variation.

4.3 Changes in Water Resources in and Around Giant Panda Habitats Water is the source of all life on Earth, and monitoring changes in the terrestrial environment of the giant panda habitat necessitates understanding the changes in water resources in the region. In March 2002, NASA and DLR jointly launched the Gravity Recovery and Climate Experiment (GRACE) gravity satellite to detect the average time variation of the Earth’s gravity field. On the Earth’s land, the monthto-month time variation of the Earth’s gravity field is mainly caused by the variation of water mass on the land, and therefore the GRACE satellite’s detected Earth’s gravity field data can be used to invert the variation of terrestrial water storage. The so-called terrestrial water storage is the sum of all forms of water on land, including surface water, soil water, groundwater, snow and ice, etc., and its changes represent the overall information of the changes of terrestrial water resources (Wahr et al. 1998; Tapley et al. 2004; Syed et al. 2008). Therefore, GRACE gravity satellite data were used to monitor water resources changes in giant panda habitat and its surrounding areas in recent years. The principle of using GRACE time-varying gravity field data to invert changes in terrestrial water reserves is based on the uptake of changes in the Earth’s surface mass into the satellite orbit. Tidal and non-tidal atmospheric and oceanic influences, as well as post-ice rebound effects, have been deducted

4.3 Changes in Water Resources in and Around Giant Panda Habitats

143

Fig. 4.6 Trends of precipitation in and around giant panda habitats from 1953 to 2012

during data processing, and thus the GRACE time-varying gravity field reflects nonatmospheric, non-oceanic mass changes, i.e., mainly changes in terrestrial water storage (Wahr et al. 1998; Xu and Zhang 2013). The Earth’s gravity field monitored by GRACE satellites is expressed in the form of spherical harmonic coefficients of a certain order (e.g., 60th order, 90th order, etc.). At present, the GRACE Earth gravity field month-by-month data are mainly provided by the Jet Propulsion Laboratory (JPL) of NASA, the Geo Forschungs Zentrum (GFZ) of Germany, the Center for Space Research at the University of Texas at Austin, CSR, Delft Institute of Earth Observation and Space Systems (DEOS), Space Geodesy Research Group (GR), France. (Jiang et al. 2014), and the Space Geodesy Research Group (GRGS). The data version is continuously updated, and the latest data version is GRACE RL05, which has been distributed since 2012. The rationale and methodology of regional terrestrial water storage based on GRACE time-varying gravity field data and the post-processing steps to eliminate data-related errors can be found in the following literature: (Swenson and Wahr 2002; Tapley et al. 2004; Wahr et al. 2004; Swenson and Wahr 2006; Chen et al. 2008; Landerer and Swenson 2012).

144

4 Analysis of Changes in Key Environmental Parameters of Land Surface …

The study uses a land water storage raster data product based on the inversion of GRACE RL05 60th order gravity field data provided by CSR from JPL (at http:// grace.jpl.nasa.gov) with a spatial resolution of 1° × 1°. During the data processing, the gravity field coefficients for each month were subtracted from the monthly average gravity field from January 2004 to December 2009 to obtain a gravity field distanceaverage variation series. Since the C20 term of GRACE cannot be obtained accurately, the results of satellite laser ranging are used to participate in the calculation. A debanding filter is used to reduce the “banding” errors in the north-south direction of the original GRACE data. Since the GRACE time-varying gravity field coefficients are only expanded to a finite order, not to infinity, there is inevitably some truncation error when calculating the water storage variation. Moreover, the error of GRACE gravity field coefficients increases with the increase of l. Therefore, the error caused by the higher-order terms in the calculation process is not negligible. In order to reduce the impact of the higher-order term errors on the results and to further reduce the spatial noise of the data information, a 200 km smoothing radius Gaussian filtering algorithm is used to smooth the data. The smoothed spherical harmonic parameters were then converted to raster data with a spatial resolution of 1° × 1°. To recover as much information as possible from the GRACE gravity field coefficients lost during truncation and filtering, the generated raster data were then multiplied by the rasterized ratio data set produced by Landerer and Swenson (2012) to finally obtain the water storage distance level time series, where the water storage is measured in centimeter equivalent water column height. Due to the low spatial resolution of the GRACE data, a study of water storage changes in the giant panda habitat and its surrounding areas was conducted using the boxed area shown in Fig. 4.1. The terrestrial water storage (expressed as equivalent water column height/cm) in the study area from January 2003 to July 2013 is shown in Fig. 4.7a. It can be seen from the figure that the terrestrial water storage in the study area shows a significant (p < 0.01) cyclic variation characteristic, and the sine fit results show an annual amplitude of 0.97 cm. Figure 4.7b shows the non-seasonal variation of water storage in the study area from January 2003 to July 2013, the signal of inter-annual cyclical variation has been deducted. In the summer of 2006, Sichuan Province experienced severe high temperature and dry weather, resulting in the most severe summer drought in Sichuan Province in more than 50 years (Pan and Liu 2006). It can also be seen from the figure that the average water storage in the study area decreased by about 8 cm in August 2006 and September 2006 compared to the average water storage in the same period, reaching the lowest value of terrestrial water storage in the past 10 years. In July 2012, heavy to very heavy rainfall occurred in Sichuan, resulting in flooding in some areas; the entire main stream section of the upper Yangtze River from Yibin to Cuntan in Chongqing exceeded the warning level, and the Three Gorges of the Yangtze River received the largest flood peak since the construction of the reservoir on July 24 (Ye et al. 2013). As can be seen in Fig. 4.7, the water storage in the study area reached the highest value in the past 10 years in July 2012. The spatial distribution of the annual rate of change in water storage for the last 10 years (2003–2012) was calculated at each raster point in the study area. From

References

145

Fig. 4.7 Month-by-month water storage variation from January to July in 2013

the result, it can be seen that the water resources in the giant panda habitat have not changed significantly in recent years.

References Chen Y, Schaffrin B, Shum CK (2008) Continental water storage changes from GRACE line-ofsight range acceleration measurements. In: Xu P, Liu J, Dermanis A (eds) Vi Hotine-Marussi symposium on theoretical and computational geodesy. Springer, New York Han W (2013) Habitat evaluation and restoration of giant pandas in Wolong Nature Reserve after the earthquake. Master’s thesis. Capital Normal University, Beijing (in Chinese) Jiang D, Wang J H, Huang Y et al (2014) The review of GRACE data applications in terrestrial hydrology monitoring. Adv Meteorol 12:758–767 Jin X (2012) Establishment of a monitoring parameter system for the survival status of giant pandas in the Qinling Mountains and its application. Beijing Forestry University, Beijing (in Chinese) Landerer FW, Swenson SC (2012) Accuracy of scaled GRACE terrestrial water storage estimates. Water Resour Res 48:W04531 Nie N, Zhang WC, Cai D (2012) Spatial and temporal climate changes and future trends in the Yarlung Tsangpo River basin from 1978 to 2009. Glac Permafr 34(1):64–71 (in Chinese) Nie N, Zhang ZJ, Zhang WC et al (2013) Remote sensing study of glacier system characteristics and analysis of typical glacier changes in the Yarlung Tsangpo River basin over the past 30a. Glac Permafr 35(3):541–552 (in Chinese) Ouyang ZY, Liu JG, Xiao H (2001) Habitat evaluation of giant pandas in Wolong Nature Reserve. J Ecol 21(11):1869–1874 (in Chinese) Pan JH, Liu XQ (2006) Analysis of rare high temperature and drought in Sichuan Province in 2006. Sichuan Meteorol 26(4):12–14 (in Chinese) Shen GZ, Li JQ, Ren YL et al (2002) Indicators of suitable habitat restoration for giant pandas. J Beijing Forest Univ 24(4):1–5 (in Chinese) Shen GZ (2002) Habitat restoration of giant pandas. Beijing Forestry University, Beijing (in Chinese) Swenson S, Wahr J (2002) Methods for inferring regional surface mass anomalies from gravity recovery and climate experiment (GRACE). Meas J Geophys Res Solid Earth 107(B9):Artn 2193

146

4 Analysis of Changes in Key Environmental Parameters of Land Surface …

Swenson S, Wahr J (2006) Post-processing removal of correlated errors in GRACE data. Geophys Res Lett 33(L08402):L08402 Syed TH, Famiglietti JS, Rodell M et al (2008) Analysis of terrestrial water storage changes from GRACE and GLDAS. Water Resour Res 44(2):Artn W02433 Tapley BD, Bettadpur S, Ries JC et al (2004) GRACE measurements of mass variability in the Earth system. Science 305(5683):503–505 Wahr J, Molenaar M, Bryan F (1998) Time variability of the Earth’s gravity field: hydrological and oceanic effects and their possible detection using GRACE. J Geophys Res Solid Earth 103(B12):30205–30229 Wahr J, Swenson S, Zlotnicki V et al (2004) Time variable gravity from GRACE: first results. Geophys Res Lett 31(11):L11501 Wang R (2008) Study on the evaluation of non-use value of giant panda habitat. Master’s thesis, Sichuan Agricultural University, Chengdu (in Chinese) Xu P, Zhang W (2013) GRACE inversions of recent changes in terrestrial water storage on the Qinghai-Tibet Plateau and the Yarlung Tsangpo River Basin. J Water Resour Water Eng 24(1):23– 29 (in Chinese) Xu W, Ouyang Z (2006) Habitat evaluation and conservation measures for giant pandas in the Daxiangling Mountain system. Biodiversity 14(3):223–231 (in Chinese) Ye D, Zhao S, Wang Y (2013) Review of major meteorological hazards in China in 2012. Disas Sci 28(3):128–132 (in Chinese) Zhang Z, Hu J (2000) Habitat selection of giant pandas. J Sichuan Norm Univ (Nat Sci Edn) 21(1):18–21 (in Chinese) Zhou J (2005) A preliminary study on the index system for evaluating the habitat of giant panda. J Cent South Forest College (3):39–44 (in Chinese)

Chapter 5

Spatial Observation and Assessment of Ecological Changes in Giant Panda Habitats

5.1 Spatial Observation and Evaluation of Habitat Ecological Landscape Diversity by Optical Remote Sensing Ecological diversity is a comprehensive indicator of the richness and uniformity of distribution of biological species inhabiting an ecosystem, i.e., the number of ecosystem components, the complexity of the structure of biological populations, the length of food chains, the complexity of food webs, the transformation of energy, and the number of ways of material cycling, reflecting the type of community structure, level of organization, stage of development, stability and habitat differences. Ecological stability refers to the ability of biological populations to recover after a period of time due to the feedback effect of biological populations when they encounter large changes in the ecological environment. Stability is reflected in the quantitative changes in the structure of biological populations, the high suitability of the ecological environment for the survival and development of biological populations within a certain range, and the basic balance of material and energy input and output of the ecosystem. Strong ecological diversity means that the ecosystem has many species of organisms, many trophic levels, long food chains, complex food webs, rich heritage gene pools, various organisms located at different trophic levels, and various organisms form intricate food webs through feeding relationships, and the system has many channels of energy, material and information input and output, which are intertwined, because the flow is large, fast and productive. Even if individual input and output pathways are destroyed, the system will ensure the normal operation of energy and information flow due to the mutualism between various species and the compensation and substitution of species with similar ecological status. Therefore, ecological diversity is one of the conditions to ensure ecological balance, and diversity is conducive to the stable and balanced development of ecosystems. In ecology, the quality structure of wildlife habitat is divided into three major influencing factors: food, water, and cover. Among them, plants in the habitat are not © Science Press 2023 X. Wang et al., Spatial Observation of Giant Panda Habitat, https://doi.org/10.1007/978-981-19-8794-6_5

147

148

5 Spatial Observation and Assessment of Ecological Changes in Giant …

only the food source of giant pandas, but also provide them with concealment and shelter, so the assessment of vegetation ecosystem is one of the important ways to understand the quality of life of giant pandas in the habitat. The exploitation, use and pollution of natural environmental resources by human activities have caused ecological changes, which are important causes of species decline and extinction, ecological landscape fragmentation and habitat destruction. Therefore, the diversity and changes in the landscape structure of graduate habitats and habitats are important for the biodiversity of species.

5.1.1 Study Area Overview 1. Location According to the third national survey report of giant pandas, the total number of giant pandas in China is about 1600, mainly distributed in the south of Qinling Mountains in Shaanxi, southwest and northwest of Sichuan and south of Gansu, with a total area of about 30,000 km2 . The actual habitat area is only about 20% of the total area, and is unevenly and discontinuously distributed in blocks. Giant pandas have a unique lifestyle and are only distributed in specific forest environments. In recent years, with changes in natural conditions such as climate, soil and vegetation, as well as the influence of human factors such as deforestation, deforestation and illegal hunting, the ecological environment of the giant panda habitat is deteriorating and the survival of the giant panda is threatened. Wolong Nature Reserve is located in the southwest of Wenchuan County, Sichuan Province, on the southeast slope of the Qionglai Mountain System and the upper reaches of the Min River, with a latitude and longitude range of 102°52' –103°24' E, 30°45' –31°25' N, spanning 60 km from east to west and 63 km from north to south, with a total area of 2000 km2 , and is one of the largest nature reserves in China. It is one of the largest nature reserves in China, mainly protecting rare and endangered plants and animals such as giant pandas, golden monkeys, bull antelopes, dove trees, light-leaved dove trees, water cyprus, Sichuan red fir, and the whole alpine ecosystem. Wolong is located in the high mountain valley area in the transition from the Sichuan Basin to the Qinghai-Tibet Plateau, and the terrain drops sharply from northwest to southeast. Due to the uplifting effect of the neotectonic movement and the erosion effect of rivers, the mountains and valleys in the area are high and deep, and the relative height difference is huge. The elevation of Sigun Mountain in the northwest is as high as 6250 m, while the elevation of Mujiangping in the east is only 1150 m, and the distance between the two places is 48 km, with a relative height difference of 5100 m. The main rivers in the area are Pijiao River, Zhenghe River, Xihe River and Zhonghe River, with many tributaries at various levels on both sides of the river, forming a dendritic water system, and the river valley is “V” shaped with a large drop. With rich hydropower resources. The area belongs to the climate zone of Qinghai-Tibet Plateau, with cool and rainy summers and cold and dry winters. The

5.1 Spatial Observation and Evaluation of Habitat Ecological Landscape …

149

average annual temperature is 9.8 °C, the lowest temperature (January) is −1.7 °C, the highest temperature (July) is 17 °C, the annual sunshine hours are 926.7 h, the annual precipitation is 1800 mm, the evaporation is 873.9 mm, and the relative humidity is over 80%. The ecological type of forest landscape in Wolong National Nature Reserve is mainly warm temperate deciduous broad-leaved forest, but due to long-term exploitation, the distribution of forest landscape types in the area is fragmented, the area ratio is uneven, and the pattern and process changes are complicated, especially in recent years, the expansion of urban area, the rapid development of economic development and tourism has aggravated the degree of habitat fragmentation, and the Caopo, which is adjacent to the northeast of Wolong Reserve, has a similar habitat landscape. The habitat landscape of the Caopo Nature Reserve in the northeastern part of Wolong Reserve is similar, so it is considered together in the study. Remote sensing images of the study area in 1994 and 2007 are shown in Fig. 5.1. 2. Vertical zoning of vegetation The Wolong Nature Reserve covers 118,000 hm2 of forest, accounting for 56.7% of the total area of the reserve, and 30,400 hm2 of scrub and meadow, and the complex and variable natural conditions have resulted in a diversity of plant species and communities. The vegetation in the study area can be divided into six types according to the vertical zoning characteristics (Fig. 5.2).

(a) 1994 Fig. 5.1 TM remote sensing image of the study area

(b) 2007

150

5 Spatial Observation and Assessment of Ecological Changes in Giant …

Fig. 5.2 Topography of the Caopo area in Wolong, Sichuan

(1) Evergreen broad-leaved forest: distributed in the section below 1600 m above sea level, the establishment species mainly include camphor, beech, camellia and holly plants. There are a small number of deciduous broad-leaved trees in the forest, such as birch, maple and walnut, and there are large areas of white bamboos, oleander bamboos and abaca bamboos in the understory. (2) Evergreen deciduous broad-leaved mixed forests: distributed at an altitude of 1600–2000 m, with evergreen species such as beech, camphor, deciduous species such as birch, walnut, maple, etc., accompanied by rare ancient relict plants such as dove, dove, water cyanotype, collar spring, etc. The understory of the forest is dominated by bamboos, with obvious seasonal changes in vegetation appearance, dark green and soft green in spring and summer, mixed with green, yellow, red and brown in autumn. The vegetation changes seasonally, with dark green and soft green in spring and summer, mixed with green, yellow,

5.1 Spatial Observation and Evaluation of Habitat Ecological Landscape …

(3)

(4)

(5)

(6)

151

red and brown in autumn, and only a small amount of green in the forest canopy in winter in a white world. Needle and broad-leaved mixed forests: distributed at altitudes from 2000 to 2600 m, with broad-leaved species such as red birch, maple, Tibetan bristlecone hazel, linden, etc., and coniferous species such as hemlock, Sichuan red fir, pine, etc. The understory of the forest is widely distributed with bamboo, and locally there are big arrow bamboo and cold arrow bamboo, with significant seasonal changes in vegetation appearance, green in spring and summer, and colorful in late autumn and early winter. Coniferous forest: It is distributed at an altitude of 2600–3700 m, and the established species include Mai Hang fir, many kinds of fir, Fangzhi cypress, Sichuan red fir, etc. There is a large area of cold arrow bamboo in the understory, accounting for about 50% of the total area of bamboo in the region, and there are also big arrow bamboo and Huaxi arrow bamboo in the local area, and the vegetation appearance is dark green, and the seasonal change is not obvious. alpine scrub and alpine meadow: distributed at an altitude of 3700–4400 m, the hardy scrub is dominated by Zingiber officinale, Oxalis willow, fine branch embroidery daisy, Huaxi silver dewberry and incense cypress, and the alpine meadow has: weed-like meadow dominated by Polygonum pearlum, grass meadow dominated by fescue, and sedge meadow dominated by dwarf artemisia. Sparse vegetation zone of alpine rocky beach: distributed at an altitude of 4400– 5000 m, mainly composed of hairy, fleshy dwarf herbaceous plants, such as many kinds of phoenix hair daisy, many kinds of saxifrage, many kinds of rhodiola, fleabane, dotted plum, in addition to a small number of lichens and mosses.

The study area is divided into six zones according to altitude, and the corresponding forest landscape types are: evergreen broad-leaved forest (624– 1600 m), mixed coniferous forest (1600–2600 m), coniferous forest (2600–3600 m), scrub(3600–3900 m), meadow (3900–4400 m), rhyolite beach and snow (4400– 5973 m). The third national giant panda survey report shows that the main area suitable for giant pandas is the gently sloping area, i.e., the area with elevation between 2400 and 3500 m. The topographical map of the study area is shown in Fig. 5.2.

5.1.2 Research Methodology 1. Landscape pattern analysis method Using RS and GIS technologies, combined with landscape pattern analysis methods, we analyze the landscape pattern characteristics and changes in the study area at two levels: patch type and landscape, providing a basis for ecological construction and protection of the landscape, pollution control, rational adjustment of the landscape structure, and improving the productivity and stability of the landscape.

152

5 Spatial Observation and Assessment of Ecological Changes in Giant …

Landscape pattern includes the type, number and spatial distribution and configuration of landscape components, which is a concrete manifestation of landscape spatial structure and an important factor in landscape function and dynamics. Organisms including humans live in habitats with different scale spatial patterns, and these spatial patterns interact with the cognition and behavior of organisms, and then become the driving force of population regulation and community succession. In general, population dynamics, biodiversity and ecosystem processes are inevitably constrained or influenced by the spatial patterns of the landscape. Landscape pattern analysis, as the basic research content of landscape ecology, can quantitatively analyze the spatial distribution characteristics of landscape components and is the basis for further research on landscape functions and dynamics. 1) The significance of studying landscape diversity Ecological diversity is a condition for ecological balance, and the vegetation in the habitat is not only a source of food for giant pandas, but also provides them with shelter, so the assessment of vegetation ecosystem is one of the important ways to understand the quality of life of giant pandas in the habitat. The exploitation, use and pollution of natural resources by human activities have caused changes in the ecological environment and are important causes of species decline and extinction, ecological landscape fragmentation and habitat destruction. The study used multi-temporal, multiple remote sensing and spatial geographic data, combined with indices and spatial analysis methods of landscape ecology, to compare the spatial change characteristics of habitat and habitat landscape diversity from 1997 to 2007, and analyze the structural characteristics, change development and impact of vegetation cover in giant panda habitat from the perspective of landscape ecology (Table 5.1). 2) Landscape diversity index selection and meaning The landscape diversity index is mainly divided into 3 categories: landscape type diversity, patch diversity and pattern diversity. Landscape type diversity refers to the richness and complexity in landscape types, often considering the number of different landscape types in the landscape and their proportion of the area, and the diversity of landscape structure mainly studies landscape heterogeneity, connectivity, spatial correlation, patchiness, porosity, contrast, landscape grain level, structure, proximity, patch size, probability distribution. The representative indexes are: diversity index, dominance, uniformity and richness index. Patch diversity refers to the diversity and complexity of the number, size and shape of patches in the landscape, and mainly identifies the proportion and distribution abundance of patch types, the landscape type of composite patches, the community structure of population distribution (richness, endemic species). Representative indicators include the number of patches, area, shape, fragmentation, number of sub-dimensions, etc.

Substrate and soil variability, slope and aspect, plant biomass and appearance characteristics, foliage density and stratification, vertical blockiness, canopy openness and clearance, species abundance, density and distribution of key natural features and elements

Ecosystem diversity Identification of relative abundance, frequency, enrichment, evenness, distribution ratios of populations, endemism, exotic, threatened, and endangered species, dominance-diversity curves, proportion of life types, similarity coefficients, C3–C4 plant species ratios

Structure Landscape heterogeneity, connectivity, spatial correlation, patchiness, porosity, contrast, landscape grain size, tectonics, proximity, patch size, probability distribution, and edge-to-area ratio

Composition

Landscape diversity Identification of the proportion and distribution abundance of patch (habitat) types, landscape types of composite patches, group structure of population distribution (richness, endemism)

Level

Biomass, resource productivity, herbivores, parasitoids and capture rates, species invasion and regional extinction rates, changes in patch dynamics (small-scale disturbances), nutrient cycling rates, human invasion rates and intensities

Disturbance processes (extent, frequency or feedback cycle, intensity, predictability, severity, seasonality), nutrient cycling rates, energy flow rates, patch stability and cycles of change, erosion rates, geomorphological and hydrological processes, land use directions

Function

Table 5.1 Methods of biodiversity survey, monitoring and evaluation indicators at two levels

Aerial and other remote sensing, ground-based observations, time series analysis, natural habitat determination and resource surveys, habitat suitability indices, field observations, censuses and species inventories, capture and other field survey methods, mathematical parameter simulations (diversity indices, heterogeneity indices, stratified dispersal, organism assemblage types)

Aerial, satellite and other remote sensing and GIS data, time series analysis, spatial statistical analysis, mathematical parametric simulation (landscape pattern, heterogeneity, connectivity, edge effects, autocorrelation, fractional dimensional analysis)

Investigation and monitoring tools and methods

5.1 Spatial Observation and Evaluation of Habitat Ecological Landscape … 153

154

5 Spatial Observation and Assessment of Ecological Changes in Giant …

In order to compare the changes that occurred in the landscape structure of the study area during 8 years, the study chose to calculate the shape index and the number of sub-dimensions of the patches, on the basis of which the Shannon diversity index and the evenness index were calculated. (1) Shape index, indicating the shape characteristics of the plaque. p/4 shape = √ s

(5.1)

where p is the perimeter of the plaque; s is the area of the plaque. (2) The number of sub-dimensions, indicating the shape complexity of the patches. fractal =

ln p/4 √ √ = s log p/4 ln s

(5.2)

(3) Shannon’s diversity index (SHDI), indicating the richness diversity of patch types. SHDI = −

m Σ

P log2 Pt

(5.3)

i=1

(4) Shannon’s evenness index (SHEI), indicating the degree of uniformity in the distribution of patch types. Σm SHEI = −

P log2 Pt log2 m

i=1

(5.4)

The study used the TM remote sensing image data of 1994 and 2007 with 30 m resolution as the basis of landscape classification (TM_1994, TM_2007), combined with the DEM data of 30 m resolution in the study area, and the WorldView remote sensing image (WV_2010) with 0.5 m resolution collected in December 2010 as the reference image for the classification results. TM_1994 and TM_2007 were geometrically and radiometrically corrected, and then classified according to the above forest landscape classification system, and the classification results were corrected by the reference images to obtain the forest landscape classification maps of Wolong Caopo Nature Reserve in 1994 and 2007. The workflow is shown in Fig. 5.3. 2. Forest cover classification methods 1) Classification method based on spatial features such as DEM, slope, texture (entropy) and slope direction Based on the DEM and TM remote sensing image data, spatial features such as vegetation index, texture entropy (entropy), slope (slope) and slope direction (aspect) were extracted respectively (Fig. 5.4) and added to the classification process.

5.1 Spatial Observation and Evaluation of Habitat Ecological Landscape …

155

Fig. 5.3 Flow chart of forest landscape classification

2) Multi-scale segmentation of remote sensing images Each forest land type has different patch scales and needs to be extracted at different segmentation scales. Therefore, we first need to perform multi-scale segmentation of the image to test whether various land types can be segmented into complete patches at different segmentation scales, and then record the segmentation scales corresponding to each land type, so that extraction can be performed at different segmentation levels. For example, the segmentation scale of cloud layer is about 14, bare land, road, etc. is about 18, and forest land type is about 20, and different parameters are chosen for the experiment until satisfactory results are obtained (Fig. 5.5). 3) Classification results Based on the multi-spatial feature classification method and multi-scale objectoriented classification method, TM_1994 and TM_2007 were classified and analyzed

156

5 Spatial Observation and Assessment of Ecological Changes in Giant …

Image

Entropy

DEM

NDVI

Slope

Aspect

Fig. 5.4 Flow chart of classification based on spatial features

Fig. 5.5 Object-oriented multi-scale segmentation method

respectively, and the forest cover classification map of the study area was obtained, which is shown in Fig. 5.6. 3. Forest cover change monitoring methods The spectral characteristics of a specific feature change constantly with seasonal and meteorological conditions, so the spectral characteristics of similar features on the before and after images may have large or small differences, especially the vegetation is greatly affected by seasonal and meteorological conditions, and the changes

5.1 Spatial Observation and Evaluation of Habitat Ecological Landscape …

157

Fig. 5.6 Forest cover classification map of Wolong Caipo study area

obtained by band calculation or spectral variation may contain a lot of false changes and noise, which is difficult to distinguish. In addition, due to the phenomenon of “foreign matter in the same spectrum”, many real change information will be missed due to phase subtraction, resulting in inaccurate change results. The study represents the change or not as a [0, 1] degree of ambiguity in the change extraction process. Firstly, the raster image is partitioned into vector polygons and the corresponding relational attribute table using a clustering algorithm, and two maps (assumed to be map 1 and map 2) are overlaid and analyzed on the basis of vector polygons, and either one of the maps (e.g., map 1) is selected as a benchmark for comparison, and the area ratio of the corresponding polygons in map1 that intersect or contain each polygon in map 2 is recorded to generate a new relational table on The new relationship table of map 2 is used to calculate the transfer probability matrix of the change, determine the threshold value of the change according to the actual situation of the study area, and finally detect the area where the change occurs (Table 5.2). Table 5.2 Classification transfer matrix for comparison 合计

Map 2 Map 1

1

2

…

c

1

P11

P12

…

P1c

P1t

2

P21

P22

…

P2c

P2t

Pc1

Pc2

…

Pt1

Pt2

… c Sum

Pcc

Pct

Ptc

1

158

5 Spatial Observation and Assessment of Ecological Changes in Giant …

1) Pixel-by-pixel extraction method A similarity value, denoted by Kappa, is calculated for the entire comparison graph during the comparison process. Carletta believes that when comparing a certain number of objects, it is appropriate to use the Kappa measure of similarity, which contains the K l and the “volume” of similarity, respectively, in terms of spatial location. K h expresses the number of occurrences of a particular type of grid cell on the whole map. And K l measures the configuration of spatial location relations of a certain type of grid. In Eq. (5.5), pij represents the probability of class i being classified as class j, P(A) is the calculated degree of consistency, and P(E) in Eq. (5.6) represents the expected value of P(A) calculated from the existing probability distribution. Kappa in Eq. (5.11) denotes the Kappa coefficient calculated by the pixel-by-pixel method. C Σ

P(A) =

pi j

(5.5)

pi T × pT i

(5.6)

i=1

P(E) =

c Σ i=1

P(max) =

c Σ

min( pi T , pT i )

(5.7)

i=1

Kh =

P(max) − P(E) 1 − P(E)

(5.8)

Kl =

P( A) − P(E) P(max) − P(E)

(5.9)

K = K h × Kl Kappa =

P( A) − P(E) 1 − P(E)

(5.10) (5.11)

2) Fuzzy set-based change extraction method Unlike the pixel-by-pixel method, a fuzzy vector approach is used in the study to describe the grid cells, and the influence of neighboring cells on the centroid is expressed as an affiliation function based on the fuzzy set. The advantage of this comparison method is that it is tolerant to spatial location errors. When comparing with this method, not only the image element being compared itself is considered, and the influence of the image element adjacent to that image element on it is quantified in the form of a decay function, and this similarity is expressed as a value between [0,

5.1 Spatial Observation and Evaluation of Habitat Ecological Landscape …

159

1], which is more suitable for describing the position change relationship of spatial data compared with the traditional method. The pixel-by-pixel comparison method in Fig. 5.7a cannot distinguish the effect of different locations of surrounding change elements and is very sensitive to the error of displacement. In Fig. 5.7b, the similarity Kappa is calculated by assigning a value of 0.5 to the decay function of directly adjacent image elements and 0 to the decay function of other neighboring image elements, so the comparison method is tolerant to small spatial displacements and small changes, and the magnitude of the changes is expressed numerically in the results. fuzzy Kappa = ⎛ Vfuzzy

Pfuzzy. A − Pfuzzy.E 1 − Pfuzzy.E

F1 = max(μ1,1 ∗ m 1 , μ1,2 ∗ m 2 , · · · , μ1,C ∗ m C )

⎜ F2 = max(μ2,1 ∗ m 1 , μ2,2 ∗ m 2 , · · · , μ2,C ∗ m C ) ⎜ =⎜ ⎜ .. ⎝.

(5.12) ⎞ ⎟ ⎟ ⎟ ⎟ ⎠

(5.13)

FC = max(μ N ,1 ∗ m 1 , μ N ,2 ∗ m 2 , · · · , μ N ,C ∗ m C ) Equation (5.12) is the fuzzy Kappa index for direct raster comparison of the original map according to the fuzzy expression. Equation (5.13) is a fuzzy vector representation of each image element on the map as a fuzzy vector containing the influence of the neighborhood, and the fuzzy set is compared in the form of a vector for superposition. Assuming a land classification thematic map with N categories, consider the fuzzy vector expression of the neighboring C image elements, where mi represents the fuzzy affiliation of the adjacent image element at the i-th image element, and the value of mi is defined by a fuzzy affiliation function decaying by the distance d i , where d i is the distance from the central image element of the window in Eq. (5.14).

(a)

(b)

Fig. 5.7 Difference between the pixel-by-pixel comparison method and the fuzzy comparison method

160

5 Spatial Observation and Assessment of Ecological Changes in Giant …

Table 5.3 Fuzzy vector expressions for each comparison grid

Grid cell raw properties

Grid cell number Vector expression

Meadow

1

(1, 0.4, 0.2, 0, 0, 0.3)

Scrub

2

(0.4, 1, 0.2, 0, 0, 0.1)

Broadleaf forests

3

(0.2, 0.4, 1, 0, 0.4, 0.6)

Driftstone beach and 4 snow and ice

(0.4, 0.4, 0.2, 1, 0, 0)

Mixed coniferous forest

5

(0.4, 0.2, 0.4, 0, 1, 0.2)

Coniferous forest

6

(0.4, 0.2, 0.4, 0, 0.4, 1)

Other

…

…

Broad-leaved forest

…

…

Meadow

…

…

…

…

…

m i (di ) = eln(1/2)×di /2 = 2−di /2

(5.14)

The fuzzy set theory is introduced for the description of each grid cell, and the grid originally belonging to a certain type is defined as a vector cell belonging to all types based on the affiliation function, as shown in Table 5.3. 3) Fuzzy set-based landscape index extraction method The landscape index has multiple definitions describing different characteristics at three levels of patches, patch types and mosaics. Only the characteristics based on descriptive patches are used to analyze the pattern in the study. A certain attribute, such as perimeter, area, shape index, etc., is selected and calculated for each polygonal patch in the classification result, and then the value is assigned to each image element point within the patch range in a minimum granularity rasterization. Taking shape index S shape and sub-dimension number S frac as examples, we first calculate the perimeter S peri and area S size for each patch, and use them as basic features to calculate other landscape indices, including shape index, sub-dimension number, Euclidean distance, diversity index, etc. Define the comparison map M containing n patches as a vector set SM. SM = {S1M , S2M , . . . , Sn M }

(5.15)

5.1 Spatial Observation and Evaluation of Habitat Ecological Landscape …

) ( peri,M shape,M Si M = Sisize,M , Si , . . . , Si , Sifrac,M shape

Si

Si

f rac

=

peri / si / sisize 4

=

(5.16)

(5.17)

( peri ) ⎧ peri size si /u s S ⎪ i i ⎪ ⎪ ⎨ if u > 4 : u 2 log 4 = ⎪ peri ⎪ ⎪ ⎩ if si = 4:2 u peri / s shape Si = i / sisize 4

(5.18)

(5.19)

C = {c1 , c2 , c3 , . . . , ci , . . . , cn }

(5.20)

NBH = {nbh1 , nbh2 , . . . , nbhi , . . . , nbhn }

(5.21)

{ } nbhi = nbhi,1 , nbhi,2 , · · · , nbhi,ti

(5.22)

| nbhi, j ∈ C |di, j ≤r

(5.23)

|| di, j = ||nbhi, j − ci ||u

(5.24)

f (d) = eln(1/2)×d/2 = 2−d/2

(5.25)

( ) wi, j = f di, j

(5.26)

} { LM = l1M , l2M , . . . , lnM

(5.27)

Σti liM

161

peri,M (lisize,M , li ,···

shape,M frac,M , li , li )

=

j=1

wi, j v(S M , nbhi, j ) Σti j=1 wi, j

(5.28)

The fuzzy affiliation degree of each feature of each image element in the neighborhood is calculated for each category, and the vector set is changed to LM. Let M be a set C containing n image elements, and the neighborhood set with radius r centered on ci is NBH. In Eq. (5.28), t i denotes the number of neighboring image elements when the central image element is ci , d i , j is the Euclidean distance between the central image element ci and the neighboring image element nbhi,j in Eqs. (5.23)

162

5 Spatial Observation and Assessment of Ecological Changes in Giant …

Fig. 5.8 Three different attenuation functions: linear, Gaussian, and exponential

and (5.24), and u is the actual area size of the grid cell, which is set to 1 for simplified calculation. On the basis of the above fuzzy comparison algorithm, firstly the landscape indices of the before and after two images are generated separately, the index value of the patch is assigned to each image element in the patch, a circular window of certain size is slid, and the geographic location relationship of the image element points is considered in the process of comparison, the value of the center point of the window is calculated by taking a weighted sliding average according to the value of each image element around it, and the weight value is calculated according to the distance between the surrounding points and the center point decay. The weight is calculated according to the distance between the surrounding points and the center point, as is shown in Fig. 5.8. The exponential decay function (exponential) was chosen for this study: M(d) = eln(1/2)×d/2 = 2−d/2

(5.29)

where d is the Euclidean distance from the center image element of the window. Then we compare graph A and graph B pixel by pixel to obtain the confidence graph M of the comparison classification result graph, and each point on the graph and its neighbors calculate the affiliation degree about y types to sum up, assuming that v (M, c) is the type defined at ci in the classification result, then the confidence degree of the type located at ci is defined as Eqs. (5.30)–(5.32). fuzzy,M

Pi

fuzzy,M

pi,k

fuzzy,M

= { pi,1

fuzzy,M

, pi,2

fuzzy,M

, . . . , pi,y

[ ] ti = max wi, j ∂ v(M, nbhi, j ), k j=1

( ∂(a, b) =

if a=b:1 otherwise :0

)

(5.30) (5.31)

(5.32)

Equation (5.33) defines the Euclidean distance comparison method for the vectors PiA and PiB at ci , and Eq. (5.34) defines the Shannon diversity index comparison. || || DiEuclid = || PiA − PiB ||

(5.33)

5.1 Spatial Observation and Evaluation of Habitat Ecological Landscape …

DiShannon =

y Σ k=1

B B pi,k ln( pi,k )−

y Σ

A A pi,k ln( pi,k )

163

(5.34)

k=1

5.1.3 Analysis of Results From 1994 to 2001, on the one hand, with the increase of human activities, the fragmentation of habitats in the landscape has increased, and the area of habitats suitable for living organisms has been drastically reduced, which has reduced the ecological functions provided by habitats for species; on the other hand, with the complex shape of patches (fragmentation), the edge effect of patches has increased, which has led to the reduction of the core area of natural habitats. On the other hand, with the complex shape of the patch (fragmentation), the edge effect of the patch increases, resulting in the reduction of the core area of natural habitat, which greatly reduces the function of habitat to protect biodiversity. Figure 5.9a–c shows the areas where changes occurred pixel by pixel, and combined with Fig. 5.10a, it is found that most of the areas with large changes in each type of patch are concentrated in the range of 1600–2600 and 2600–3600 m above sea level, the distribution of mixed coniferous forests and coniferous forests is more variable according to the vertical distribution of vegetation in Fig. 5.2. This area is the main distribution area of wild giant panda. From the changes of Shannon’s diversity index, the changes in ecological diversity were concentrated in the mixed coniferous forests and alpine meadows, while the shape index and edge density, which indicate the fragmentation of patches, changed more intensely in most areas, indicating that the shape of patches in the habitat tended to be more complex in the last decade, and the edge effect of patches was stronger and the ecological pattern was fragmented. Comparing Figs. 5.10b and 5.11b, the area of each patch in scrub (3600–3900 m) and meadow (3900–4400 m) was more variable. In mixed coniferous forests (1600–2600 m) and evergreen broad-leaved forests (below 1600 m), the shape of the patches became more complex and the fragmentation of the landscape pattern increased (Table 5.4). From the analysis of each landscape index of the habitat, it can be seen that human disturbance in the giant panda habitat is strong, and the edges of each landscape patch tend to be regular and gentle, but the vegetation type of the main giant panda habitat, temperate mixed coniferous forest (2600–3600 m), is widely distributed with the least fragmentation and becomes the landscape matrix (Table 5.5). From the whole mountain system, the fragmentation of the habitat is low, and the diversity of various types of patches in the gently sloping areas is relatively gentle, which can satisfy the survival and reproduction of the giant panda population. At the same time, landscapes unsuitable for giant panda habitat, such as warm deciduous broadleaved scrub, deciduous broad-leaved scrub, scrub and grasses, which were formed by repeated deforestation, existed in a large area in the habitat, among which the

164

5 Spatial Observation and Assessment of Ecological Changes in Giant …

(a) European distance index

(b) Edge density index

(c) Type by type fuzzy comparison

(d) Shape index

Fig. 5.9 Type-by-type extraction of landscape index variation (decay radius 3 p, sliding window size 5 p)

warm deciduous broad-leaved scrub was the most disturbed by humans and had the highest fragmentation, and the survival of giant panda still faced many problems from the perspective of effective protection of species and their habitats.

5.1 Spatial Observation and Evaluation of Habitat Ecological Landscape …

(a) Type by type fuzzy comparison

165

(b) Shannon diversity index (SHDI)

Fig. 5.10 Area-based type-by-type variation diagram (decay radius 3 p, sliding window size 5 p)

(a) Type by type fuzzy comparison

(b) Shannon diversity index (SHDI)

Fig. 5.11 Perimeter-based type-by-type variation diagram (decay radius 3 p, sliding window size 5 p)

166

5 Spatial Observation and Assessment of Ecological Changes in Giant …

Table 5.4 Pixel-by-pixel comparison of the change image element classification transfer matrix Meadows

Scrub

Broad-leaved forests

Driftwood and snow

Other

Aggregates

Meadows

1900

1410

327

0

1301

4938

Scrub

1243

6669

509

2

2408

10,831

Broad-leaved forests

104

43

4342

7

1138

5634

Driftwood and snow

31

105

55

484

108

783

Other

2266

2490

6229

462

9543

20,990

Aggregates

5544

10,717

11,462

955

14,498

33,324

Table 5.5 Fuzzy comparison of Kappa values for each type of extracted changes Meadows

Scrub

Broad-leaved forests

Driftwood and snow

Other

Kappa

0.275

0.492

0.404

0.548

0.233

kl

0.294

0.496

0.688

0.609

0.335

kh

0.934

0.993

0.587

0.899

0.697

5.2 Spatial Observation and Assessment of Habitat Ecological Changes by Microwave Remote Sensing The active imaging radar in microwave remote sensing is suitable for spatial observation and quantitative assessment of ecological environment changes in rare animal habitats due to its all-day, all-weather working characteristics and high spatial resolution data acquisition capability, combined with data processing and information inversion special capabilities such as radar polarization and radar interference, with terrain analysis, forest dynamics survey and landslide monitoring as the entry point.

5.2.1 Background of the Study As one of the world icons of biodiversity conservation, the giant panda has been listed as an important endangered animal in the International Union for Conservation of Nature (IUCN) Red Line. At present, giant panda habitats are mainly located on the edge of the Tibetan Plateau, i.e. more than 20 connected or isolated blocks of forested mountains located in Sichuan, Shaanxi and Gansu. Among them, Sichuan Province has the highest number of giant pandas in China, and the 924500 hm2 habitat, including Wolong Nature Reserve, was listed as a World Natural Heritage Site in 2006; Wolong Nature Reserve is the most concentrated place for giant pandas, with 100–150 pandas inhabiting there.

5.2 Spatial Observation and Assessment of Habitat Ecological Changes …

167

On May 12, 2008, an 8.0 magnitude earthquake struck Wenchuan County, Aba Tibetan and Qiang Autonomous Prefecture, Sichuan Province, killing 69,227 people, injuring 374,643 and leaving 17,923 missing. The epicenter of the earthquake was located at the junction of Yingxiu and Xuankou towns in Wenchuan County, 92 km to the west-northwest of Chengdu, the capital of Sichuan Province, adjacent to Wolong Reserve, the core area of Sichuan Giant Panda World Heritage Site, which had a great impact on the ecological environment of the giant panda habitat living in Minshan and Qionglai Mountains, as published in Frontiers in Ecology and Environment report shows that 23% of the giant panda habitat in the area covered by the study has been destroyed, and the remaining, fragmented habitat has destroyed ecological corridors between habitats and increased fragmentation within the habitat, potentially affecting the reproduction of giant pandas. The 2008 Wenchuan earthquake had a major impact on the sustainability of panda habitat. Due to their proximity to the epicenter, habitats in Minshan and Qionglai Mountains were the most severely damaged. In order to assess the impact of the disaster on giant panda habitat and the survival of rare animals in a timely manner, a series of studies were initiated and carried out, including field field surveys (Cheng et al. 2012; Zheng et al. 2012) and remote sensing assessment methods. Despite the positive contribution of these methods to the conservation of giant panda habitats and medium- and long-term protection, they still have limitations in the face of future natural hazard events (Sichuan Province is prone to earthquakes and secondary hazards due to the subduction and impact of the Indian and Asian plates and the development of typical seismogenic structures): (i) Ground-based field studies are difficult for human access to the high mountainous environment of giant panda habitats. The difficulty of human access increases after the occurrence of earthquakes and other disasters; (ii) The technical means of assessment dominated by optical remote sensing is difficult to obtain data in near-real time in the cloudy areas of Sichuan (especially during the rainy season from May to September each year), which leads to the incompleteness of habitat area assessment and errors in the assessment.

5.2.2 Research Data To quantitatively assess the impact of the Wenchuan earthquake on giant panda habitat, multi-band satellite radar data (Envisat ASAR data in C-band and ALOS PALSAR data in L-band) were used to select Minshan and Qionglai Mountains, which were most significantly affected by the earthquake, as case studies. The northeastern end of Qionglai Mountain was covered by Envisat ASAR and ALOS PALSAR (Fig. 5.12). Interference coherence maps were used as a quantitative analysis tool to assess the impact of the Wenchuan earthquake on typical habitats in the World Heritage Site. The epicenter of the Wenchuan earthquake was close to the Wolong Biosphere Reserve; the Caopo Reserve, part of the Sichuan Giant Panda Habitat World Heritage Site, is also located a short distance (approximately 25 km) north of the epicenter.

168

5 Spatial Observation and Assessment of Ecological Changes in Giant …

Fig. 5.12 Envisat ASAR and ALOS PALSAR covering the Minshan and Qionglai Mountain test areas. White star-shaped marker at the epicenter of the Wenchuan earthquake, near Wolong Biosphere Reserve

The 25-view Envisat ASAR and 17-view ALOS PALSAR single-view complex data were selected for the coherence timing analysis. The C-band (5.6 cm wavelength) Envisat ASAR with the downlink mode Track 290 was acquired from December 24, 2007 to August 30, 2010; the data acquisition parameters are shown in Table 5.6. The incidence angle of data imaging is about 23°, corresponding to a spatial resolution of 20 m on the ground. L-band (23.6 cm wavelength) ALOS PALSAR data in elevated orbit mode with an imaging incidence angle of 34° were acquired from February 2, 2007 to December 29, 2010; the data acquisition parameters are shown in Table 5.7. PALSAR data are acquired in two modes. Fine beam singlepolarization (FBS) with HH polarization and a distance-to-bandwidth of 28 MHz, and fine beam dual-polarization (FBD) with HH/HV polarization and a distance-tobandwidth of 14 MHz. The two data acquisition modes have the same distance-tocenter bandwidth frequency, and support both FBS/FBD mixed-mode interferometric processing after the distance-to-center oversampling of the FBD mode data using the Sinc function. To evaluate and remove the topographic phase from the interferometric data processing, 3 s (90 m) SRTM digital elevation data from the United States

5.2 Spatial Observation and Assessment of Habitat Ecological Changes …

169

Geological Survey (USGS) are used; this data can also be used for the geocoding of the later radar interferometric satellite products (i.e., conversion from the distanceDoppler coordinate system of the imaging radar to the universal transverse Mercator coordinate system). (i.e., conversion from the distance-Doppler coordinate system of the imaging radar to a common horizontal Mercator map coordinate system). Table 5.6 Envisat ASAR data acquisition parameters No

Acquire date

Get mode

Track pass

No

Acquire date

Get mode

Track pass

1

2007-12-24

IS2

Descending track

14

2009-07-06

IS2

Descending track

2

2008-03-03

IS2

Descending track

15

2009-09-14

IS2

Descending track

3

2008-06-16

IS2

Descending track

16

2009-10-19

IS2

Descending track

4

2008-07-21

IS2

Descending track

17

2009-11-23

IS2

Descending track

5

2008-08-25

IS2

Descending track

18

2009-12-28

IS2

Descending track

6

2008-09-29

IS2

Descending track

19

2010-02-01

IS2

Descending track

7

2008-11-03

IS2

Descending track

20

2010-03-08

IS2

Descending track

8

2008-12-08

IS2

Descending track

21

2010-04-12

IS2

Descending track

9

2009-01-12

IS2

Descending track

22

2010-05-17

IS2

Descending track

10

2009-02-16

IS2

Descending track

23

2010-06-21

IS2

Descending track

11

2009-03-23

IS2

Descending track

24

2010-07-26

IS2

Descending track

12

2009-04-27

IS2

Descending track

25

2010-08-30

IS2

Descending track

13

2009-06-01

IS2

Descending track

Table 5.7 ALOS PALSAR data acquisition parameters

No

Acquire date

Get mode

Track pass

1

2007-02-02

FBS

Elevated track

2

2007-06-20

FBD

Elevated track

3

2007-08-05

FBD

Elevated track

4

2007-09-20

FBD

Elevated track

5

2007-12-21

FBS

Elevated track (continued)

170

5 Spatial Observation and Assessment of Ecological Changes in Giant …

Table 5.7 (continued)

No

Acquire date

Get mode

Track pass

6

2008-02-05

FBS

Elevated track

7

2008-05-07

FBD

Elevated track

8

2008-06-22

FBS

Elevated track

9

2009-02-07

FBS

Elevated track

10

2009-06-25

FBD

Elevated track

11

2009-08-10

FBD

Elevated track

12

2009-09-25

FBD

Elevated track

13

2009-11-10

FBS

Elevated track

14

2009-12-26

FBS

Elevated track

15

2010-02-10

FBS

Elevated track

16

2010-06-28

FBD

Elevated track

17

2010-12-29

FBS

Elevated track

5.2.3 Research Methodology Given the proximity to the epicenter of the Wenchuan earthquake (as marked by the white star in Fig. 5.12), the landscape of giant panda habitat in Minshan and Qionglai Mountains was dramatically altered by earthquake-induced landslides, debris flows, stripping of ground cover, and collapse of man-made structures. Post-earthquake habitat forest degradation could further exacerbate habitat fragmentation around the currently existing Wolong Biosphere Reserve. The study introduces coherent time series changes to monitor post-earthquake induced forest degradation. Typically, radar decoherence generally occurs in densely vegetated areas due to the body scattering mechanism; therefore, when natural disasters lead to forest degradation and surface exposure, the radar coherence coefficient values of exposed areas can be significantly enhanced; this is the basic principle of the study for forest degradation analysis. Considering the difficulties in aligning the pre/post-earthquake interferometric images to the sub-pixel level, coupled with the severe temporal decoherence, the interferograms obtained from the pre/post-earthquake crossover data acquisition mode are of poor quality, which in turn hinders the extraction of information on forest degradation. In order to further suppress the errors caused by the data acquisition interval, and the different imaging geometry patterns, the selection of interferometric pairs is particularly important. In view of this, three major guidelines for interferogram image pairs were designed: (i) The temporal and spatial baselines of pre- and post-earthquake coherence map acquisition must be consistent to overcome the errors caused by different baselines; (ii) To improve the quality of coherence maps, small baseline set interferogram pairs were selected; (iii) The seasons used to compare interferogram pairs were consistent as much as possible to suppress the decoherence errors caused by seasonal physical changes.

5.2 Spatial Observation and Assessment of Habitat Ecological Changes …

171

Forest degradation based on coherence analysis can be obtained by differential calculation of pre- and post-earthquake coherence maps. The zero-centered floating of the differential interferogram characterizes the trend of forest degradation. Usually, for natural stochastic phenomena, the histograms generated by differential coherence maps for two time periods follow the zero point centered in the normal Gaussian distribution, i.e., the trend of coherence value gain and coherence value decrease in differential interferograms are comparable. However, when external forces such as natural disasters or human activities intervene, the equilibrium will be broken, resulting in a zero-point shift in the histogram corresponding to the differential coherence map, which can be used to quantitatively assess the pre/post-earthquake forest degradation caused by the Wenchuan earthquake. Synthetic aperture radar interferometry or radar differential interferometry is a quantitative remote sensing analysis tool for digital elevation model production, change monitoring, and surface deformation monitoring (Chen et al. 2013). Coherence is an important quality indicator for radar interferometric data processing, as it relates to phase unwinding, and the accuracy of topography and deformation inversion. Unlike optical remote sensing, SAR actively emits microwave signals and receives backscattered pulse signals from the observed scene. Thanks to its all-day, all-weather capability, radar remote sensing has become an important tool for continuous and effective observation in cloudy and rainy areas. The emergence of multimode, multi-polarization, multi-temporal and high-resolution radar data has further expanded the applications of radar remote sensing; among which the monitoring and assessment of natural and cultural heritage is one of the potential directions. Considering the limitations of optical remote sensing in imaging cloudy and rainy areas and the advantages of radar remote sensing observation, a large number of change detection-based radar remote sensing methods have been developed, including differential methods (Cable et al. 2014; Gong et al. 2014), systematic statistical methods, modeling methods, coherence analysis methods (Meyer and Sandwell 2012), and other (Schmitt et al. 2014). The study introduced radar coherence analysis to quantitatively assess the impact of the Wenchuan earthquake on forest degradation in giant panda habitats in order to overcome the limitations of directions such as field surveys and optical remote sensing assessments. To objectively analyze multi-band radar electromagnetic waves, the study introduced C-band Envisat ASAR and Lband ALOS PALSAR data; the results showed that long-band radar interferometry has good potential for assessing post-disaster forest degradation.

5.2.4 Degradation Monitoring 1. Envisat ASAR Degradation Monitoring In order to improve the quality of interferograms, the above interferometric image pair matching criterion ii-small baseline, i.e., the spatial baseline is less than 250 m and the temporal baseline is less than 180 m, was first applied to obtain 49 pairs of

172

5 Spatial Observation and Assessment of Ecological Changes in Giant …

Fig. 5.13 Envisat ASAR coherence map image small baseline set pairing. The spatial vertical baseline is less than 250 m and the temporal baseline is less than 180 days. The post-earthquake coherence map image pairs of candidate areas are shown in blue and cyan. The final selected pre/post-earthquake coherence maps are marked by the pink lines

coherence maps, as shown in Fig. 5.13. Among them, there is only one pre-earthquake coherence map, which corresponds to the interferometric pair data acquisition time of December 24, 2007 and March 3, 2008, respectively. For the sake of simplicity, the data acquisition times are indicated by year-month-day numbers as shown in Tables 5.6 and 5.7; therefore, the pre-earthquake coherence maps can be renamed as 2008-03-03 to 2007-12-24 (spatial vertical baseline 26.18 m, temporal baseline 70 days, and data acquisition season corresponding to the spring-winter combination mode). Then, the coherence map image pairing criteria i (similar baseline combination) and iii (same seasonal combination) were applied to select post-earthquake interferograms. In order to improve the efficiency of the selection, the spatial and temporal baseline thresholds of [−40, 40] m and 70 days, respectively (shown in blue cyan in Fig. 5.13), can be set with reference to the values of the coherence map parameters from 2008-03-03 to 2007-12-24 before the earthquake. Second, the candidate coherent image pairs can be constrained by the combination of spring-winter season. Finally, the post-earthquake period from 2009-01-12 to 2008-11-03 (spatial vertical baseline 17.33 m, temporal baseline 70 days) was selected for the coherence comparison analysis (Fig. 5.14). The comparison results show that the forest degradation along the Minjiang River was serious after the earthquake, and the coherence value was significantly enhanced. In addition, the phenomenon of decreasing coherence value can be found in the lower right corner of the post-earthquake coherence map,

5.2 Spatial Observation and Assessment of Habitat Ecological Changes …

173

Fig. 5.14 Envisat ASAR, a pre-earthquake 2008-03-03 to 2007-12-24; b post-earthquake 200901-12 to 2008-11-03 coherence map comparison plots. The red oval marks the post-earthquake coherence values along the Min River to the area of significant gain. c Corresponding human settlements (pink polygons), agricultural land (green polygons), and river terraces (blue polygons) in the plain area

174

5 Spatial Observation and Assessment of Ecological Changes in Giant …

which is explained by the disturbance of human activities such as settlement, agricultural land and riverbed area, as well as the temporal phase change of soil moisture (Fig. 5.14). 2. ALOS PALSAR Degradation Monitoring Thanks to the long wavelength, the L-band ALOS PALSAR data have a strong penetration and temporal coherence retention capability. Therefore, the high quality coherence maps and the significant coherence value changes characterized in the coherence maps after the earthquake can be used for accurate information extraction and assessment of post-earthquake forest degradation compared to Envisat ASAR data in C-band. It is in view of the above-mentioned superiority of L-band that the study focuses on the application of PALSAR data for quantitative analysis of forest degradation. Similar to the Envisat ASAR data processing, first, the small baseline criterion ii (spatial vertical baseline less than 2000 m and temporal baseline less than 322 days) was applied to generate 48 combinations of coherence maps, as shown in Fig. 5.15. To further suppress the decoherence effect caused by the spatial vertical baseline, a threshold of [−800, 800] m was selected as the pre/post-earthquake coherence map image pair selection constraint (see Fig. 5.15 marked in blue and cyan). Therefore, for the post-earthquake 2010, only 2010-06-28 to 2010-02-10 (summer-winter

Fig. 5.15 ALOS PALSAR coherence map image pairing based on a small baseline set (spatial vertical baseline less than 2000 m, temporal baseline less than 322 days). The spatial baseline constraint threshold range for searching pre/post-earthquake coherence maps is marked by blue cyan, and the final selected contrast coherence map image pairs are marked by pink lines

5.2 Spatial Observation and Assessment of Habitat Ecological Changes …

175

combination, spatial vertical baseline 563 m, temporal baseline 138 days) satisfy the above constraints, which is referred to as the coherence map 2010. The coherence map is used to select other pre- and post-earthquake coherence maps. By applying the coherence map image pairing criteria i (similar baseline combination) and iii (same seasonal combination), the pre-earthquake coherence map 2007-06-20 to 2007-0202 (summer-winter season combination, spatial vertical baseline −746 m, temporal baseline 138 days, and abbreviated as coherence map 2007) and post-earthquake coherence map 2007 were further obtained for comparison. From 2009-06-25 to 2009-02-07 (summer-winter season combination, spatial vertical baseline 755 m, time baseline 138 days, and abbreviated as coherence map 2009). The final selected contrasted coherence map image pairs are shown in Fig. 5.16 with pink lines. The corresponding generated coherence map, as shown in Fig. 5.17, clearly shows that the landslides, mudflows and landslides caused by the post-earthquake have destroyed a large amount of forest cover, which is characterized as significant coherence value gain in the post-earthquake coherence map, especially in the area marked by the red ellipse in Fig. 5.16. The intensity of the earthquake in the study area is VII–XI. Because of the steep terrain, high mountains and deep hollow valleys in the Sichuan panda habitat, and the complex geological and tectonic environment, the earthquake triggered a large number of landslides and landslides in the area, especially in Gaojiagou, Happiness Gully and Hongchun Gully. The forest degradation caused by landslides and other disasters mainly includes two major parts: (i) Landslides and landslides directly triggered by the main earthquake and aftershocks, and the landslide movement directly destroys the forest cover along the route; (ii) The land cover surface is soft after the earthquake, and under the influence of strong precipitation, surface erosion intensifies and shallow debris flow slides, which can converge to form large landslides. In order to compare the forest degradation before/after the earthquake more intuitively, a coherence map differencing technique was used to generate differential coherence maps before and after the disaster event from 2009 to 2007 and from 2010 to 2007, as shown in Fig. 5.17. Similar to the Envisat ASAR treatment, in addition to the change in coherence values due to forest degradation, other human activities and natural phenomena can also produce a gain or decrease in coherence values, such as the gain in coherence values due to the reconstruction of post-earthquake settlements marked by red polygons and the decrease in coherence values due to changes in soil moisture in river terraces or low-lying areas marked by blue polygons in Fig. 5.17. 3. Quantitative assessment and cross comparison The impact of the Wenchuan earthquake on the forest degradation of giant panda habitat in Minshan and Qionglai Mountains in the test area was assessed using the histogram zero offset quantification of the differential coherence map, after excluding the effect of alluvial plains (not related to the forest degradation of giant panda habitat in Sichuan) using the mask technique. Figures 5.18 and 5.19 correspond to the assessment results using C-band Envisat ASAR and L-band PALSAR, respectively. The results of Envisat ASAR zero-point (47.96%) offset monitoring showed that only 2.04% of the forests in the test area were degraded by earthquake damage. On the

176

5 Spatial Observation and Assessment of Ecological Changes in Giant …

Fig. 5.16 Pre/post-earthquake ALOS PALSAR coherence plot. The red ellipse characterizes the post-earthquake forest degradation and the area of significant gain in coherence value

contrary, the PALSAR 2009-2007 differential coherence plot corresponds to a zero point of 29.34%, indicating that about 20.66% of the forests were still degraded by the Wenchuan earthquake damage until 2009 after the earthquake; 2010 the differential coherence plot corresponds to a zero point of 32.66% from 2009 to 2007, indicating that 17.34% of the forests were still degraded by 2010. The apparent inconsistency between the estimation results of Envisat SAR and ALOS PALSAR can be explained as follows. (i) The two types of satellite data have different spatial coverage of the test area, which may introduce inconsistent trends in coherence values due to different observation areas; (ii) The time baselines used for generating coherence maps by Envisat ASAR and ALOS PALSAR, as well as the different seasonal combination patterns, may introduce inconsistencies

5.2 Spatial Observation and Assessment of Habitat Ecological Changes …

177

Fig. 5.17 Change in coherence values in differential coherence diagram except for forest degradation. Coherence value reduction (dark tone areas marked by blue polygons) and coherence value gain (bright tone areas marked by red polygons) subplots “1,2,3” show the coherence value gain and reduction detail information

178

5 Spatial Observation and Assessment of Ecological Changes in Giant …

Fig. 5.18 Envisat ASAR difference coherence map (Coherence Map 2008-03-03 to 2007-12-24 vs. 2009-01-12 to 2008-11-03). The zero point of the corresponding coherence histogram is shifted, i.e., the zero point is located at 47.96%, indicating that the post-earthquake forest degradation only 2.04 occurred in the area not significantly affected by the earthquake due to the poor quality of the coherence map and the coverage of too many areas

Fig. 5.19 PALSAR differential coherence map corresponding to the zero shift of the coherence histogram

in temporal and physical information; (iii) The two data bands are different, and thus have different coherence retention capabilities; for example, C-band Envisat ASAR data are sensitive to vegetation, except for completely bare forest degraded areas, which can be missed by C-band radar data for moderate to lightly damaged forest degraded areas because they are still characterized by severe decoherence and no significant coherence gain; in addition, the inferior coherence map makes Envisat ASAR results more susceptible to other random noise; thus causing an overall underestimation of forest degradation (e.g., 2.04% in the study). In contrast, the L-band ALSO PALSAR coherence map is sensitive to the gain in coherence values due to forest damage and can effectively overcome these limitations. Therefore, comparing

5.2 Spatial Observation and Assessment of Habitat Ecological Changes …

179

the two performances, PALSAR is considered more suitable for quantitative assessment of post-earthquake forest degradation, such as the direct post-earthquake degradation of forest degradation of up to 20.66% in the giant panda habitat in Minshan and Qionglai Mountains obtained in the study.

5.2.5 Validation and Discussion In the past 5 years, two field test expeditions were conducted in 2009 and 2013 to evaluate and verify the correctness and accuracy of radar interference monitoring of forest degradation in giant panda habitats in Sichuan. The field survey found that the post-earthquake forest degradation was mainly in the area of mountain collapse and landslide along the Minjiang River in the Yingxiu-Wenchuan County section (Figs. 5.16 and 5.17). As seen from the field field photographs taken (Fig. 5.20), the ground cover vegetation was stripped away after the landslides, leaving bare or bare rock bodies. The hypothesis that forest degradation can produce a gain in coherence values was adopted in the study. To support and verify this hypothesis, the aerial photographic image mosaic land use information acquired after the earthquake Fig. 5.21c was applied to verify and compare the PALSAR pre-earthquake coherence maps 2007-0620 to 2007-02-02 (Fig. 5.21a) and the post-earthquake coherence maps 2010-06-28 to 2010-02-10 (Fig. 5.21b). In Fig. 5.21, two phenomena can be observed: (i) The high coherence values in the radar coherence map occur only in the bare land or residential areas before Fig. 5.20 Photographs of typical landslides along the Min River in Wenchuan County

180

5 Spatial Observation and Assessment of Ecological Changes in Giant …

Fig. 5.21 PALSAR radar interferometric coherence map of forest degradation monitoring against aerial imagery

the earthquake, while the bare land where landslides or other geological hazards occurred can be characterized as high coherence values after the earthquake; (ii) The spatial distribution of high coherence values in the post-earthquake coherence map is strictly consistent with the mountain areas found in the aerial images. Figure 5.21 Coherence value gain anomalies and land cover anomalies monitored in the postearthquake subplots “1” and “2”. When the aerial image map and the radar coherence map have the same image coordinate system, 100 sampling points were randomly selected from the post-earthquake 2010-06-28 to 2010-02-10 coherence map and the aerial image map for quantitative analysis; after manual interpretation and analysis, 82% of the radar coherence value gain anomalies were found to correspond to the landslide area. The small differences between the two verifications can be explained

5.2 Spatial Observation and Assessment of Habitat Ecological Changes …

181

as follows: (i) The spatial resolution of the radar coherence map and the aerial image map are different, i.e., the resolution of the radar coherence map is 8 m while that of the aerial image is 2 m; (ii) The imaging geometry modes used in the two data are different (radar oblique distance imaging and aerial center projection imaging), which can cause the geometric deviation of the imaging; (iii) The time period between the two data acquisition is different, which can introduce the temporal difference of the vegetation ground cover. Fortunately, the areas of severe forest degradation were mainly concentrated in unsuitable or generally suitable habitats with steep slopes and sparse bamboo consumption by giant pandas, and this result is generally consistent with the study 2012 of Zheng et al. At the request of the UNESCO World Heritage Committee, IUCN initiated an assessment of the damage to giant panda habitats in Sichuan from the Wenchuan earthquake from 12 to 17 April, 2010. The results concluded that the earthquake had little impact on giant panda staple bamboo throughout the world heritage Site, and that the significant impact area was mainly located in the northeast corner of the site, near Wolong Biosphere Reserve. However, the significant impact area with substantial surface stripping and vegetation loss will still generally constrain or change the normal migration and lifestyle of giant pandas in the habitat. In addition, considering that giant panda staple bamboo is accompanied by fir and other arboreal forests, the vegetation ecosystem changes caused by the earthquake could affect the generation and spread of medium- to short-term staple bamboo, which in turn could affect the carrying capacity of the local habitat for giant panda population. In general, radar remote sensing monitoring and assessment of habitat changes in giant panda habitats has the following limitations: (i) At present, giant panda habitats are located in Sichuan, Shaanxi and Gansu Provinces; the single-view width of radar images is not wide enough to cover the entire habitat and thus complete a systematic assessment of habitat changes in the habitat. For example, the study only monitored and evaluated Minshan and Qionglai Mountains, which were close to the earthquake epicenter. In addition, although the results of the study generally reflected the negative effects of the earthquake on the habitat, trends such as fragmentation and isolation within the habitat could not be quantitatively assessed due to the limitations of resolution and data sources. (ii) Coherent time-series analysis requires more stringent radar data sources. For example, in order to suppress time-space decoherence and seasonal effects as much as possible, a large amount of archived data in the observation area should be obtained in advance; then a number of interferometric image pairs that meet the requirements for quantitative assessment should be obtained according to the interferometric pair selection criteria. (iii) Short-wavelength radar data, such as the study of C-band Envisat ASAR, can lead to significant underestimation of forest degradation due to severe decoherence in habitats with high forest cover.

182

5 Spatial Observation and Assessment of Ecological Changes in Giant …

5.2.6 Conclusion The study introduces radar coherence time series analysis technique the first time, and makes full use of the L-band PALSAR penetration and strong coherence retention ability to monitor and assess the forest degradation in the Sichuan giant panda habitat after the 2008 Wenchuan earthquake, and quantitatively obtains the forest degradation in the habitat near the epicenter of Minshan and the edge of Qionglai Mountain. The results showed that the earthquake triggered landslides, landslides and mudslides caused serious forest degradation along the Minjiang River, and PALSAR estimated that the post-earthquake degradation rate was still 20.66% in 2009 by applying the histogram zero-point shift method. The study also shows that the long-wave L-band PALSAR data have better performance than Envisat ASAR at C-band in monitoring dense vegetation areas due to the significant coherence value gain in the vegetationdamaged areas, and the results are validated by aerial optical remote sensing image interpretation and control. The study implies the potential application of satellitebased radar interference for forest degradation monitoring and assessment in areas susceptible to natural hazards.

5.3 Fine Spatial Observation and Assessment of Habitat Ecological Changes by LiDAR Remote Sensing The topography of forest areas, vegetation cover and its characteristic parameters, such as tree height, diameter at breast height and storage volume, are important parameters for ecological research and forestry management. Traditional manual sample site surveys are not only time-consuming and imprecise, but also cannot obtain fine parameters of forest areas. LIDAR, by transmitting and receiving laser signals, can acquire fine 3D point clouds of the environment and provide rich 3D features of the features. In recent years, people began to apply ground-based LiDAR to forest area surveys, which gradually became an important means to obtain the fine structure of vegetation, and the research mainly includes site erection scheme, terrain data extraction, vegetation point cloud processing algorithm and parameter inversion scheme, etc. At present, the research on the extraction of forest area information based on ground-based LiDAR mainly focuses on the sparse forest area (less than 1500 plants/hm2 ), while unlike the sparse forest area, the aggregated plants, dense branches and slender trunks will greatly enhance the shading effect, which directly affects the observation range of single-station LiDAR and causes the problem of vegetation parameter inversion, such as increasing the difficulty of stalk identification, and how to distinguish between closely adjacent tree trunks, has become a new challenge. As an important vegetation within the giant panda habitat, bamboo forests have high ecological and industrial values. However, both naturally growing bamboo forests and planted bamboo forests tend to have high plant density (higher than 5000 plants/hm2 ), and few studies have focused on the application of terrestrial LiDAR

5.3 Fine Spatial Observation and Assessment of Habitat Ecological …

183

in high-density forest areas, so the huge potential of LiDAR applications in highdensity vegetation areas is yet to be explored. In this section, based on ground-based laser point cloud data, we study trunk mapping and average tree height and topographic parameter inversion in high-density bamboo forest areas (greater than 7500 plants/hm2 ) to provide a demonstration of LiDAR for fine observation in heritage sites.

5.3.1 Single-Station Ground Point Cloud-Based Trunk Mapping of Bamboo Forests The trunk or stalk is the main support structure of terrestrial vegetation and is an important part of vegetation material transport, structural support and energy storage. Trunk parameters play an important role in ecological research and forestry management, and parameters such as diameter at breast height and plant density, which are directly related to the trunk, are important parameters for biomass estimation and carbon stock calculation (Yen et al. 2010; Li et al. 2014), and trunk mapping and trunk measurements are also necessary for forestry resource surveys (Maltamo et al. 2006; Liang et al. 2014). However, traditional methods for measuring vegetation parameters (e.g., diameter at breast height, plant density, LAI, etc.) rely mainly on small-scale manual measurements, which are not only time-consuming and laborious, but also easy to introduce human errors, and it is difficult to accurately obtain the characteristics of tree trunk parameters, such as diameter at different heights and trunk curves. Therefore, it is necessary to study automatic, accurate and effective methods for trunk information measurement. In recent years, ground-based LiDAR has been widely studied in forestry surveys, and the high-density and high-precision 3D point clouds it records are important for the extraction of vegetation fine structure (Yang et al. 2013; Zheng et al. 2013). Currently, many scholars in China and abroad have focused on how to detect tree trunks from ground-based laser point clouds (Maltamo et al. 2006; Astrup et al. 2014). In 2008, Maas et al. first extracted horizontal slices above the ground surface from multi-station aligned point clouds, and then performed clustering and circle fitting on the sliced point clouds to detect trunks. This method was tested in a sample plot with vegetation density below 600 plants/hm2 and was able to accurately detect 97% of the tree trunks. Similarly, cylindrical/circular fitting methods have been widely used in studies related to trunk detection. This type of method usually requires information on the understory topography, but it is difficult to obtain the topography of densely forested areas from point clouds, especially in areas with large topographic relief and dense surface vegetation cover. In 2012, Liang et al. performed trunk mapping based on single-station ground-based LiDAR point clouds: first, potential trunk point clouds were screened using the flat and vertical distribution of trunk points within a certain range; then, the trunk point cloud and trunk locations are accurately determined using a robust cylindrical fit growth algorithm. The method was tested in a sample plot

184

5 Spatial Observation and Assessment of Ecological Changes in Giant …

with a density of 1500 plants/hm2 and the mapping accuracy reached 73%; applying the method to mobile laser scanning (MLS) for trunk mapping in forest areas, the mapping accuracy reached 87.5% in a sample plot with a density of less than 500 plants/hm2 (Liang et al. 2014). Distance images and waveform data (Yang et al. 2013) have also been applied to trunk detection, but the acquisition of these data requires specific instrumentation and additional data pre-processing. Related studies also include tree trunk detection in urban environments. Hetti et al. (2013) extracted potential trunk points from the original point cloud based on the shape index and geometric features of the trunk point set based on the MLS point cloud, and obtained independent trunk point clouds using a layer-by-layer growth method based on orientation. The experimental results show that this method can extract about 90% of the street tree trunks. Lehtomäki et al. proposed a method to extract pole features from MLS point clouds in 2010, which mainly includes four steps of point cloud segmentation, point cloud clustering, cluster merging and classification, and the experiment shows that this algorithm can accurately extract 77.7% of the pole features (street lights, tree trunks, etc.), but the study does not evaluate the accuracy of The accuracy of tree trunks was not evaluated in this study. Unlike forest areas, street trees are generally evenly spaced and do not grow in clusters, and the ground surface in urban areas is flatter, and these differences make the trunk detection in forest areas and urban areas differ significantly. In conclusion, most of the relevant trunk detection studies are mainly focused on sparse forest areas (less than 1500 plants/hm2 ), and the diameter at breast height (poplar, pine, etc.) of trees in the study area is often larger than 10 cm, and few studies focus on higher density forest areas (greater than 5000 plants/hm2 ). In high-density forests, the shading effect of branches and leaves will be more significant, and the diameter at breast height of highly aggregated growing plants is often smaller, which not only increases the difficulty of trunk detection, but also makes the distinction between adjacent trunks a new problem for mapping. Therefore, it is necessary to study how to use ground laser point cloud for effective trunk mapping in high-density forest. In this section, a novel and effective algorithm for high-density trunk mapping based on single-site ground laser point cloud data is proposed and tested in a dense bamboo forest area (about 7500 trees/hm2 ). Unlike previous studies, the algorithm proposed in this section does not require topographic information or circle or cylinder fitting, and the mapping algorithm involves the distinction of adjacent trunks, providing an effective method for trunk mapping in sparse forests. 1. Experimental area and data The study area is located in a bamboo forest in Ya’an Giant Panda Reserve, Sichuan (30.06°N, 103.01°E). Bamboo belongs to the family Gramineae and is found on all continents except Antarctica, and plays an important role in biodiversity conservation, biomass and carbon storage. The plant density of the bamboo forest in the study area is about 7500 plants/hm2 , and the point cloud data of the sample sites were collected in January 2013 using a Leica ScanStation C10 terrestrial 3D laser scanner. The equipment uses the green band of 532 nm and has a nominal ranging

5.3 Fine Spatial Observation and Assessment of Habitat Ecological …

185

Fig. 5.22 Diagram of two plots. The side near the LiDAR position is defined as front and its opposite is defined as back. Circles indicate relative stem locations

accuracy of 4 mm at a distance of 50 m. Due to the clustering of trees and slender bamboo poles, the final point cloud was obtained at an interval of about 3 mm using a high density scanning method. Two sample plots were selected from the original point cloud, Sample Plot A and Sample Plot B. Based on the field sampling measurements, the trees in the sample plots ranged from 6 to 13 m in height and 3–8 cm in diameter at breast height, and Fig. 5.22 shows a sketch of the sample plots, where Sample Plot A was about 100 m2 and about 4 m from the scanner, and Sample Plot B was about 120 m2 and about 5 m from the scanner. Based on the original point cloud and manual measurement, sample plot A and sample plot B included 82 and 84 bamboo plants, respectively. Considering the topographic relief, plant density and number of experiments, Sample Plot A and Sample Plot B are representative. 2. Trunk mapping method The trunk mapping method is divided into three steps, as shown in Fig. 5.23. First, the trunk laser points are screened from the original point cloud, which can be achieved by a two-scale classification algorithm of the point cloud, then the trunk points are merged into separated trunks based on the spatial distance, and finally the trunk point clouds belonging to the same tree are merged by a simplified trunk model.

Fig. 5.23 Phases of stem detection algorithm

186

5 Spatial Observation and Assessment of Ecological Changes in Giant …

1) Trunk point identification The geometric characteristics of the point cloud can be obtained by analyzing the spatial distribution of the neighboring point sets. For a laser point, the points within a certain radius of the point as the center can be used as its neighborhood point set. Principal component analysis of the point set can be performed to obtain three eigenvalues (λ1 , λ2 , λ3 ) (λ1 ≥ λ2 ≥ λ3 ), and the corresponding three eigenvectors (e1 , e2 , e3 ). The eigenvectors represent the main directions of the spatial √ distribution of the set of neighborhood points, corresponding to the lengths δi = λi i = 1, 2, 3. In this section, three geometric features are used to represent the distribution states of local points, which are defined in order as shown in Eq. (5.35). Depending on the maximum of the three features, each point can be labeled as linear (a1D ), planar (a2D ) and scattered (a3D ). a1D =

δ1 − δ2 δ1

a2D =

δ2 − δ3 δ1

a3D =

δ3 δ1

(5.35)

The spatial distribution of local points is not only related to the feature class, but also to the choice of scale. Multi-scale features are more effective in distinguishing different classes of artificial features (Bremer et al. 2013). Since the spatial distribution of vegetation (e.g., tree trunks, branches, grasses, and shrubs) is irregular and often aggregated, multi-scale features of point clouds are rarely applied in the inversion of vegetation parameters. In this section, a two-scale classification method is proposed to monitor the trunk laser points in the original point cloud. The algorithm uses the two-scale features of the trunk to gradually filter out the trunk points by the difference with other structures of the vegetation (e.g. branches, leaves, etc.). When the scale is small, the trunk points are usually labeled as planar, and vice versa, as linear. Table 5.8 shows the results of two-scaled labeling of different types of feature point clouds (4 and 12 cm): when the scale is 4 cm, most of the trunk points are labeled as planar; when the scale is 12 cm, they are labeled as linear. As the scale changes, other types of features show different labeling results, for example, most ground points are labeled as flat at small scales, and most ground points are still labeled as flat as the scale increases. The trunk diameter varies from plant to plant, and the diameter of the same trunk varies with height, so it is important to determine the scale size of markers adaptively. Table 5.8 Results of two-scale markers for different feature types Sample

Tree trunks

Radius/cm

4

Linear/%

2.8

Flat/% Scatter/%

Branches

Leaves

Grass

Ground

12

4

12

4

12

4

12

4

12

100

100

100

10.5

7.7

49.9

14.3

6.2

2.8

96.4

0

0

0

11.5

13.3

16.8

5.9

84.3

97.2

0.8

0

0

0

78

79

33.3

79.8

9.5

0

5.3 Fine Spatial Observation and Assessment of Habitat Ecological …

187

The two intervals [r1 , r2 ] and [r3 , r4 ] in this section represent the adaptive selection range of each point size scale, respectively. The unit increments in the intervals are set to 0.5 cm, and the size of the two intervals depends on the distribution of tree species in the experimental area, but an excessively large interval range will significantly increase unnecessary calculations. For each laser point in each scale interval, the entropy function corresponding to each radius is calculated as shown in Eq. (5.36), and the radius corresponding to the minimum entropy in the interval is the optimal radius of the point at the corresponding scale. E = −a1D ln(a1D ) − a2D ln(a2D ) − a3D ln(a3D )

(5.36)

While most of the studies have been conducted to calculate multi-scale features based on the original point cloud to calculate geometric features at different scales, this section adopts a decreasing scale by scale method of point cloud to calculate multi-scale features. After calculating the small-scale features, only those points marked as planes are retained and participate in the large-scale feature calculation. In particular, to ensure the stability of the feature calculation, when the number of laser points in the neighborhood of a point is less than 5, then this radius will not be used as the optimal radius. If no optimal radius exists in the interval, the point is marked as a scattered distribution. 2) Clustering After two-scale labeling, most of the trunk points are preserved, while a large number of other feature points are eliminated. Due to the ambiguity of the feature analysis and the noise interference in the spatial distribution analysis, some branch and leaf point clouds are still left. However, in the two-scale classification process, the nontrunk points are gradually removed, so the residual non-trunk points are sparser than the trunk points. These characteristics of spatial distribution can be applied to the refinement of trunk point cloud. In this section, the Euclidean clustering algorithm is used to aggregate the discrete unorganized trunk points into trunk point sets by distance. The Euclidean clustering algorithm merges point clouds by spatial distance. For any two points, they are considered to belong to the same class if the distance between them is less than de; otherwise, they belong to different clusters. The trunk point cloud is traversed and the individual clusters are updated until all laser points are assigned to a certain cluster. The point spacing and distance thresholds determine the number of points contained in the clusters. Overall, the trunk points are larger than the set of non-trunk points. Therefore, point sets with set sizes below N c are excluded, and the remaining clusters can be considered as trunk point clouds. The clustering threshold d e is the only threshold to be determined by the clustering algorithm. To improve the automation of the algorithm, the clustering threshold uses the mean value of the best radius of the planar class point cloud when calculated at small scales.

188

5 Spatial Observation and Assessment of Ecological Changes in Giant …

3) trunk clustering merging Through cluster analysis, disordered trunk points are aggregated into scattered point sets, and this section proposes to merge the point sets belonging to the same trunk. In fact, the same trunk point cloud is often divided into multiple point sets due to the occlusion caused by the dense branches and leaves, and the mutual occlusion between the aggregated trunks. In sparse forests, disconnected trunks can be spatially merged by setting a distance threshold, but considering the situation that trunks are aggregated and bent and interlaced in bamboo forests, etc., a suitable and effective method for clustering and merging trunk point clouds needs to be investigated. In this section, we propose a clustering and merging method based on directional growth, which also sets only a distance threshold, but can meet the application needs of trunk merging in high-density bamboo forests. First, to simplify the problem, assume that the trunk can be approximated by a spatial second-order curve, as in Eq. (5.37). ⎧ ⎨ x = x(z) y = y(z) ⎩ z=z

(5.37)

where x(z) and y(z) are horizontal coordinate changes as a function of tree height z. Figure 5.24a represents a typical trunk curve in three dimensions, and ouz is the local two-dimensional coordinate system of the trunk curve, where u is parallel to the horizontal projection of the curve ab, which can be a straight line in any direction in the plane. Equation (5.38) represents the expression of the trunk in the ouz plane, where a, b, and c are function parameters. When the relationship between x and y is linear, the line ab can be expressed as a function of x and y, as shown in Eq. (5.39). If Eq. (5.40) is substituted into Eq. (5.39), the value of y in equation can be eliminated, and then the line ab can be expressed as Eq. (5.41). Then the left side of Eq. (5.38) is replaced by the right side of Eq. (5.41) to obtain Eq. (5.42), and the horizontal coordinate x(z) can be obtained as a function of elevation. Similarly, an expression for y(z) can be obtained. Note that Eq. (5.40) is generally expressed as y = kx + m, where k denotes the slope and m denotes the intercept. When a cluster is being merged, the cluster can be moved to the origin by a left translation and translated back to the original coordinates at the end of the calculation. Therefore, the intercept can be considered very close to 0 and is rounded off in the calculation. u = az 2 + bz + c u=

√

x 2 + y2

(5.38) (5.39)

5.3 Fine Spatial Observation and Assessment of Habitat Ecological …

189

Fig. 5.24 The clustering of trunk point clouds

y = kx

(5.40)

√ u = ± (k 2 + 1)x = K x

(5.41)

x(z) = Az 2 + Bz + C ( A= K , B= K , C= K , K /=0) a

b

c

(5.42)

The direction T (Tx , Ty , Tz ) of the trunk can be expressed as the first-order derivative of Eq. (5.41), as shown in Eq. (5.43). Where Ai and Bi are direction parameters. ⎧ ⎨ Tx = x ' (z) = A1 z + B1 T = y ' (z) = A2 z + B2 ⎩ y Tz = z ' = 1

(5.43)

The main idea of the directional growth algorithm is that if a cluster grows along the trunk direction and can intersect with another cluster, then both clusters belong to the same trunk. The directional parameters of the clusters can be obtained by PCA analysis, where the eigenvector corresponding to the largest eigenvalue is approximated as the main direction of the cluster, and the elevation z is used as the elevation of the cluster center. In the “growth” process, the highest point of the lower cluster

190

5 Spatial Observation and Assessment of Ecological Changes in Giant …

in the two clusters is used as the growth point, and as the elevation increases, the coordinates of the growth point are continuously updated according to Eq. (5.44) until the elevation of the growth point reaches the elevation of the lowest point of the higher cluster. The elevation increment for elevation growth can be set to 1 cm. When the growth is finished, if the horizontal distance from the growth point to the higher cluster is less than the threshold dstem , the two clusters are considered to belong to the same trunk; otherwise, they belong to different trunks. Figure 5.24b represents an example of directional growth where the two higher clusters have similar orientations, but when the lower cluster grows, the cluster to the left will be merged. The pseudo-code for cluster merging is shown in Algorithm 1. ⎧ Tx (z) ⎪ ⎨ x = x + Tz (z) · Δz T (z) y = y + Tyz (z) · Δz ⎪ ⎩ z = z + Δz

(5.44)

Algorithm 1. Trunk clustering merging algorithm Input: Point set C Output: Trunk list C s Initialize an empty C s for each ci ∈ C do for each trunk cs j ∈ Cs do Find the closest cluster ck ∈ cs j to ci Calculate the direction vectors of ck and ci Solve Eq. (5.42) According to Eq. (5.43), the lower clusters grow until the elevation is not less than the higher clusters. if horizontal distance (ck , ci ) < dstem then add ci to cs j ; break; end if end for if ci cannot be added to either cs j ∈ Cs then Create a new trunk and add to Cs. end if end for

5.3 Fine Spatial Observation and Assessment of Habitat Ecological …

191

3. Experiments and results The two-scale intervals were set to [1, 4] and [7, 15] cm for both sample sites, and the size threshold Nc was set to 50 indicate that only point sets containing more than 50 points were retained after filtering clusters at both scales, which depended on the density of the original point cloud. The last cluster merging parameter, dstem is set according to the plant density in the forest area, and 8 cm is chosen as the distance threshold to distinguish adjacent tree trunks in this section. It should be noted that since there are still some felled bamboos on the ground, those tree trunks with elevations below 30 cm after cluster merging are excluded. Figure 5.25 represents the front view of the original point cloud and the results of bamboo trunk detection. Figure 5.25a, c show the results of the point cloud rendering based on elevation, while Fig. 5.25b, d show the overlay of the original point cloud and the detection results, and the colors of the bamboo stems are randomly chosen to distinguish from each other. From Fig. 5.25, it can be seen that most of the bamboo pole point clouds are detected. To evaluate the accuracy of the detection, the accuracy (correctness) and completeness (completeness) of the detection are defined as follows: correctness =

T D

completeness =

Fig. 5.25 Original point cloud and detection results

T R

(5.45) (5.46)

192

5 Spatial Observation and Assessment of Ecological Changes in Giant …

Table 5.9 Analysis of trunk test results Sample

Reference bamboo

Detected bamboo

Type I error

Type II error

True value

Accuracy/%

Integrity/%

A

82

78

1

5

73

93.6

89.0

B

84

79

2

6

73

92.4

86.9

166

157

3

11

146

93.0

88.0

Sum

where D is the number of trunks automatically detected by the algorithm; real trunks (T ) are the number of trunks accurately detected. The reference trunk (R) is the number of trunks manually marked in the original point cloud. Error detection mainly consists of two types: Type I error indicates undetected trunks; Type II error indicates trunk merging error. Table 5.9 shows the accuracy evaluation results of the two sample sites. Among all the 157 detected bamboos, 11 belong to Type II errors, and after error merging, 6 trunks contain point cloud clusters of other trunks, 4 detected trunks contain two adjacent clusters, and one detected trunk contains 3 neighboring trunks. In summary, the accuracy of trunk mapping was 93%, while the completeness was 88% 4. Discussion 1) Trunk point identification and Type I error In this section, the trunk points are identified by stepwise filtering through two-scale features. Unlike the related multi-scale feature classification methods, the point cloud in this section retains only the “flat” point set after the first feature calculation, and the advantages of stepwise rejection in this order are: (i) since the branch diameter is much lower than the trunk, then the branch points will be (ii) as the points are removed, the density and spacing of non-trunk points will increase, which helps to finely distinguish trunk points from non-trunk points; (iii) point removal directly reduces the computational effort of the data and increases the labeling speed. In fact, in the small scale, only about 38% of the original point cloud is retained for the large scale feature computation. Figure 5.26 shows the results of two-scale feature computation and labeling, where blue indicates flat feature points, red indicates linear points, and green indicates scattered points, where the red points on the ground in Fig. 5.26b are felled bamboos. In addition, this section uses the optimal radius determination method based on the entropy function, which has been applied to various datasets, and more discussion on the adaptive selection of feature radius can be found in (Yang et al. 2013). Type I error is the number of trunks present in the original point cloud that are completely missing from the detected bamboos. Missing results of this type occur mainly during the scale classification or clustering process. The reasons for the

5.3 Fine Spatial Observation and Assessment of Habitat Ecological …

193

Fig. 5.26 An example of two-scale classification: a classification results at the first scale; and b classification results at the second scale

misclassification are mainly that the missed trunks are distributed at locations far away from the scanner where the occlusion effect is obvious and the point clouds are widely spaced, so they are likely to be marked as scattered points in the classification labeling process or rejected in the clustering analysis due to the small number of points. In the two experimental sample sites, only three bamboo plants were completely ignored and were not the main source of error. 2) Trunk merging and Type II error The study uses the directional growth method to merge the trunk point sets. Despite the large tilt of the bamboo in the test area, many point clouds of bamboo poles divided into multiple subsets appear due to shading, and this method can still detect and correctly merge the majority of trunks. Figure 5.27 shows some local details of the detection. However, 11 bamboo plants were still incorrectly merged, resulting in Type II errors. Figure 5.28 gives several possible scenarios of incorrect merging. Figure 5.28a

194

5 Spatial Observation and Assessment of Ecological Changes in Giant …

Fig. 5.27 a Side view of plot A: Bamboo on the hillside is inclined and stem clusters are fragmentized. b Details of plot B: Stems are clumped and some are seriously inclined

shows that two neighboring bamboo stems are so closely spaced that they are even partially adhered together. If the spacing is less than the threshold dstem , there is a high probability of merging errors during directional growth, resulting in the merging of neighboring stems into one. In the clustering process, closely adjacent trunks are also likely to be merged into one cluster, which directly leads to Type II error. The vast majority of other errors are shown in Fig. 5.28b–e. If there are two tree trunks that lie approximately in the same plane, then merging errors may arise during the growth of seed points. Figure 5.28b shows two independently growing trunks, while Fig. 5.28c, d show that the erroneous growth results in the extracted trunk containing a neighboring part of the trunk. Figure 5.28e indicates that the erroneous growth may result in the neighboring trunks merging into one. In the study, 10 detected erroneous trunks belonged to this category. Although the growth merge error due to small spacing seems to be unavoidable, two approaches can be taken to reduce this error. The first approach is to set a smaller dstem which is more effective in reducing Type II error. For example, Fig. 5.29a shows that three bamboo plants in sample plot B were merged into one plant (magenta), and one bamboo plant (red) was separated from the three plants by setting dstem = 5 cm. The second approach is to use a larger threshold N c to remove the fine trunk set, because when the trunk set is small, the error in its direction estimation tends to be larger, thus affecting the growth process of the growing points. In Fig. 5.29c, the three merged trunks were separated automatically by setting N c = 80. However,

5.3 Fine Spatial Observation and Assessment of Habitat Ecological …

195

Fig. 5.28 Error types of trunk growth merging. Error types: a Bamboos are too close to each other; b two independent stems; c mutual containment of two stems after growing; d one stem (in red) contains part from the other one; and e two stems are totally merged as one

changing these parameter thresholds could potentially affect the detection results of other plants. A possible approach is to extract the trunk point clouds with detection problems from the original point clouds and set feasible threshold parameters for the problems. Both Type I and Type II errors are influenced by the absence and density of point clouds, and the occlusion effect is the root cause of the absence and density variations. In fact, the occlusion effect is unpredictable and unavoidable in different scenarios. Choosing a station location with less occlusion can reduce the occlusion to some extent, and combining point clouds from multiple stations can also reduce the effect of the occlusion effect. However, it has also been shown that simply increasing the point cloud density or the number of stations does not have much effect on reducing the occlusion effect (Liang et al. 2014; Seidel and Ammer 2014). In fact, the density of the point cloud was already very high in the study. Perhaps, using the detected trunk point clouds, rebuilding the trunk model, and then searching for possible trunk point clouds in the original point clouds can somewhat reduce the false and missed detection rate and increase the density of the extracted trunk point clouds. 5. Conclusion In this section, a new method for extracting tree trunks from single-station groundbased LiDAR point clouds is proposed and applied to high-density bamboo forest areas. The method does not require a circle or cylinder fitting process, nor does it need to consider complex ground surfaces. A two-scale feature labeling method is applied to trunk identification, and Euclidean clustering is used to cluster trunk points into small sets of trunks, and finally, the sets attributed to the same trunk are merged using

196

5 Spatial Observation and Assessment of Ecological Changes in Giant …

Fig. 5.29 Solution to reduce errors

a simplified trunk model. Experimental results show that the completeness of bamboo pole detection is up to 88% at a vegetation plant density of about 7500 plants/hm2 in the measured area. This section also discusses the error sources of trunk detection, the effects of parameter adjustment on detection results and shading effects, and proposes possible error reduction schemes. With reference to the detection accuracy in sparse forest areas, the multi-scale effect of vegetation cannot be ignored, and the method proposed in this section can effectively extract the trunk point cloud in dense forest areas, expanding the application of single-station LiDAR in forestry and can be applied to other single-trunk tree species.

5.3 Fine Spatial Observation and Assessment of Habitat Ecological …

197

5.3.2 Inversion of Average Tree Height of Bamboo Forest Based on Ground-Based Point Cloud Section 5.3.1 mainly studied the fine trunk mapping, however, most of the current remote sensing applications, especially large scale (large area) resource surveys, environmental monitoring, etc., require more low and medium scale ground verification data. Therefore, it is very important to study the high-resolution tree height inversion at the sample-site scale, which provides high-precision validation data for the inversion results of airborne and satellite-based LiDAR data on the one hand, and on the other hand, the sample-site scale tree height products are easier to be fused with multi-source remote sensing data and can be used as accurate input parameters for various models. So this section focuses on the inversion of understory topography and mean tree height in bamboo forest areas. 1. Data collection and multi-site cloud alignment The plant density in the bamboo forest area is very high, considering the tilt of bamboo, branches and other elements, it is difficult to set up ground 3D laser scanning sites inside the bamboo forest, so when setting up the sites, in addition to setting up as many sites as possible inside the bamboo forest, we also try to set up sites around the measurement area to reduce the shading of the measurement area and improve the completeness of the data, and finally align each measurement site cloud to get the same coordinate system under The complete point cloud data is obtained under the same coordinate system. Point cloud alignment is actually a rigid transformation of spatial coordinates between site data, and the spatial transformation generally adopts the 7-parameter method, including three angular elements (Ω, σ, κ), three translation parameters (Δx, Δy, Δz) and a scale factor (λ, fixed to 1 in TLS data), which is actually 6 parameters. The study employs the least squares principle to solve for the six parameters by manually selecting control points interactively, and this process is known as coarse alignment. However, the error of manual point selection and the measurement error of the same name at each station make the alignment accuracy not meet the requirements, so it is necessary to carry out a more accurate alignment, which is the fine alignment. Currently, the most studied and used method for fine alignment is the iterative closest point (ICP) algorithm. This method first establishes the set of corresponding points in the cloud of two stations according to certain criteria, and then calculates the optimal coordinate transformation, i.e., rotation matrix and translation vector, using least squares iteration to minimize the error. The multi-station alignment criterion first selects two stations for alignment, and then the point clouds of other stations are aligned with the stitched point cloud set in turn. The ICP algorithm based on KD-tree accelerated indexing is used to solve the shortcomings of the traditional ICP algorithm and improve the computational efficiency. Figure 5.30 shows the detailed algorithm flow of the point cloud alignment.

198

5 Spatial Observation and Assessment of Ecological Changes in Giant …

Fig. 5.30 Flow chart of point cloud registration

2. Research methodology 1) Point cloud denoising In the process of point cloud data acquisition, there are many noise points and outliers in the point cloud data due to the defects of the scanner itself, environmental interference and other factors. These points have an important impact on DEM and bamboo forest height inversion. The point cloud denoising method can be divided into the following processes. (i) KD tree indexing of point clouds. (ii) Calculate the total distance dist for each point in the point cloud to the surrounding K points. (iii) Frequency histogram of the total statistical local distances. (iv) Set the threshold to reject noise points. The frequency histogram is assumed to be Gaussian distributed, and a point is considered noisy if dist is greater than its overall mean and t times the standard deviation. Where t can be adjusted according to the data characteristics. As shown in Fig. 5.31, the point cloud is rendered based on the elevation. Figure 5.31a shows the point cloud before denoising, and the points inside the red circles are considered as noise points, while Fig. 5.31b shows the point cloud after denoising, where t is taken as 3.0. 2) Point cloud filtering Ground-based LiDAR can acquire discrete point cloud data on the surface of a feature, and the process of separating the points from the real surface from the original point cloud is called point cloud filtering. The point cloud filtering is the basis of DEM calculation in forest area. The filtering algorithm used in this section is morphological filtering. Morphological methods are based on the principle of set operations to extract features in images. The basic idea of mathematical morphology is to measure and

5.3 Fine Spatial Observation and Assessment of Habitat Ecological …

199

Fig. 5.31 Example of 3D point cloud denoising

extract the corresponding shapes in an image using structural elements with certain morphology (rectangles, circles, rings). There are four basic morphological operations in mathematical morphology, namely, the erosion operation, the expansion and opening operation and the closing operation. Morphological corrosion can obtain the minimum value within the neighborhood structure, while morphological expansion can obtain the maximum value within the neighborhood structure. The morphological open operation is corrosion followed by expansion, and the open operation can remove isolated targets smaller than structural elements, such as isolated points or isolated surfaces, while the closed operation is expansion followed by corrosion, and the closed operation can fill or connect isolated voids or gaps. Morphological filtering applies morphological operations to point cloud data processing, but requires rasterizing the original point cloud into a two-dimensional image, and the lowest value of the point cloud in each raster space range is used as the pixel value of that point. Specifically, if there are multiple data points in a raster cell, it is more likely that the lowest elevation point is a ground point, so only the elevation of the point with the smallest elevation value is recorded. This is because the filtering is to separate ground points from feature points, which are often lower in elevation than feature points. Since each grid of the raster image has clear geographic coordinates, the rasterized image can correspond precisely to the original point cloud.

200

5 Spatial Observation and Assessment of Ecological Changes in Giant …

In the filtering by morphological method, the point cloud raster image is first corrupted and then inflated. A structure window of size w × w is opened with the discrete points as the center, and the minimum elevation value in the structure window is taken as the corrupted elevation value. After the corrosion, the same structure window is used to traverse the image and the maximum elevation value is taken as the expanded elevation value. If the absolute value of the difference between the expanded elevation value of a discrete point and its corresponding original elevation value is less than or equal to the given elevation difference threshold, the point is a ground point, otherwise it is a non-ground point. It should be noted that a fixed size (w × w) window is difficult to get a good filtering effect, because if the window is too small, most ground points are retained, but large features such as large buildings cannot be removed, while if the window is too large, many ground features will also be removed, such as hills and mountains. In order to get a better filtering effect, a multi-scale filtering window is generally used for iterative operations, and the size of the structural elements is gradually increased during the iterative process. In order to ensure that all buildings are removed, the window size of the last iteration must be larger than the area of the largest building in the data area, and the number of iterations is usually 5–6. The overall flow of morphological filtering is shown in Fig. 5.32.

Fig. 5.32 Slope-based point filtering process

5.3 Fine Spatial Observation and Assessment of Habitat Ecological …

201

3) Calculation of understory terrain Digital elevation model (DEM) is a solid ground model that represents the ground elevation in the form of a set of ordered numerical arrays. Using DEM, some thematic maps such as contour maps and cross-section maps can be easily derived using certain algorithms. In the study, DEM is an important basis for the inversion of vegetation height in sample sites. The 3D laser scanning can obtain the high precision and high density 3D coordinate information of the object, which is very suitable for the production of DEM, and has the characteristics of fast acquisition speed and low processing cost. The essence of laser point cloud generation DEM is to resample the original scattered point cloud and convert it into standard grid data. Currently there are two main methods of resampling (Dietrich and Wilson 1993; Moore et al. 1991): the first is to obtain the elevation of each grid center point in a regular arrangement by interpolation; the second is the block averaging method, where the elevation of each grid center is represented by the average elevation of each point in the grid area. The interpolation method requires high accuracy of node data and low resampling accuracy when there is a large observation error; the selection criteria and search algorithm of interpolation nodes are complicated, and for the case of fixed number of nodes, it requires both nodes to satisfy the interpolation conditions and at the same time to select the nearest node, making the node search algorithm complex and poor in real time. The block averaging method divides the terrain into a square grid, and the elevation of each sampling point falling into the grid is averaged as the elevation of the center point of the grid. When each sampling point is closer to the center of the grid, the block averaging method has a certain anti-noise ability for the random noise with zero-mean distribution due to the noise averaging effect, so the accuracy is higher, and the advantage is that the algorithm complexity is low and it is easy to process in real time. Based on the ground point cloud after filtering and classification, the main process of DEM generation is shown in Fig. 5.33. Fig. 5.33 DEM generation process

202

5 Spatial Observation and Assessment of Ecological Changes in Giant …

Among them, the compression of point cloud data is essentially the simplification of point cloud data, which is currently achieved mainly by filtering algorithms. Data filtering methods are generally Gaussian, average or median and curvature filtering. The average filtering method uses the average value of each data point in the filter window; the median filtering method, as the name implies, takes the statistical median value of each data point in the filter window; the advantages of both filtering methods are simple calculation and small computation, but there is a possibility of losing important topographic information because the influence of topographic factors is not taken into account. The surface filtering is to decide the point trade-off according to the change of curvature, keeping more points where the change of curvature is large and filtering out more points where the change of curvature is small; this method can retain the terrain characteristics well, but its algorithm is too complicated. The Gaussian filter has a Gaussian distribution of weights in the specified domain, and its average effect is smaller, so it can keep the shape of the original data better while filtering, and its algorithm complexity is moderate, which can meet the application demand of vegetation height inversion. The shape of the experimental area is a circle of 35 m radius with the center coordinates of the station as the origin, as shown in Fig. 5.34, where the topographic resolution is set to 0.5 m. In tree height inversion, the average tree height is obtained by DSM (digital surface model) minus topographic DEM, and a high precision DEM is the primary condition for tree height inversion, while the accuracy of constructing a DEM based on ground-based LiDAR point clouds is mainly determined by two factors, namely, the horizontal resolution of the DEM and the point cloud density. The horizontal resolution of DEM is one of the direct factors for the good or bad terrain description of grid DEM, which reflects the amount of information contained in DEM. Although the high resolution DEM is more accurate to describe the real ground surface, it will increase the data computation, and the low resolution DEM cannot meet the application requirements of high precision tree height inversion, Fig. 5.34 Ground point cloud generation DEM

5.3 Fine Spatial Observation and Assessment of Habitat Ecological …

203

therefore, choosing the DEM resolution that can maximize the simulation of real ground relief and avoid data redundancy becomes a necessary prerequisite for DEM terrain analysis. By calculating the slope median error of each resolution DEM, we can quantify the accuracy of different resolution DEMs to simulate the surface microtopography, and then use the principle of terrain factor information content analysis to select and extract the main terrain factor index—roughness index. Finally, it is supplemented by the comparison of the time cost required for microtopography DEM generation, and the best resolution of forest terrain DEM is analyzed comprehensively. In this section, the terrain resolution is set to 0.5 m as needed. Point cloud density is an important factor that affects the accuracy of DEM. Theoretically, the higher the point cloud density, the higher the DEM accuracy. Moreover, in the process of point cloud data acquisition, in order to record the elevation data of the terrain surface as detailed as possible, the scanned point cloud often has a high density. Usually the DEM based on the original point cloud can meet the accuracy requirement, but considering the redundant data in the original point cloud data (e.g., the terrain with flat slope, few point clouds can describe the terrain information). Therefore, it is necessary to choose a reasonable data compression method, which not only can guarantee the accuracy of DEM, but also can obtain the 3D elevation data which is easier to handle and operate. In fact, due to factors such as occlusion and undulation of the forested terrain, ground points will have large missing and large variations in point spacing, so a suitable indexing scheme is also needed in the process of DEM generation. Considering these factors, this section firstly establishes the grid management method and integrates the K-neighborhood finding and radius finding methods for the DEM grid calculation of the point cloud. 4) Average tree height inversion Similar to the generation process of DEM, DSM is produced based on the denoised raw point cloud data (Fig. 5.35). Using the difference between DSM and DEM, the nDSM (normalised digital surface model) of the measurement area can be obtained, as shown in Fig. 5.36. The nDSM appears to be below 0, which is caused by the point cloud filtering error, which is acceptable in practice considering the very small error value (−0.075 m). 3. Analysis of results The bamboo forest in the survey area was mainly distributed in a circular area with a radius of 15 m centered on the survey station. This section crops out the nDSM of the corresponding area, as shown in Fig. 5.37, and calculates the average tree height in this area to be 9.3 m, while the average tree height of the manual measurement sample in the sample site is 8.8 m. Due to the dense forest area, it is difficult to determine the tops of trees accurately when measuring manually, and considering that the measuring instrument relies on human eye interpretation, the point cloudbased inversion of bamboo forest height can be considered to have a higher confidence level. The average tree height of the inversion measurement area by this method also has certain errors. It mainly includes: (i) the mutual shading of tree branches and

204

5 Spatial Observation and Assessment of Ecological Changes in Giant …

Fig. 5.35 Create DSM from denoised point clouds

Fig. 5.36 nDSM: the difference between DSM and DEM

leaves, resulting in the top of the bamboo far from the measurement station part cannot be measured, which leads to the error of tree height calculation; (ii) the error when performing DEM and DSM interpolation calculation, such as the negative value appeared in Fig. 5.36. In practical application, these errors can be reduced by erecting multiple stations and DEM smoothing. In addition, the data collection should try to choose a windless weather to reduce the measurement errors caused by canopy oscillation.

5.4 Fine Observation of Habitat of Giant Panda Based on High-Resolution …

205

Fig. 5.37 Inversion of bamboo height within 15 m of the scanner

5.3.3 Summary The study applies ground-based LiDAR to bamboo forest parameter inversion for the first time and consists of two main parts: the first part proposes a new fully automated method for automatic trunk detection from single-site clouds and for trunk mapping in bamboo forests. The method introduces multi-scale analysis into the vegetation structure classification and proposes a method to distinguish adjacent trunks based on directional growth. The algorithm was tested in bamboo forests with plant density greater than 7500 plants/hm2 , and the overall accuracy of the mapping reached 88%. The second part focuses on the extraction of vegetation parameters at the sample site scale, including the DEM and DSM of the bamboo forest area, and then the average tree height of the survey area, which mainly includes the design and practice of the acquisition scheme, point cloud denoising, multi-site cloud stitching, and the calculation of the DEM and nDSM of the survey area. The algorithms and results of the study also provide an effective solution for remote sensing monitoring of the ecological environment of giant panda habitat and the accurate estimation of bamboo production for giant panda.

5.4 Fine Observation of Habitat of Giant Panda Based on High-Resolution Remote Sensing Images Vegetation is an important component of the environment and one of the important signs reflecting the regional ecological environment, as well as a sign for the interpretation of soil, hydrology and other elements, which is of great indicative significance to the habitat quality evaluation of giant panda habitats. Remote sensing can quickly

206

5 Spatial Observation and Assessment of Ecological Changes in Giant …

and effectively monitor the distribution, species, growth and other kinds of information of large area vegetation, and has unique advantages in vegetation information extraction and vegetation biomass estimation. Therefore, the use of remote sensing technology for vegetation information extraction is one of the important tools for habitat monitoring of giant panda habitats. In recent years, high spatial resolution earth observation technology has been rapidly developed, and high spatial resolution remote sensing images have detailed texture and spatial shape information, rich spectral information and clear detail information, which make it possible to extract fine information of features from the images. The high resolution remote sensing images have been widely used in vegetation information extraction, forest resources survey and animal habitat observation and other related fields. Wildlife habitat observation is the evaluation of the suitability of animal habitats in terms of physical environmental factors, biological environmental factors, and human activity factors. Information on the distribution of vegetation types is an important basis for judging the range of animal activities. Animals have different habits and food habits, and their distribution conditions are different. For example, the main food of giant pandas is bamboo, so the distribution information of bamboo in mountainous areas plays a vital role in determining the suitable habitat of giant pandas. The medium- and low-resolution remote sensing images can identify large areas of lush, bare bamboo forests using image spectral and texture information, while it is difficult to identify bamboo that grows shorter, is hidden by bushes, and is mixed with shrubs. Therefore, fine identification and information extraction of bamboos by high-resolution remote sensing images are expected to be an effective means to delineate suitable habitats for giant pandas. In addition, factors such as roads and their influence areas, tree depression, elevation, and water system are also important factors in the evaluation of habitat suitability for giant pandas.

5.4.1 Study Area Overview Wolong Nature Reserve is the first batch of giant panda nature reserves established in China, and is also the largest nature reserve in Sichuan Province with the most complex natural conditions and the most rare plants and animals. The reserve straddles Wolong and Gengda townships, with a total area of about 200,000 hm2 , mainly protecting the natural ecosystem of the southwest alpine forest area and rare animals such as giant pandas. The reserve is located at the western edge of Sichuan Basin, southeast slope of Qionglai Mountain Range, southeast of Aba Tibetan and Qiang Autonomous Prefecture, west of Yingxiu Town, Wenchuan County, upstream of Minjiang River and deep valley of Chengdu Plain to Qinghai-Tibet Plateau transition, 102°52' –103°25' E, 30°45' –31°25' N, 60 km wide from east to west, 63 km long from north to south. It is connected with Yingxiu Township of Wenchuan County in the east, Baoxing and Xiaojin County in the west, adjacent to Dayi and Lushan Counties in the south, and adjacent to Li County and Cao Po Township of Wenchuan County in the north. The Wolong Nature Reserve is in danger of declining due to the

5.4 Fine Observation of Habitat of Giant Panda Based on High-Resolution …

207

expansion of farmland reclamation, forest harvesting and industrial facilities. After the devastation of the 2008 Wenchuan earthquake, the survival environment of wild pandas in Wolong Nature Reserve is even worse. The scattered and fragmented distribution of bamboos in Wolong Nature Reserve, mostly shaded by tall trees and easily mixed with bushes, poses a great challenge to the identification and extraction of bamboo forest information. The use of WorldView2 images in this area is expected to improve the accuracy of bamboo information extraction and mapping, and thus to effectively identify suitable habitats for giant pandas. The WorldView-2 remote sensing images used in the study cover the area of Wuyipeng Giant Panda Field Ecological Observatory in Wolong Nature Reserve, as shown in Fig. 5.38. The data was acquired on January 14, 2014, and the coverage area ranged from 30°58' 41.1'' –31°0' 9.6'' N, 103°8' 57.2'' –103°10' 34.1'' E. The data contained one panchromatic band with 0.5 m spatial resolution and eight multispectral bands with 2 m spatial resolution. The eight multispectral bands were as follows: Coastal band (0.400–0.450 µm), blue band (0.450–0.510 µm), green band (0.510–0.580 µm), yellow band (0.585–0.625 µm), red band (0.630–0.690 µm), red-edge band (0.705–0.745 µm), and NIR band 1 (0.770–0.895 µm) and NIR band 2 (0.860–1.040 µm). The elevation range of the study area is from 1900 to 3450 m, and the land cover types are divided into bamboo, coniferous forest, broadleaf forest, mixed coniferous and broadleaf forest and shrubland, as well as and shade. Most of the bamboos that giant pandas can eat grow in the mixed coniferous and broad forest areas above 2000 m elevation.

Fig. 5.38 The study area in Wuyipeng (the true color composition of the WV-2 MS Bands 5, 3 and 2 as red, green and blue channels, respectively)

208

5 Spatial Observation and Assessment of Ecological Changes in Giant …

5.4.2 Overview of Fine Observation Studies of Giant Panda Habitats from High-Resolution Remote Sensing Images In recent years, scholars at home and abroad have conducted a lot of studies on remote sensing information extraction and mapping of bamboo and tree species (Chernet 2008; Wang et al. 2009; Du et al. 2010). Most of the studies have used low- and medium-resolution remote sensing images to map large areas, such as Landsat TM/ETM+ (Bai et al. 2012; Carvalho et al. 2013; Fan et al. 2014) and MODIS data (Sun 2011), and other data also include multi-temporal data, such as IRS data and hyperspectral data. In recent decades, with the rapid development of high-resolution satellite sensor technology, more and more studies have focused on using high-resolution multispectral satellite images (e.g., IKONOS, QuickBird, OrbView, GeoEye, RapidEye, WorldView, etc.) to address tree species classification problems (Ghosh and Joshi 2014). In addition, some progress has been made in using high-resolution images for class-specific land cover classification studies, such as Kamagata et al. (2005) using image-based and object-oriented classification methods and IKONOS images for forest landscape determination, Ouma and Tateishi (2006) using texture-based classification methods for QuickBird images, and Araujo et al. (2008) used QuickBird images for bamboo forest mapping. The use of highresolution remote sensing images for fine vegetation mapping will be one of the hot spots of vegetation remote sensing research in the future. It has been shown that the use of WorldView-2 images has great potential for tree species classification mapping. For example, Immitzer et al. (2012) investigated the recognition ability of eight bands of WorldView-2 images for more than 10 tree species in Australia; Pu and Landry (2012) explored the tree species extraction potential of WorldView-2 for urban areas and compared the recognition ability with IKONOS images; Ghosh and Joshi (2014) compared various methods for bamboo classification using WorldView-2 images. Efficient classification methods are crucial for high-resolution image processing, and their algorithm research has received great attention. The object-oriented image analysis method has been developed since the late 20th century and has been successfully applied to high-resolution remote sensing data processing. This technique overcomes the shortcomings of traditional methods such as inability to accurately classify detailed high-resolution images and reduces salt-and-pepper noise in classification results, and makes it possible to finely classify features. When interpreting images, the human eye automatically aggregates a piece of area with the same color and determines its category attributes based on experience, instead of resolving image pixels one by one. The object-oriented classification method refers to the function of the human eye in aggregating information, and firstly, it uses image segmentation technology to extract connected, spectrally and texturally homogeneous image pixels at a certain scale by certain computational laws, which are called segments. The image pixels within the segments have minimum heterogeneity. The object-oriented image analysis method takes these segments as the basic unit for classification, and then detects and extracts various features (such as spectrum, shape, size, structure, texture,

5.4 Fine Observation of Habitat of Giant Panda Based on High-Resolution …

209

shadow, spatial location, etc.) of the target feature according to the specific requirements of remote sensing image classification, and finally classifies all segments. The essence of this technique is to analyze remote sensing images from a higher level with segments as the smallest unit for classification or detection, so that the extraction results contain richer semantic information. The commonly used object-oriented classification methods are broadly classified into two categories: simple discriminant rule-based classification methods and pattern-based classification methods. The former obtains the category of segments by comparing the attribute information of segmnts in different bands (or features), given the feature range value of the attributed category; the latter’s discriminant rule is similar to that of image-based classification, which is divided into supervised and unsupervised classification, and the commonly used methods are maximum likelihood, K-nearest neighbor, fuzzy classification, artificial neural network, support vector machine, etc. The object-oriented classification algorithms are developing continuously, such as Wu et al. (2007) used vector data to assist the determination of segement boundaries in object-oriented classification; Kim et al. (2009) combined texture analysis based on grayscale co-occurrence matrix with object-oriented classification for forest mapping; Bhaskaran et al. (2010) proposed an integrated imageoriented and object-oriented classification method, which improved the accuracy than the simple object-oriented classification method. Polychronaki et al. (2013) used an object-oriented method with normalized difference index to detect fire areas; Huang et al. (2014) used sparse decomposition of high-resolution remote sensing data to detect line features. The current mainstream methods of object-oriented classification discriminant rules still use the spectral information or geometric shape features of segments, often ignoring the spatial relationships between segments or making insufficient use of spatial relationships. In the absence of auxiliary data, how to improve the classification accuracy by using the spatial location information provided by known sample points is a key issue in current object-oriented classification. Some related techniques of object-oriented classification algorithms have also received considerable attention. For example, Chen et al. analyzed the spatial scale of object-oriented methods in 2003; Grenier et al. analyzed the sampling strategy and used object-oriented classification methods to extract wetland categories in 2008; Sun et al. (2009) used a combination of multiple classifiers for high-resolution remote sensing target recognition; Aguilar et al. (2012) pointed out that selecting different features for different categories would achieve better results by classifying GeoEye-1 images of urban areas; Saibaiti and Du Peijun proposed an object-oriented classification method based on multiple example learning in 2012; Yu et al. proposed an unsupervised object-oriented classification method in 2012; Zhang Xuehong studied the spatial heterogeneity and scale effects of eucalyptus forests using high-resolution remote sensing in 2012; Gao et al. (2013) analyzed coastlines using an object-oriented spatio-temporal feature model of remote sensing. MyBurgh and van Niekerk (2013) compared three common object-oriented classification methods; Zhuo et al. (2013) used object-oriented method to extract 3D information from buildings; Zhu et al. (2014) used object-oriented technique to extract wetland information in layers; Liu

210

5 Spatial Observation and Assessment of Ecological Changes in Giant …

et al. (2014) used random forest model for high-resolution remote sensing classification; Pang et al. (2014) studied the effect of high-resolution image alignment error on classification. There are also many comparative studies based on image pixel and object-oriented classification methods (Rittl et al. 2013). In these techniques, the issue of the scale of the segments has not received sufficient attention. The properties that one observes and the laws that one summarizes at one scale may be valid and similar at another scale, more often they need to be corrected (Wu et al. 2015; Zhang et al. 2014). In supervised classification, segments with a certain face size are usually treated as known sample points, and a series of issues such as how to convert the scales in this process, what effects will be brought by scale inconsistency, and how to measure the spatial distance of segments with different face sizes when introducing spatial relations are to be studied in depth. Object-oriented classification is mainly applied in the fields of change detection, urbanization mapping, ecological habitat identification, urban biodiversity assessment, landscape epidemiology studies, open pit mine information extraction, vegetation classification, glacier classification (Rastner et al. 2014), water body and wetland vegetation monitoring, bamboo forest extraction, coastal zone extraction, impervious surface extraction, and engineering monitoring. In addition, some useful explorations have been carried out by domestic scholars for specific experimental areas. For example, High-resolution image data classification experiments in Nanjing urban area by Du et al. (2004), in Dujiangyan region by Zhang et al. (2010), and in Inner Mongolia grasslands by Yu et al. (2014). Hou et al. (2010) extracted feature information such as residential land in Li County, Sichuan; Wang (2008) used Futian District, Shenzhen for vegetation extraction; Liu et al. (2012) identified and mapped rubber forests in Xishuangbanna; Zhang et al. 2014 carried out the recommended application of nitrogen fertilizer for winter wheat using GeoEye imagery; Chen and Cheng (2015) extracted wetland vegetation along Hangzhou Bay using the spectral features of the fused image of HSPA-1, etc. In general, existing object-oriented classification methods do not make sufficient use of spatial relationships compared to image-based classification methods, Tobler’s first law of geography states that all objects are related to other objects, and objects that are close together are more closely related than objects that are far away (Tobler 1970). According to this law, image pixel-based classification algorithms can use spatial and textural information to aid in classification. Classification methods that incorporate texture information into spectral information are collectively referred to as spatial contextual feature-based classification methods (Franklin and Peddle 1990). The basic idea of this classification method is to improve the classification accuracy by reassigning categories to the results of spectral classification based on spectral classification by borrowing spatial relationship modeling of adjacent image pixels with the point to be classified. Studies have shown that this method has higher accuracy than classification methods based solely on spectral information (Zhang et al. 2014). The common classification methods based on contextual features are auto-correlation function, grayscale co-occurrence matrix, fractal dimension method, Markov random field model, and Gabor filter model. Among these methods, the

5.4 Fine Observation of Habitat of Giant Panda Based on High-Resolution …

211

grayscale covariance matrix is the most widely used, but it is computationally intensive, inefficient to implement, and difficult to choose parameters. Moreover, when calculating the grayscale co-occurrence matrix, the grayscale of the image is often compressed and the image information is lost. The fractal dimension method is a classification method based on fractal geometry. The fractal-based representation is qualitatively related to the roughness or texture of human perception, while the calculation of the fractal dimension provides an automatic quantitative analysis method, but the fractal dimension method often has the same fractal dimension value with different textures. The Markov random field model only considers the correlation between the current image pixels and the neighboring image pixels, which reveals the high-frequency features of the texture but ignores the more low-frequency features, and is suitable for textures with relatively small neighborhoods. Currently, the use of spatially dependent remote sensing image classification methods has become a new research hotspot.

5.4.3 Data Acquisition Two field trips were conducted to the Wuyipeng for fine information extraction of the area. The first expedition was conducted on June 11, 2014, to collect feature points for geometric correction of the images and training sample points required for image classification. The second visit was on September 11, 2014, to check the classification results. The instrument was a handheld GPS (GeoXH TM 6000) from Tiaobao, with a positioning accuracy of 10 cm. Because of the poor signal on the mountain and the tree canopy, an antenna was attached to the GPS and fixed to a 2 m alignment pole to ensure that the signal of at least 3 stars could be detected even under the trees. Eight feature points (including road intersections and house corners) were collected during the first expedition for geometric correction of the images. In addition, the coordinates and category information of 37 points were collected in the field. Since the images had not been geometrically corrected at the time of the first sample acquisition, precise coordinate information was lacking, resulting in some points falling in the shadow area and could not be used as training samples. The final selected samples for typical categories are: 3 points for bamboo, and 4 points for each of the remaining categories (coniferous forest, broadleaf forest, mixed forest, shrubland, clearing and shadow), for a total of 27 points as the initial training samples. For the object-oriented classification method, the first multi-resolution segmentation was performed on the experimental area images, and all 8 bands were involved in the segmentation with a scale parameter of 10 and shape and tightness factors of 0.1 and 0.5, respectively, resulting in 62,979 image segments. 27 training samples are obviously insufficient compared to this number of segments, so it is necessary to increase the number of training samples for classification.

212

5 Spatial Observation and Assessment of Ecological Changes in Giant …

5.4.4 Research Methodology The new bands added to WorldView-2 have unique advantages for vegetation identification. Before classification, the best combination of classification bands can be found by principal component analysis. In the principal component analysis of the 8 WorldView multispectral bands, the importance of each principal component is shown in Table 5.10, and the band loadings of each principal component are shown in Table 5.11. Table 5.10 shows that the first three principal components are critical, and the cumulative weight of the variance is nearly 1.0, while the variance of the first principal component is almost six times that of the second principal component, and the weight of the variance of the first principal component has reached 0.96. Table 5.11 shows that bands 6 to bands 8 in the first principal component have the largest band loadings, so we choose these three bands (red-edge band and two near-infrared bands) as the bands involved in the Therefore, we selected these three bands (red-edge band and two NIR bands) as the bands for subsequent classification. Based on the selected 27 training samples, the box plots of the 8 multispectral bands for the 7 categories are shown in Fig. 5.39, and the spectral values are taken as the mean values of the segments. The top and bottom horizontal lines of the box plot are the upper and lower quartiles, respectively, and the middle line is the median. It can be seen that the red-edge band and the two NIR bands are more class-separable than the other bands. The spectra of bamboo can be completely separated from other categories in these three principal component bands, but the spectra of the two categories of mixed forest and shrubland have partial overlap. Based on the above spectral analysis, the spectral mean and variance of each category in the principal component bands can be used to further select samples. For the mean μ and variance σ, given a parameter t, the selected sample spectra range is μ ± tσ. The selection follows two principles: (i) for different categories, there is no or little overlap in the spectra between each band; and (ii) the sample size of each category should be relatively appropriate with respect to the total number of segments. Figure 5.40 shows the spectral distribution of the training samples on the three principal component bands, and the arrows on the bars indicate the spectral ranges of the expanded training samples. The parameters t for the expanded samples are given in Table 5.12. As shown in Table 5.12, the total number of training samples obtained by expanding the training samples is 748, which accounts for 1.2% of the total number of image segments (62,979). Figure 5.41 shows the spatial distribution of the expanded samples. It can be seen that the number of samples in the categories of open space and shade is not as many as other vegetation categories, but the area of segments in these two categories is mostly larger than other categories because the distribution of these two categories is more aggregated and continuous, and it is easy to form large segments in the image segmentation process, while the distribution of bamboo and other vegetation is more fragmented and scattered.

33.78

182.27

0.96

0.96

Variance

Weight of variance

Cumulative weight of variance

0.99

0.03

Main components 2

Main components 1

Item

1.00

0.01

14.21

Main components 3

Table 5.10 Importance ranking of principal components

1.00

0.00

8.31

Main components 4

1.00

0.00

5.35

Main components 5

1.00

0.00

3.05

Main component 6

1.00

0.00

2.13

Main components 7

1.00

0.00

1.74

Main components 8

5.4 Fine Observation of Habitat of Giant Panda Based on High-Resolution … 213

−0.753

0.274

0.127

−0.570

−0.637

Band 7

Band 8

0.542

0.253

−0.474

Band 6 0.486

−0.805

0.101 0.298

−0.547

−0.426

−0.126

Band 4

Band 5

−0.227

−0.518

−0.137

Band 3

0.187

−0.146

−0.411

−0.312

0.117

0.589

Main component 5 0.560

−0.120

Main components 4

−0.198

Main components 3

−0.336

Main components 2

Band 1

Main components 1

Band 2

Item

Table 5.11 Band loadings for each principal component

−0.147

0.326

−0.705

0.459

0.125

−0.369

Main components 6

−0.159

−0.596

0.171

0.643

−0.365

−0.180

Main components 7

0.297

−0.220

0.150

−0.638

0.658

Main components 8

214 5 Spatial Observation and Assessment of Ecological Changes in Giant …

5.4 Fine Observation of Habitat of Giant Panda Based on High-Resolution …

215

Fig. 5.39 Distributions of the reflectance of different land cover types across eight MS bands

Fig. 5.40 Distributions of the reflectance of different land cover types across five bands and expanded training data

For the test samples, which corresponded to the bamboo category were verified in the second field trip, and the last 48 points were selected as the test samples for the bamboo category. However, due to the limitation of mountain roads in the Wuyipeng area, the expedition route was only distributed in the upper left band of the image. Therefore, for the other categories, their test samples were obtained by manual selection on the image. To ensure the accuracy of the test samples, image fusion was used to enhance the spectral information of the images. The Gram-Schmidt method was used for the fusion, and the test points were selected directly on the synthetic image. The two categories of empty space and shadow are easier to distinguish,

216

5 Spatial Observation and Assessment of Ecological Changes in Giant …

Table 5.12 Parameter selection for expanded training samples with expanded spectral range and number of samples Category

t

Spectral range (red-edge band, NIR1, NIR2)

Sample number

Bamboo

0.15

(122.6, 128.1), (118.7, 124.9), (134.8, 141.5)

32

Coniferous forests

0.4

(335.8, 360.9), (377.6, 406.1), (401.5, 435.8)

186

Broadleaf forest

0.3

(293.7, 321.0), (318.0, 347.7), (342.8, 375.7)

155 146

Mixed forests

0.15

(210.7, 238.8), (224.6, 257.6), (239.3, 274.7)

Shrubs

0.15

(161.2, 171.0), (168.3, 180.5), (188.6, 204.3)

95

Open space

0.8

(79.2, 108.8), (56.6, 81.3), (53.1, 78.8)

40

Shade

0.8

(49.4, 52.1), (35.5, 38.8), (35.3, 40.5)

94

Fig. 5.41 Spatial distribution of the samples: a training data and b testing data

so only a few test points were selected for these two categories. The numbers of test samples for each category were 48 for bamboo, 38 for coniferous forest, 27 for broad-leaved forest, 39 for mixed forest, 34 for shrub, 6 for open space, and 9 for shading. The spatial distribution of the test samples is shown in Fig. 5.41. Although some points appear to be close in spatial distribution, they do not belong to the same segment, so that the accuracy calculation is not disturbed by the cumulative accuracy of multiple points falling in the same segment. Since there is no large distribution of bamboo in the study area, it is difficult to use bamboo texture information to assist in classification. Here, a spatial weightbased K nearest neighbor classifier (kg-NN) is used for the object-oriented classification method to extract bamboo information. In the traditional K-nearest neighbor approach, the classifier criterion is to classify the image pixel to be classified into

5.4 Fine Observation of Habitat of Giant Panda Based on High-Resolution …

217

the class of image pixels whose samples are at the nearest distance in the feature space. For the K-nearest neighbor classification method based on spatial weights, the probability that an image pixel u belongs to the class m can be calculated by the following equation. ΣK [

] Sg × pm,m (h uk ) × ωuk + (1 − Sg ) × ωuk pgK−NN [c(u) = m] = Σ M Σ K [ ] m ' =1 k=1 Sg × pm,m ' (h uk ) × ωuk + (1 − Sg ) × ωuk (5.47) k=1

where h is a vector measuring the spatial distance; the subscript uk of h is the distance between the image pixel u and its neighbor k; Pm,m' (huk ) is the fitted model of the spatial covariance function (also called the category conditional probability). The symbol m' (m' = 1, …, M) is the category code and m is the category of interest; S g is a weighting factor taking values from 0 to 1. A larger S g indicates a larger spatial weighting component of the probability. Given the category m' and a neighbor K with spatial distance h, the category conditional probability Pm,m' (huk ) that the image u belongs to the category m can be calculated by the following equation. ΣN pm,m ' (h uk ) =

i=1

| ] [ I c(h) = m ' |c(u) = m ΣN i=1 I [c(u) = m]

(5.48)

N is the total number of training samples; c(h) is the category of the neighbor K with distance h. The indicator function I takes the value of 1 if the condition is satisfied, otherwise it is 0. The commonly used models for fitting the conditional probability map of the category are spherical model, exponential model and Gaussian model. The advantage of the K-nearest neighbor method based on spatial weights is that the method takes into account the spatial dependence between the unknown point locations and the nearest neighboring sample points in the feature space, so the spectral and spatial information will jointly determine the classification results. For comparison with the K-nearest neighbor method based on spatial weights, two other more commonly used classification methods are used: classification and regression tree (CART) and support vector machines (SVM). The classification and regression tree algorithm is a nonparametric classification method proposed by Breiman et al. (1984). The algorithm uses a dichotomous recursive partitioning technique that divides the current sample set into two subsample sets such that each non-leaf node of the generated decision tree has two branches. Therefore, the decision tree generated by this algorithm is a binary tree with a simple structure. The CART algorithm assigns categories to each node considering that each node has the possibility of becoming a leaf node. The method of assigning categories can use the most occurring category in the current node, or it can refer to the classification error of the current node or other more complex methods. the CART algorithm uses post-pruning, where some information is found by unfolding one more layer during the tree generation, and the CART algorithm runs to the position where no more branches can grow, thus

218

5 Spatial Observation and Assessment of Ecological Changes in Giant …

obtaining a maximum decision tree, and then pruning this large tree. Support vector machine is a classification boundary based method. The basic principle is (taking two-dimensional data as an example): if the training data are points distributed on a two-dimensional plane, they are clustered in different regions according to their classification. The goal of a classification boundary-based classification algorithm is to find, through training, the boundaries between these classifications (linear boundaries are called linear divisions and curved boundaries are called nonlinear divisions). For multidimensional data (e.g., N-dimensional), they can be considered as points in an N-dimensional space, and the classification boundaries are the faces in the N-dimensional space, called hyperplanes (hyperplanes are one dimension less than the N-dimensional space). Linear classifiers use hyperplane type boundaries, and nonlinear classifiers use hypersurfaces. According to the classification method introduced above, the spatial relations are first modeled by estimating the category conditional probabilities of each category in order to incorporate the spatial relations into the K-nearest neighbor classifier. To obtain the spatial distance between each segment, the center of gravity of each segment is first extracted, and the Euclidean distance between any two segments can be deduced from the coordinates of the center of gravity positions. The category conditional probability map with spatial covariance fitted model calculated by 748 sample points for each category is shown in Fig. 5.42, each anisotropy is not considered, and the fitted parameters of the model are shown in Table 5.13. According to the established spatial relationship model, the conditional probability is a function of spatial distance. The K-nearest-neighbor classification method based on spatial weights is calculated by expanding Eq. (5.47), where the training sample is a segmented segment and the three principal component bands are involved in the classification. The conditional probability of each category is obtained from the nearest neighbor segments, where K = 5, the spatial distance measure is the center of gravity of the segments, and the spatial weight parameter Sg is set to 0.4. For comparison with this method, the conventional K-nearest neighbor classification method, CART, and SVM classifier are each implemented once in this study area based on the same classification conditions. The tree depth of the CART method is 6. Figure 5.43 shows the rules of the decision tree, and it can be seen that most of the categories are divided by the rules of NIR band 1, and the red-edge band distinguishes the two categories of open space and shadow, and NIR band 2 can distinguish between bamboo and shadow.

Fig. 5.42 Category conditional probability diagram and fitted model for each category

5.4 Fine Observation of Habitat of Giant Panda Based on High-Resolution … 219

220

5 Spatial Observation and Assessment of Ecological Changes in Giant …

Table 5.13 Conditional probability fitting models for each category (units of variation are the number of pixels) Category

Models

Abutment value

Variable range value

Block gold value

Bamboo

Exponential models

0.80

70

0.20

Coniferous forests

Exponential models

0.75

25

0.25

Broadleaf forest

Exponential models

0.85

30

0.15

Mixed forests

Exponential models

0.80

15

0.20

Shrubs

Exponential models

0.85

35

0.15

Open space

Exponential models

0.80

120

0.20

Shadow

Exponential models

0.90

70

0.10

Legend

Yellow < 87.19

Feature and threshold Shadow

Yellow < 80.62

Yellow < 112.39

Bamboo

Mixed woodland

Red < 64.93 Coniferous

Broadleaved

Brush

Bamboo

NIR1 < 319.19

Coniferous

Yellow < 131.37

NIR1 < 200.35

Brush

(a)

FALSE

NIR1 < 354.93

Yellow < 92.51

Red edge < 244.09

Barren land

TRUE

NIR1 < 130.42

Barren land

Red edge < 118.92

Shadow

NIR1 < 94.15

Mixed woodland

Broadleaved

Mixed woodland

(b)

Fig. 5.43 Decision rules of the CART classification based on a original training data and b expanded training data

5.4.5 Analysis of Results Figure 5.44 show in order the spatial weight-based K-nearest neighbor classification method, the traditional K-nearest neighbor classification method, the classification results of CART and SVM methods. As can be seen, Fig. 5.44a, b are very similar in the spatial distribution of the categories, but the bamboo category in red appears more in the K-nearest neighbor classification results based on spatial weights. Figure 5.44c, d show that the classification results of CART and SVM have more pixels classified into shrub categories, and the SVM has significantly more bamboo categories than the other methods. Figure 5.44 gives the classification error matrix for the four classification methods. The error matrix, also known as the confusion matrix, is an array used to represent

5.4 Fine Observation of Habitat of Giant Panda Based on High-Resolution …

Fig. 5.44 Classification results of the Wuyipeng area

221

222

5 Spatial Observation and Assessment of Ecological Changes in Giant …

Table 5.14 Classification accuracy of K-nearest neighbor method based on spatial weights (Kappa = 0.729) Item

Bamboo Coniferous Broadleaf Mixed Shrubs Open forest forest forests space

Shadows Mapping accuracy/%

Bamboo

39

0

0

0

2

0

0

95.12

Coniferous 0 forest

31

8

1

0

0

0

77.50

Broadleaf forest

1

0

19

6

3

0

0

65.52

Mixed forests

0

4

0

28

5

0

0

75.68

Shrubs

3

3

0

3

24

0

0

72.73

Open space 3

0

0

1

0

6

0

60.00

Shadows

0

0

0

0

0

9

81.82

81.58

70.37

71.79

70.59

100.00 100.00

2

Mapping 81.25 accuracy/%

77.61

the number of pixels classified into a category compared with the number of ground tests for that category. Usually, the columns in the array represent the reference data and the rows represent the category data obtained from the classification. The error matrix can reflect the overall classification accuracy, producer accuracy (omission error) and user accuracy (misclassification error). The overall classification accuracy is equal to the total number of pixels correctly classified divided by the total number of pixels; the mapping accuracy is the ratio of the number of pixels that the classifier correctly classifies into a certain class to the total number of true references of that class; the user accuracy represents the ratio of the total number of pixels correctly classified into a certain class to the total number of pixels that the classifier classifies into that class. In addition, the Kappa coefficient statistically reflects the extent to which the classification result is better than the random classification result and can be used to compare the error matrix of two classifiers to see if they are significantly different. From Tables 5.14, 5.15, 5.16 and 5.17, it can be seen that the overall precision and Kappa coefficient of the K nearest neighbor classification methods based on spatial weights reached 77.61% and 0.729, respectively, which were the highest among the four methods, while the overall precision of the other three methods did not reach 70%. For the category of bamboo, the two nearest neighbor methods have higher producer accuracy (81.25%) than CART and SVM methods, while the user accuracy of all four methods reaches 85%, with the K nearest neighbor classification method based on spatial weights having the highest user accuracy (95.12%) and the CART method ranking second in user accuracy (90.91%). The four methods showed different advantages for other vegetation types: coniferous forests obtained the highest user accuracy with the K-nearest neighbor classification method based on spatial weights; all methods, except SVM, achieved 81.58% producer accuracy.

5.4 Fine Observation of Habitat of Giant Panda Based on High-Resolution …

223

Table 5.15 Classification accuracy of K nearest neighbor method (Kappa = 0.620) Item

Bamboo Coniferous Broadleaf Mixed Shrubs Open forest forest forests space

Shadows Mapping accuracy/%

Bamboo

39

0

0

0

5

0

0

88.64

Coniferous 0 forest

31

20

1

0

0

0

59.62

Broadleaf forest

1

0

7

7

5

0

0

35.00

Mixed forests

0

5

0

27

4

0

0

75.00

Shrubs

3

2

0

3

19

0

0

70.37

Open space 3

0

0

1

0

6

0

60.00

2

0

0

0

1

0

9

75.00

81.58

25.93

69.23

55.88

100.00 100.00

Shadows

Mapping 81.25 accuracy/%

68.66

Table 5.16 Classification accuracy of CART method (Kappa = 0.559) Item

Bamboo Coniferous Broadleaf Mixed Shrubs Open forest forest forests space

Shadows Mapping accuracy/%

Bamboo

30

0

0

0

3

0

0

90.91

Coniferous 0 forest

31

21

1

0

0

0

58.49

Broadleaf forest

1

0

2

0

1

0

0

50.00

Mixed forests

0

5

4

32

9

0

0

64.00

Shrubs

12

2

0

5

18

0

0

48.65

Open space 5

0

0

1

0

6

0

50.00

Shadows

0

0

0

3

0

9

75.00

81.58

7.41

82.05

52.94

100.00 100.00

0

Mapping 62.50 accuracy/%

63.68

The producer accuracy of broadleaf forest is lower with K-nearest neighbor, CART and SVM methods. In contrast, the producer accuracy of shrubs was higher with CART and SVM methods than with both K-nearest neighbor classification methods. Overall, the K-nearest neighbor classification method based on spatial weights has higher producer and user accuracies than other methods for coniferous, broadleaf and shrub categories. Some issues derived from the classification results of this study area are discussed below. Since the bamboo in this region is largely covered by a tall tree canopy, the bamboo spectrum does not show a clear distinction from the surrounding vegetation. Although the use of WorldView-2 high-resolution images showed advantages in

224

5 Spatial Observation and Assessment of Ecological Changes in Giant …

Table 5.17 Classification accuracy of SVM method (Kappa = 0.492) Item

Bamboo Coniferous Broadleaf Mixed Shrubs Open forest forest forests space

Shadows Mapping accuracy/%

Bamboo

31

0

0

0

4

0

0

88.57

Coniferous 0 forest

29

19

0

0

0

0

60.42

Broadleaf forest

1

6

5

1

2

0

0

33.33

Mixed forests

13

2

2

34

22

0

0

46.58

Shrubs

0

0

1

3

3

0

0

42.86

Open space 2

0

0

1

0

6

0

66.67

1

1

0

0

3

0

9

64.29

76.32

18.52

87.18

8.82

100.00 100.00

Shadows

Mapping 64.58 accuracy/%

58.21

Fig. 5.45 Tree crown photos taken using a fisheye camera at the testing locations. a The bamboo class surrounded by brush and was correctly classified, b the bamboo class covered by mixed woodland and was misclassified as brush

vegetation extraction, it is worth exploring to what extent the bamboo is exposed under the trees to make it possible to extract the bamboo under the canopy from the high-resolution images. Figure 5.45 shows two images taken vertically from the ground to the sky with a fisheye lens at the location of the test sample site in the May Day shed. The point in Fig. 5.45a is a bamboo surrounded by bushes, which is correctly classified in the K-nearest neighbor method based on spatial weights; Fig. 5.45b is a bamboo covered by a mixed forest, but misclassified as bushes. The degree of closure is the cover of the tree layer, i.e. the percentage of the vertical projection of the highest layer of plants in a forest over the entire forest area. Visually it is clear that the bamboo in Fig. 5.45a is less closed than in Fig. 5.45b. A simple estimation of the closeness of the two photos was made by first removing the background area in the outer ring of the fisheye lens and then binarizing the vegetation with the background of the sky, as shown in Fig. 5.46, and the closeness of the two photos was 0.82 and 0.67, respectively. Although the closeness of each

5.4 Fine Observation of Habitat of Giant Panda Based on High-Resolution …

225

Fig. 5.46 The canopies shown in binary maps of the photos. a The bamboo class surrounded by brush and was correctly classified, b the bamboo class covered by mixed woodland and was misclassified as brush

test sample location was not estimated, the following conclusions can be tentatively drawn: in areas with high closeness (closeness greater than 0.7) it is difficult to extract It is very difficult to extract bamboo from areas with high levels of closure (>0.7), and the possibility of extracting bamboo from areas with medium levels of closure (0.2–0.7) and sparse forests (1000 × 104 m3 ) volumes accounting for 70% of the total volumes (Figs. 6.7 and 6.8). Large-scale landslides change the face of mountains and rivers, wash away road traffic, destroy the ecological environment, and have an impact on wildlife such as giant pandas. 3) Causes a large number of unstable slopes The earthquake caused a large number of mountain cracks and formed a large number of unstable slopes. According to field investigation, there are a large number of secondary landslides, debris flows and rock piles caused by earthquakes around the S303 provincial highway running through Wolong Nature Reserve, and the highway along the way happens to pass through the front edge of the slope body causing secondary slope cutting, which increases the sliding force of the landslide and leads to a great reduction in stability, and in case of precipitation or induced factors such as earthquakes, secondary landslides will easily occur and cause casualties (Fig. 6.9). Figure 6.9 Landslide is located near Xinhua Village, about 70 m long, 50 m wide and 40 m high. The landslide back wall of the slope was caused by the 5–12 Wenchuan earthquake, but on June 19, 2014, the day before the inspection, continuous heavy Fig. 6.7 Geological hazard area statistics of Wolong Reserve

Fig. 6.8 Volume statistics of geological hazards in Wolong Reserve

244

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

Fig. 6.9 Site survey map of Xinhua Village

6.1 Research on Distribution Types and Patterns of Seismic and Geological …

245

rainfall caused a change in the stress of the landslide body, and a large amount of accumulation on the surface of the slope slid down, blocking the adjacent road. 3. Impact factors and regional distribution characteristics of secondary geological hazards The development of landslide is controlled by many factors, such as topography, stratigraphic lithology (easy sliding stratum), geological structure (special tectonic site, fault fragmentation zone, etc.), valley cutting density (linear gully distribution in the region), etc. In-depth study of the relationship between each main controlling factor and landslide development is very important to study the occurrence and distribution pattern of landslide. Therefore, the study uses relative density to quantitatively describe the relationship between geological hazards and each influencing factor in terms of different aspects such as slope, slope direction, elevation, stratigraphic lithology, distance from faults, distance from water systems, distance from roads, NDVI, etc. P=

Ni Si / N S

(6.1)

where N is the total number of geological hazards in the study area; S is the total area of the study area; N i is the total number of geological hazards distributed in category x; S i is the area containing evaluation factor x i in the study area. Relationship with slope: Slope is an important factor affecting the stability of slope, which controls the stress distribution of slope. With the increase of slope, the stress is concentrated to the slope angle, which provides the necessary stress conditions for the generation of effective critical surface of slope. At the same time, the slope also restricts the redistribution of surface materials and energy, the recharge and discharge of surface water runoff, slope groundwater, the thickness of loose material accumulation, vegetation cover, etc. play a decisive controlling role (Lee and Min 2004; Saha et al. 2005; Ercanoglu and Gokceoglu 2002). Therefore, slope is the main control factor that directly controls the stability of slopes. By grading the slope map of the study area and statistically analyzing the hazard points and slope, it is obtained that most of the geohazard points are distributed within the slope interval of 20°–50°, as shown in Table 6.2, and the distribution density increases with the increase of slope (Fig. 6.10). Table 6.2 Relationship between hazard sites and slope Slope

Relative area/%

Frequency of occurrence of disaster sites/%

Relative density

0–10

7.30

2.56

0.3517

10–20

15.81

9.37

0.5925

20–35

42.08

33.27

0.79

35–50

29.36

42.70

1.45

5.43

12.02

2.21

>50

246

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

Fig. 6.10 Slope distribution in the study area

The relationship with slope orientation: Different slope orientations have different solar radiation intensities, which affect evapotranspiration, vegetation cover, slope erosion and other factors, thus affecting the distribution of groundwater pore pressure and geotechnical characteristics of slopes, and thus further affecting the stability of slopes and landslides (Chen et al. 2009), and this directional effect is particularly evident in earthquakes. In gully slopes that are nearly perpendicular to the seismic rupture zone, the density of landslide development on the back slope side of seismic wave propagation is significantly higher than that on the facing side, which is generally referred to as the “back slope effect” (Xu et al. 2009). The results are shown in Table 6.3, which shows that the relative density of most hazard sites is more on the

6.1 Research on Distribution Types and Patterns of Seismic and Geological …

247

Table 6.3 Relationship between hazard sites and slope Slope direction Baseline

Relatice area/% 0.00001

Frequency of occurrence of disaster/%

Relative density

0

0

North

11.70

10.80

0.859

Northeast

11.80

5.98

0.5064

East

15.50

12.43

0.798

Southeast

14.20

21.10

1.4853

South

12.35

16.80

1.364

Southwest

12.63

11.06

0.875

West

10.48

8.40

0.808

Northwest

11.17

13.96

1.248

south-facing slope than on the north-facing slope, with the southeast and south-facing slopes being the largest (Fig. 6.11). Relationship with elevation: Strictly speaking, there is no direct relationship between a single elevation factor and the deformation and instability of landslides. However, the elevation can control the magnitude of stress value within the slope, which increases significantly with the increase of slope height. Secondly, different elevations affect different problems, such as different degrees of water system development, different vegetation cover and different human interference. Wolong Nature Reserve, with mountainous development, large height difference and steep slope, has an elevation change range of 1194–5789 m. From the relationship between hazard points and elevation, it can be seen that the hazard points are mainly distributed in 1179–3500 m, the relative density is greatest at 1500–2000 m, and the number of disaster sites is greatest at 2000–2500 m, as shown in Table 6.4 and Fig. 6.12. Relationship with the lithology of strata: The lithology of strata is the material basis for landslides, and landslides in certain areas occur in certain strata. The type and hardness of the rocks and the inter stratigraphic structure determine the physical and mechanical strength, weathering resistance, stress distribution and deformation and damage characteristics of the geotechnical body, which in turn affect the stability of the slope body and the ease of surface erosion, and are one of the important influencing factors and intrinsic conditions for the formation of landslides (Xiang et al. 2010) The stratigraphy of the Pre-Paleogene to Mesozoic Triassic is well developed in the study area, while the stratigraphy of the Mesozoic Jurassic, Cretaceous and Cenozoic Paleoproterozoic and Neoproterozoic is missing. Based on the statistical analysis, it can be seen that the most frequent geological hazards occur in the oldest Yanshan granite strata (Table 6.5). Relationship with fault distance: Tectonics is an important factor influencing landslide development. It is not only a necessary condition for the development of individual landslides, but also a direct control factor for the characteristics of regional landslides, which often occur in clusters near large tectonic fracture zones (Foumelis et al. 2004). It has been proved that geological hazards often arise in areas with strong

248

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

Fig. 6.11 Slope distribution map of the study area Table 6.4 Relationship between hazard points and elevation Elevation/m 1179–1500

Relative area/% 0.319

Frequency of occurrence of disaster sites/% 0.6

Relative density 1.880

1500–2000

5.02

12.4

2.483

2000–2500

12.45

26.0

2.094

2500–3000

16.27

24.3

1.497

3000–3500

17.11

18.3

1.069

3500–4000

18.51

17.1

0.926

4000–5789

30.29

0.98

0.0324

6.1 Research on Distribution Types and Patterns of Seismic and Geological …

249

Fig. 6.12 Elevation distribution map of the study area

geological formations, fractures and folds, and broken rock formations. The closer to the fracture zone, the more fragmented the rocks are, which is conducive to weathering and the formation of banded weathering crust, thus reducing the integrity of the slope and providing favorable conditions for the occurrence of geological hazards. Statistics show that most of the geological hazards in the study area occur within 0– 2000 m from the fracture zone, with the most distribution in the range of 800–2000 m, see Table 6.6 and Fig. 6.13. The relationship with the distance from the water system: the erosion of rivers is also one of the important factors affecting landslides, mainly in the form of weakening of the resistance of the leading edge of the slope by erosion and the increase of the critical surface, thus affecting the stability of the slope (Gokceoglu and Aksoy

250

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

Table 6.5 Relationship between hazard sites and stratigraphy Stratigraphic Relative area/% Frequency of occurrence of disaster sites/% Relative density η51b

0.296

0

0

ζ51b ( γo2 4) 2 γ5 b

0.296

0

0

2.47

17.40

7.06

3.10

0

0

Pthn1

0.065

0.08

13.47

( γδ2 4)

1.56

6.33

4.057

δ2

2.13

12.90

6.051

(4) γ2

0.136

0.24

1.806

(3)

Smx2

0.494

0.40

0.829

Smx3

2.30

2.26

0.984

Smx4

9.403

3.90

0.415

Smx5

2.27

1.58

0.696

D 1y1

0.109

0.24

2.24

D 2y1

0.03

0.08

2.68

D2y

0.603

0.13

0.226

D2g

0.386

0.60

1.554

1 Dwg

4.831

2.97

0.616

13.5

1.046

2 Dwg

12.92

D2+3

0.156

0

0

P1

0.832

0

0

P2

0.067

0

0

T1b

6.52

9.00

1.39

T2z

17.05

10.81

0.634

T3zh

16.21

0.40

0.025

Table 6.6 Relationship between hazard sites and fault distance Distance from fracture zone/m

Relative area/%

Frequency of occurrence of disaster sites/%

Relative density

0–200

2.88

1.72

0.59

200–400

2.62

1.93

0.739

400–600

2.64

1.74

0.662

600–800

2.40

800–2000

12.32

17.2

1.396

>2000

77.1

74.6

0.967

2.75

1.154

6.1 Research on Distribution Types and Patterns of Seismic and Geological …

251

Fig. 6.13 Distribution of distance from the fault in the study area

1996; Saha et al. 2002). The statistics of the study area showed that most of the geological hazards are located at a distance of 100–500 m from the river with the largest distribution at 300–400 m (Table 6.7, Fig. 6.14). Relationship with roads: As the intensity and frequency of human modification of nature increases, a series of human activities such as road construction, excavation, filling, engineering blasting, construction loading, deforestation, etc. will also become factors that induce geological hazards. According to the statistics, there are more geological hazards at the distance of road 200–400 m (Table 6.8, Fig. 6.15). Relationship with vegetation cover: Vegetation also has a profound influence on the development and stability of geological hazards. The deeper the roots of plants penetrate into the soil, the stronger the soil surface can be, which is equivalent to the

252

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

Table 6.7 Relationship between the distance of the disaster site and the water system Distance from river/m Relative area/% Frequency of occurrence of disaster Relative density sites/% 0–100

13.9

8.0

0.575

100–200

11.28

14.0

1.244

200–300

11.6

19.4

1.674

300–400

9.3

17.0

1.828

400–500

9.19

13.4

1.468

27.9

0.626

>500

44.6

Fig. 6.14 Distance distribution of the study area from the water system

6.1 Research on Distribution Types and Patterns of Seismic and Geological …

253

Table 6.8 Relationship between disaster sites and road distance Distance from road/m Relative area/% Frequency of occurrence of disaster Relative density sites/% 0–100

6.13

5.80

0.956

100–200

4.94

5.51

1.116

200–300

5.10

7.45

1.460

300–400 >400

5.07 73.50

7.24 73.9

Fig. 6.15 Distribution of distance from roads in the study area

1.427 1.004

254

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

Table 6.9 Relationship between hazard sites and vegetation cover NDVI ←0.2

Relative area/% Frequency of occurrence of disaster sites/% Relative density 0.018

−0.2 to 0.063 19.8

0

0

11.60

0.585

0.063–0.108

15.1

6.15

0.405

0.108–0.28

9.9

25.80

2.610

0.28–0.45

20.8

33.24

1.592

0.45–0.76

34.2

23.15

0.6763

role of anchors, and the larger and deeper the roots are, the stronger the role they play, thus effectively inhibiting or weakening the slope deformation and reducing the probability of landslides. In addition, the vegetation cover area, which can reduce the influence of climatic factors (such as rainfall, etc.), reduces the encroachment on the soil body and therefore reduces the probability of slope sliding. An important indicator of vegetation cover is usually the normalized vegetation index (NDVI) value, and a larger NDVI value means a higher vegetation growth force and a lower probability of landslides, and vice versa (Xiao et al. 2003; Banair 1995). It can be seen from the statistics that the relative density of geological hazards is highest in the area with NDVI between 0.0.108–0.28 (Table 6.9, Fig. 6.16). Relationship with profile curvature: Profile curvature theoretically refers to the change in slope of a sloping body along its maximum slope, which can be referred to as the slope of the slope (Wilson and Gallant 2000). A positive curvature indicates that the surface of the image element is convex upward, a negative curvature indicates that the surface opening of the image element is concave upward, and a value of 0 indicates that the surface is flat. Slope curvature influences the process of river erosion (Ercanoglu and Gokceoglu 2002; Oh and Pradhan 2011). Statistically, geohazards are distributed in both concave and convex profile curvature, and less in flat places (Table 6.10, Fig. 6.17).

6.1.3 Spatial Vulnerability Evaluation of Geological Hazards The Wenchuan earthquake caused great damage to the reserve’s giant pandas and related scientific research infrastructure, and many staff members were killed in the earthquake. In particular, the giant panda enclosures in the Walnut Ping Giant Panda Research Center were almost buried by a landslide and could no longer be used (Fig. 6.18). After the earthquake, one panda living in the research center was killed and one was injured, and six were missing. Five were later recovered in the mountains, and one is still missing (Cheng and Song 2008a, b), and the number of dead wild pandas is not known at this time. In order to minimize the impact of geological hazards on rare animals such as giant pandas, it is necessary to start from

6.1 Research on Distribution Types and Patterns of Seismic and Geological …

255

Fig. 6.16 Distribution of vegetation cover in the study area Table 6.10 Relationship between hazard points and profile curvature Profile curvature

Relative area/%

Frequency of occurrence of disaster sites/%

Relative density

Flat

80.00

61.4

0.767

Convex

9.76

12.3

1.265

Concave

10.22

26.2

2.564

256

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

Fig. 6.17 Curvature distribution of the study area profile

the characteristics of geological hazards themselves, grasp their spatial distribution patterns, locations, scales, frequencies and their interrelationships, and assess the spatial susceptibility of geological hazards, so as to figure out which areas are more prone to geological hazards and what the probability of occurrence is, and the results obtained will provide a scientific basis for the protection of wild giant pandas, as well as for post-earthquake recovery and reconstruction. The results obtained will provide a scientific basis for scientific site selection for wild panda conservation and post-earthquake recovery and reconstruction.

6.1 Research on Distribution Types and Patterns of Seismic and Geological …

257

Fig. 6.18 Post-earthquake wildlife training base of the Chinese giant panda Research Center in Walnut Ping

1. Impact factor selection The occurrence of geological hazards is the product of the combined effect of various internal and external influencing factors, and the study of landslide formation conditions and influencing factors is an important part of the study of landslide formation mechanism and its evaluation. Aleotti and Chowdhury (1999) once clearly pointed out that the identification of factors that lead to slope instability and cause landslides is a very important basic work when evaluating the vulnerability of geological hazards. Nine factors such as slope, slope direction and elevation as described above were studied to participate in the evaluation as the main control factors affecting the occurrence of geological hazards. 2. Evaluation model selection At present, evaluation models can be summarized in general into two categories: qualitative and quantitative. The qualitative method, also known as the expert scoring method, is the most widely used method in landslide hazard evaluation, which mainly relies on the personal experience of the evaluator and varies from person to person and from place to place. Generally, when the scale is small and the information is incomplete, it is necessary to have a macroscopic understanding of the vulnerability of a region to geological hazards. This method is often used in the case of the occurrence profile. However, in order to compensate for the relatively high subjective component of the expert scoring process, the parameters of the composite factors of the instability process are determined by means of statistical analysis and then applied to the quantitative or semi-quantitative evaluation of experts in areas with

258

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

the same conditions but not yet affected by landslides. For example, binary statistics (Brabb et al. 1972; Yilmaz and Yildirim 2006; Constantin et al. 2011), multivariate statistics (Carrara 1983; Chung et al. 1995; Piegari et al. 2009; Pradhan 2010a; Nandi and Shakoor 2010) and logistic regression (LR) (Lee and Pradhan 2007; Nefeslioglu et al. 2008a, b; Yilmaz 2009; Pradhan 2010a, b, 2011a, b; Süzen and Kaya 2012; Felicisimo et al. 2013), these statistical analysis methods take the premise that landslides will occur under similar topographic, geological and seismic conditions, and carry out studies on the effects of different seismic landslide impact factors on seismic landslides based on the actual occurrence of landslides, so that objective results of seismic landslide susceptibility or hazard evaluation can be obtained. Artificial neural network models and support vector machine models developed in recent years have overcome the problem that statistical methods require the dependent variable to be binary must be dichotomous and are also widely used in geohazard evaluation (Pradhan et al. 2010a, b; Sezer et al. 2011; Oh and Pradhan 2011; Tien et al. 2012; Micheletti et al. 2013). In fact, each method has its own advantages and disadvantages, and the study evaluates and compares the results of the study using three models: logistic regression, hierarchical analysis, and fuzzy support vector machine, respectively. 3. Evaluation results Based on three different evaluation models (logistic regression, hierarchical analysis and fuzzy vector machine F-SVM) and three impact factors (slope, slope direction, elevation, stratigraphy and lithology, distance from fault, distance from road, distance from water system, vegetation cover, and curvature of profile)9affecting geological hazards such as landslides, 70% of the geological hazard points were used as training samples (the remaining 30% were used as test samples for accuracy verification). The evaluation of geological hazard susceptibility and the classification of susceptibility (Figs. 6.19, 6.20 and 6.21) were carried out (Meng et al., 2015). According to the standard deviation of the geological hazard susceptibility index, it was classified into 5 levels of very high, high, medium, low and very low (Fig. 6.22). In terms of the distribution of different areas in the study area, the distribution of the evaluation results obtained by the logistic regression model and the fuzzy support vector machine are relatively consistent, with 5% of the study area classified as very high susceptibility area, 18% as high susceptibility area, and 27% as medium susceptibility area, which is similar to the results obtained by the analytic hierarchy process (AHP). The results based on AHP classify 30.9 and 9.8% of the area as low and very low susceptibility areas, while the results of logistic regression and F-SVM classify 20 and 28% of the area as low and very low susceptibility areas. From the geographical distribution of geological hazards, the geological hazard-prone areas based on logistic regression and F-SVM are clearly divided, mainly concentrated in the ancient Yanshan-age granite area in the northeast of Gengda Township at the entrance of the reserve, on both sides of the provincial highway S303, and in the valley basins of the Pijiao, Zhonghe and Xihe rivers.

6.1 Research on Distribution Types and Patterns of Seismic and Geological …

259

Fig. 6.19 Geological hazard susceptibility zoning map of Wolong based on logistic regression model

4. Accuracy verification To evaluate the merit of the method needs to be verified with certain accuracy, the study cites the success rate verification method, and the area under the curve is selected (area under curve, AUC) to quantitatively measure and compare the evaluation accuracy of the model, and the closer the AUC value is1, the better the performance of the model and vice versa (Chung et al. 1995). The X-axis represents the cumulative percentage of geological hazard susceptibility risk index and the Y-axis represents the cumulative percentage of geological hazard occurrence (Fig. 6.23).

260

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

Fig. 6.20 Geological hazard susceptibility zoning map of Wolong based on hierarchical

From the cumulative frequency curve and the area surrounded by X-axis, the corresponding values of F-SVM, logistic regression, and AHP are 85.73%, 84.55%, and 78.84% respectively, which indicates that the evaluation accuracy of logistic regression and fuzzy vector machine model is not much different and matches the actual situation, among which the fuzzy vector machine model is slightly better.

6.2 Post-earthquake Landslide Monitoring and Assessment

261

Fig. 6.21 Geological hazard susceptibility zoning map of Wolong based on fuzzy support vector machine mode

6.2 Post-earthquake Landslide Monitoring and Assessment Landslide refers to the phenomenon that the rock or soil on a slope is affected by rainfall, earthquake or human activities, etc., and slides as a whole or scattered under the action of gravity on a large scale. The occurrence of landslide is closely related to geotechnical type, geological structure, topography, hydrology, and many triggering factors. The occurrence of landslides in China is characterized by high frequency and large scale, causing great damage to the natural environment and people’s lives and properties every year. The most serious landslide disasters in Sichuan Province, including the World Natural Heritage Site of the Giant Panda, are due to active faults,

262

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

Fig. 6.22 Spatial vulnerability zoning distribution of geological hazards based on three different models

Fig. 6.23 Comparison of accuracy of geological hazard susceptibility evaluation methods

frequent earthquakes, deep ravines, broken rocks, and high summer precipitation. As of 2008, the number of recorded landslides in Sichuan Province exceeded 100,000, affecting more than 100 cities and almost half of the townships (Wu et al. 2008). The Wenchuan magnitude 8.0 earthquake of May 12, 2008, not only caused hundreds of thousands of casualties and millions of homeless people, but also triggered more than 5600 landslides covering an area of many 41,750 square kilometers (Cui et al.

6.2 Post-earthquake Landslide Monitoring and Assessment

263

2009). The earthquake loosened and fragmented rocks and made a large number of slopes very unstable, so that landslides occurred much more frequently and on a larger scale in the region after the earthquake than before (Huang and Li 2014). To reduce the impact of landslide hazards, risk analysis and condition assessment are very important (Dai and Lee 2002). The deformation of unstable slopes is the most commonly used indicator for landslide characterization, landslide mapping and prediction (Chen et al. 2014). Analyzing the dynamics before and after the occurrence of disasters is the key to understanding the mechanism of landslide occurrence. However, traditional survey methods, such as level survey, GPS and geophysical sounding, are difficult to carry out in mountainous areas with poor road access, and point data are obtained with very limited coverage. Differential Interferometric Radar (DInSAR) technology has very great potential for deformation monitoring and has the advantages of all-weather, all-day, wide coverage, and high spatial and temporal resolution. Its earliest application dates back to the mid-1990s (Fruneau et al. 1996), and in the 21st century, the multi-temporal DInSAR (MTInSAR) technique has been widely used in landslide monitoring because of its finer (millimeterscale) deformation monitoring capability (Bovenga et al. 2012; Chen et al. 2014). Monitoring landslide phenomena with the existing MTInSAR technique encounters many challenges, such as rugged terrain, dense vegetation, large deformation gradients, atmospheric delays, geometric distortions in mountain radar images, and the number and spatio-temporal baseline distribution of SAR images, which can introduce large errors, and the estimation and interpretation of slope deformation are often erroneous considering the complexity of landslide motion. This experiment combines the advantages of the existing small-baseline subset (SBAS) (Berardino et al. 2002) technique and distributed interferometric SAR (Ferretti et al. 2011) technique, and proposes a method based on highly coherent points and distributed scatter points, which is suitable for mountainous areas with low vegetation cover and complex topography. Some unstable slopes were detected by this technique to obtain part of the deformation field of Sichuan Giant Panda Natural Heritage Site. Most previous studies in this region have focused on previous landslides, such as the distribution characteristics of landslides and the impact of landslides, while this experiment focuses on the detection of potential landslides, which is more important for disaster prediction and risk reduction.

6.2.1 Experimental Area and Data Introduction The experimental area is located in Dujiangyan City and Wenchuan County, Sichuan Province (Fig. 6.24), which is close to the center of the Wenchuan earthquake and is among the most severely affected areas. Especially in Yingxiu town, about 80% of the houses collapsed in the earthquake (Zhang et al. 2014a, b). The total area of the experimental area is about 4200 km2 , and the terrain is very rugged, with elevations of 500–5000 m and slopes of 72° in some areas. The Minjiang River flows from north to south through the canyon and the Zipingpu Reservoir and eventually into the Chengdu

264

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

Fig. 6.24 Topographic map of the study area (generated by SRTM DEM)

Plain. The northeast-trending Longmenshan fault zone, including the Maowei fault, the Yingxiu-Beichuan fault, and the Anxian fault in Gou County, runs parallel to the ridge and canyon through the entire experimental area. The rock types mainly include shale, sandstone, limestone, slate, granite and basalt, dating from Precambrian to Cretaceous. The area has a subtropical monsoon climate with an average annual temperature of about 15 °C and an annual precipitation of 1200 mm. 50–70% of the precipitation occurs between June and September, which is an important factor in slope instability and landslide hazards (Handwerger et al. 2013). As a famous world heritage site for giant pandas, the area was severely fragmented after the Wenchuan earthquake, and the number of giant panda habitat patches increased approximately six times compared to the pre-earthquake period (Wang et al. 2008b). Post-earthquake disasters have also threatened the survival of giant pandas, and there have been incidents of giant pandas dying from landslides.

6.2 Post-earthquake Landslide Monitoring and Assessment

265

The data used for landslide monitoring are 11 views L-band ALOS/PALSAR acquired after the Wenchuan earthquake, dated from June 22, 2008 to December 13, 2011. Google Earth images were used for the presentation and analysis of the results.

6.2.2 MTInSAR Technical Methodology In this experiment, an improved MTInSAR technique is proposed for landslide monitoring in large areas of mountainous regions. The basic flow is shown in Fig. 6.25, which is divided into two main parts. 1. Convergence of CS and DS points The sparse number of measurable targets and their uneven distribution are the main factors that hinder the application of MTInSAR technology in mountainous areas. Usually, a density of more than one 5point per square kilometer is required to ensure the correct estimation of atmospheric phase delay and deformation parameters. The

Fig. 6.25 Flow chart of MTInSAR technology

266

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

Fig. 6.26 Spatio-temporal baseline distribution of interference pairs (circles represent SAR images)

number of permanent scatters is very small in the experimental area, and the extraction of code segment (CS) points is affected by the decoherence of vegetation. Data segmet (DS) points refer to image elements with the same temporal backscattering characteristics in a certain area, and they are widely distributed on slopes covered by low vegetation or debris. In order to increase the density of monitorable points, CS and DS points were jointly extracted, with a total number of 1.8 million, covering most of the experimental area except for radar shadowing and stacking masks. In order to extract the CS points, firstly, a 2 × 6 multi-view of the interferometric pairs is performed, and the image element corresponds to a ground distance of about 20 m × 20 m. Setting a spatial baseline threshold of 1500 m and a time interval threshold of 365 days, a total of 36 interferometric pairs are obtained (Fig. 6.26). Then the coherence of each interference pair is estimated, and the average coherence coefficient above 0.3 is selected as the CS point. The results show that the distribution of CS points is extremely uneven, with hundreds of points per square kilometre in plain areas and none in some mountainous areas. DS points are selected by using a 20 × 20 moving window with the same number of backscattering feature pixels in the monitoring window, and those with more than 180 can be used as DS points. The coherence and signal-to-noise ratio of DS points are low, and complex phase filtering is performed to improve the reliability of the parameter inversion. Ferretti et al. (2011) used the maximum likelihood method to estimate the most probable phase to replace the original phase. [ (| | ) ] −1 ˆλ = arg min Λ H ||Cˆ || ◦ Cˆ Λ λ

(6.2)

6.2 Post-earthquake Landslide Monitoring and Assessment

267

λˆ = [0, ϑ2, · · · , ϑ N ]T is the optimal estimation of phase information for Nscene images, the first view assumes a value of 0. Λ = exp(i λ), λ is the initial phase; Cˆ is the complex coherence coefficient matrix based on homogeneous image elements. To solve the unknowns, this experiment uses the LBFGS (limited memory broyden-fletcher-goldfarb-shannon) algorithm, which has a significant advantage in large nonlinear problems. The results of the phase optimization estimation are shown in Fig. 6.27. 2. SBAS technique for parameter estimation After identifying CS and DS, unknown parameters (e.g., deformation and residual elevation) are estimated by satellite based augmentation system (SBAS) method.

Fig. 6.27 Comparison of before and after differential interference phase filtering, a–c is the initial phase before filtering; d–f is the phase after filtering

268

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

augmentation system (SBAS) method. First, the differential interferogram is obtained by removing the topography and the flat earth phase and then estimate the 2π integer part of the entangled phase, i.e., phase deconvolution. Out-of-phase, the phase discontinuity caused by too sparse points discontinuities above 2π, large topographic errors, or fast deformations can all cause errors in the deconvolution. In this experiment, a 3D phase decoupling algorithm (time and space dimensions) is used (Pepe and Lanari 2006), which is more It is suitable for a large number of interferometric pairs with short baselines. The reference point is chosen in the city of Dujiangyan, which is far from the landslide, and its phase and deformation are assumed to be 0, and the relative values of the interferometric results can be converted to absolute values. After the conventional interferometric treatment, the deformation The phase change is still confused with the orbital error, atmospheric phase, and residual topography. For long-band PALSAR, with long-band PALSAR data, the orbital error can reach 30 cm (Nakamura et al. 2007), which causes bias in the estimation of topographic phase. This can cause bias in the topographic phase estimates, which is more serious in mountainous areas. The atmospheric delay can be divided into turbulent and laminar components. The laminar flow is associated with elevation and is often mistaken for topography or deformation. The main error of topography is related to the vertical baseline of the interferometric pair. In this experiment, the terrain error is first corrected and removed by least-squares fitting, then the orbital error is removed by quadratic polynomial, and the residual terrain error is removed by linear function, and the assembled model is as follows. (6.3) where x and y are azimuthal and distance coordinates, h is the elevation, ε is the residual error phase, and ai is the parameter to be estimated. After the correct entanglement and removal of various phase errors, the least-squares solution is applied to solve for the deformation rate and time-series deformation information. 3. Analysis of the applicability of SAR monitoring of landslides In addition to point selection, point density, removal of various errors and phase unwinding, the relationship between the imaging geometry of SAR images and the terrain also needs careful analysis. The DInSAR technique can only detect the distance change from the sensor to the line of sight (LOS), which is the result of the projection of the real surface change onto the LOS direction. This unidirectional measurement mode limits the sensitivity to deformation to a plane approximately perpendicular to the orbital direction (Wasowski and Bovenga 2014). The ALOS satellite is in a near-polar orbit with an inclination of 98.2° and a right-viewing scan mode, with data in the experiment in ascending orbit. This imaging geometry dictates that deformations on the gently sloping near east-facing slopes are easily detected, while deformations occurring in the north-south direction are difficult to detect. The hypersensitive region (the region where deformations are difficult to detect) on the SAR image can be determined based on the geometric relationship with the expression:

6.2 Post-earthquake Landslide Monitoring and Assessment

269

] [ Facing north: (β > θ/2) & 336.8◦ < δ < 360◦ ||δ < 6.8◦

(6.4)

[ ] Facing south: (β > θ/2) & 156.8◦ < δ < 186.8◦

(6.5)

where, β is the slope slope; θ is the LOS direction; δ is the slope direction (calculated clockwise from north). The slopes facing west and the steep slopes facing west and east are susceptible to shadowing and overshadowing, and these regions can be extracted using the GEO module of GAMMA software. The GEO module of the GAMMA software extracts these areas, whose signals are not useful for interferometric processing. It is found that the hypersensitive areas and stacking/shadowing are very widely distributed in the experimental area. The distribution of these regions is very wide, with 12% and 9% of the image elements with slopes larger than 10°, respectively. This implies that if the landslides are uniformly distributed, at least 1/5 of the landslides cannot be monitored in this experiment. In addition, the slope gradient and direction of the slope also affect the distribution of measurable points. In this experiment, about 70% of the points were distributed. In this experiment, about 70% of the points are located in the slope direction of 10°–50° and 30°–180° (northeast to south). Because of the effect of stacking mask, very few points fall in the slope direction of 180°–300°. The incidence angle of 34° means that PALSAR data are more sensitive to vertical deformation, and to obtain 3D deformation information, at least three different types of irradiation are required. To obtain 3D deformation information, at least three types of data with different exposure geometries (including lift tracks) are required. For landslide monitoring, deformation in the LOS direction is usually transferred to the slope direction, which is the most important factor in the development of the slope. For landslide monitoring, the LOS directional deformation is usually transferred to the slope direction, assuming that the landslide is translating along the slope. In this experiment, the LOS direction for the analysis.

6.2.3 Landslide Analysis In InSAR products, backscatter coefficients, coherence maps and DEM can be used to identify some of the more intense landslide hazards. The deformation maps can be used to monitor potential landslides that are slowly changing over time. The MTInSAR technique was used to obtain surface deformation information for about 4200 km2 in the Dujiangyan area, with deformation rate values ranging from −7 to 1.5 cm/a. The negative sign means that the ground target is moving away from the radar sensor, and considering the imaging geometry and slope direction, the area with negative deformation rate can be considered as a potential landslide area. In total, dozens of potential landslides were extracted from this experiment. They are mainly distributed along the Minjiang River and Longmenshan Fault Zone, and some are also located in high altitude mountainous areas.

270

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

1. Minjiang River The Min River is the most important tributary on the upper left bank of the Yangtze River, with a main stem of about 1279 km, a catchment area of many130000 km2 , a high annual runoff, and a natural drop of more than 3 km. Dozens of hydropower stations have been built on the Min River, including the famous ancient Dujiangyan irrigation system (a world heritage site) and the modern Zipingpu Reservoir. Earthquakes and landslides occur very frequently in the upper reaches of the Min River north of Yingxiu town. The lithology of the area is dominated by volcanic rocks, clastic rocks and carbonate rocks. Images show that almost every hillside is covered with landslides, the ground is fragmented, and vegetation cover has also been severely damaged. The movement of these landslides is dominated by large-scale translations with little rotation and retrogradation. Due to the SAR imaging geometry (stacked mask on the east bank of the river and shadows on the west bank) and surface decoherence (caused by rapid changes), there are relatively few points to monitor in the area, concentrating on the unbroken surface and on the landslide areas that have remained relatively stable during the SAR image acquisition time. The rate of deformation is shown in Fig. 6.28a, where dozens of landslides are distributed in a canyon of approximately 40 km in length, ranging in elevation from 1200 to 2000 m, and varying in size, with a tendency for small landslides to converge to form large landslides. In Fig. 6.28b, the most severe hazards occur at lower elevations near river valleys, and landslides have occurred in almost half of the total area. The largest of these landslides is 400 m wide and 700 m long. The average surface deformation rate is about −3 cm/a, which indicates that the slopes in this area are still unstable and there is a possibility of recurrence of disasters. The slope in Fig. 6.28c is southeast-facing, with an average slope of about 40°. The number of landslides on it is extremely large, with the largest length exceeding 1 km, and the whole slope has become fragmented. The deformation results show that the slope at the lower end of the previous landslide is more unstable than the slope at the lower position near the river. The slope in Fig. 6.28e is located near Ginkgo Township, facing east with an average slope of about 50°, and the surface on both the left and right sides is severely damaged by the massive landslide. The lower boundary of the previous landslide body (shown as the black line in Fig. 6.28e) is essentially parallel to the boundary of the unstable area in the InSAR inversion, approximately the deformation rate of −5 cm/a indicates that the previous landslide area is still in an unstable state and the possibility of another large-scale landslide is very high. The probability of another large-scale landslide is very high. This can be explained by the fact that after the last landslide, a large amount of rocks and soil accumulated on the lower edge of the landslide body, and under the effect of gravity, induced by precipitation and other factors, there is a possibility of forming a disaster again. Three profile lines were selected from it for deformation analysis, and the results are shown in Fig. 6.29.

6.2 Post-earthquake Landslide Monitoring and Assessment

271

Fig. 6.28 Deformation rate along the Minjiang River. a Shows the overall rate map, and the bottom map is the SAR image average intensity map; b, c, e shows three typical landslides, corresponding to the location of (1)(2)(3) in Figure (a); d shows the deformation history of three points A, B, and C; the black line in e is the lower boundary of the previous landslide, and the blue dashed lines (EE, , FF, , and GG, ) are the three selected profiles in Fig. 6.29

272

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

Fig. 6.29 Deformation rates of points on three profile lines (EE, , FF, and GG, ). The area between the two adjacent red dashed lines is a potential landslide area

The area between the adjacent red dashed lines in Fig. 6.29 is the potential landslide area, so the potential landslides can be identified by the difference in deformation. In addition, there is an area on the lower left side of Fig. 6.28e with deformation rate of −2 to −4 cm/a, which is mainly caused by human activities such as road construction and housing repair. Figure 6.28d shows the deformation histories of three unstable points with deformation rates of −3 cm/a, −1.7 cm/a, and −3.4 cm/a, respectively. The deformation history shows that the deformation of these three points tends to become slower after the Wenchuan earthquake in 2008. In conclusion, the Minjiang Canyon is the area where the most landslide hazards occur, which may be related to the fragile lithology and steep topography. Fortunately, most residential areas are located on gentle slopes away from landslides, but there are still some roads and other infrastructure in the river valley that can be threatened by landslides. 2. Longmenshan fracture zone The Longmenshan fault zone consists of the Mao-wen fault, the Yingxiu-Beichuan fault, and the Guanxian-Anxian fault. In the Wenchuan earthquake, the surface rupture was the largest on the Yingxiu-Beichuan fault, followed by the GuanxianAnxian fault, and there was no surface rupture on the Mao-wen fault (Chigira et al. 2010). The lithology along the fault zone is dominated by volcanic rocks, clastic

6.2 Post-earthquake Landslide Monitoring and Assessment

273

rocks, limestone, gray sandstone and mudstone. The part of the Mao-wen fault from Wenchuan to Caopo Township is affected by the undercutting of the Minjiang River, and the rocks are exposed and subject to serious erosion. There are many works studying the geology and geomorphology of this area and landslide hazards, which have been analyzed and illustrated in the previous section (along the Min River). Figure 6.30 shows the area south of the Grassy Slope. The distribution of landslides that can be seen on Google Earth is characterized by a small number, discontinuity, and small scale, which is determined by the topographic features of the area. The shallow canyon and gentle slope reduce the probability of landslide hazard. On the east side of the canyon, the number of points that can be monitored is low because of the influence of stacking masks, while there is good monitoring on the gentle slope on the west side. Figure 6.30b, c show two areas of instability with an average deformation rate of about −5 cm/a. The canyon in Fig. 6.30b is about 6 km long, with a maximum slope gradient of more than 60°, dozens of landslides ranging from a few hundred meters to 1 km in length, and more than 10 villages located on high, gentle slopes. The canyon in Fig. 6.30c is 8 km long, with elevations ranging from 1600 to 2400 m and slopes ranging from 20° to 50°. The area used to be safe for human habitation because no trace of landslides can be found on Google Earth images. Because of this, many homeless people were relocated to this area after the Wenchuan earthquake. However, the deformation results show that the slopes in this area are becoming unstable, especially at lower elevations near the construction sites. This instability is the result of human intervention. Figure 6.30d shows a south-facing slope with a slope of 40°, divided by a deep trough in the middle, with the blue line marking the area of previous landslides. The points on the western half of the slope are concentrated on the upper side of the previous landslide, and the average deformation rate is about − 2.3 cm/a. The closer the points are to the landslide area, the more unstable they are. The same phenomenon occurs on the eastern half of the slope, where the points are densely distributed around the landslide body. It should be noted that the deformation information on this slope is grossly underestimated, because the radar is extremely insensitive to the deformation in the north-south direction. The area shown in Fig. 6.31 includes Yingxiu Town and Zipingpu Reservoir. It is believed that the Yingxiu-Beichuan Fault and Zipingpu Reservoir were a major factor in triggering the Wenchuan earthquake, and the landslides in this area are mainly of the seismically induced type, usually large in size, reaching 2 km in length and 800 m in width. The distribution of landslides varies spatially, with the number of landslides south of Yingxiu on both sides of the Minjiang River being much lower than that in the north, and the number of landslides on the south bank of Zipingpu Reservoir being much lower than that on the north bank. The deformation results show that some areas are still unstable, most notably the town of Yingxiu (Fig. 6.31b), which was also the most severely damaged area during the Wenchuan earthquake. No obvious landslide marks were found on the slopes on both sides of the canyon traces, deformation results also show that they are stable, while there are a large

274

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

Fig. 6.30 Deformation rate diagram of the lithology of the Mawra fault a is the total deformation map, and the bottom map is the SAR image average intensity map; b–d are the three typical landslide areas, corresponding to the location of 1–3 in Figure (a). d The blue line in (d) shows the area where the previous landslides were concentrated

6.2 Post-earthquake Landslide Monitoring and Assessment

275

Fig. 6.31 Deformation rates along the Yingxiu-Beichuan and Gouxian-Anxian faults a is the total deformation map, and the bottom map is the SAR image average intensity map; b–d are the three typical landslide areas, corresponding to the location of 1–3 in Figure (a), e is the deformation rate of the AA, profile in Figure (b)

number of negative deformation points exist in the canyon, which is not related to landslide disaster, but caused by surface subsidence. The reason is that after the Wenchuan earthquake, a large number of resettlement houses were hastily built here, and the location close to the river makes the surface soft and unable to withstand the heavy pressure of the houses over time. Figure 6.31e shows the deformation rate of the profile AA, through Yingxiu, and the red circle in the figure indicates the area where the resettlement houses are located, which is also the settlement area. The north shore of Zipingpu Reservoir is steep, and the earthquake-induced landslide is very large in scale, and large areas of vegetation have been severely damaged. Small landslides caused by precipitation and other factors, or rock falls, are also everywhere. Figure 6.31c shows the Zipingpu Dam, which is stable in terms of deformation, and Fig. 6.31d shows a small landslide on the south bank of Zipingpu Reservoir.

276

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

3. High altitude mountains In addition to the Minjiang River valley and along the Longmen Mountain fault, signs of potential landslides were found on some high elevation hilltops, such as the two mountainous areas and the four hilltops where landslides occurred shown in Fig. 6.32. The area shown in Fig. 6.32a is located about 9 km southeast of Wenchuan and has rugged terrain with elevations up to 4500 m. Some of the mountain tops have snow cover in the back shaded areas throughout the year. Many of the slope surfaces are fragmented by landslides of various sizes, and the situation is more severe at the summit than at the foot of the mountain. From the deformation results (Fig. 6.32c–f), the points with negative deformation rates are also concentrated in the summit area, which predicts that potential landslides also develop from the summit, consistent with the previous landslide characteristics. Figure 6.32c shows the hilltop at an elevation of 4000 m. As can be seen on the Google Earth image, there are severe landslide traces on the slopes in all orientations. The deformation monitoring results show an average deformation rate of about −3 cm/a, with some areas exceeding − 5 cm/a, the former landslide is still in a highly unstable state. The hilltop shown in Fig. 6.32d can also be seen as clear landslide traces on the Google Earth image, such as the area marked by the blue line in the figure. The western side of the hill has fewer monitorable points due to the influence of stacking masks, while the eastern side has detected a large number of unstable points. The topographic setting of the area shown in Fig. 6.32b is similar to that of Fig. 6.32a, but no large-scale landslides are observed. The deformation results show that the area is relatively stable overall except for individual hillsides, as shown in the area in Fig. 6.32e, f. The hillside in Fig. 6.32e faces southeast and is covered with a very thin layer of snow in winter. Some of the monitored instability points are located on some previously small landslides, and the deformation history of one of them is also given in the figure. It should be noted that the thin snow cover (less than 10 cm) does not affect the deformation monitoring results, since the L-band SAR has a strong penetrating ability for dry snow. However, the water from melting snow can cause some rocks to loosen and slide, and even steep slopes to collapse. In contrast, the slope in Fig. 6.32f is oriented to the north and covered with snow all year round. Hundreds of meters of landslide traces can be seen on the Google Earth image, and there is also the possibility of another landslide in the future. 4. Impacts of landslides on giant panda heritage sites Giant panda habitats in Sichuan Province are being affected by human activities (e.g. logging, grazing, road construction, poaching, etc.) and geological disasters (earthquakes, landslides, etc.). According to the 3rd China Giant Panda Habitat Survey (state forestry administration), the areas affected by landslide disasters accounted for about 1.06% as of 2006, and the scale of impact is mainly small and medium. As a result, the quality of survival of giant pandas has been reduced and some individuals have been killed in the disaster. The Wenchuan earthquake in 2008 resulted in the loss of 5.95% of giant panda habitat (Ouyang et al. 2008), fragmentation of habitat, destruction of corridors between habitats, and the cutting off of some water sources

6.2 Post-earthquake Landslide Monitoring and Assessment

277

Fig. 6.32 Landslide monitoring results in some high-altitude areas. a, b Shows the total deformation map, and the bottom map shows the average intensity of SAR images; c–f show the typical landslide areas; the blue line marks the area where previous landslides were concentrated

by landslides, and these impacts are expected to continue for some time. There are two major giant panda habitats in the experimental area, namely the Minshan habitat and the Qionglai Mountain habitat. They account for 41.66% and 26.47% of the total area of giant panda habitat in China, with densities of one 0.074 panda/km2 and 0.072 panda/km2 , respectively. Approximately 50 potential landslides were extracted from the deformation monitoring results, 20 of which are located in the giant panda natural heritage site, mainly along river valleys and faulted canyons. The impact of these

278

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

landslides on giant pandas is low because they mainly live in higher altitude areas with dense vegetation and gentle slopes. Combined with Google Earth images, the distribution of landslides in giant panda habitats can be summarized as follows: (i) The number of landslides in Minshan Mountain is much higher than that in Qionglai Mountain, which means that giant pandas living in Minshan Mountain are more likely to be affected by landslides; (ii) The number of landslides is related to the altitude, and landslides are mostly distributed between river valleys or canyons, and the number of landslides in higher altitudes is lower. In conclusion, monitoring potential landslides in giant panda natural heritage sites by MTInSAR technology can outline dangerous areas that need to be focused on prevention and assist in improving the management of heritage sites.

6.2.4 Summary This experiment monitors the potential landslides in the Giant Panda World Natural Heritage Site in Sichuan through multiple ALOS/PALSAR data. MTInSAR technology has a strong capability in small deformation monitoring, and also shows great potential in monitoring landslides in mountainous areas, but the existing technology will be limited in the application to large areas, complex geographic environments and small data sets. The algorithm proposed in this experiment concentrates the advantages of the existing technology. The algorithm proposed in this experiment combines the advantages of existing techniques and successfully monitors the habitat of giant pandas through joint detection of CS and DS points, phase filtering and other improvement measures. Dozens of potential landslides were detected in a range of about 42,002 km, which were mainly distributed along the Minjiang River valley, the Longmen Mountain fault canyon, and some in the high-altitude mountainous areas. The results suggest that previously occurring landslides will continue to be a hazard in the short to medium term, with the potential for new large-scale slides to be triggered at any time. In some areas, small, densely distributed landslides are at risk of converging into large landslides. Although the impact of landslides on giant panda habitat is limited, continuous monitoring and regular assessments are still necessary.

6.3 Microwave Remote Sensing Monitoring Model of Animal Habitat Ecology As mentioned earlier, radar interference-based coherence map change detection based on histogram zero shift estimation can be used for quantitative assessment of forest degradation in the Wenchuan earthquake. When post-earthquake radar data are abundant, it is expected to establish a post-earthquake forest restoration monitoring and assessment model based on coherence interferometric time series change

6.3 Microwave Remote Sensing Monitoring Model of Animal Habitat Ecology

279

using differencing and comparison techniques, subject to the three criteria of coherence map pairing (see Sect. 5.2) and the availability of post-earthquake coherence map time series information.

6.3.1 Post-Earthquake Forest Restoration Monitoring and Assessment Methods In order to obtain the most accurate estimation results as possible, it is recommended to use L-band long-wave radar data as a way to improve the quality of coherence maps and coherence timing retention. The forest restoration and assessment radar interferometric model data processing steps and methods are shown in Fig. 6.33. 1. High accuracy alignment of long time series single vision radar complex images The alignment of the mono-view complex image is the first step of the interferometric processing, and its good or bad affects the quality of the generated interference stripes. Usually, when two SAR images are accurately aligned, their phase difference images will show clear stripes and the stripes will contain surface topographic information; on the contrary, if the two images are not aligned, the stripes will be blurred or not generated at all, showing noisy information. Interference processing requires the primary and secondary images to be within 1/10 to 1/8 pixel accuracy. In order to suppress data processing errors, the secondary images should be resampled losslessly using the sinc function. 2. Small baseline set interference pair generation In order to improve the quality of the posterior coherence map, a small baseline interferometric image pair combination is the optimal solution. Here, the small baseline contains two dimensions: (i) A short spatial vertical baseline, which can suppress the decoherence introduced by different imaging geometric poses; (ii) A short temporal baseline, i.e., the interferometric pair revisit time interval between primary and secondary images is as short as possible to suppress the effect of temporal decoherence. 3. Pre- and post-event interference pairs selected In order to suppress the time-space decoherence before and after the earthquake, as well as the discrepancy error caused by physical changes, three major guidelines need to be followed for the selection of pre- and post-event interferometric image pairs, namely (i) the time and space baselines of the pre- and post-earthquake coherence maps must be the same, which is used to overcome the error caused by different baselines; (ii) in order to improve the quality of the coherence maps, a small baseline set of interferometric images should be selected; (iii) Interferometric images used for comparison are as consistent as possible with the season of data acquisition and are used to suppress the seasonal physical variation caused by decoherence errors.

280

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

Fig. 6.33 Post-earthquake forest recovery monitoring and assessment based on radar coherence time series analysis

6.3 Microwave Remote Sensing Monitoring Model of Animal Habitat Ecology

281

4. Time-series coherence diagram generation After selecting the pre- and post-event image interferometric pairs, the interferogram can be obtained by conjugate multiplication of the collocated primary and secondary timing image pairs. Assuming that the speckle noise has been suppressed using multiview processing during the interferogram generation process and assuming that the scatters within the multi-view estimation window are all-state ephemeral, the coherence map can be obtained pixel-by-pixel by calculating the coherence coefficients within the window. 5. Estimation of the zero offset of the coherence value of the differential coherence map The pre-earthquake/post-earthquake coherence maps are obtained by taking the preearthquake coherence map as the reference and differencing the post-earthquake coherence map from it. The histogram analysis and zero-point location of the differential coherence map can be used to estimate the current forest degradation rate at the post-earthquake observation time relative to the pre-earthquake reference (see Sect. 5.2 for detailed methodology and technical details). The offset of the histogram zero point characterizes the earthquake triggered forest degradation rate. 6. Time-series analysis of forest restoration When the zero-point offset method of differential coherence map is used to obtain the forest degradation rate values in the post-earthquake time series observation year relative to the pre-earthquake base year, the degradation rate differential comparison technique can be used to calculate the post-earthquake forest restoration time series change information, and then support the post-earthquake forest restoration monitoring and quantitative assessment.

6.3.2 Forest Restoration Monitoring Results and Interpretation Using the Sichuan giant panda habitat in Minshan and Qionglai Mountains around the epicenter of Wenchuan as a demonstration, we used L-band ALOS PALSAR data and Fig. 6.33. Post-earthquake forest restoration monitoring and assessment data processing and methodological procedures were used to obtain pre-earthquake/postearthquake coherence maps. The pre/post-earthquake coherence maps were obtained from 2007-06-20 to 2007-02-02 (pre-earthquake 2007) and 2009-06-25 to 2009-0207 (post-earthquake 2007). The pre/post-earthquake coherence maps were obtained from 2007-06-20 to 2007-02-07 (pre-earthquake 2007), 2009-06-25 to 2009-0207 (post-earthquake 2009), and 2010-06-28 to 2010-02-10 (post-earthquake 2010). Then the pre-earthquake 2007 coherence map is used as the base, and the postearthquake 2009 and post-earthquake 2010 coherence maps are differenced from each other to obtain the differential coherence maps from 2009 to 2007 and from

282

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

2010 to 2007. The differential coherence maps were obtained from 2009 to 2007 and 2010 to 2007. The histogram analysis of the differential coherence maps shows that the zero points are located at the histogram analysis of the differential coherence maps showed that the zero points were located at 29.34 and 32.66% (Fig. 5.19), which corresponded to 20.66 and 17.34% of the forest degradation rate. Using direct time-series comparison, the It is easy to find that the Wenchuan earthquake had a significant impact on forest degradation in the region, and the forest ecological recovery was slow in the first three years after the earthquake. The forest recovery rate was about 3% (20.66–17.34%) in the one-year cycle from 2009 to 2010. The low forest habitat recovery rate can be explained as follows: (i) After the earthquake, on the steep slopes near the Minjiang River and other streams, the ground cover together with vegetation was stripped away by landslides, landslides and debris flows, leaving only the bare surface or rock mass. (ii) The physical properties of the ground surface in the affected area can change significantly after the earthquake, including saturated water content, moisture carrying capacity, cover layer Poor surface conditions of the cover layer may inhibit vegetation recovery and growth.

6.4 Habitat Evaluation of Post-earthquake Giant Panda Habitats During the evolutionary process, individual organisms or populations constantly seek to balance their own survival needs with the resources of the surrounding habitat, selecting those habitats that maximize reproductive success and maximize their fitness. How animals select and adapt to habitats has been a hot topic in animal ecology research (Kernohan et al. 2001; Powell 2012). In the past century, the fragmentation and even loss of wildlife habitats due to global environmental changes, anthropogenic disturbances, and natural disasters have led to a dramatic reduction in the range and numbers of the entire giant panda population, which is on the verge of extinction (Macdonald and Rushton 2003; Viña et al. 2010; Hull et al. 2011). According to ancient texts, giant pandas were once distributed in southern central and southwestern China (Sichuan, Hubei, Hunan, Yunnan, Guizhou), as far north as Zhoukoudian in Beijing, and as far south as northern Vietnam The giant panda was once distributed in southern, central and southwestern China (Sichuan, Hubei, Hunan, Yunnan and Guizhou), as far north as Zhoukoudian in Beijing, and as far south as northern Vietnam and northern Myanmar (Zhu and Long 1983; Hu 1981, 1990, 2001). After geological changes, climate change, human activities, deforestation the current population of giant pandas is confined to the northern part of Sichuan, Gansu, and Shaanxi Provinces, driven by geological changes, climatic changes, human activities, deforestation, and agricultural land disturbances. The population is now confined to the Minshan, Qionglai, Liangshan, Daxiangling, Xiaoxiangling and Qinling mountains in Sichuan, Gansu and Shaanxi Provinces (Fig. 6.34).

6.4 Habitat Evaluation of Post-earthquake Giant Panda Habitats

283

Fig. 6.34 Distribution range of giant pandas

Ouyang et al. (2008), Hu (2001), Pan and Lv (2001) suggested that the cause of the extensive disappearance of habitats on which giant pandas depend in recent centuries is from external disturbances (Fig. 6.35). According to the information of the third national survey of giant pandas, the higher disturbance factors are human disturbances such as logging, grazing, medicine collection, roads, bamboo cutting and shooting, and poaching (State Forestry Administration 2006). However, natural disasters such as earthquakes, hurricanes, flash floods, fires, and landslides pose a greater threat to giant panda habitats than human disturbances (Cheng and Song 2008a, b), especially geological disasters triggered by sudden tectonic earthquakes, which can destroy giant panda habitats on a large scale within a short period of time, change habitat structure and quality, and have a profound and long-term impact on the survival and reproduction of giant pandas (Lin et al. 2004).

284

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

Fig. 6.35 Disturbance factors affecting giant pandas

6.4.1 Impacts of Different Types of Geological Hazards on the Habitat of Giant Pandas Giant Pandas, forests, and bamboos have interacted and evolved synergistically, and have formed strong links over long periods of evolution (Shen et al. 2004). Geological disasters such as landslides rapidly strip vegetation cover, resulting in extensive forest and staple bamboo damage, affecting giant panda species richness (Qin 1993; Hu 2001), and continuing to disturb the ecosystem for a long time to come. However, due to the complexity of the ecosystem itself, not all geological hazards play a negative role (Geertsema et al. 2009; Zhang et al. 2014a, b). In addition, landslides may expose buried soils and improve soil aeration, drainage, and temperature, which in turn may be more conducive to vegetation recovery, drive forest structural turnover, and increase biodiversity (Vittoze et al. 2001). Some foreign scholars have demonstrated that surface landslides can increase biodiversity and improve habitat environment to some extent by changing topography and soil physicochemical properties (Lundgren 1978; Myster and Walker 1997; Dale and Adams 2003; Wells et al. 2001; Claessens 2005, 2006; Geertsema and Pojar 2007). In addition, different types of geological hazards differ in the way they damage habitats due to different mechanisms of occurrence. 1. The manner of collapse damage to giant panda habitat Most of the landslides in the protected area are mainly distributed at elevations of 1320–1800 m, with steep slopes, locally reaching 70°–80°. The exposed bedrock is mainly Chengjiang-Jinning stage granite, and the tectonic fissures are developed. Some of the bedrock is exposed all year round, and the weathering of the rocks on the hollow surface is relatively strong, and the fissures are developed, the lower collapse

6.4 Habitat Evaluation of Post-earthquake Giant Panda Habitats

285

area is slow, and there are less human activities on the slope, and the vegetation cover is mostly lush woods and shrubs. According to the damage pattern of the collapse, it is mostly divided into dumping type and sliding type (Qiao 2014). Dumping type collapse is mainly cut by rock cracks to form a dangerous rock body, located on the steep cliff, the dangerous rock body is subject to root splitting, hydrostatic pressure of groundwater in the root system, etc. to form the overturning moment, the dangerous rock is subject to the overturning moment and dumps and rolls down, when affected by earthquake, precipitation, human activities, etc., the dangerous rock body is very easy to collapse (Fig. 6.36). Avalanches at steep cliffs and slopes completely strip the vegetation cover, completely destroying the root system of the vegetation and leaving the bare rock exposed for years. The impact of this type of collapse on vegetation is more serious. Some scholars even believe that it takes at least 50 years for this type of landslide to recover vegetation. Some scholars even believe that it takes at least 50 years for this type of collapse to recover, which is the most serious damage to the ecological environment. Slip collapse is a landslide body controlled by the outwash structural surface, under the action of earthquake, groundwater freezing and thawing and self-weight rock damage along a certain structural surface direction shear, slip to form a rock collapse. The landslides are mainly angular boulders with small quantity but large volume, which are distributed in the slope corners or scattered in the rivers and block the rivers to some extent. There is an obvious shear surface of the collapse body,

Fig. 6.36 Dumping type collapse

286

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

Fig. 6.37 Slip collapse

and the slope surface is distributed with more loose accumulation. These landslides have damaged the vegetation on the slope surface, but some vegetation has started to grow back around the landslides and at the gentle slope corners (Fig. 6.37). 2. Modes of landslide damage to giant panda habitat Most of the landslides in the reserve are of the cracking-faulting type. These landslides have obvious morphological characteristics, with a large number of loose boulders accumulating on the slope surface and slope corners and moving at a long distance and fast speed, often forming secondary hazards such as landslides and debris flows, which are highly dangerous. It is found through the field investigation that this type of landslide does not completely destroy the vegetation cover, and a small amount of vegetation can be seen in the upper part of the slope, and it tilts in one direction (commonly known as “drunkard’s forest”) (Fig. 6.38). The stripping effect of the landslide has reset the sequence of soil layers, and the clay layer rich in Mg, Fe, Ca and other trace elements, which was originally in the lower part of the slope, has been exposed to the surface, and the aeration, drainage and temperature of the soil layer have increased significantly, which promotes the mineralization and decomposition of organic matter in the soil and changes the chemical composition of the soil, which is conducive to the recovery of plants. A large amount of scrub has been restored to the back wall, slope angle, boundary, and mounding area of the landslide in the post-earthquake years, but the recovery of vegetation in the derogation area of the landslide is still rare.

6.4 Habitat Evaluation of Post-earthquake Giant Panda Habitats

287

Fig. 6.38 Seismic-faulting landslide and profile

3. The way debris flow and debris flow damage the habitat of giant panda The debris flow and debris flow in the protection area are mostly made by the effect of avalanche, loose debris accumulation along the terrain in the gully or slope, and the slope of the source area is steep, and the accumulation formed by the slope of the gully converges to the gully from the high level when precipitation occurs, and it is very easy to generate high level debris flow. Because of the high exit gully, compared with the general debris flow, the scale is larger and more destructive. At the same time, such geological hazards are unstable and long-lasting, taking at least 5–10 years to recover (Cui et al. 2011), and have a greater impact on human living environment and wildlife habitat. As shown in the field study, vegetation restoration is mainly located in the gentle slope angle of debris flow and on the middle and lower sides of the gully, but less in other areas (Fig. 6.39).

6.4.2 Impacts of Geological Hazards on Giant Panda Habitat The Wenchuan earthquake geological disaster brought great losses and impacts to Wolong Nature Reserve habitat, mainly in terms of (i) geological hazards have largely changed the landscape pattern; (ii) geological hazards have accelerated habitat fragmentation; (iii) geological hazards have affected giant panda behavior patterns; and (iv) geological hazards have blocked giant panda gene exchange. 1. Changes in vegetation cover by geological hazards The study analyzed the spatial and temporal variation of vegetation cover types at different times using TM remote sensing image data covering the whole region in 3 scenes, taken in the following order of year 2007–09, 2009–09 and 2015–09. The trends of landscape indices such as the number of patches, patch density and average patch area of giant panda habitats caused by the earthquake and geological disasters were counted to reveal the fragmentation process of giant panda habitats by combining the landscape pattern theory. By combining SVM supervised classification and manual discrimination, the landscape of the reserve was classified

288

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

Fig. 6.39 Debris flow

into 5woodland, grassland, bare ground, water body, and snow (Figs. 6.40, 6.41 and 6.42). In terms of the total area of the study area occupied by each landscape type, Wolong Nature Reserve before the Wenchuan earthquake consisted of 1088.9 km2 (about 53% of the total area) of woodland, 108 km2 (about 5.3% of the total area) of grassland, 654 km2 (about 32% of the total area) of bare land, 175.8 km2 (about 8.6% of the total area) of snow, and 6.39 km2 (about 0.3% of the total area) of water (0.3% of the total area) of water. As a result of the geological disaster, the forested area decreased significantly to 946.9 km2 in 2009, while the bare land and grassland increased to 716.6 and 175.7 km2 (Table 6.11), indicating that the geological disaster caused part of the forested area to be stripped into bare land or destroyed the tree trunks into grassland. From 2009 to 2015, the 6 post-earthquake years, forest land increased and bare land and grassland decreased, indicating that the post-earthquake landscape pattern was changing, and most of the disaster bodies had recovered vegetation growth under the dual effect of natural recovery and artificial planting, especially the bare land in the area of West and Middle Rivers of Sanjiang Township had been covered with vegetation, but some bare land in the area near the transhumance building of Gengda Township still had not been recovered, and the overall habitat recovery in the protected area was good. 2. Impacts of geological hazards on habitat fragmentation of giant pandas According to the different landscape pattern types in three different periods, the landscape spatial pattern indices such as number of patchres (NP), mean patch size

6.4 Habitat Evaluation of Post-earthquake Giant Panda Habitats

289

Fig. 6.40 Ground cover map of Wolong Nature Reserve in September 2007

(MPS) and edge density (ED) were selected, the spatial pattern indices of forest, grassland, bare ground, water system, snow and other landscape categories were calculated by software Fragstat 4.0 to reflect the habitat fragmentation in the study area (Table 6.11). As can be seen from Table 6.11, the area of forested land, which accounted for 53.5% of the total area of the study area before and after the earthquake (between 2007 to 2009), decreased to 946.9 km2 , but the number of patches increased from 285 to 448, and the average area of patches decreased from 574.6 to 319.3 hm2 , and the edge density increased from 15.3 to 16.9, indicating that the earthquake

290

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

Fig. 6.41 Ground cover map of Wolong Nature Reserve in September 2009

triggered geohazard caused the forest habitat patches to be divided into fine plaques with complex margins. From 2009 to 2015, the area of forested land recovered to 1063.1 km2 , and the number of patches was reduced to 341, and the average area of patches increased by 550.5 hm2 , and the edge density also decreased to the preearthquake level. From 2007 to 2009, The grass area, number of patches, and edge density of grassland were increasing and the average patch area was decreasing, which indicating that the geological disaster destroyed part of the forest land to become grassland and the ecological function was reduced, but with the natural recovery of vegetation, the area of grassland was decreasing, the number of patches

6.4 Habitat Evaluation of Post-earthquake Giant Panda Habitats

291

Fig. 6.42 Ground cover map of Wolong Nature Reserve in September 2015

and edge density were also decreasing, and the average patch area was increasing. The change trend of bare land is similar to that of grassland, the area of bare land, the number of patches and the density of patches are increasing and then decreasing, and the average patch area is decreasing and then increasing. In general, the number of patches in woodland is 56 more than that before the earthquake, but the total area, average patch area, and edge density are all close to the pre-earthquake level, indicating that the woodland habitat has recovered best. The total area of grassland recovered to 69.6% of the pre-earthquake area. The number of patches was 445 less than that before the earthquake, indicating that the grassland ecosystem had recovered

292

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

Table 6.11 Landscape classification and fragmentation index of Wolong Nature Reserve Type Wood land

Grass land

Year

Area/km2

Waters

Snow

NP/pc

MPS/hm2

ED/(pcs/m2 )

2007

1088.9

53.5

285

574.6

15.3

2009

946.9

46.6

448

319.3

16.9

2015

1063.1

52.3

341

550.5

15.1

2007

108.0

5.3

1051

15.1

10.3

2009

175.7

8.6

1157

10.3

15.1

3.7

606

12.4

12.4

2015 Bare ground

Proportion of study area/%

75.18

2007

654.5

32.1

414

158.1

17.3

2009

716.6

35.2

756

94.8

20.6

2015

663.4

32.6

606

109.5

18.6

2007

6.39

0.31

13

49.2

0.9

2009

6.58

0.32

20

32.9

0.8

2015

7.58

0.37

18

42.1

1.1

2007

175.87

8.64

484

36.3

8.4

2009

187.9

9.24

448

41.9

8.1

2015

224.71

546

41.5

10.3

11.0

to a good level. Although the total area of bare land is close to the pre-earthquake level, the number of patches is still more than 192, and the average patch area is less than 48.6 hm2 before the earthquake. The average area of the patches was less than 48.6 hm2 , indicating that the fragmentation of ecological functions of some bare land had not been restored. 3. Effects of geological hazards on behavioral patterns of giant pandas Collecting the spatial distribution maps of giant panda footprint sites before and after the earthquake and the spatial distribution maps of geological hazards, simulating the migration path of giant pandas through the least-cost distance model, and analyzing the correlation between the distance of the migration path and the surrounding geological hazard sites (Fig. 6.43). The results showed that the migration distance of giant pandas increased as the number of hazard sites increased, and the geological hazard sites within 1 km2 of giant pandas had the strongest influence on the migration distance of giant pandas, with a correlation coefficient of r = 0.647. 4. Geological disaster blocks giant panda gene exchange From Fig. 6.43, it can be seen that near Zhonggang and Renchugou (Area A) in Gengda Township, a high-density area of geological hazard distribution, the before the earthquake, many giant panda tracks were distributed, but no tracks appeared after the earthquake, while more tracks were found in the vicinity of fire burning shelter in the northwestern part of Gengda. This indicates to a certain extent that Area A has become unsuitable for the survival of giant pandas after the earthquake

6.4 Habitat Evaluation of Post-earthquake Giant Panda Habitats

293

Fig. 6.43 Simulated migration distance based on least-cost path analysis

and that pandas have the instinct to actively avoid the disaster and move to the area. The increase in the number of tracks in the northwestern part of Gengda indicates to a certain extent that area A is no longer suitable for survival and that pandas have the instinct to avoid disasters and move to more suitable environments. At the same time, this area is a high risk area for geological disasters, and the possibility of cutting off or blocking the channels for exchange of giant panda populations is high, and giant pandas will have high cumulative costs of migration and difficulties in genetic exchange with other giant panda populations. Without human intervention, panda populations living in this area may be reduced to “reproductive islands”.

294

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

6.4.3 Evaluation of Habitat Suitability Considering Geological Hazard Factors The habitat evaluation of giant pandas in Wolong Reserve is to identify the limiting or dominant factors affecting the behavior of giant pandas, and to conduct a comprehensive habitat analysis and evaluation based on certain criteria through threedimensional or hierarchical descriptions of specific indicators, based on individual indicators. In the previous habitat evaluation system, geological hazards were hardly considered in the environmental disturbance factors. The disturbance of panda habitat is more considered as anthropogenic disturbance factor. Therefore, the study integrated geological hazard factors as environmental disturbance factors into the habitat suitability evaluation of giant pandas for the first time, and the main technical route of the evaluation work is shown in Fig. 6.44. 1. Evaluation factor selection Natural environmental factors refer to the geographical location, spatial state, and survival environment in which the species are located. Environmental factors in

Fig. 6.44 Technical roadmap for post-earthquake giant panda habitat suitability evaluation

6.4 Habitat Evaluation of Post-earthquake Giant Panda Habitats

295

giant panda habitats The complexity of the interactions among environmental factors and the differences in regional characteristics make the comprehensive selection of environmental factors affecting giant panda habitats with some difficulty. Therefore, considering the evaluation purpose, data availability, and interrelationship among the factors, the study selected elevation, slope, slope direction, and water source as natural environmental factors for the evaluation of the habitat suitability of giant pandas in Wolong Nature Reserve. The elevation, slope and slope direction were obtained from the DEM, and the water source was obtained from the topographic map of 1:50,000. Biological environmental factors refer to species-related biological factors such as food, vegetation, companion species, competing species, and natural enemies. According to research studies, giant pandas mainly feed on cold arrow bamboo, short-cone yucca bamboo, and walking stick bamboo (Yi 1997; Hu 1981, 1985). Mixed coniferous and coniferous forests are the optimum vegetation for giant pandas (Hu 1981; Tang 1983). Competing species mainly include bamboo rats, red pandas and other bamboo-eating animals, but in Wolong, competing species and natural enemies have no significant adverse effects on giant panda habitat quality (Ouyang et al. 1996a, b). In this book, two factors, vegetation type and food source distribution, were selected as the bioenvironmental factors. In this study, two factors, vegetation type and food source distribution, were selected as bioenvironmental factors and vector rasterized in ArcGIS. Environmental disturbance factors are factors that put pressure on the species’ living environment, such as anthropogenic and natural disturbances. Due to infrastructure such as various giant panda research centers, townships, schools, residential houses, transportation and communication facilities in the reserve after the earthquake were heavily collapsed and a large number of residents relocated to other safe areas, direct disturbances from humans were relatively few, so this book road (S303) was chosen to represent human disturbance. For quantitative description of the impact of geological hazards on the habitat of giant pandas, this book uses the probability density values of geohazards based on kernel density estimation (KDE). Based on the above evaluation factors, an evaluation index system was established (Table 6.12). 2. Evaluation methods The habitat suitability model (HCM) was developed by using the location of animal occurrence and environmental factors to The model expresses the relationship between habitat suitability and environmental factors. With the development of GIS and remote sensing technology, the rapid acquisition and analysis of large amounts of environmental factor data to characterize the environment in which animals live has become possible, and spatial information The application of spatial information technology (3S) in spatial analysis and simulation of wildlife, habitat evaluation, etc. has been increasingly studied (Sanchez et al. 2007; Wang and Chen 2004; Bo et al. 2008; Chen and Liu 1999; Liu et al. 1997, 2002). The frequency of occurrence among species-environment relationships was counted by logistic regression methods to establish the regression relationship between species and environment

296

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

Table 6.12 Post-earthquake Wolong giant panda habitat suitability evaluation system Target layer

Factor layer

Post-earthquake Wolong giant panda habitat suitability evaluation

Natural Elevation environmental factors Height

Indicator layer

Slope

Description method Topographic map, DEM, water system distribution map

Slope direction Water Source Bioenvironmental factors

Environmental disturbance factors

Vegetation type

Habitat vegetation distribution map

Food source distribution

Bamboo distribution map

Human interference

Distance from the road

Geological hazard disturbance

Density of geological hazard sites

was established by logistic regression, and then predicted where the species might occur, i.e. the species suitability area (Fig. 6.45). 3. Evaluation results According to the standard deviation of habitat suitability of giant panda, 14.5, 15.9, 20.5, 47.6 and 1.5% of the entire area of Wolong Nature Reserve were classified

Fig. 6.45 Habitat suitability evaluation map for giant pandas

6.5 Giant Panda Habitat Recovery After Earthquake

297

as very high, high, medium, low and low suitability zones. 47.6 and 1.5% were divided into five suitability zones: very high, high, medium, low, and low, of which nearly half of the high altitude areas are considered to be unsuitable for giant pandas, mainly concentrated in Gengda Qilcenglou, Tangfang Zhongpenzi and Qitouyan in the western part of Zhonghe. The distribution of geological hazards was compared with that of the giant panda. By overlaying the distribution map of geological hazards with the evaluation map of habitat suitability of giant pandas, it is easy to that a large number of geological hazards are distributed in Gengda and Tangfang, indicating that the habitat degradation in these two areas is mainly caused by geological hazards. The geological hazards play a dominant role in the habitat degradation of the two areas, while the Qitouyan area in the western part of Zhonghe is due to the lack of bamboo, a food source for giant pandas. The most suitable habitat for giant pandas is mainly located in the southeastern part of the reserve, which consists of a large amount of coniferous forest, mixed coniferous and broad coniferous forest, and bamboo forest. The Pijiao River, Zheng River and Provincial Highway S303 cut the entire habitat into three parts. The first part is mainly located in the northwest of Gengda Township in the area of transhumance building—Walnut Ping—Cangwanggou—Kengbangzi. The second part is located in the western part of Provincial Road S303, scattered in Yinchanggou, Fengxiangping, Huahongshu and Sandaoqiao, etc. The spatial distribution of this area tends to be fragmented due to human activities such as firewood harvesting and road construction. The third part is distributed on the east side of road S303, which is the largest and most complete habitat in the whole area, connecting Zhonghe Basin, Xihe Basin and Pijiao River Basin, where about 73% of the giant panda tracksites occur. The geological hazard distribution map shows that this area is less traumatized, with a small number of geological hazards mainly located at Luyanqiao in Sanjiang, but they do not pose a threat to the giant panda population.

6.5 Giant Panda Habitat Recovery After Earthquake Giant pandas mainly inhabit mountain forests, and the relative abundance of pandas in coniferous forests is 54.44 and 40.04% in broad-leaved forests according to statistics (State Forestry Administration 2006). The restoration of giant panda habitat after the earthquake is mainly the restoration of forest area and depression, thus achieving the restoration of forest ecosystem functions. There are various means to protect and restore giant panda habitat, including natural ecological restoration, artificial intervention, establishment of ecological corridors, and strengthening post-disaster reconstruction and ecological protection.

298

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

6.5.1 Natural Ecological Restoration The surface vegetation of the reserve has been severely damaged by the earthquake and geological disaster. In the Wolong region, for example, the largest area of coniferous forest was damaged, accounting for 70% of the entire forest area (Meng 2016). These areas are critical for restoring giant panda habitats. Due to the humid climate and abundant precipitation in the habitat of giant pandas, the natural recovery of vegetation is faster. Zhang et al. (2014a, b) found that most of the vegetation-damaged areas had recovered 70% within one4 year after the earthquake with the help of multiperiod remote sensing satellite image tracking analysis, and Meng et al. (2016) found that the back wall, slope angle, boundary, and mounding areas of landslides in the protected area had recovered a large amount of scrub growth within 5 years after the earthquake, but the recovery of vegetation in the landslide’s detracted areas was still rare. For the restoration of this post-earthquake habitat, natural ecological restoration is one of the most dominant ways.

6.5.2 Human Intervention For areas where vegetation cover has been completely stripped by excessive human logging or partial collapse, and the root system of the vegetation has been completely destroyed, the habitat can be planted with high-density tree species (spruce, dove, red birch, etc.) and suitable bamboo species (walking stick bamboo, cold arrow bamboo, Huaxi arrow bamboo, etc.) through supplementary measures such as flysowing afforestation and artificial afforestation, so as to form a unique bamboo forest ecological structure suitable for giant pandas and promote their succession to potential habitats. The ecological structure of the endemic bamboo forest is suitable for the giant panda and promotes its evolution to potential habitat.

6.5.3 Establishing Ecological Corridors The most significant impact of seismic and geological hazards on giant panda habitats is the formation of “reproductive islands”, which limit and divide. The construction of ecological corridors can effectively improve the construction of ecological corridors can effectively improve the habitat dispersal of giant pandas. The construction of ecological corridors can effectively improve the dispersal of giant pandas in the habitat and promote the genetic exchange of species. According to the simulated giant panda migration corridor based on the least-cost distance model shown in Fig. 6.43, it is possible to consider the construction of ecological corridors in Gengda. The construction of ecological corridors can effectively improve the dispersal of giant

References

299

pandas in the habitat and promote the genetic exchange of species. It is important to link the scattered giant panda habitat patches.

6.5.4 Strengthening Post-disaster Reconstruction and Ecological Protection The entire post-disaster reconstruction project started in 2009 and was completed and put into use by the end of 2012, including the re-planning, siting and reconstruction of the infrastructure of the China Research Center for the Protection of Giant Pandas, the Ya’an Bifengxia Base of the China Research Center for the Protection of Giant Pandas and the Dujiangyan CDC of the China Conservation Giant Panda. The center will be re-designed, re-located and rebuilt, and the ecological migration and long-term monitoring of giant panda habitat through information technology will be completed. The ecological migration and long-term monitoring of the habitat of giant panda species by means of information technology, deepening field patrols and joint prevention work, establishing an effective evaluation mechanism, and further improving the ecological protection of the reserve protection.

References Aleotti P, Chowdhury R (1999) Landslide hazard assessment: summary review and new perspectives. Bull Eng Geol Env 58:21–24 Banair AA (1995) Review of vegetation indices. Remote Sens Rev 13:95–120 Berardino P, Fornaro G, Lanari R et al (2002) A new algorithm for surface deformation monitoring based on small baseline differential SAR interferograms. IEEE Trans Geosci Remote Sens 40:2375–2383 Bo T, Zhou Y X, Zhang L Q et al (2008) A GIS and remote sensing based analysis of migratory bird habitat suitability for Chongming Dongtan Nature Reserve, Shanghai. Acta Ecol Sinica 28:3049–3059 Bovenga F, Wasowski J, Nitti DO et al (2012) Using COSMO/SkyMed X band and ENVISAT C band SAR interferometry for landslides analysis. Remote Sens Environ 119:272–285 Brabb EE, Pampeyan EH, Bonilla M (1972) Landslide susceptibility in the San Mateo Country, California. Miscellaneous Field Studies Map, USGS Carrara A (1983) Multivariate models for landslide hazard evaluation. J Int Assoc Math Geol 15(15):403–426 Chen FL, Lin H, Hu X (2014) Slope superficial displacement monitoring by small baseline SAR interferometry using data from L band ALOS PALSAR and X band TerraSAR: A case study of Hong Kong, China. Remote Sens 6:1564–1586 Chen L, Liu X (1999) Habitat fragmentation of giant pandas in Wolong Nature Reserve. J Ecol 19(3):291–297 (in Chinese) Chen X, Ran H, Qi S (2009) 1976 Analysis of the susceptibility of influencing factors of landslides induced by the year Longling earthquake. J Peking Univ (Nat Sci Edn) 45(1):104–110 (in Chinese) Chen Z, Liu Y, Zhao J et al (1998) GPS measurements of crustal movements on the eastern Tibetan Plateau. Geol China 5:32–35 (in Chinese)

300

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

Cheng S, Song HT (2008a) Impact of Wenchuan earthquake on giant panda habitat in Wolong Nature Reserve. Mt Sci 26:65–69 Cheng S, Song H (2008b) Impact of the Wenchuan earthquake on the habitat of giant pandas in Wolong National Nature Reserve, Sichuan. J Mt Sci 26(s1):65–69 (in Chinese) Chigira M, Wu X, Inokuchi T et al (2010) Landslides induced by the 2008 Wenchuan earthquake, Sichuan, China. Geomorphology 118:225–238 Chung CJF, Fabbri AG, Westen CJV (1995) Multivariate regression analysis for landslide hazard zonation. Adv Nat Technol Hazards Res 5:107–133 Claessens L (2005) Modelling landslide dynamics in forested landscapes: addressing landscape evolution, landslide soil redistribution and vegetation patterns in the waitakere ranges, West Auckland, New Zealand. Wagening University, The Netherlands Claessens L (2006) Contribution of topographically based landslide hazard modelling to the analysis of the spatial distribution and ecology of kauri (Agathis australis). Landsc Ecol 21:63–76 Constantin M, Bednarik M, Jurchescu MC (2011) Landslide susceptibility assessment using the bivariate statistical analysis and the index of entropy in the Sibiciu Basin (Romania). Environ Earth Sci 63(2):397–406 Cui P, Chen XQ, Zhu YY et al (2011) The Wenchuan Earthquake (May 12, 2008), Sichuan Province, China, and resulting geohazards. Nat Hazards 56(1):19–36 Cui P, Zhu YY, Han YS et al (2009) The 12 May Wenchuan earthquake induced landslide lakes: distribution and preliminary risk evaluation. Landslides 6:209–223 Dai FC, Lee CF (2002) Landslide characteristics and slope instability modeling using GIS, Lantau Island, Hong Kong. Geomorphology 42(1):213–228 Dale VH, Adams WM (2003) Plant reestablishment 15 years after the debris avalanche at Mount St. Helens, Washington. Sci Total Environ 313:101–113 Ercanoglu M, Gokceoglu C (2002) Assessment of landslide susceptibility for a landslide prone area (North of Yenice, NW Turkey) by fuzzy approach. Environ Geol 41:720–730 Felicisimo A, Cuartero A, Remondo J et al (2013) Mapping landslide susceptibility with logistic regression, multiple adaptive regression splines, classification and regression trees, and maximum entropy methods: a comparative study. Landslides 10:175–189 Ferretti A, Fumagalli A, Novali F et al (2011) A new algorithm for processing interferometric data stacks: SqueeSAR. IEEE Trans Geosci Remote Sens 49:3460–3470 Foumelis M, Lekkas E, Parcharidis I (2004) Landslide susceptibility mapping by GIS based qualitative weighting procedure in Corinth area. Bull Geol Soc Greece 46:904–912 Fruneau B, Achace J, Delacourt C (1996) Observation and modeling of the Saint Etienne de Tinée landslide using SAR interferometry. Tectonophysics 265:181–190 Geertsema M, Highland L M, Vauegois L. 2009. Environmental impact of landslides. In: Sassa K, Canuti P (eds) Landslides-disaster risk reduction. Springer, Berlin Geertsema M, Pojar JJ (2007) Influence of landslides on biophysical diversity-A perspective from British Columbia. Geomorphology 89(1):55–69 Gokceoglu C, Aksoy H (1996) Landslides susceptibility mapping of the slopes in the residual soils of the Mengen region (Turkey) by deterministic stability analyses and image processing techniques. Eng Geol 44:147–161 Handwerger AL, Roering JJ, Schmidt DA (2013) Controls on the seasonal deformation of slow moving landslides. Earth Planet Sci Lett 377:239–247 He H, Sun Z, Wang S et al (2008) Surface rupture zone o f the Wenchuan MS8.0 earthquake. Earthq Geol 30(2):359–361 (in Chinese) Hu J (1985) Ecogeographic distribution of giant pandas. J Nanchong Norm College (Nat Sci Edn) 2:7–15 (in Chinese) Hu J (1990) Research and progress in giant panda biology. Sichuan Science and Technology Press, Chengdu; Hu J (1981) Biology of the giant panda. Science 38(3):181–191 (in Chinese) Hu J (2001) Giant panda research. Shanghai Science and Technology Education Press, Shanghai (in Chinese)

References

301

Huang RQ, Li WL (2014) Post earthquake landsliding and longterm impacts in the Wenchuan earthquake area. China Eng Geol 182:111–120 Huang RQ, Weile L (2009) Development and distribution of geohazards Triggered by5·12 Wenchuan Earthquake in China. Sci China (Series E): Technol Sci 52(4):810–819 Huang RQ, Li WL (2008) Study on the development and distribution pattern of geological hazards triggered by the “512” Wenchuan earthquake. J Rock Mech Eng 27(12):2585–2592 (in Chinese) Huang R, Pei X, Li T (2008) Analysis of the basic characteristics and formation mechanism of the giant landslide of Daguangbao triggered by Wenchuan earthquake. J Eng Geol 16(6):730–741 (in Chinese) Huang R (2008) Basic characteristics of geological hazards of the “5-12” Wenchuan earthquake and its suggestions for post-disaster reconstruction. Chinese Geol Educ 17(2):21–24 (in Chinese) Hull V, Xu W, Wei L et al (2011) Evaluating the efficacy of zoning designations for protected area management. Biol Cons 144(12):3028–3037 Jing F, Shen XH, Hong Shun Y et al (2008) Application of remote sensing technology in seismic science research. Remote Sens Land Resour 2:5–8 (in Chinese) Kernohan BJ, Gitzen RA, Millspaugh JJ et al (2001) Analysis of animal space use and movements. In: Millspaugh JJ, Marzlu JM (eds) Radio tracking & animal populations. Elsevier, Amsterdam, pp 125–166 Lee S, Min JK (2004) Probabilistic landslide hazard mapping using GIS and remote sensing data at Boun, Korea. Int J Remote Sens 25(25):2037–2052 Lee S, Pradhan B (2007) Landslide hazard mapping at Selangor, Malaysia using frequency ratio and logistic regression models. Landslides 4(1):33–41 Li Y, Allen PA, Zhou RJ et al (2006) Dynamics of the Middle Cenozoic Longmenshan foreland basin on the eastern margin of the Tibetan Plateau and its coupling with continental collision. J Geol 80(8):1101–1109 (in Chinese) Lin CY, Lo HM, Chou WC et al (2004) Vegetation recovery assessment at the Jou Jou Mountain landslide area caused by the 921 Earthquake in Central Taiwan. Ecol Model 176(s1–2):75–81 Liu C (2008) Earthquake hazards and geoenvironmental safety in Wenchuan, Sichuan. Geol Bull 7(11):1907–1912 (in Chinese) Liu J, Yang J, Wei C et al (2004) History, status and trends of remote sensing acquisition technology for earthquake damage information. J Nat Hazard 13(5):46–52 (in Chinese) Liu XH, Bronsveld MC, Toxopeus AG et al (1997) GIS application in research of wildlife habitat change—A case study of the giant panda in Wolong Nature Reserve. J Chin Geogr 7(4):51–60 Liu XH, Skidmore AK, Wang TJ et al (2002) Giant panda movement pattern in Foping Nature Reserve, China. J Wildlife Manag 66(4):1179–1188 Liu Z, Yu X, Xu X et al (1980) Basic geological features of the Qinghai-Tibet Plateau. J Earth Sci 2:25–48 (in Chinese) Lundgren L (1978) Studies of soil and vegetation development on fresh landslide scars in the Mgeta Valley, western Uluguru Mountains, Tanzania. Geogr Ann Ser B 60:91–127 Luo Y, Yu R, Hou L et al (1998) The longmen mountain-jinping mountain intra-land orogenic belt. Sichuan Science and Technology Press, Chengdu (in Chinese) Macdonald DW, Rushton S (2003) Modelling space use and dispersal of mammals in real landscapes: a tool for conservation. J Biogeogr 30(4):607–620 Meng Q, Miao F, Zhen J et al (2015) GIS based landslide susceptibility mapping with logistic regression, analytical hierarchy process, and combined fuzzy and support vector machine methods: a case study from Wolong giant panda Natural Reserve, China. Bull Eng Geol Env 75(3):923–944 Meng Q, Miao F, Zhen J et al (2016) Analyzing and evaluating landslide impacts on giant pandas and their habitats in the Wolong Natural Reserve, China. J Mt Sci 13(10):1789–1805 Meng Q (2016) Geological hazard impact assessment of Wolong giant panda habitat based on 3S technology. Chengdu University of Technology, Chengdu (in Chinese) Micheletti N, Kanevski M, Bai S et al (2013) Intelligent analysis of landslide data using machine learning algorithms. Springer, Berlin, Heidelberg

302

6 Long-Term Remote Sensing Monitoring of Post-earthquake Habitat …

Myster RW, Walker LR (1997) Plant successional pathways on Puerto Rican landslides. J Trop Ecol 13:165–173 Nakamura R, Nakamura S, Kudo N et al (2007) Precise orbit determination for ALOS. In: Proceedings of the 20th international symposium on space flight dynamics, Annapolis, MD, USA Nandi A, Shakoor A (2010) A GIS based landslide susceptibility evaluation using bivariate and multivariate statistical analyses. Eng Geol 110(1):11–20 National Disaster Reduction Committee, Ministry of Science and Technology, Earthquake Disaster Relief Expert Group (2008) Comprehensive analysis and assessment of Wenchuan earthquake disaster. Science Press, Beijing (in Chinese) Nefeslioglu HA, Duman TY, Duemaz S (2008a) Landslide susceptibility mapping for a part of tectonic Kelkit Valley (Eastern Black Sea region of Turkey). Geomorphology 94:401–418 Nefeslioglu HA, Gokceoglu C, Sonmez H (2008b) An assessment on use of logistic regression and artificial neural networks with different sampling strategies for the preparation of landslide susceptibility maps. Eng Geol 97(3–4):171–191 Oh HJ, Pradhan B (2011) Application of a neuro fuzzy model to landslide susceptibility mapping for shallow landslides in a tropical hilly area. Comput Geosci 37(9):1264–1276 Ouyang CY, Wang RS, Fu G (1996a) Ecological niche suitability model and its application to land use planning in Taojiang. J Ecol 16(2):113–120 (in Chinese) Ouyang ZY, Yang Z, Tan YC et al (1996b) Application of geo-graphical information system in the study and management in Wolong Biosphere Reserve. MAB China’s Biosphere Res Ann (special), pp 47–55 Ouyang Z, Xu W, Wang X et al (2008) Impacts of the Wenchuan earthquake on the ecosystem. J Ecol 28(12):5801–5809 (in Chinese) Pan W, Lv S (2001) A chance for continued survival. Peking University Press, Beijing (in Chinese) Pepe A, Lanari R (2006) On the extension of the minimum cost flow algorithm for phase unwrapping of multitemporal differential SAR interferograms. IEEE Trans Geosci Remote Sens 44:2374– 2383 Piegari E, Cataudella V, Di Maio R et al (2009) Electrical resistivity tomography and statistical analysis in landslide modelling: a conceptual approach. J Appl Geophys 68(2):151–158 Powell RA (2012) Diverse perspectives on mammal home ranges or a home range is more than location densities. J Mammal 93(4):887–889 Pradhan B (2010a) Remote sensing and GIS based landslide hazard analysis and cross validation using multivariate logistic regression model on three test areas in Malaysia. Adv Space Res 45:1244–1256 Pradhan B (2010b) Application of an advanced fuzzy logic model for landslide susceptibility analysis. Int J Comput Intell Syst 3:370–381 Pradhan B (2011a) Manifestation of an advanced fuzzy logic model coupled with geoinformation techniques for landslide susceptibility analysis. Environ Ecol Stat 18(3):471–493 Pradhan B (2011b) Use of GIS based fuzzy logic relations and its cross application to produce landslide susceptibility maps in three test areas in Malaysia. Environ Earth Sci 63(2):329–349 Qiao J (2014) Study on the Distribution Law of Landslides Induced by large earthquakes and the method of hazard evaluation. Science Press, Beijing (in Chinese) Qin Z (1993) Dynamic succession of bamboo and forest in Wolong giant panda ecosystem. China Forestry Press Tang, Beijing (in Chinese) Saha AK, Gupta RP, Arora MK (2002) GIS based Landslide hazard Zonation in the Bhagirathi (Ganga) valley, Himalayas. Int J Remote Sens 23(2):357–369 Saha AK, Gupta RP, Sarkar I et al (2005) An approach for GIS based statistical landslide susceptibility zonation with a case study in the Himalayas. Lanslides 2:61–69 Sanchez Hernandez C, Boyd DS et al (2007) Mapping specific habitats from remotely sensed imagery: support vector machine and support vector data description based classification of coastal saltmarsh habitats. Eco Inform 2(2):83–88

References

303

Sezer EA, Pradhan B, Gokceoglu C (2011) Manifestation of an adaptive neuro fuzzy model on landslide susceptibility mapping: Klang valley, Malaysia. Expert Syst Appl 38(7):8208–8219 Shen GZ, Jun Qing LI, Jiang SW (2004) Structure and dynamics of subalpine forests in giant panda habitat. Acta Ecol Sin 24(6):1294–1299 State Forestry Administration (2006) Report of the third national giant panda survey. Science Press, Beijing (in Chinese) Süzen ML, Kaya BS (2012) Evaluation of environmental parameters in logistic regression models for landslide susceptibility mapping. Int J Digital Earth 5(4):338–355 Tang Z (1983) Bamboo and nutrition of giant pandas. Wildlife (5):1 (in Chinese) Tien BD, Pradhan B, Lofman O (2012) Landslide susceptibility mapping at Hoa Binh Province (Vietnam) using an adaptive neuro fuzzy inference system and GIS. Comput Geosci 45(4):199– 211 Viña A, Tuanmu MN, Xu W et al (2010) Range wide analysis of wildlife habitat: implications for conservation. Biol Cons 143(9):1969 Vittoze P, Stewart GH, Duncan RP (2001) Earthquake impacts in old growth Nothofagus forests in New Zealand. J Veg Sci 12:417–426 Wang JL, Chen Y (2004) Applications of 3S technology in wildlife habitat researches. Geogr GeoInf Sci 20(6):44–47 Wang X, Xu W, Ouyang Z et al (2008a) Impacts of Wenchuan Earthquake on giant panda habitat in Dujiangyan region. Acta Ecol Sin 28:5856–5861 Wang X, Wang L, Wang Y et al (2008b) Remote sensing earthquake damage assessment of 8.0 Wenchuan earthquake. Earthq Def Technol 3(3):251–258 (in Chinese) Wasowski J, Bovenga F (2014) Investigating landslides and unstable slopes with satellite multi temporal Interferometry: current issues and future perspectives. Eng Geol 174:103–138 Wells A, Duncan RP, Stewart GH (2001) Forest dynamics in Westland, New Zealand: the importance of large, infrequent earthquake induced disturbance. J Ecol 89:1006–1018 Wilson JP, Gallant JC (2000) Terrain analysis: principles and applications. Wiley, New York Wu Y, He SM, Li XP (2008) Characteristics, causes and prevention measures of the landslides in Sichuan Province. J Anhui Agric Sci 36:7387–7390 Xiang L, Cui P, Zhang J et al (2010) Analysis of the susceptibility of Wenchuan County to earthquake-induced landslide hazards. J Sichuan Univ Eng Sci Edn 42(5):105–112 (in Chinese) Xiao S, Ye GS, Ni ZR et al (2003) Application of satellite remote sensing images to analyze surface vegetation changes in Xiamen. Forest Sci 39(1):129–133 (in Chinese) Xu Q, Pei X, Huang R (2009) Study of large landslides in the Wenchuan earthquake. Science Press, Beijing (in Chinese) Xu Z, Hou L, Wang Z et al (1992) Mountain building processes in the Songpan-Ganzi orogenic belt. Geological Press, China. Beijing Yi T (1997) Sichuan bamboo flora. China Forestry Press, Beijing (in Chinese) Yilmaz I, Yildirim M (2006) Structural and geomorphological aspects of the Kat landslides (Tokat Turkey) and susceptibility mapping by means of GIS. Environ Geol 50(4):461–472 Yilmaz I (2009) Landslide susceptibility mapping using frequency ratio, logistic regression, artificial neural networks and their comparison: A case study from Kat landslides (Tokat Turkey). Comput Geosci 35(6):1125–1138 Yin Y (2008) Study on geological hazards of Wenchuan magnitude 8 earthquake. J Eng Geol 16(4):433–444 (in Chinese) Zhang J, Hull V, Huang J et al (2014a) Natural recovery and restoration in giant panda habitat after the Wenchuan earthquake. For Ecol Manag 319(3):1–9 Zhang Y, Cheng Y, Yin Y et al (2014b) High position debris flow: a long term active geohazard after the Wenchuan earthquake. Eng Geol 180:45–54 Zhu J, Long Z (1983) The rise and fall of the giant panda. J Zool (1):96–107 (in Chinese)

Chapter 7

Detailed Evaluation of Giant Panda Habitats and Countermeasures Against the Future Impacts of Climate

7.1 Impact and Assessment of Climate Change 7.1.1 Impact of Climate Change on Ecosystem Since the 1950s, many observed climate changes were unprecedented in previous decades to thousands of years. In recent decades, climate change has affected natural and human systems on all continents and oceans, which shows that natural and human systems are very sensitive to climate change (IPCC 2014). Climate change and other pressures such as habitat loss and fragmentation will lead to changes in species distribution range, population composition, phenological period and ecosystem function. Species with small population, narrow distribution range, weak migration ability, single feeding habit and weak genetic ability may face the risk of extinction due to climate change (Li et al. 2015; Root et al. 2003). According to the fifth assessment report released by IPCC in September 2013, under a series of greenhouse gas emission scenarios, taking the 20-year interval as the assessment unit and taking 1986–2005 as the comparison benchmark, the global average surface temperature in this section will increase by 0.3–0.7 °C from 2016 to 2035, and by the last 20 years of the 21st century, the global average surface temperature will increase by 0.3–4.8 °C. It is estimated that by the end of the 21st century, the earth’s temperature increase is likely to exceed the 2 °C threshold promised by governments. This also means that there will be a greater risk of extinction of most species due to climate change in the 21st century and beyond. In most landscapes, most plant species cannot naturally change their geographical range quickly enough to keep up with the current estimated high-speed climate change. In the 21st century, in the flat landscape, most small mammals and freshwater mollusks will not be able to keep up with the estimated speed under the scenarios of RCP4.5 and above. What’s worse, with the growth of global warming in the future, the risk of species extinction will accelerate. It is estimated that in the current environment, one of the four global mammal species is at risk of extinction, and 1/2 of the species are decreasing. Rapid climate change is posing severe challenges to ecosystem security and global Biodiversity by © Science Press 2023 X. Wang et al., Spatial Observation of Giant Panda Habitat, https://doi.org/10.1007/978-981-19-8794-6_7

305

306

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

changing a single species and its surrounding environment (Fan et al. 2014). Global climate change affects almost all organisms, and 80% of species are threatened by climate warming. The vast territory and far-reaching impact of global climate change have greatly impacted wildlife, showing signs of change everywhere from polar to equatorial tropics, from ocean to inland (Peng et al. 2002). How to interpret these changes and put forward targeted measures and suggestions is the concern of experts and scholars at home and abroad, which urgently needs to explore the interaction between climate system and biological system. Therefore, a new discipline climate change, biology was born in the late 1980s, which is a discipline to study the impact of climate change on natural system, the focus is on understanding the future impact of human induced climate change. Since the early 1980s, relevant research on global change has been brewing and planned, and the specific implementation can be traced back to the middle and late 1980s. The Intergovernmental Panel on Climate Change, a well-known intergovernmental scientific and technological organization specializing in climate change, was established in 1988 by the World Meteorological Organization, and was jointly established by the World Meteorological Organization and the United Nations Environment Programme. IPCC issued The First Assessment Report in 1990, which became a standard reference book to guide the follow-up research on climate change and its impact. Since then, reports on the distribution and migration of animals and plants caused by climate change have increased all over the world. Pounds et al. (1999) found that 20 of the 50 tailless animals (frogs and toads) in the 30 km2 sample plot of Costa Rica alpine forest in Central America disappeared with the sharp increase of population in the same period. It is considered that the changes of bird, reptile and amphibian populations in this region may be related to the recent global warming. Gian-Reto Walther et al. (2002) discussed the recent ecological response of animals and plants to climate change. The author firstly gave a lot of evidence of the advance of spring events, and discussed the drift of species distribution under climate change, including the rise of tree lines in Europe and New Zealand to higher altitudes; Alaska’s polar shrub vegetation expands to places where there was no such vegetation before; Alpine plants in Europe move upward 1–4 m every 10 years in the altitude direction; 39 species of butterflies in North America and Europe have moved north by 200 km in 27 years, which are related to global warming. This study adopts the method of comparison between investigation and historical data, and its research results are highly reliable. Parmesan and Yohe (2003) used climate change prediction, meta-analysis and cluster analysis to analyze more than 1700 species around the world based on long-term and large-scale multi species data sets. The results show that the recent trend of Biological response is consistent with the prediction of climate change. The global meta-analysis indicates that the average rate of species drift in the distribution range in the polar direction is 6.1 km/10a (or meter level/10a in the direction of high altitude), and the average advance of events in spring is about 2.3d/10a. This study removes the noise from a large number of research data obtained in the past, and obtains reliable evidence of the impact of global warming on species distribution. Root et al. (2003) collected 143 research data all over the world. The meta-analysis method showed that the change direction of the response of more than 80% of species to global warming

7.1 Impact and Assessment of Climate Change

307

was consistent with the direction predicted based on the physiological constraints of known species. At the same time, it is pointed out that there may be four types of changes in species due to climate warming. First, the species density in a certain area may change, and the species range may migrate to the polar or high altitude, because species have to move to the temperature area where their metabolism can tolerate. Second, because the natural historical characteristics of species are triggered by temperature related events, changes may occur in phenology, such as migration, flowering or spawning. Third, morphological changes, such as changes in body size and behavior. Fourth, the gene sequence may change (Root et al. 2003). At present, a large number of studies and observations show that the distribution pattern of species has changed due to climate change. Migration to high latitude or high altitude is the adaptation strategy of most species. The reason may be that with global warming, in order to seek a cooler environment similar to that before, the distribution area will move north in the latitudinal direction, while in the mountainous areas with dense animal and plant distribution, it will move higher in the altitude direction. In order to adapt to the impact of climate change, different species will make specific responses in combination with their own ecological characteristics. Global warming may be beneficial to some species and gradually expand their populations, such as some alien species, pests and pathogens; some species with poor adaptability and migration ability will be subject to this change, and the population will gradually shrink and even face the threat of extinction. Many rare and endangered animals and plants belong to the latter. Overall, global warming will make more wild animals at a loss. Compared with low altitude areas, climate change will be more prominent in alpine areas, especially temperature rise, precipitation pattern change and other extreme climate events. The research of Harald et al. (2007) shows that the warming of the European Alps is more than twice the increase of the global average temperature in the past 50 years, and the warming rate of the Qinghai Tibet Plateau is faster, equivalent to three times the global warming rate. Chen et al. (2011) predicted that the average speed of species distribution moving towards high altitude due to climate warming in recent years is about 11 m/10a, and the average speed moving towards high latitude is about 16.9 km/10a. Professor Chris D. Thomas, a British biologist, and 18 scientists from all over the world collected the species distribution and regional climate ground data of six biodiversity rich areas covering 20% of the global land area. According to the model prediction, 15–37% (165–408 species) of 1103 species in the sampling area will be at risk of extinction by 2050 under the scenario of moderate climate warming (Thomas et al. 2004). If the study is extended to the global scale, researchers estimate that more than 1 million species in the world will face the risk of extinction by 2050. Although there is some uncertainty in the climate change prediction model, it is very necessary to take some measures to prevent the negative effects of climate change (Ekins and Speck 2014). Global climate change is an important part of global change, and global change has the characteristics of large-scale and long-period spatio-temporal evolution. It is a complex system and needs to be studied with a variety of theories and methods. The macro, dynamic, fast and accurate detection characteristics of earth observation technology make it have unique advantages in global change research. Therefore,

308

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

how to rely on the rapid development of earth observation technology to study and analyze the change trend of species habitat under the background of climate change, whether the existing protected areas can continue to maintain the integrity of species and ecosystem, and how to adjust to mitigate and reduce the impact of climate change on species is an urgent problem to be solved in the field of ecological protection.

7.1.2 Impact of Climate Change on Giant Panda Habitat Giant panda is a rare and endangered animal in China. It is known as “national treasure” and “living fossil”. As one of the recognized endangered species, it was listed in the IUCN Red List in 1990. At the same time, it is also the “flagship species” of Biodiversity conservation in the world. It is on the flag and emblem of WWF as well as the symbol of world nature conservation. Giant pandas have high ecological value, scientific research value, economic value and ornamental value. They also play a very unique role in politics, economy, diplomacy and other fields. Their reputation, influence, survival and protection status have attracted universal attention from the international community. According to fossil research, giant pandas were widely distributed in the Yellow River, Yangtze River and Pearl River Basins in Southeast China, Beijing Zhoukoudian in the north and some Southeast Asian countries in the south, such as Thailand, Laos and Myanmar. Its quantity and distribution range are sharply reduced with the continuous expansion of climate change, geology and human activities. At present, it is only limited to the high mountains and dense forests at the junction of Sichuan, Shaanxi and Gansu. Its endangered status is mainly due to human activities, such as poaching and habitat disturbance in the past 3000 years. These effects, coupled with the unique ecological constraints of the species, have been further exacerbated, including feeding specificity, low reproductive rate and restricted gene exchange. Giant pandas were large carnivores at first. After millions of years of evolution and adaptation to environmental changes, they had to gradually evolve into obligatory animals. Now 99% of their food sources are composed of Bamboo under the forest. The distribution of giant pandas has shrunk sharply in the past 300 years or so. Due to the sharp rise of population and the growth rate of climate warming, especially the rapid industrial expansion in the early 20th century, the continuous expansion of the scope of human production and management activities and the excessive logging and utilization of forest resources, the living environment of giant pandas has deteriorated day by day and the population has decreased sharply. By the middle of the 20th century, Giant pandas are confined to six isolated mountain systems in Sichuan, Shaanxi and Gansu Provinces (Zhu et al. 2013; Hu et al. 2011), and the protection situation is becoming increasingly severe. In addition, due to the increasingly fierce human resource development and economic activities, such as grazing, traffic roads, medicine, logging, hunting, mining, hydropower station construction, tourism development and transmission line laying, the degree of habitat fragmentation is becoming more and more serious, and the local species groups are further differentiated from

7.1 Impact and Assessment of Climate Change

309

24 in the 1990s to 33 now, Among them, 24 local populations have high survival risk (accounting for about 12.0% of the total wild population, involving 223 giant pandas). At the same time, giant pandas have poor genetic ability and single diet, and 99% of their food comes from bamboo. These factors make giant pandas face more severe survival challenges (Kang and Li 2016; State Forestry Administration 2015). It is very necessary and urgent to accurately analyze and evaluate the habitat change trend of giant panda species under the future climate change scenario and take active protection strategies to mitigate or reduce the adverse impact of future climate change on the species. Rapid climate change has been widely recognized as a serious threat to Biodiversity and is expected to interact with other environmental factors. Climate change may increase the risk of extinction of endangered species due to small populations, habitat specialization or limited geographical scope. Most endangered species are limited to specific habitats, have poor tolerance to environmental change, poor migration/dispersion ability, and cannot track climate change. A major challenge in conservation planning for these species is to integrate the impact of climate change into species conservation strategies. At present, the assessment of impact of climate change on giant panda habitat mostly focuses on a larger scale, such as the research on the overall giant panda habitat, or on a larger mountain scale, such as Qinling Mountain, Minshan Mountain and Daxiangling Mountain, and most studies focus on predicting the direction in which habitat will increase or decrease under the background of climate warming. However, few studies gave more specific and clear protection measures for specific research areas in combination with the current protection network. This is precisely the information and suggestions urgently needed for the current giant panda habitat management in order to more effectively protect species continuity. Songer et al. (2012) used the maximum entropy model (MaxEnt). Under scenario A2, GCM3 and had CM3 were selected, 19 climate factors selected from 35 were used to predict habitat suitability and change, and AUC was used to evaluate the model accuracy (AUC = 0.752). The study area is the whole giant panda habitat, including six mountain systems of Qinling Mountains, Minshan Mountains, Qionglai Mountains, Daxiang Mountains, Xiaoxiang Mountains and Liangshan Mountains. The time scale is 2080. The research conclusion is that under the influence of climate change, the distribution range of giant pandas will shift to high altitude and high latitude, which is consistent with many previous research conclusions. In the future, about 90% of the suitable habitats will appear in the three mountains in the north, while the mountains in the South will reduce more than 80% of the suitable habitats. In addition, the predicted results under the two climate models show that 60% of the habitat will disappear, but the predicted degree of habitat loss under different climate models and regions is also different. Among them, Qinling is the most optimistic, with only 1% habitat loss under CGCM3 and 17% under HadCM3, and 70% of the reserves are still valid in 2080; the loss of other mountain habitats is 60–97%, and only 1–28% of the effective protected areas in 2080. The prediction results under the two climate models show that by 2080, few suitable habitats will be reserved in the three mountains in the south. Under both models, the potential habitats outside the current

310

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

distribution are predicted, but only 12–14% of the new habitats are in the reserve. Habitat fragmentation has also increased—the mean patch size (MPS) has decreased significantly [from 505 km2 reduced to 67 km2 (CGCM3) and 38 km2 (HadCM3)], mean nearest neighbor distance (MNN) increased except Qionglai Mountains and Xiaoxiang Mountains. Jian et al. (2014) used multivariable linear regression model to calculate habitat suitability index and drew habitat suitability index map. The study area is the range after the expansion of 30 km buffer zone (101°19, –109°10, E, 27°52, –34°18, N) outside the distribution area of giant pandas in Qinling Mountains in three national surveys. The time scales are 2002–2050 and 2050–2099. The results show that from 2002 to 2050, the temperature will increase by 1–2 °C, the humidity will increase by 10–20%, and the precipitation will also increase, which will cause 2.64% of the unsuitable area to become the suitable area and 1.5% of the suitable area to become the unsuitable area. Higher temperature will increase Biomass to a certain extent and turn the unsuitable area in the southwest of the study area into a suitable area. However, the increase of precipitation will also change the regional environment, resulting in some suitable areas in the Northeast becoming unsuitable areas. From 2050 to 2099, the temperature will rise again by 1–3 °C, the humidity will decrease, and the precipitation will be stable, which will cause 3.43% of the unsuitable area to become the suitable area and 6.59% of the suitable area to become the unsuitable area: too high temperature and reduced humidity will reduce the biomass in Southwest China and turn the suitable area into the unsuitable area. While in the northeast, due to the suitable climate and stable precipitation, it can promote the local Biomass, so the suitable area will increase. In general, climate change will help the habitat in the previous stage (2050), and with its further change, the giant panda habitat tends to move from south to north. But the overall change trend still poses a threat to the survival of giant pandas. Fan et al. (2014) used the mechanistic model to predict the habitat under two climate scenarios (A2 and B2) with three climate factors (monthly average temperature of January, May to October, and annual precipitation) instead of Biological factors. The study area is Qinling Mountains, and the time scale is 2070–2100. The research conclusion is that compared with the current stage (1990–2007), the climate will become warm and humid from 2070 to 2100—this change will lead to the reduction of suitable habitats (62% in A2 scenario and 37% in B2 scenario), the increase of altitude and the emergence of new habitats in the northwest, which means that the habitat will move to the northwest as a whole. There are some problems in this method: some other factors such as vegetation type and its distribution, sunshine and so on are not considered; The evaluation criteria are uncertain—more parameters need to be added to improve the effectiveness of the evaluation criteria; The meteorological data provided by the international meteorological center may not accurately describe the local meteorological conditions; Extreme weather conditions were not considered. In addition, such models rarely combine the real distribution points of species in the calculation of habitat suitability, so the results are only the predicted suitable habitat, and cannot reflect the actual utilization of the habitat by the giant panda.

7.1 Impact and Assessment of Climate Change

311

Li et al. (2015) used the Maximum Entropy Model (MaxEnt) to consider the bamboo distribution map, eight bioclimatic factors with correlation coefficient lower than 0.7, environmental factors (bamboo availability, slope and aspect), and human disturbance factors into the model, and calculated them respectively under three emission scenarios (RCP 2.6, 4.5 and 8.5) of the five climate models. The study area is the whole giant panda habitat, including three provinces (Sichuan, Gansu and Shaanxi). Time scale is 2070. The research conclusion is that with climate change, by 2070, the suitable habitat will be reduced by 50–70%, and the quality of suitable habitat will decline, and there will be a serious trend of fragmentation. Among them, the habitat in Minshan Mountains will move westward, and the suitable habitat in Qinling Mountains and Qionglai Mountains will be narrowed. There will be a gap in the habitat of Qionglai mountain system, which may cause population isolation of giant pandas, and the suitable altitude generally increases. In short, climate change has the greatest impact on habitats in Qinling Mountains, Qionglai Mountains and Daxiangling Mountains. Li et al. (2015) used the MaxEnt model to add eight Biological factors (Bio 2, Bio 4, Bio 10, Bio 11, Bio 15, Bio 17, Bio 18, Bio 19) with correlation coefficients lower than 0.7 into the model to predict the distribution of bamboo under three emission scenarios (RCP 2.6, RCP4.5 and RCP 8.5) under five climate models (CCSM4, CNRM-CM5, Had GEM2-ES, MICRO5 and MPI-ESM-LR). The study area is the whole giant panda habitat, and the time scale is 2070. The results show that climate change affects the distribution of bamboo—six species of bamboo will disappear from their habitats. At the same time, the forest area of a single bamboo species will increase. The number of bamboo species decreases most obviously in Qinling, Daxiangling and Qionglai Mountains, while it will increase in Minshan Mountain in the northwest and Liangshan Mountain in the south. Liu et al. (2016) used the habitat suitability model, using temperature data and representative habitat factor data (altitude, vegetation type and bamboo species), selected A2 as the current greenhouse gas emission scenario, RCP 2.6 as the CO2 emission model in 2050, and BCC-CSM1-1 as the climate model to evaluate the habitat suitability and predict its future situation. In addition, the MaxEnt model is also used to predict the same study area to compare the results of the two models. Referring to Songer et al. (2012), regarding the mean temperature of the hottest quarter as the strongest influence factor, and 2 km2 grid was used to explore the preference of giant pandas for climate. Model validation: calculating Pearson correlation coefficient between occurrence times and suitability of giant pandas in each grid (ρ = 0.175, P < 0.05). The study area was set in the north of Minshan Mountains, and the time scale is to 2050. The results show that 15–24 °C is the suitable temperature for giant panda, and 6–14 and 25–39 °C are the second suitable temperature; The occurrence data of giant pandas are negatively correlated with temperature (ρ = 0.133, P < 0.05), which reveals that giant pandas are more suitable to live in places with low temperature. However, WorldClim predicts that the temperature will rise to 28.9 ± 11 °C in 2050, much higher than the current 20.7 °C, so the future temperature will pose a threat to the survival of giant pandas; The order of contribution degree of influencing factors: altitude (49.3%), temperature (25.1%); At present, the suitable

312

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

and sub suitable areas account for 70.7% of the study area, and the East (76.7%) is higher than the Northwest (66.3%); Under scenario A2, the habitat suitability will increase by 2.7% and decrease by 7.1% in the East, especially in the eastern edge, and increase by 9.8% in the northwest. The prediction results of maximum entropy model show that the suitability is 56.1% in the East and 43.4% in the northwest, and will increase by 2% by 2050; The suitable and sub suitable areas in the East will decrease by 5.01% and the northwest will increase by 7.5%. Model validation: training AUC = 0.89, test AUC = 0.79. The overall change trend is the same as that calculated by habitat evaluation model. Wu et al. (2009) used CART model to simulate and analyze the changes of distribution range and spatial pattern of giant pandas under the background of climate change under four greenhouse gas emission scenarios (A1, A2, B1 and B2). The results show that the current suitable distribution range of giant pandas will be further reduced under the background of climate change, in which the change is the largest under scenario A1 and the smallest under scenario B1. Under the background of climate change, the new suitable distribution area will mainly expand to the west of the current suitable distribution area, while some suitable ranges in the East, northeast and south of the current suitable distribution area will no longer be suitable.

7.1.3 Fine-Scale Evaluation of Climate Change Impact on Giant Panda Habitat The country and relevant departments have always attached great importance to the protection and management of giant pandas and invested huge labor power, material and financial resources to strengthen the protection and management of giant pandas and their habitats. So far, China has completed four nationwide giant panda surveys, including the First Giant Panda Survey organized and carried out from 1974 to 1977, the Second Giant Panda Survey organized and carried out from 1985 to 1988, the Third Giant Panda Survey organized and carried out from 1999 to 2003, and the Fourth Giant Panda Survey completed from 2011 to 2014. The results of the previous two surveys can be used as an important reference due to the long time and the limitations of technology, labor power and other conditions at that time. The third and fourth surveys basically adopted the same methods and techniques, and the survey time is about 10 years apart. The comprehensive comparative analysis of the two survey results is an important data source for researchers engaged in giant panda and its habitat. The research of this book also takes the survey data of these two times as an important input. According to the results of the Fourth National Giant Panda Survey (2011–2014), by the end of 2013, the population of wild giant pandas in China had reached 1864, an increase of 268 compared with 1596 in the third survey, with a growth rate of 16.8%, and the population of wild giant pandas increased steadily. Second, the habitat range has been significantly expanded. At present, the areas of wild giant panda habitat

7.1 Impact and Assessment of Climate Change

313

and potential habitat in China are 2.58 million hm2 and 0.91 million hm2 respectively, which are distributed in Sichuan, Shaanxi and Gansu Provinces. According to the statistics of administrative divisions, the number of cities (prefectures), counties (cities, districts) and towns involved are 17, 49 and 196 respectively. Compared with the third survey, the number of counties (cities and districts) increased to 4, and the growth rates of habitat area and potential habitat area were 11.8% and 6.3% respectively. In addition, the protection and management capacity of wild giant pandas has been gradually enhanced. Compared with the third survey completed about 10 years ago, there are 27 new reserves, with an area of 1.18 million hm2 . The number of reserves with giant pandas and their habitats has reached 67, with a total area of 3.36 million hm2 . The current nature reserve network covers 66.8% of the wild giant pandas and 53.8% of the giant panda habitats, and the population and habitat protection rate are constantly improving. IUCN also reduced the protection status of giant pandas in the red list from endangered to vulnerable (Swaisgood et al. 2016), which is based on the increase in the number of adult giant pandas between the second tone (1985–1988) and the fourth tone (2011–2014). On the one hand, the growth of giant panda population and the increase of suitable habitat are due to a series of major forestry ecological protection projects successively carried out by the country, including natural forest protection, returning farmland to forest and grassland, wildlife protection and nature reserve construction. The implementation of these projects has played a very positive role in promoting the protection of giant pandas. In addition, it is also related to the attention and efforts of many scholars at home and abroad caused by the degradation and fragmentation of the suitable habitat of giant pandas in the early years. Since the mid-1960s, Chinese scientists have begun to study the ecology of wild giant pandas. Wolong and other first giant panda nature reserves have been established here, and giant pandas have been listed as prohibited animals. However, giant pandas still face a major threat: the reduction of suitable habitats. In addition to protecting the current giant panda habitat and existing nature reserves, predicting the climate and environment in order to understand in advance the impact of climate conditions on the potential suitable habitat of giant pandas is also one of the important tasks of protecting species diversity. Accurate and reliable prediction results can provide information for relevant departments to prepare in advance, such as establishing ecological corridors to connect existing and future habitats, establishing protected areas in future habitats, etc. On the other hand, the distribution area of wild giant pandas is located in the source area of the main tributaries of the upper reaches of the Yangtze River and the places flowing through and the hot spots of biodiversity protection in the world. The effective protection of giant pandas and their habitats is not only conducive to the construction of ecological barriers in the upper reaches of the Yangtze River, but also conducive to the “umbrella” of Sichuan golden monkeys, lesser pandas, takins, dove trees, ginkgo and many other national key protected wildlife species. Constructing an effective evaluation system for the suitability of giant panda habitat and carefully evaluating the possible change trend of giant panda habitat under the background of

314

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

current and future climate change are important measures to effectively protect giant pandas and their habitat. In addition, in recent years, ecological niche models (ENMs) are increasingly used to predict the impact of climate change on species distribution around the world (Tanner et al. 2017). ENM is a new research field based on niche theory. ENMs uses the known distribution data of the research object and its related environmental variable data to calculate the actual niche of species according to a certain theory or model, and then projects the calculation results to the new environmental variable space in different time and space to predict the potential distribution of species. Although niche model was originally mainly used to study biogeography, more recent studies have focused on using this model to calculate the potential distribution of species (Anderson and Martinez-Meyer 2004), species vulnerability to climate change (Li et al. 2017), population size (LeGault et al. 2013), population density (Oliver et al. 2012) Reproduction parameters (Brambilla and Ficetola 2012) and species abundance (Howard et al. 2014) have been widely used in species distribution prediction and evaluation, invasion biology, conservation biology, and the study of spatial transmission of infectious diseases. The concept of niche was put forward by Grinnell (1917), a pioneer in this field in the 1900–1920. It has a development process of more than 100 years, and has experienced three main stages: concept establishment, differentiation and unification and quantitative modeling. This kind of models and algorithms have made great progress in recent twenty or thirty years. At present, there are more than 20 commonly used algorithms, such as maximum entropy model, BioClim analysis system and ecological niche factor analysis (ENFA). Although various models have different theoretical basis, data requirements, data analysis and expression methods, these basic algorithms make it possible to accurately evaluate the changes of giant panda habitat under the background of climate change by using niche model. To sum up, facing the urgent need of protecting the national treasure giant panda and its habitat more effectively under the background of climate change, this book intends to adopt the ecological niche model widely used at present, combined with the data of bioclimatic factors closely related to the distribution of giant pandas, supplemented by multi-scale, long-time series Multi source remote sensing data and related products are used to carry out fine assessment of giant panda habitat change under the background of climate change. The study plans to assess the impact of climate change on giant panda habitat at two levels. Firstly, the habitat of Sichuan Giant Panda, which accounts for more than 74% of the number of wild giant pandas in China, is evaluated on a macro scale. Secondly, the core area of giant panda habitat—Ya’an area is predicted and evaluated on a more detailed spatial scale. At the same time, the feasibility of the recommendations is verified with high-resolution remote sensing data, DEM, LUCC and other thematic data, GIS and ground survey data on a fine scale, in order to provide information support for the protection and management departments. It is worth noting that the research methods and technical routes in this book are also applicable to the analysis and evaluation of protected animals and plants in other countries, with a view to

7.1 Impact and Assessment of Climate Change

315

making a modest contribution to China’s biodiversity protection and the construction of ecological civilization.

7.1.4 Technical Roadmap for Assessing the Impact of Climate Change on Giant Panda Habitat In the study, we used the MaxEnt model (version 3.3.3 k) based on niche theory to analyze and predict the suitability of giant panda habitat. This model is a more commonly used niche model in recent years. Compared with other niche models, it has higher accuracy and better prediction ability. Especially when there are few sample points and only point data, it is considered to be one of the algorithms with the best prediction ability, and its prediction results are better than many other algorithms. In addition, studies have shown that MaxEnt model has been used to study the potential distribution of invasive species, reserve planning and the impact of global climate change on species distribution for more than 2000 times, which also shows the effectiveness of MaxEnt from another side. At present, in the existing analysis of the impact of climate change on giant panda habitat, there are many studies using maximum entropy model. However, most of them are large-scale research, and combined with the existing protection network is not specific enough, and the operability for management needs is not strong enough. Under the dual influence of climate change and human economic activities, the biological suitable habitat will change accordingly. In MaxEnt model, by replacing the environmental variables reflecting the current environment with the environmental variables reflecting the future environment, the purpose of predicting the distribution range of species in the future can be achieved. Since the suitable habitat of species is declining all over the world, one of the effective measures to strengthen species protection is to plan and establish effective nature reserves. The maximum entropy model can evaluate the current and future habitat suitability of species, so as to provide the management and protection department with the habitat suitability of species, and provide reference for planning nature reserves, building corridors and designing buffer zones. Firstly, we obtain the species sampling points and single factor layers corresponding to the two scale study areas through data preprocessing, and then conduct comparative research on uncertain data sources during climate prediction on the scale of the whole Sichuan Giant Panda habitat, including the optimization of benchmark climate data sets, environmental variables The optimal combination scheme is determined by the comparative analysis of different global climate models and different typical concentration paths. The maximum entropy model is used to carry out the current and future habitat transformation under the background of climate change at two scales. In the simulation process, 75% of the sample points of giant pandas are randomly selected as training samples, and the remaining 25% are the test model results. In each scenario, the model is run with 10 randomly generated sample sets,

316

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

and the average value of the results of 10 runs is the model operation result. At the same time, we measure the accuracy of the model by the AUC (area under the curve) value of the receiver operating characteristic (ROC). The output value of the maximum entropy model is a continuous value of 0–1. The AUC value of 0.5 is similar to the random distribution result. The closer it is to 1, the better the prediction accuracy of the model. The AUC value greater than 0.75 is considered to be useful. The results of the maximum entropy model are imported into ArcGIS 10.2. Different threshold separation methods are used to reclassify the model results for different study area scales. For the large-scale study area, the operation results are divided into suitable and unsuitable by using the logical value corresponding to the 10th percentile training presence. For Ya’an research area on a regional scale, in order to make the research results correspond to the results of the fourth giant panda survey in China, the book uses the logical value corresponding to the minimum training presence as the low breakpoint, and the logical value of maximum training sensitivity plus specificity as the high breakpoint to divide the operation results into appropriate Sub suitable and unsuitable. So far, the habitat suitability evaluation map in the future 2050 under the climate change scenario is obtained, and the habitat change trend and amplitude are obtained through in-depth analysis. Then, for the buffer zone with roads, mining sites, hydropower stations, scenic spots and transmission lines as interference factors on the basis of considering only the habitat suitability in the future 2050 under the climate change scenario, the radius of the buffer zone is 3 km from the expressway, and the rest are 2 km (Sichuan Forestry Department 2015), comprehensively analyze the future habitat change after human disturbance. On this basis, comprehensive analysis is made to put forward possible active protection suggestions and formulate the scope of priority protected areas. Finally, the feasibility of protection suggestions is verified by using high-resolution remote sensing images, land cover maps, DEM and other remote sensing data, combined with field survey data, and finally feasible decision-making suggestions for giant panda habitat protection are obtained. The research technology roadmap is shown in Fig. 7.1.

7.2 Introduction to Dual Scale Study Area In order to realize the fine assessment of the impact of climate change on the habitat of giant pandas, the study set up a study area at macro and regional scales. On the macro scale, Sichuan Giant Panda habitat with the highest population number and density in China is selected as the study area. On the regional scale, Ya’an area, the core area of Sichuan Giant Panda habitat, is selected for further detailed evaluation. This area is not only located at the core of the national giant panda habitat in terms of geographical location, Moreover, it spans Qionglai mountain system and Daxiangling mountain system, and is also the city (prefecture) with the largest habitat area of giant pandas in Sichuan Province.

7.2 Introduction to Dual Scale Study Area

317

Fig. 7.1 Technical roadmap

7.2.1 Sichuan Giant Panda Habitats Giant panda is a unique and precious wild animal in China. It is known as “national treasure” and “living fossil”. It is also the flagship species of biodiversity conservation in the world. It is listed in the red list of endangered species of the World Conservation Union (Swaisgood et al. 2016). Historically, giant pandas have been widely distributed in China. Their fossil remains have been found in the Yellow River, Yangtze River and Pearl River basins. They are distributed from Zhoukoudian in Beijing to the South and to the north of Vietnam, Laos and Myanmar. However, with the continuous expansion of climate change, geology and human activities, they have shrunk sharply. The large area reduction of giant panda distribution area began more than 1000 years ago, and retreated sharply in recent 100–200 years. At present, giant pandas are only limited to the six mountain systems of Qinling, Minshan, Qionglai, Daxiang, Xiaoxiang and Liangshan from north to South (Fig. 7.2). Among the six mountain systems, Minshan Mountain system and Qionglai mountain system in Sichuan Province are the largest distribution areas. In order to further understand

318

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

the dynamic changes of giant panda population and its habitat, in accordance with the requirements of laws and regulations such as the Wildlife Protection Law and the Regulations of Terrestrial Wild Animals Protection, the State Forestry Administration organized and carried out the latest Fourth National Giant Panda Survey from 2011 to the end of 2014. In February 2015, the State Forestry Administration announced the results of the four surveys. The survey results show that by the end of 2013, the number of wild giant pandas in China had reached 1,864, increased 16.8% compared with the third survey. The habitat area reached 2.58 million hm2 , increased 11.8% (State Forestry Administration 2015). In addition, according to the fourth survey report on giant pandas in Sichuan Province, as the main distribution place of wild giant pandas, Sichuan Province has 1387 giant pandas, accounting for 74.4% of the total number in China. The total habitat area of giant pandas in Sichuan is 2438500 hm2 , accounting for 94.5% of the total habitat area in China. Both the population number and population density in Sichuan are at the highest level in China. Compared with the second and third giant panda surveys, the number of wild giant pandas in Sichuan increased by 52.59% and 15.01% respectively, showing a sustained and restorative growth trend. In addition, the fourth survey report on giant pandas in Sichuan Province was published in December 2015, which systematically describes the population, habitat, socio-economic status and relevant protection and management measures of giant pandas in Sichuan. The survey report is a comprehensive comparative study with the third national survey report on giant pandas published by the State Forestry Administration in 2006. The changes of giant pandas and their habitats in recent 10 years can be obtained. Therefore, the giant panda habitat in Sichuan Province is selected as a large-scale research area. According to the mountain system, Minshan Mountain system has the largest number of wild giant pandas in Sichuan Province, with a total of 666, accounting for 48.02% of the total number of wild giant pandas in Sichuan Province. The second is Qionglai Mountain system, with 528 giant pandas, accounting for 38.07% of the total in the province; Liangshan Mountain system ranks third among the six mountain systems, with 124 wild giant pandas, accounting for 8.94% of the total number of wild giant pandas in the province. The number of wild giant pandas in Daxiangling and Xiaoxiangling Mountains is the same, 38 and 30 respectively, accounting for 2.74% and 2.16% respectively. The number of wild giant pandas in Qinling Mountains (part of Sichuan) is the least, only one. The population density of giant panda is in order of Minshan Mountain system, Qionglai Mountain system, Liangshan Mountain system, Daxiangling Mountain system, Xiaoxiangling Mountain system and Qinling Mountain system (Sichuan part) from high to low (Sichuan Forestry Department 2015). This ranking is consistent with the order of habitat area from large to small. See Fig. 7.3 for the number of giant pandas in each mountain range in Sichuan Giant Panda Habitat and its comparison with the number in the second and third transfer. According to the technical regulations for the fourth national giant panda survey Sichuan provincial implementation rules, the giant panda habitat and potential habitat in Sichuan Province are distributed in 11 cities (prefectures) and 41 counties in the province, with a total area of 2,438,500 hm2 , and the geographical location is between 101°55, –105°27, E and 28°12, –33°34, N, The habitat area and potential habitat area

7.2 Introduction to Dual Scale Study Area

319

Fig. 7.2 Geographical distribution of the two scale study areas

of giant pandas in the province are 2,027,200 hm2 and 411,300 hm2 respectively. Compared with the results of the second and third giant panda surveys conducted previously, it can be seen that the area of giant panda habitat in Sichuan Province has maintained a restorative growth trend since 1985. According to the statistics of giant panda habitat area by administrative region in Sichuan Province, Ya’an City is the city (prefecture) with the largest giant panda habitat area in the province, with a giant panda habitat area of 547,700 hm2 . According to the statistics of county-level administrative divisions, Pingwu is the county with the largest giant panda habitat area in the province. The county is known as “the first giant panda county”, with a giant panda habitat area of 288,300 hm2 . Baoxing of Ya’an City ranks second, with a giant panda habitat of 192,800 hm2 . Wenchuan, which is seriously affected by the earthquake, ranks third, with a giant panda habitat area of 148,300 hm2 . According to the proportion of giant panda habitat area in the land area of the county (city, district), Baoxing of Ya’an City ranks first, its giant panda habitat accounts for 61.91% of the county area, and Fengtongzhai National Nature Reserve is also located in this county, followed by Tianquan County of Ya’an City and Pingwu County of Mianyang City, accounting for 59.68% and 48.47% respectively. According to the results of the third giant panda survey, the

320

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

Fig. 7.3 Geographical distribution map of Ya’an research area

city (prefecture) with the largest increase in giant panda habitat area is Liangshan Yi Autonomous Prefecture, with an increase in giant panda habitat area of 84,700 hm2 ; the next are Ya’an City and Mianyang City, with an increase of 65,500 hm2 and 25,100 hm2 respectively. It can be seen that Ya’an City is in the forefront in terms of the area of giant bear habitat, the proportion of giant panda habitat and the increase in number. Therefore, the study area is set as Ya’an area on a regional scale. This area is not only located at the core of giant panda habitat in China, but also across the Qionglai Mountain system and Daxiangling Mountain system, the pink area in Fig. 7.2 is Ya’an research area.

7.2.2 Introduction to Ya’an Research Area As China’s “national treasure” and “living fossil”, the fossil of its ancestor (ailurartos) can be traced back to 8–9 million years ago. After a long historical evolution, the giant panda has gradually adapted to the surrounding specific environment and reached a dynamic balance. Giant panda populations in different mountain ranges have similar and different characteristics in the selection of habitats. Therefore, in addition to analyzing the impact of climate change on the habitat of giant pandas in Sichuan on a large scale, this book also sets up a regional research area for more detailed evaluation. The selected study area is the core distribution area of giant pandas—Ya’an area. The core research area is introduced as follows.

7.2 Introduction to Dual Scale Study Area

321

Ya’an is the place where the world’s first giant panda was discovered and the hometown of giant pandas. Ya’an, located in the middle of Sichuan Province, is the transition zone from Sichuan basin to Qinghai Tibet Plateau, and also the transition zone between Qionglai Mountain system and Daxiangling Mountain system among the six mountain systems where giant pandas are distributed. According to the latest China’s fourth giant panda survey report, Ya’an City has a giant panda habitat area of 547,700 hm2 , which is the largest area of giant panda habitat in China. This area is the core area of Sichuan Giant Panda World Natural Heritage site and one of the most continuous areas of existing giant panda habitat, accounting for 21.3% of the country. There are 340 wild giant pandas living in Ya’an area, accounting for 18.24% of the total number of wild giant pandas in China. Compared with the results of the third survey, the number and population density of wild giant pandas in this area have increased. Our study area focuses on the complete area with a large number of wild giant pandas and many investigation records, including Baoxing, Tianquan, Lushan and Yingjing counties. Study area (102°15, –103°23, E, 29°28, –30°56, N) is distributed with 315 wild giant pandas, accounting for 92.65% of the number of wild giant pandas in Ya’an City, covering most of the wild giant panda activity areas, including Fengtongzhai, Labahe and Daxiangling nature reserves. The location of the study area is shown in Fig. 7.3. The geographical distribution map of Ya’an research area shows that the terrain of the whole city is high in the north, West and South and low in the East. The mountains in Baoxing County and Lushan County in the north are the areas where giant pandas are densely distributed, the mountains in Tianquan County in the West also have a large number of wild giant pandas, and the mountains in Yingjing County in the south are relatively low in altitude, which is the main distribution area of giant pandas in Daxiangling Mountain system. The climate of the whole city belongs to subtropical monsoon mountain climate. Due to the large change of elevation difference in the area, the vertical change of temperature is obvious, and the annual average temperature is 14.1–17.9 °C. The precipitation is large, known as “Rain City”, with an average annual precipitation of about 1800 mm, which is also the region with the largest precipitation in Sichuan. Ya’an is rich in forest resources, and the vegetation types mainly include evergreen broad-leaved forest, deciduous forest and mixed forest (Nanmu, Glaber, Betula, Banyan, Quercus glauca, etc.) from low to high altitude; Coniferous and broad-leaved mixed forest (hemlock, etc.); Subalpine coniferous forest (fir, spruce, birch, etc.), and alpine shrub (Rhododendron, etc.).

7.2.3 Landform of the Study Area The existing giant panda habitat spans the Yangtze River and the Yellow River, and is located in the transition zone from the first step of China’s terrain—the Qinghai Tibet Plateau to the second step—the transitional zone of plateau, mountain and basin. The whole habitat has vertical and horizontal mountains and rugged terrain. The distance

322

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

between North and south is about 750 km, and the width from east to west is 50– 180 km. It is distributed in a narrow and long island shape. The terrain is high in the West and low in the East. The altitude suddenly changes from hundreds of meters in the western edge of the eastern Sichuan basin to more than 3000 m in the eastern edge of the Qinghai Tibet Plateau. Most areas in the habitat are high mountains and canyons, with a relative height difference of more than 2000 m, such as Xuebaoding, the main peak of Minshan Mountains, with an altitude of 5588 m. Siguniang Mountain, the main peak of Qionglai Mountains, is 6250 m above sea level. The main peak of Qinling Mountain is Taibai Mountain, with an altitude of 3767 m. The highest peak of Daxiangling Mountain system is Xiaoliangshui well at the junction of Hongya, Hanyuan and Yingjing counties, with an altitude of 3552 m. The vertical band spectrum of vegetation in the habitat is obvious, forming a special landform of high mountains and valleys. Research and application of SRTM 90 m provided by CGIAR-CSI (http://srtm.csi.cgiar.org, version4) the digital elevation data is used to make the digital elevation model map of Sichuan Giant Panda Habitat and surrounding areas. The spatial resolution of the data is 30 s (about 1 km). The global DEM data is downloaded in blocks, and each data is downloaded by 6000 × 6000 pixels. The study area is composed of six DEM mosaics, as shown in Fig. 7.4. Based on the obtained 30 s spatial resolution DEM of Sichuan Giant Panda Habitat and surrounding areas, with the help of the powerful spatial analysis function of ArcGIS, the corresponding slope map of the area is obtained by using the 3D analyst tool/grid surface/slope command in ArcToolbox. At present, many researchers have described the physical environment of giant panda habitat from altitude, slope and slope direction based on mountain scale or reserve scale. In order to reproduce the habitat selection law of giant panda from macro scale and correspond to the previous research results, the slope map has been reclassified with an interval of 5, That is, the slope in the study area is divided into 9 grades: 0°–5°, 5°–10°, 10°–15°, 15°–20°, 20°– 25°, 25°–30°, 30°–35°, 35°–40° and 40°–45°. The slope map after reclassification is shown in Fig. 7.5. The slope direction data of Sichuan Giant Panda habitat are processed in a similar way. Use the 3D analyst tool/grid surface/aspect command in ArcToolbox of ArcGIS software to execute according to the default settings of the software to obtain the slope aspect map corresponding to the habitat. See Fig. 7.6 for details. The habitat of wild giant pandas includes terrain conditions such as altitude, slope and aspect, as well as various environmental factors such as climate, soil, vegetation and water area. These environmental factors do not exist independently. They are interrelated and restrict each other. The survival, population, distribution and reproduction of giant pandas are affected by these environmental factors. However, in different historical periods or under different conditions, one of various environmental factors is always dominant, which is the main contradiction that determines the adaptation of giant pandas to environmental factors. Therefore, in different areas of giant panda habitats, there are both similarities in spatial configuration and differences caused by continuous adaptation due to long-term isolation. The results of four surveys show that the wild giant panda in Sichuan Province is divided into 22 local populations within the six mountain ranges. The first and second largest

7.2 Introduction to Dual Scale Study Area

323

Fig. 7.4 Digital elevation model of Sichuan Giant Panda Habitat

populations are Huya local population in Minshan Mountain range and Xiling Snow Mountain Jiajin Mountain local population in Qionglai Mountain range respectively. The smallest local populations are Heihe, Xiaojin and Maozhai local populations, which are all composed of a wild giant panda. These local small populations with few data will be more affected under the background of future climate change, and they should be paid more attention to in climate change assessment. As mentioned earlier, many scholars have discussed the elements of giant panda habitat selection from the perspective of mountain system or nature reserve based on ground survey data, Ouyang et al. (2001) and others are representative in quantitative description when evaluating the giant panda habitat in Wolong Nature Reserve, they pointed out that the habitat quality of Wolong giant panda mainly depends on the physical environmental factors including altitude and slope, as well as the distribution of bamboo, the staple food of giant panda, the type of vegetation, the interference degree of human activities and other factors. Through long-term observation, they found that the most suitable habitat for Wolong Giant Panda is mainly located at an altitude of 2300–2800 m. The vegetation types are subalpine coniferous forest and

324

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

Fig. 7.5 Slope map of Sichuan giant panda habitat

coniferous broad-leaved mixed forest. Under the forest, there are dense bamboos, the staple food of giant pandas, including cold arrow bamboo, short cone Yushan bamboo and crutch bamboo. The suitability evaluation criteria used in this study are summarized in Table 7.1. Based on the fourth survey of trace points of Sichuan giant pandas, superimposed with DEM, slope and aspect data, this book reproduces the selection preference of wild giant pandas for three representative physical environment indicators of altitude, slope and aspect on a macro scale by spatially analyzing the frequency of giant panda trace points in different intervals of three basic physical environment factors. See Fig. 7.7 for statistics of occurrence frequency of giant panda trace points at different altitudes, slopes and slope directions in the whole habitat of Sichuan Giant Panda. Figure 7.7 shows that during the fourth tune, the traces of wild giant pandas in Sichuan are concentrated at an altitude of 1600–3600 m, of which 2400–3000 m is the most dense. According to the comparison of the vertical distribution of giant pandas in Qionglai Mountain system drawn by Hu (1986b), the altitude section is mainly coniferous and broad-leaved mixed forest and subalpine coniferous forest belt. And

7.2 Introduction to Dual Scale Study Area

325

Fig. 7.6 Slope direction map of giant panda habitat in Sichuan Table 7.1 Evaluation criteria for natural environmental factors of Wolong Giant Panda Habitat Factor

Highly suitable habitat

Suitable habitat

Moderately suitable

Low suitability

Physical environmental factors

Elevation/m

2250–2750

1500–2250; 2750–3250

≤1500 3250–3750

>3750

Slope/(°)

≤15

15–30

30–45

>45

Biological environmental factors

Vegetation

Coniferous and broad-leaved mixed forest, subalpine coniferous forest

Evergreen deciduous broad-leaved forest, coniferous forest

Hardy shrub, Low mountain secondary shrub

Alpine meadow, Alpine quickstone beach and sparse vegetation

Note According to the habitat assessment of giant pandas in Wolong Nature Reserve by Ouyang et al. (2001)

326

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

Fig. 7.7 Frequency distribution of giant panda trace points at different altitudes

the two vegetation types are Ouyang et al. (2001) classified as the most suitable habitat type. It can be seen that the statistical analysis from the macro perspective can still find the habitat selection characteristics similar to those of wild giant pandas on the regional scale. The utilization degree of wild giant pandas on different slopes is shown in Fig. 7.8. It can be seen from the figure that giant pandas mainly use gentle slopes with slopes of 5°–20°, of which 10°–15° has the highest utilization rate. This is consistent with the research results of Hu et al. (1980) and Wei et al. (1999). In addition, according to the ecological biology research of Wuyi shed giant panda in the late 1970s, ditch tails, flat ponds, and river valley terrace with a slope of less than 20° are the best food bases for wild giant pandas. This is related to the ecological characteristics of giant pandas studied the nest area of 6 giant pandas in Wolong (Hu et al. 1990). It was found that the nest area of giant pandas is 3.9–6.4 km2 , and the nest area of male animals is slightly larger than that of female animals. Each giant panda rarely uses its nest area more than 25% of its total area every month. In addition, the average linear distance of daily movement of most giant pandas is less than 500 m. Giant pandas are large in size, therefore, foraging on relatively flat hillsides is conducive to reducing energy consumption. Figure 7.9 shows the selectivity of the Sichuan wild giant panda to the aspect during the four seasons. It can be seen from the figure that the difference of the giant panda,s selection of the aspect is not very obvious. The frequency of the selection of the east aspect, the southeast aspect and the south aspect is slightly higher, and the frequency of the selection of other aspect is similar. Hu et al. (1980) during the

7.2 Introduction to Dual Scale Study Area

327

Fig. 7.8 Frequency distribution of giant panda trace points on different slope intervals

observation and statistics in Wuyi shed, the aspect was divided into sunny slope, shady slope and semi shady slope. The sunny slope included the south, Southeast and southwest; the shady slope included the north, northeast and northwest; and the semi shady slope included the East and West. According to the statistics, it was found that the giant panda mainly operated in the environment of shady slope and semi shady slope in a year. This is because the sunny slope is relatively dry, which is a hydrothermal strip for the growth of bamboo, it is not as good as shady slope and semi shady slope, so it is concluded that giant pandas like to move in a humid and cool environment. Wei et al. (1999) analyzed the habitat selection of giant pandas in Xiangling Mountain system. It shows that giant pandas like to move on the sunny south slope, almost randomly choose the east slope, and don’t like to choose the west and north slope. Zhang and Hu (2000) analyzed the habitat selection data of giant pandas in many nature reserves such as Mabian Dafeng, Mianning, Foping, Tangjiahe and Fengtongzhai. It is considered that giant pandas like warm and humid, and generally prefer to move in the southeast slope or sunny slope and semi cloudy and semi sunny slope habitat in terms of aspect selection. At the same time, the article also points out that except Qinling Mountain system, the other five mountain systems are generally strike north-south and are longitudinally distributed between the Sichuan Basin and the Qinghai Tibet Plateau. Therefore, the southeast monsoon from the Pacific Ocean and the southwest monsoon from the Indian Ocean can go deep into the east and south slopes of these mountains, which means the sunny slope can form a warm and humid habitat conducive to the survival of giant pandas.

328

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

Fig. 7.9 Frequency distribution of giant panda trace points in different slope directions

In conclusion, we can find that the habitat selection of giant pandas not only has very significant similarity between mountain systems, but also shows local differences as a strategy for long-term adaptation of wild giant pandas to the surrounding environmental system. Therefore, when evaluating the habitat suitability of this species, especially considering the habitat suitability determined by climatic conditions (temperature, precipitation, etc.) on a large scale, in addition to the bioclimatic factors on a large scale, it is also necessary to comprehensively consider many physical environmental factors closely related to species adaptability, such as altitude, slope, aspect, etc.

7.2.4 Soil Vegetation in the Study Area Due to the influence of terrain, the climate and soil vertical band spectrum of giant panda habitat are extremely obvious. The main types of soil are mountain yellow brown soil and brown soil. The vertical band spectrum of natural vegetation is obvious. Along the altitude direction, it is evergreen broad-leaved forest, evergreen and deciduous broad-leaved mixed forest, coniferous and broad-leaved mixed forest, coniferous forest, subalpine and alpine shrub, meadow, etc. the altitude range of different vegetation types will be locally different due to the different location of the mountain system and the influence degree of human activities. Because Qionglai Mountain system is located in the middle of the giant panda habitat, and Sichuan Giant Panda natural heritage site is built, the impact of human activities is moderate, so the vertical zonal altitude range of vegetation distribution can be used

7.2 Introduction to Dual Scale Study Area

329

as a reference. According to the vertical band spectrum of “Giant Panda in Qionglai Mountain” published by Hu (1986a, b), see Fig. 7.9, and the text description is sorted as follows: (1) Below the altitude of 1600 m, it is subtropical evergreen broad-leaved forest. The altitude range overlaps with human social and economic activities, so giant pandas visit occasionally. (2) It is mixed evergreen and deciduous broad-leaved forest at an altitude of 1600– 2000 m. If there is no interference in this altitude range, it may be transformed into a food base for the giant panda under disaster conditions, such as bamboo flowering. (3) It is coniferous and broad-leaved mixed forest at an altitude of 2000–2600 m. Crutches bamboo and arrow bamboo are widely distributed under the trees at this altitude and are in a dominant position among the shrubs under the forest. This section is the main food source for giant pandas in case of disaster. (4) It is subalpine coniferous forest at an altitude of 2600–3600 m. The representative tree species are Minjiang Abies and Larix. Under the forest are Bashania fangiana and Huaxi Arrow Bamboo, and Bashania fangiana is dominant. (5) At an altitude of 3600–4400 m is alpine shrub meadow, which is occasionally visited by giant pandas living without bamboo. (6) At an altitude of 4400–5000 m is Alpine screes vegetation, and there is no giant panda visit. (7) Above 5000 m above sea level is permanent snow zone. In addition, the study also used the land use data in 2010 in China’s 1:100,000 scale land use status remote sensing monitoring database to make the land use map of Sichuan Giant Panda Habitat and surrounding areas. The main data source of this data set is Landsat TM/ETM Remote Sensing Images of each period, which are generated by manual visual interpretation. Since 1990, it has been issued every five years. Since the duration of the four surveys is close to that in 2010, this book selects the 2010 land use data as the basic data. Land use types include 6 primary types (cultivated land, forest land, grassland, water area, residential land and unused land) and 25 secondary types. According to the comprehensive analysis of the figure, the habitat of giant pandas in Sichuan is mainly distributed in woodland and sparse woodland, as well as a small part of shrub grass and meadow. According to the results of the fourth survey of Sichuan Province, 3995 traces of wild giant panda activities have been found in the giant panda habitat in Sichuan Province, of which 3987 are located in the vegetation, 8 are located in the non-vegetated river beach and around the town. At different vegetation classification levels, the coniferous forest vegetation type group, cold temperate coniferous forest vegetation type, cold temperate evergreen coniferous forest vegetation subtype and fir forest group have the most wild giant panda trace points, accounting for 45.12% and 40.46% of the total giant panda trace points in the giant panda habitat vegetation in the province respectively 39.98 and 32.05% (Sichuan Forestry Department 2015).

330

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

7.2.5 Climatic Characteristics of the Study Area The giant panda habitats are located in the transition zone from Sichuan basin to Qinghai Tibet Plateau, with high mountains and dense forests. Due to the longterm effect of atmospheric circulation between the basin and the plateau, the unique mountain climate effect of giant panda distribution area is formed, that is, abundant rainfall and obvious dry and wet seasons. It is also the geographical boundary between north and south, east and west of China. Affected by the Pacific and Indian Ocean monsoon, the climate in the overall habitat is warm and humid. Except Qinling Mountain system, other mountain systems are located in the rain screen belt in Western China, with an annual precipitation of 850–1500 mm and up to 2000 mm in some areas. For example, Ya’an City within Qionglai Mountain system is known as “Rain City”, with an annual precipitation of more than 1800 mm. As the giant panda takes bamboo as its staple food, and bamboo takes the monsoon area as the distribution center, the giant panda has become a wet animal in the monsoon area. The annual average temperature in the distribution area of giant pandas is 10–15 °C, the average temperature in January is −6 to 1 °C, and the average temperature in July is 11–17.5 °C. The relative humidity of the air is 70–85%, the sunshine hours are 1040–1830 h, and the fog days are 5–322 days.

7.2.6 Interference of Human Activities in the Study Area A total of 13,737 field survey sample lines were completed in the fourth giant panda survey in Sichuan Province, and 17 types of interference factors of human activities were recorded. The interference factors with the highest encounter rate of transects are grazing and traffic roads, and the encounter rate of other transects is relatively low. According to the impact of human disturbance factors on habitat, it can be divided into two types: general disturbance for production and operation and large disturbance such as facility construction. The general disturbances of production and operation mainly include grazing, cutting, bamboo cutting, medicine picking, bamboo shoot picking, farming, etc. this book briefly introduces several representative general disturbances of production and operation according to the fourth investigation report. Grazing is the traditional production mode of the indigenous people in the giant panda habitat and surrounding communities in Sichuan Province. It is an important source of meat food and economic income for some indigenous people. Therefore, grazing is one of the interference factors with the highest sample line encounter rate. Among the mountain systems, the grazing disturbance of Minshan Mountain system, Qionglai Mountain system, Xiaoxiangling Mountain system and Liangshan Mountain system is the most serious. Compared with the third survey, grazing in the fourth survey in Sichuan Province rose from the second to the first. The main grazing species in the giant panda habitat are cattle, sheep, pigs and horses. There

7.2 Introduction to Dual Scale Study Area

331

are differences in grazing species in different regions. Liangshan Mountain system mainly raises sheep and pigs. Yaks are mainly raised in high-altitude areas of other mountain systems, and sheep are mostly raised in low-altitude areas. The disturbance degree of grazing varies greatly in each mountain system, and the two mountain systems in the South have a greater impact. The Liangshan Mountain system has the highest sample line encounter rate, which is as high as 0.5248/piece, and the whole mountain system is serious. The second is Xiaoxiangling Mountain system, and the sample line encounter rate is 0.4120/piece. Minshan Mountain system and Qionglai Mountain system are concentrated in the north and west respectively, and there are relatively few grazing activities in Daxiangling Mountain system. Since the implementation of the natural forest protection project in 1998, the commercial cutting of natural forests in the habitat has been prohibited, so there is no large-scale cutting of natural forests at present. Compared with the third survey, logging also decreased from the first to the fourth. The Liangshan Mountain system and Xiaoxiangling Mountain system in the South have the highest encounter rate, and their spline encounter rates are 0.1036 and 0.0759, respectively. Medicine collection is the traditional production mode of the indigenous people in and around the giant panda habitat in Sichuan Province, and is also the main source of finance (cash) income. The collected varieties are mainly Gastrodia elata, Paris polyphylla, Panax notoginseng, Tianqi, Angelica sinensis, Rhubarb, Polyporus umbellatus and other Chinese medicinal materials with high economy value. Herb harvesting ranks third among the interference factors in the province, mainly in Minshan Mountain system and Liangshan Mountain system. Bamboo shoot picking is the traditional production mode of indigenous people in and around the giant panda habitat in Sichuan Province, and it is the main source of finance (cash) income for some indigenous people. There are many bamboo species collected, but the bamboo species collected on a large scale are mainly August bamboo and March bamboo. The main purpose of picking bamboo shoots is to sell them to obtain economic income. At the same time, it is also an important food source for some community residents. Other collection refers to the collection of other non-wood forest products other than medicine and bamboo shoots. The main collection varieties include wild vegetables, wild fungi, forest fruits, raw lacquer, etc. it is the traditional production mode of the aborigines in and around the giant panda habitat in Sichuan Province and the source of some economic income and food for the aborigines. Fuelwood is the main energy source for the indigenous people in the giant panda habitat in Sichuan Province, which is used to meet their production and living needs such as making meals, heating and processing feed. Wood cutting is common in high mountains and remote mountains in the habitat of giant pandas. With the economic and social development and the extension of public services to remote areas, the energy structure of residents in and around the giant panda habitat is changing, and their dependence on firewood in the giant panda habitat is gradually decreasing. Hunting used to be the traditional production and life style of the aborigines in and around the giant panda habitat in Sichuan Province, providing some economic

332

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

income and meat food for the aborigines. With the promulgation and implementation of relevant international and local laws and regulations and the severe crackdown on illegal hunting, as well as the significant reduction in the number of young and middle-aged people in mountainous areas, illegal hunting activities have been reduced, and poaching activities specifically for giant pandas have been basically eliminated. However, for the purpose of obtaining high economic income, illegal hunting of wildebeest, black bear, musk deer, hyenas, crested deer Wild animals such as impala still occur, and illegal hunting by residents outside the habitat of giant pandas for entertainment purposes has increased significantly. Farming is the traditional production mode of indigenous people in the giant panda habitat and surrounding communities in Sichuan Province. It mainly occurs in the low-altitude giant panda habitat and marginal areas, including planting crops, traditional Chinese medicine, wood raw material forest, fruit forest, etc. Farming ranks seventh among the interference factors in the province, mainly in Qionglai Mountain system and Minshan Mountain system. Large scale disturbances such as facility construction mainly include roads, hydropower stations, high-voltage transmission lines, mines, etc. The giant panda habitat in Sichuan Province is located in the transition zone from Sichuan basin to Qinghai Tibet Plateau, with large river fluctuation and rich hydraulic resources. At the same time, the forest vegetation is well preserved, and there are still a few largearea primitive forests with rich and diverse landscapes and high tourism value. In addition, the habitat is rich in mineral resources and various varieties. Guoba rock and marble in Qionglai Mountain system are well-known far and near. With the rapid development of China’s economy and the in-depth promotion of the western development strategy, the development of natural resources in the giant panda habitat has been expanding day by day. According to the incomplete statistics of the fourth survey, there are 314 hydropower stations, 1192 km roads, 149 km high-voltage transmission lines, 453 mines (mining sites) and 14 scenic spots in the habitat of giant pandas in Sichuan Province. The construction of these infrastructure not only increases the income of local residents, but also has varying degrees of impact on the habitat which listed as follows. Traffic roads refer to the roads for motor vehicles, including expressways, national roads, provincial roads, county and village roads, mining and hydropower station roads, tourism and nature reserve patrol roads within the habitat of giant pandas in the province. In the context of the western development of China, with the rapid development of infrastructure construction, eco-tourism and other tertiary industries, the interference of traffic roads is widely distributed in various mountain systems, among which Daxiangling Mountain system and Qionglai Mountain system are the most serious. Traffic roads increase the division of giant panda population and the fragmentation of habitat. At present, the sample line encounter rate of traffic roads is second only to grazing and ranks second among the interference factors in the province. Therefore, when assessing the quality of habitat, traffic roads must be an important impact factor. In the follow-up evaluation process, the book will set differentiated buffer zones for analysis according to the level and impact degree of traffic roads and the results of the fourth survey report.

7.2 Introduction to Dual Scale Study Area

333

Hydropower stations are basically built in the rivers in the giant panda habitat in Sichuan Province, most of which are small hydropower stations with diversion power generation, and most of the dam types are bottom grid fence dams or low dams. Due to the successive operation of hydropower stations in Sichuan Giant Panda Habitat and its surrounding areas in recent years, the output projects required for supporting them have increased significantly, and some of them cross the giant panda habitat. The number and density of giant pandas and bases around the hydropower station are low. Among the mountain systems, except that there are no hydropower stations in Qinling Mountain system, the rest are Qionglai Mountain system, Daxiangling Mountain system, Liangshan Mountain system, Xiaoxiangling Mountain system and Minshan Mountain system according to the number of hydropower stations. The hydropower station will also be used as a comprehensive analysis of human impact factors in the subsequent climate assessment of this book. The giant panda habitat in Sichuan Province is rich in mineral resources and diverse in species. As other areas with better mining conditions are basically mined, mining and exploration are more common in and around the giant panda habitat with poor mining conditions. At the same time, illegal and illegal mining also exists in some areas of the giant panda habitat. In each mountain system. According to the number of Mines (mining sites), they are Minshan Mountain system, Qionglai Mountain system, Daxiangling Mountain system, Xiaoxiangling Mountain system and Liangshan Mountain system. There are no mines in Qinling Mountain system (part of Sichuan). Among all counties (cities and districts), the “first county of giant panda” has the largest number of Mines (mining sites) It is called Pingwu County, followed by Baoxing County and Tianquan County in Ya’an City. It can be seen that the density of Mines (mining sites) is highly overlapped with that of wild giant pandas. Therefore, mines (mining sites) are also comprehensively analyzed as an important human disturbance factor in the subsequent climate change assessment of this book. There is great potential for hydropower generation in Western China. With the promotion of the “West to East Power Transmission” project, the construction of high-voltage transmission lines has also become a major infrastructure construction content in the giant panda habitat. At present, there are 129 km long high-voltage transmission lines in the habitat of giant pandas in Sichuan, except Qinling Mountains. In addition to the absence of high-voltage transmission lines (in Sichuan), Qionglai Mountain system, Daxiangling Mountain system, Liangshan Mountain system, Xiaoxiangling Mountain system and Minshan Mountain system are ranked in order according to the length of transmission lines. The number and density of giant panda trace points around the high-voltage transmission lines are low, which may be related to the direct occupation of giant panda habitat by the high-voltage transmission lines, the cleaning of transmission corridors, and the decline of habitat quality, transmission line noise affects the utilization of giant pandas’ habitat. The two counties with the longest high-voltage transmission lines are Tianquan County and Yingjing County of Ya’an City. Therefore, this book also takes the high-voltage transmission line as an important human impact factor for the subsequent habitat suitability analysis of giant pandas.

334

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

With the convenient transportation and the rise of eco-tourism, the giant panda habitat in Sichuan Province has attracted a large number of domestic and foreign tourists with its rich and diverse landscape. In recent years, all localities have generally accelerated the pace of tourism development in and around the giant panda habitat, among which Jiuzhaigou, Huanglong and Xiling Snow Mountain have been built into famous scenic spots. At present, there are 14 large-scale scenic spots in the giant panda habitat in Sichuan Province, covering an area of 320,000 hm2 , with an annual total of 8.19 million tourists. Jiuzhaigou is the largest scenic spot and the largest number of tourists. Scenic spots are distributed in 20 reserves in the province, including Wolong, Tangjiahe and Siguniang Mountain. The quality of giant panda habitat around the scenic spot is low, which is also an important reason why the book takes the scenic spot as an important human impact factor into the subsequent analysis of giant panda habitat suitability.

7.3 Research Methods and Data 7.3.1 Maximum Entropy Model (MaxEnt) Domestic and international scholars have carried out a lot of research work on species habitat suitability evaluation and prediction, hoping to find an evaluation method with stable prediction performance, high accuracy and good universality, so as to provide a scientific basis for the protection and restoration of biodiversity. At present, the commonly used methods or models for predicting and evaluating species habitat suitability mainly include: maximum entropy model (MaxEnt), niche factor analysis (ENFA), bioclimate analysis and prediction system model (BIOCLIM), etc. different models have different prediction results and have their own advantages and disadvantages, which may be related to the factors, quantities and set parameters involved in model operation, or the selection of models (one study showed that 36% of the studies using the maximum entropy model only because it does not need to collect non distributed data, and better results can be achieved by using other models that can deal with “presence-absence” data). The mentioned models [Niche factor analysis model (ENFA model), bioclimate analysis and prediction system model] will not be discussed in this part, please see Sect. 7.1.3 in details. The following focuses on the maximum entropy model selected in the study. This model is based on the niche theory that has gradually become popular in recent years. In ecology, there are a lot of studies on the prediction of species diversity and its distribution, Ecological niche models (ENMs) play a prominent role in this research field. Niche model assumes that each species has its corresponding suitable environment, i.e. niche. This niche can be simulated by a certain model or algorithm through the actual distribution data of species and the surrounding environment data. Each species has its climate and physical tolerance, and this determines where

7.3 Research Methods and Data

335

they can survive. Most species start their internal processes according to climate clues, including temperature, precipitation and the seasonality of precipitation. The comprehensive factor to determine where a species can survive is the concept of niche, Niche is defined as the position of the population in the ecosystem in time and space and its relationship and role with other populations. Niche model is based on the concept of niche. People often participate in niche model with parameters such as ecology and climate to predict the impact of climate change on the number and distribution of species. At present, niche model has been widely used to study the impact of climate change on species distribution, prediction of invasive areas of alien species, prediction of potential distribution areas of key species, planning and adjustment of protected areas, planning and construction of ecological corridors, etc. Through extensive literature research, this book finally determines to use the maximum entropy model based on niche concept to carry out subsequent giant panda habitat prediction and assessment. The model will also be introduced below. Maximum entropy model is a niche model based on maximum entropy theory. The model predicts the distribution of species by finding the probability distribution with the largest entropy under the constraint of available actual data (that is, the mean value of each variable is close to the mean value of observed data when predicting the distribution) (Phillips et al. 2006). The model was built by Phillips et al. (2004). It is a niche model widely used in recent years. It has good prediction ability and higher accuracy than other niche models. Especially when there are few sample points and only point data, it is considered to be one of the algorithms with the best prediction ability (Elith et al. 2011; Phillips and Dudík 2008). In addition, studies have shown that MaxEnt model has been used to predict the impact of climate change on species distribution and potential distribution of invasive species for more than 2000 times, which shows the effectiveness of the model from another aspect (Xu et al. 2015). In the existing analysis on the impact of climate change on giant panda habitat, there are many studies using maximum entropy model (Gong et al. 2017; Li et al. 2017, 2015). However, most of them are large-scale studies, and the existing protection network is not specific enough, and the operability for management needs is not strong enough. The research will be based on the maximum entropy model in macro (Sichuan Giant Panda Habitat). To study the impact of climate change on giant panda habitat on two scales (Ya’an research area), this chapter will focus on the maximum entropy model and relevant research data.

7.3.2 Maximum Entropy Principle The concept of MaxEnt model is: when learning probability model, the model with the largest entropy among all possible models is the best model; if the probability model needs to meet some constraints, the maximum entropy principle is to select the maximum entropy model in the set of conditions that meet the known constraints. The maximum entropy principle points out that when predicting the probability distribution of a random event, the prediction should meet all known constraints

336

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

without making any subjective assumptions about the unknown. At this time, the probability distribution is the most uniform, the predicted risk is the smallest and the entropy is the largest. In short, in the maximum entropy, it is necessary to provide a set of samples in a spatial distribution, The idea of maximum entropy is to determine the distribution of maximum entropy (i.e. closest to uniformity) and limited by the constraint that the expected value of each feature under the distribution matches its empirical average to predict the distribution of the target species. In particular, for the species distribution model, the activity site of the species is used as the sampling point, the region of distribution to be calculated is the geographical region of interest, and the feature is the environmental variable (or its function). The predicted potential distribution of species (i.e. potential habitat) does not have spatial correlation, so the correlation does not need to be taken into account in the modeling process.

7.3.3 Formula and Description X the region of interest; x i distribution point data; f i the original environment variable or a higher-level characteristic parameter obtained by the environment variable; πˆ the simulation species distribution; π˜ the true species distribution. Entropy is defined as H ( p) = −

Σ

p(x)ln p(x)

(7.1)

x∈X

The Poisson/π distribution close to the distribution law of species and the criterion that the expected value of fi under π distribution is the same are used to approximate the simulated distribution to the real situation. Of course, there are many simulated distributions that meet this criterion. At this time, the maximum entropy principle is applied—Select a Poisson distribution closest to uniform distribution as the simulated species distribution. Alternatively, consider the Gibbs distribution: qλ(x) = eλ· f (x) /Z λ While Z λ =

Σ x∈X

eλ· f (x)

(7.2) (7.3)

7.3 Research Methods and Data

337

the distribution is a normal distribution. Therefore, according to Della Pietra and Lafferty (1997), the maximum entropy model and the maximum likelihood Boltzmann distribution (the Boltzmann distribution that minimizes the relative entropy) can be used instead of each other. However, the simulated value cannot be completely consistent with the actual value, so there is an expression: | | |πˆ [ f i ] − π˜ [ f i ]| ≤ β j

(7.4)

β j is the difference between the simulated distribution and the real distribution. The parameters in the model are mainly divided into two categories: continuous and categorical. The former mainly describes irregular values in reality, such as precipitation and maximum temperature, and the categorical mainly describes categories, such as soil type or vegetation type. At present, MaxEnt has six characteristics, namely linear-L, quadratic-Q, hinge-H, product-P, threshold-T and category-C. The more parameters, the more complex the model is and the easier to over fit.

7.3.4 Model Result Evaluation There are two kinds of errors in general prediction models: one is underestimation, which predicts the actual positive area as negative area and false negative area, and the other is overestimation, which predicts the actual negative area as positive area and false positive. These two kinds of errors are closely related to the judgment threshold. The commonly used model evaluation indexes include sensitivity, specificity, TSS (true skill statistic), kappa statistics and AUC (Wang et al. 2007). The area under the ROC curve (subject working characteristic curve) is mainly used as the measurement index of the model prediction effect in the evaluation of MaxEnt model. The value range of AUC is [0–1], the larger the value, the better the prediction effect of the model. It is generally believed that the AUC value of the model is greater than 0.7, which is considered to achieve acceptable performance (Swets 1988).

7.3.5 Advantages and Limitations MaxEnt model algorithm is clear that it only needs point data and environmental variable data affecting species distribution, and its regularization program can prevent over fitting in the case of small samples. Therefore, MaxEnt is more suitable for simulating species with limited distribution data and narrow niche. In the current commonly used multiple species distribution models, Maximum entropy (Phillips et al. 2004) is considered to be a very effective method for modeling rare species with narrow distribution range and only a small amount of occurrence data. It is worth noting that the model uses very limited existence point data and combined with

338

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

an algorithm or rule to predict the potential suitable area of species. This prediction result only represents that the predicted suitable area has similar environmental conditions with the current distribution area, and does not represent the actual distribution boundary of species. Giant pandas are mainly distributed in six mountain systems in Sichuan, Gansu and Shaanxi Provinces: Minshan Mountain, Qionglai Mountain, Daxiang Mountain, Xiaoxiang Mountain, Liangshan Mountain and Qinling Mountain. In order to study the impact of climate change on giant panda habitat, scholars have adopted different models and investigated different study areas. Through incomplete statistics, the models used for giant panda habitat evaluation and prediction mainly include: expert system, neural network, multiple linear regression, niche factor analysis, maximum entropy, etc. in recent five years, most scholars have used the maximum entropy model for giant panda habitat evaluation and prediction; The survey scope is mainly divided into large scale (six mountain systems), large and medium scale (around the giant panda habitat shown in the national giant panda survey report), medium scale (individual mountain systems) and small scale (individual districts, counties or protected areas). The prediction results vary with the study area. For the same study area, the use of different meteorological models and emission scenarios will only differ at the digital level, However, the overall prediction trend is not affected (Songer et al. 2012). Some scholars also use the method of controlling variables to use two models for prediction, and the results of the two models have the same trend. However, some research results still have different prediction trends because of different models, which may also be related to the time scale studied by scholars. For example, Fan et al. (2014) used the mechanism model to predict the habitat of Qinling Mountains. The selected time scale is 2070–2100. The results show that under the influence of climate change, the giant panda will move to the altitude, which means that the suitable habitat in the future will move to the northwest, while Gong et al. (2017) predicted the habitat of Qinling Mountains by using the maximum entropy model. The results show that by 2050, the suitable habitat for giant pandas will move about 11 km eastward; Jian et al. (2014) used different time periods for prediction (2002–2050 and 2050–2099), different climate change trends are obtained, resulting in different habitat prediction results. In addition to the selection of region, climate factors and time scale, Qi et al. (2012) also studied that data on different spatial scales participate in model operation, keep other variables consistent, and obtain different prediction trends. In conclusion, in the process of giant panda habitat assessment and prediction, we should not only pay attention to the geographical area and the selected method model, but also pay attention to other factors, such as meteorological model, emission scenario, time scale and spatial scale.

7.3 Research Methods and Data

339

7.3.6 Evaluation Index System of Giant Panda Habitat Suitability Evaluating the habitat of organisms, especially rare and endangered animals and plants, is an important means to analyze the reduction and endangerment of species. It can also provide basis for formulating reasonable protection and management countermeasures (Ouyang et al. 2001). Ouyang et al. (2001), Zhou (2008) divided the main factors affecting the survival and population reproduction of giant pandas into three categories: physical environmental factors, biological environmental factors and human activity factors. In recent years, the growth rate of climate change has highlighted its impact on global biodiversity, and climate factors have been gradually introduced into the habitat evaluation of rare and endangered species. Therefore, based on the existing research results, the research will affect giant pandas Habitat factors can be divided into four types: physical environmental factors, biological factors Human disturbance factor and climate factor (Hull et al. 2016; Liu et al. 2016). The physical environmental factors are described by three factors: altitude, slope and slope direction; the distribution range and distance from the water source of giant panda edible bamboo are taken as biological factors; human interference factors are five types of interference factors: Roads, mines, hydropower stations, transmission lines and scenic spots with high encounter rate in the habitat; climate factors are determined by WorldClim The 19 bioclimatic factors (WCBIO for short) downloaded from the website and the 19 similar climatic factors obtained by remote sensing (abbreviated as RSBIO) compare and analyze the selected benchmark climate data set. At the same time, compare and analyze the climate factors involved in the model calculation by using the correlation/contribution method and principal component analysis method to select the final selected climate factors. These climate factors are the climate factors related to the distribution of animals and plants obtained through a certain calculation combination of annual average temperature and precipitation, including annual average temperature The hottest month, the highest temperature, annual precipitation, the wettest season precipitation, etc. Because the study evaluates and predicts the impact of climate change on habitat at the macro scale and the core regional scale, the determined evaluation index system has been fine-tuned, as follows. (1) Physical environment factors: altitude, slope and aspect; (2) Biological factors: the distribution range of giant panda edible bamboo (core research area) and the distance to water source are taken as biological factors; (3) Human interference factors: at present, there are many human activities interfering with the habitat of giant pandas. The fourth giant panda survey in Sichuan Province recorded 17 types of human activity interference factors in the habitat, including logging, roads, grazing, medicine harvesting, bamboo cutting and shoot shooting, hydropower stations, power transmission lines, etc. among them, grazing and traffic roads are the interference factors with the highest sample line encounter rate. Deforestation caused by logging. The reduction of area also leads to the loss and fragmentation of giant panda habitat. In only 15 years

340

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

from the early 1970s to the mid-1980s, the habitat of giant pandas decreased by more than 56%. Roads lead to habitat fragmentation, hinder and reduce gene exchange between populations, and may eventually lead to habitat loss. Ya’an is rich in mineral resources and bamboo forest resources, so mining, bamboo cutting and bamboo shoot shooting are serious in this area, which is one of the main interference factors in the habitat of giant pandas. (4) Bioclimatic factors: 19 bioclimatic factors (WCBIO for short) downloaded from WorldClim website and 19 similar climatic factors obtained by remote sensing (abbreviated as RSBIO) compare and analyze the selected benchmark climate data set. At the same time, the climate factors involved in the model calculation are compared and analyzed by correlation/contribution method and principal component analysis method to select the final selected climate factors.

7.3.7 Research Data According to the determined habitat suitability evaluation index system, the study involves the following six types of data. (1) Physical environmental factors. According to investigation and research, giant pandas in Wolong Nature Reserve of Qionglai Mountain system usually move at an altitude of 1400–3600 m, and like to feed on ridges and platforms with gentle terrain slope and generally rising gently below 20°. In addition, giant pandas only exist in the dense alpine forests of six mountain systems, and the vegetation in their habitat is obviously vertical. Different vegetation types alternate with different altitudes. Therefore, altitude, slope and aspect are selected as physical environmental factors. The altitude is obtained from the digital elevation model. Large scale DEM research applies SRTM 90 m provided by CGIAR-CSI (http:// srtm.csi.cgiar.org/, version 4) digital elevation data production, the spatial resolution of the data is 30 s (about 1 km). Based on DEM, the corresponding slope and aspect data can be obtained by using the geographic analysis function. The DEM data in the core research area adopts the new aster GDEM geo electronic terrain product. The spatial resolution of the DEM data is 30 m and the vertical accuracy is 20 m. The two physical environment factors of slope and aspect are generated by DEM through the ArcToolbox of ArcGIS 10.2 “slope” and “aspect” tools. (2) Biological factors. The giant panda is an omnivore, 99% of its food comes from bamboo, so the distribution of staple food bamboo has a great impact on its habitat suitability. The spatial distribution map of staple food bamboo in the scale study of the core area of this book is derived from the third national giant panda survey and partially revised with the classification results of GF-01 image on March 7, 2017. The water system data are mainly from the country Basic GIS data, supplemented by remote sensing image interpretation and recorded data of early field investigation. The distance from the water source is calculated

7.3 Research Methods and Data

341

through ArcGIS 10.2 calculate the Euclidean distance from each pixel point to the nearest water source. (3) Human interference factors. According to the fourth giant panda survey report, this book obtains five large-scale disturbances with high encounter rate and great impact in the area, specifically roads, hydropower stations, scenic spots, mines and transmission lines. (4) Climate factors. Global climate model (GCM) is used to describe the possible future climate changes under preset scenarios. It can provide analysis of the impact, adaptation, vulnerability and emission reduction of climate change on the study area. The newly introduced preset scenarios are called “representative concentration pathways (RCPs)” scenarios, RCP8.5. RCP6.5. RCP4.5 scenario and RCP2.6 scenario. Compared with the emission scenarios defined by IPCC in 2000 (SRES), RCPs take into account the impact of social economy and national climate policies on greenhouse gases and aerosols, and more scientifically describe the prediction results of future climate change. Based on different RCPs scenarios, different models can be used to predict the future climate change trend. At present, compared with the more commonly used models are MRI-CGCM3, NORESM1-M, MIROC5 and BCC-CSM1.1. In another way, analyzing the simulation ability of each climate model in a certain region, determine the simulation weight of each model firstly, and then do ensemble simulation for multiple models. Climate and environmental data available from the World Meteorological Database (http://www.worldclim.org). The database was created by Hijmans et al. (2005) by collecting the detailed meteorological information recorded by meteorological stations around the world from 1950 to 2000 and generating it by interpolation. Its spatial resolution is 30 s (about 1 km), and 19 meteorological grid data corresponding to more than 10 meteorological models can be selected according to the demand. The four typical concentration paths are RCP8.5, RCP6.5, RCP4.5 and RCP2.6 from high to low. Two emission scenarios Rcp4.5 and Rcp2.6 are selected in this study. In the study, two benchmark climate data sets are used for comparative study in the large-scale study area, one of which is from the commonly used WorldClim website (http://www.worldclim.org), and another is the estimated temperature and precipitation data obtained by RS and edited by Debrauwe et al. The resolution is 0.05° (about 6 km), and its temperature time range is 2001–2013 and precipitation time range is 1981–2013. The monthly average value is used to extract the same 19 bioclimatic factors as WC. Bioclimatic factors can be obtained from https://vdebla uwe.wordpress.com. Four climate models and two typical concentration paths are used for comparative analysis in a large-scale range; BCC-CSM1-1 climate model and RCP4, which are more suitable for the geographical environment of Southwest China, are adopted in the core study area 5 emission scenario. The specific meanings of the 19 climate factors are shown in Table 7.2. The spatial resolution of WorldClim climate data is 30 arc-seconds (equivalent to about 0.86 km2 in equatorial regions, usually described as about 1 km), which is also the highest resolution climate data available on WorldClim website. Hijmans

342

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

Table 7.2 Names and meanings of bioclimatic factors Bioclimatic factors

Names of environment variables

Unit

BIO1

Annual mean temperature

°C

BIO2

Mean diurnal range [mean of monthly (max temp-min temp)]

°C

BIO3

Isothermality (BIO2/BIO7) (×100)

°C

BIO4

Temperature seasonality (standard Deviation × 100)

°C

BIO5

Max temperature of warmest month

°C

BIO6

Min temperature of coldest month

°C

BIO7

Temperature annual range (BIO5–BIO6)

°C

BIO8

Mean temperature of wettest quarter

°C

BIO9

Mean temperature of driest quarter

°C

BIO10

Mean temperature of warmest quarter

°C

BIO11

Mean temperature of coldest quarter

°C

BIO12

Annual precipitation

mm

BIO13

Precipitation of wettest month

mm

BIO14

Precipitation of driest month

mm

BIO15

Precipitation seasonality (coefficient of variation)

mm

BIO16

Precipitation of wettest quarter

mm

BIO17

Precipitation of driest quarter

mm

BIO18

Precipitation of warmest quarter

mm

BIO19

Precipitation of coldest quarter

mm

Note BIO1–BIO11 are temperature-related climatic factors, and the result is the actual temperature value × 10. BIO12–BIO19 are precipitation related data, in mm

et al. (2005) used the meteorological information recorded by a large number of meteorological stations around the world, most of which were from 1950 to 2000, to generate global climate grid data through integration and interpolation (Hijmans et al. 2005). The spatial resolution of the data is higher than that of the previously applied global land surface interpolated climate data (except Antarctica) with a resolution of 10 arc min (equivalent to 18.5 km for equatorial regions) is more than 400 times higher, and the improved elevation data is applied in the interpolation process. The reason why we chose THE global climate model BCC-CSM1-1 in the core area study is that this model was developed by Beijing Climate Center and participated in the Fifth International Coupled Model Comparison Plan (CMIP5 Plan). Meanwhile, this model is also one of the models selected in China’s National Climate Impact Assessment report. It is also proved that the evaluation results are highly reliable and robust (Shen et al. 2015). RCP 4.5 comprehensively considers China’s development trend as a future climate emission scenario. At the same time, temperature and precipitation data of Ya’an research area from 1951 to 2016 were obtained from the National Meteorological Science Data Sharing

7.3 Research Methods and Data

343

Service Platform (http://data.cma.cn) to assist in the analysis of climate change trend and change rate of giant panda habitat in this area. (5) Multi-source remote sensing data. The quality of optical remote sensing data is not good because of cloudy and rainy weather in southwest China. Landsat data with less cloud cover on June 13, 2001 and July 19, 2011 were selected for this study, considering the time periods of the third (2000–2001) and fourth (2011–2014) ground survey of giant pandas. Landsat data were used to study LUCC changes during the two giant panda ground surveys. At the same time, Gaofen-1 image data on March 7, 2017 was used to assist classification and carry out feasibility analysis and verification of later protection network. In addition, with the rapid development of remote sensing technology, some indices that can reflect the characteristics of specific ground objects are derived from the operation between different bands of sensors, among which vegetation indices are widely used. Vegetation index (NDVI) is an important index to reflect vegetation coverage, growth, biomass and is a good way to distinguish plant species. It can also reflect the intensity of vegetation activity on the surface. At present, a variety of vegetation indices have been developed to reflect the characteristics of vegetation change. The commonly used normalized vegetation index (NDVI), as an indicator of surface vegetation activity, can reflect the information of vegetation cover on a large spatio-temporal scale, and is widely used in land cover monitoring, land cover classification, crop yield estimation and other fields. The commonly used large-range normalized vegetation index can be divided into monthly mean, quarterly mean, annual mean and multi-year average according to the statistical cycle, because this index can better reflect the changes of vegetation growth state, vegetation coverage and human activities in different time intervals, and the giant panda habitat and vegetation coverage are closely related. Therefore, this book also uses NDVI index to evaluate the suitability of giant panda habitat. As an effective reflection of vegetation characteristics, NDVI data is widely used in species distribution models (SDMs). In this study, the time series NDVI product data are MODIS MOD13A3 NDVI data monthly from January 2001 to December 2015, with a resolution of 30 s (about 1 km). The MODIS NDVI product data are considered to be the continuation and upgrade of AVHRR NDVI, eliminating the interference of atmospheric water vapor, improving the spatial resolution and chlorophyll sensitivity, and adjusting the synthesis method. The product data are developed by NASA MODIS Land Product Group according to the unified algorithm. Figure 7.10 shows the change trend of the mean NDVI of the Sichuan giant panda habitat from 2001 to 2015. As can be seen from Fig. 7.10, the overall trend of mean NDVI is increasing year by year, which indicates that the vegetation coverage in this region is increasing year by year. This is due to the continuous attention paid by the state to the habitat of giant pandas for a long time, as well as a series of protection measures such as returning farmland to forest and natural forest protection project. At the same time, it is obvious from the figure that the mean VALUE of NDVI in 2008 suddenly decreased, which was caused by the destruction of large areas of green vegetation such as forest and

344

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

Fig. 7.10 Annual mean NDVI variation of Sichuan Giant Panda Habitat from 2001 to 2015

grassland after the Wenchuan earthquake in May 2008, resulting in a significant reduction of green vegetation. Moreover, it took a period of time for vegetation to recover in the earthquake area, and it was not until 2014 that it basically recovered to the original level. The Ya’an earthquake in 2013 is not clearly reflected in this curve, so it can be seen that the impact of Ya’an earthquake on the panda habitat is not significant. See Fig. 7.11 for the monthly mean NDVI variation of Sichuan giant panda habitat from January 2001 to December 2015. As can be seen from Fig. 7.11, the mean value of NDVI in winter and summer of the giant panda habitat in Sichuan basically shows a gradual increase trend, which is consistent with the annual trend. As can be seen from the figure, the mean VALUE of NDVI in winter of 2011 and 2012 was significantly lower than that of adjacent years, which was also reflected in the mean value curve of NDVI in Fig. 7.10, which may be related to the gradual restoration of vegetation after the earthquake. In addition to NDVI, which can indirectly reflect the vegetation status within the habitat, land cover type data is also used in the model calculation on a large scale. Although MaxEnt can support both continuous environmental factors and classified data, such as land use cover, continuous variable budgeting results are better. Therefore, the land cover/use data adopted in this study is the global 1 km continuous land cover data made by Tuanmu and Jetz (2014). The data is calculated by using the consistency of four global land cover products, and the continuous probability estimation of 12 land cover types corresponding to each pixel is given. These data have been shown to be better than their original land cover counterparts at

7.3 Research Methods and Data

345

Fig. 7.11 Monthly NDVI variation of Sichuan giant panda habitat from January 2001 to December 2015

predicting species distribution. The data are available on site: http://www.earthenv. org/landcover, is a valuation of land cover from 1992 to 2006, for the recent data to give greater weight (Tuanmu and Jetz 2014). (6) Giant panda habitat survey data. The state Forestry Administration completed the fourth giant panda survey from 2011 to 2014. The traces of giant pandas, human disturbance factors (roads, hydropower stations, tourist attractions, power transmission lines, etc.) and protected areas in this book are all from the fourth giant panda survey report. In addition, around the study area, the team member has carried out three times a field investigation and verification work, from north to south, covering the Wanglang National Nature Reserve, the Wolong Nature Reserve, and Ya’an region, main work contents including the classification accuracy of vegetation types and the giant panda habitat suitability evaluation results, human disturbance factor for field investigation, visit the local community, As well as the exchange and sharing of information and data with local research departments. Table 7.3 lists the data used in this book at the two study area scales. All data were unified in UTM WGS84 coordinate system. Since the habitat of giant panda is high in mountains and dense in forests, and the climate observation stations are sparse, the resolution of all grid data is unified at 1 km × 1 km to match the climate data.

346

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

Table 7.3 A list of data used in the study Data type

Sichuan giant panda habitat

The core area of giant panda habitat-Ya’an

Physical environmental data

Altitude (SRTM 90 M DEM version 4, 30 s,, resolution (−1 km))

Altitude (ASTER GDEM, 30 m resolution)

Biological factor

SLOPE

SLOPE

ASPECT

ASPECT

The distance to water source

The distribution range of giant panda edible bamboo The distance to the water source

Human disturbance factor

Bioclimatic factor

Multi-source remote sensing data

Road (Expressway, national highway and provincial highway)

Road (Expressway, national highway and provincial highway)

Mine

Mine

Hydropower station

Hydropower station

Power transmission line

Power transmission line

Scenic spot

Scenic spot

Basic dataset: ➀ the 19 Bioclimatic factors downloaded from WorldClim website, 30 s,, resolution (−1 km); ➁the 19 Bioclimatic factors acquired by remote sensing data, 0.05° resolution (−6 km)

The 19 Bioclimatic factors downloaded from WorldClim website, 30 s,, resolution (−1 km)

Time (2): at present and in 2050

Time duration (2): at present and in 2050

GCMs (4): BCC-CSM1-1 (BC),CCSM4 (CC),CNRM-CM5 (CN), and HadGEM2-ES (HE)

GCMs: BCC-CSM1-1 (BC)

RCP (2): RCP2.6和RCP4.5

RCP: RCP4.5

NDVI: Monthly NDVI Variation of Sichuan Giant Panda Habitat From January 2001 to December 2015, 30 s,, resolution (−1 km)

Landsat data and thematic classification data

Continuous land cover/use data, spatial resolution 1 km, totally 12 layers

GF-1: to verify the feasibility of the proposal (continued)

7.4 Future Habitat Change Assessment of Giant Pandas Under Climate …

347

Table 7.3 (continued) Data type

Sichuan giant panda habitat

The core area of giant panda habitat-Ya’an

Giant panda habitat survey data

The fourth investigation report of Sichuan Province, giant panda trace points, population number, nature reserves, human disturbance factors, etc.; our project team twice field survey data in giant panda habitat

The fourth investigation report of Sichuan Province, giant panda trace points, population number, nature reserves, human disturbance factors, etc.; our project team twice field survey data in giant panda habitat

7.4 Future Habitat Change Assessment of Giant Pandas Under Climate Change Scenarios Climate change has been a frequent event throughout earth’s history, with at least four major climate change events occurring in the last 420,000 years (Petit et al. 1999). The current rate of climate change could pose serious threats to a wide range of species. Despite the uncertainties of climate change models, measures must also be taken now to prevent the negative effects of the climate change process. The effects of a large number of climate change scenarios on the geographical distribution of plants and animals have been assessed on a wide range of taxa by niche models. The vulnerability of each species depends on a combination of exposure, sensitivity and adaptation. Although climate change has been predicted to be a major threat to biodiversity in the 21st century, accurate predictions and effective solutions have proved very difficult to implement (Dawson et al. 2011). At present, niche models have been used to predict the possible suitable distribution areas of species under the background of climate change and to provide specific suggestions and measures based on the current conservation status. It is important to note that uncertainty is an important consideration for all climate change assessments. Ignoring or minimizing the importance of uncertainty can negatively affect the usefulness of evaluation results used in decision-making and planning (Winkler 2016). Uncertainty is a particular concern for future species distribution assessment in the context of climate change, because sensitivity analysis shows that uncertainty from multiple sources can significantly affect the assessment results. These sources include species the accessibility and quality of information, develop niche model, able to capture the selection of species distribution influence environment prediction variables, the possibility of continuous running model species is converted to binary species occurring predicted results of the threshold, the climate change assessment model parameter setting and adjustment, as well as the choice of the future climate simulation data and so on. Although few assessments explicitly consider all these sources of uncertainty, often because of source constraints, the importance of these uncertainties in interpreting and applying these assessment findings to conservation planning has been well described in the literature.

348

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

Research on climate change predictions may import the uncertainty in the process of multiple links are analyzed, including the comparative study of benchmark climate data sets, environment variables, the preferred method of comparative study, and four kinds of models of the climate, the concentration of the two typical path of comprehensive analysis, a detailed analysis process see literature (Zeng 2018). In this study, the optimal combination scheme after comparative study was used to predict the habitat of the giant panda in Sichuan and Ya’an respectively at the time point of 2050, only the habitat suitability under the background of climate change was considered. That is, through the correlation variable contribution method, CVC method) six variables (bio4, BIO3, BIO14, BIO6, BIO12 and BIO7 with correlation coefficients less than 0.8 in 2050) screened from WorldClim data set were used as bioclimation factors, and 16 single factor layers from remote sensing data, such as DEM, slope and aspect, were involved in model calculation. Each prediction was based on four GCMs (BC model, CC model, CN model, and HE model) and two typical concentration paths (RCP2.6 and RCP4.5), and was based on a 10-run averaging strategy. On this basis, the prediction results of the two research areas are analyzed, as detailed below.

7.4.1 Habitat Change Assessment of Giant Panda in Sichuan According to the above calculation strategy, the suitability of Sichuan giant panda habitat in 2050 corresponds to 80 possible prediction results (4GCMs * 2RCPs * 10), and the corresponding average AUC mean and variance are shown in Table 7.4. Since the ROC curves of each climate model are similar in shape under a specific typical concentration path, this book takes the ROC curves of the 10 calculations of BC model under the scenario of typical concentration path RCP 2.6 as an example to illustrate the calculation accuracy of the model, as shown in Fig. 7.12. According to the comprehensive analysis of Table 7.4 and Fig. 7.12, the average AUC value of the predicted results of 80 models is 0.943 or 0.944, and all of them are greater than 0.9. The variance of AUC is 0.004 or 0.005, indicating that the prediction results are highly converged and concentrated near the mean. These two indicators indicate that the prediction results of the model are good and reliable. Table 7.4 Evaluation table of the accuracy of the prediction results of Sichuan giant panda habitat in 2050

Climate model

Typical concentration path AUC mean and variance (RCP2.6)

AUC mean and variance (RCP4.5)

BC model

0.943 ± 0.005

0.944 ± 0.005

CC model

0.943 ± 0.005

0.944 ± 0.004

CN model

0.944 ± 0.004

0.944 ± 0.005

HE model

0.943 ± 0.004

0.943 ± 0.004

7.4 Future Habitat Change Assessment of Giant Pandas Under Climate …

349

Fig. 7.12 ROC curve of 10 times random sampling operation results under RCP2.6 of BC model

The 10 percentile training presence was used as the threshold value to discretize the results of different models, and the suitable habitat area and corresponding change rate of giant pandas under 8 combination states (4GCMs * 2RCPs) were obtained, as shown in Table 7.5. See Fig. 7.13 for habitat suitability evaluation of giant pandas in Sichuan at the current stage, and see Fig. 7.14 for the prediction results (CN GCMs, RCP 4.5) of the largest habitat suitability reduction of giant pandas in Sichuan by 2050. Since the four models have the same prediction trend, the BC model independently developed in China is still used in the study to predict 2050 under the scenario of RCP 4.5, and its suitability and the current habitat suitability of the giant panda in Sichuan are analyzed by spatial superposition using ArcGIS. The change diagram of habitat suitability of giant pandas in Sichuan from 2050 is obtained, as shown in Fig. 7.15. Table 7.5 Prediction results and change rate of giant panda habitat in Sichuan in 2050 Suitable habitat at present/km2 37,167

Climate model (GCMs)

In 2050 (RCP2.6)

In 2050 (RCP4.5)

Change rate/% (RCP2.6)

Change rate/% (RCP4.5)

BC model

36,775

35,470

−1.05

−4.57

CC model

36,168

34,557

−2.69

−7.02

CN model

35,212

34,230

−5.26

−7.90

HE model

36,854

34,546

−0.84

−7.05

350

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

Fig. 7.13 Habitat suitability evaluation for giant pandas at present (Comprehensive evaluation results of 6 preferred WC climate factors and 16 remote sensing thematic layers)

In the figure, green points are areas with increased fitness, while red points are areas with decreased fitness. Figure 7.13 shows 6 climate factors selected by CVC method and 16 remote sensing thematic data factors. MaxEnt model is used to calculate the initial continuous probability graph of giant panda appearance, and then the logic value of the 10th percentile training point is used as the threshold value to discretize the model results. According to the calculation, the current suitable area of the giant panda habitat in Sichuan is 37167 km2 , while the national fourth giant panda survey report obtained the national giant panda habitat area is 25800 km2 . The model prediction result is obviously larger than the actual survey result, which may be due to the high mountains and dense forests predicted in the suitable area, the ground survey is difficult to reach; On the other hand, although the comprehensive prediction of natural conditions is suitable for these areas, pandas cannot actually use these areas due to the influence of road construction, power transmission lines and other human disturbances. Therefore, although the model evaluation results are different from the field survey in terms of the total amount of suitable habitat, the predicted change

7.4 Future Habitat Change Assessment of Giant Pandas Under Climate …

351

Fig. 7.14 Prediction of habitat suitability for giant pandas in Sichuan in 2050

trend and change range are of guiding significance to the development of wild giant panda protection strategy, reserve planning and corridor construction in the future. As can be seen from Table 7.5, the suitable area of the giant panda habitat in Sichuan in 2050 will be reduced no matter in which climate model or typical concentration path. Under the ideal RCP 2.6 greenhouse gas emission scenario, the HE model predicted the least reduction of suitable habitat, only 0.84% of the current suitable area, while the CN model predicted the largest reduction of suitable habitat, 5.26%. The BC model developed by China predicted 1.05% reduction of suitable habitat for giant pandas in the future. It is second only to HE model and smaller than CC and CN model. For greenhouse gas emissions scenarios RCP4.5, our country developed BC models to predict the future of 2050 giant pandas suitable habitat loss minimum amplitude, is 4.57%, less than 5% of the current suitable habitat, the other three model of the giant panda suitable habitat loss rate is over 7%, which reduce the sharpest CN model, close to 8%, the overall reduction of panda habitat within 10%. In addition, it can be seen from Fig. 7.15 that climate change has different impacts on different mountain systems. In the northwest part of the Minshan Mountain

352

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

Fig. 7.15 Change trend of habitat suitability of giant pandas in Sichuan from 2050

system, there is a relatively dense improvement in fitness, while in the middle and west parts of the Minshan Mountain system, there is a small increase in fitness, which means that the suitable habitat for giant pandas has a trend of spreading westward to northward, which is the same as most previous research results. At the same time, there will be a large area of suitability reduction in the northeastern part of the Daxiangling Mountain system and the Xiaoxiangling Mountain system in the south, especially in the Daxiangling Mountain system, almost the whole mountain system will experience the suitability reduction, and the relevant protection and management departments should pay attention to it. Meanwhile, climate change has little impact on the eastern and central parts of the Qionglai Mountain system, which is also the core area of the giant panda habitat and the location of the Sichuan Giant Panda World Natural Heritage. In conclusion, considering climate change scenarios only, the suitable habitat for giant pandas in Sichuan will decrease by 2050, but the reduction is not significant. Under the typical concentration path of RCP 4.5, the reduction is less than 10%.

7.4 Future Habitat Change Assessment of Giant Pandas Under Climate …

353

7.4.2 Assessment of Changes in Core Areas of Giant Panda Habitat At the regional research scale, the book still adopts the maximum entropy model to conduct fine assessment and prediction of Ya’an Research area, the core area of giant panda habitat under climate change scenarios. The specific steps are as follows: (1) Selection and processing of sample points. MaxEnt model has a low requirement on sample size (>5), requiring only occurrence points of species. This book is based on the fourth National Giant Panda survey report to obtain the giant panda traces in the research area, and input the species name, longitude and latitude in the order of Excel software, stored as a*.csv file. (2) Acquisition and processing of environmental variables. MaxEnt model can deal with both continuous and discrete environmental variables, but continuous variables have better prediction effect. MaxEnt requires strict alignment of singlefactor layers when dealing with environmental factor variables from different sources, that is, to ensure strict consistency of study area boundaries, coordinate system and grid cells, and to convert them into *.ASC format. Research in the maximum entropy model, using 313 points and 24 environment variables involved in model operation, environment variables, including three physical environment variables (altitude, gradient, slope direction), two biological

Fig. 7.16 Habitat suitability evaluation of giant pandas in Ya’an under climate change at present and in 2050

354

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

factor (staple food bamboo distribution, distance from water) and six climatic factors (by correlation coefficient—contribution method from 19 biological climate factor selected). When processing data from different sources, the maximum entropy model requires that all raster data must maintain the same resolution and be strictly aligned geographically. In this study, all environmental variables are unified as 1 km × 1 km spatial resolution, and the coordinate system is unified as UTM WGS84 coordinate system. Among the samples participating in model operation, 75% are randomly selected for model training and 25% of the data are used for accuracy verification. The 10 randomly generated sample sets are used for operation and the average value is taken as the model operation result. The format conversion and reclassification of the output continuous prediction results were carried out with the support of GIS software, and the evaluation chart of giant panda habitat suitability in Ya’an region under the background of climate change at present and in 2050 was obtained (Fig. 7.16). (3) Result verification. AUC (area under ROC curve) is used to evaluate the performance of MaxEnt model. AUC value is one of the most commonly used methods to evaluate niche model results. The larger the AUC value is, the farther it is from the random distribution, indicating the better the simulation effect of the model. In this study, AUC value was used to evaluate the accuracy of model prediction. It is generally believed that the prediction results are reliable when AUC value is greater than 0.7. The average AUC value of the maximum entropy model is 0.74 for training samples and 0.72 for test samples. The model running results can be used as the basis for prediction. After processing and analyzing the prediction, the research results show that. Only under the background of climate change, the area of potential suitable habitat in the study area increased as a whole. The results differ from most previous predictions of habitat loss. The areas with more concentrated increase in suitability are located in the northwest, northeast and southeast directions of the study area, and the changes are shown in Table 7.6. The possible reasons for this trend can be explained by the expansion of broadleaved forest, coniferous and broad-leaved mixed forest and coniferous forest due to climate change. The meteorological observation data for 65 years from 1951 to 2016 show that the annual average temperature in Ya’an presents a continuous upward trend with a change rate of 0.114 °C/10a. The trend of temperature warming is Table 7.6 Change in the suitability of giant panda habitats in the study area from the present to 2050

Habitat type

Suitable

Moderately suitable

Low suitability

Present/km2

3038.00

3922.00

1515.00

2050/km2

4057.00

3241.00

1177.00

Change/km2

1019.00

−681.00

−338.00

33.54

−17.36

−22.31

Change rate/%

7.4 Future Habitat Change Assessment of Giant Pandas Under Climate …

355

Fig. 7.17 Variation of air temperature and precipitation in Ya’an from 1951 to 2016

accelerated especially since the 1990s. The annual precipitation decreased slowly, and the precipitation change rate was −33.22 mm/10a, with an overall trend of “warm and dry”. This conclusion is consistent with existing research results (Wang et al. 2010), as shown in Fig. 7.17. At the same time, it has been pointed out that forest ecosystems will benefit from climate change. The natural ecosystems in southwest China are particularly affected by climate change, with the migration of ecological vegetation belts, the elevation of forest lines, and the invasion of shrub species into alpine meadows. In the vegetation type group of giant panda habitat in Sichuan Province, the area of broadleaf forest is the largest, which is 748600 hm2 , followed by coniferous forest, which is 703300 hm2 . The area of two vegetation type groups accounts for 71.88% of the total vegetation area of giant panda habitat in Sichuan Province. Our research group has shown that in the past 30 years, the mountain vertical belt of Wolong and Wanglang panda habitat has increased (Chang et al. 2015; Liao 2016). Therefore, with the warming of climate, the forest line will rise, resulting in the change of broad-leaved forest, coniferous mixed forest, coniferous forest and even the expansion of some vegetation zones, and the corresponding change and even expansion of the habitat range of giant panda, which may be one of the reasons for the overall increase of potential suitable habitat area in the research area.

7.4.3 Comparison of Habitat Change Trends at Different Scales Through comparative analysis of the two scale research areas of the whole Sichuan giant panda habitat and the core area of the habitat, we found that: (1) the impact of climate change on the Sichuan panda habitat under different scales do not affect the outcome and degree in the same way, in terms of the whole Sichuan panda habitat, the next 2050 years under the background of climate change is only considered overall will reduce its suitable habitat area, according

356

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

to the four models of the climate and the reducing trend of two typical concentration under the path of comprehensive evaluation, At the large scale, habitat area is reduced within 10% under the GHG emission scenario with the typical concentration path of RCP 4.5. (2) At the large scale level of the giant panda habitat in Sichuan, the impact of climate change on different mountain systems is not the same. Figure 7.15 shows the projected changes in habitat suitability to 2050. It can be seen from the figure that in the northwest part of Minshan Mountain system, there is a relatively dense improvement in fitness, that is, the suitable habitat for giant pandas has a trend of spreading westward to northward, which is the same as most previous research results. At the same time, there will be a large area of suitability reduction in the northeastern part of the Daxiangling Mountain system and the Xiaoxiangling Mountain system in the south, especially in the Daxiangling Mountain system, almost the whole mountain system will experience the suitability reduction, and the relevant protection and management departments should pay attention to it. (3) In the southern Liangshan Mountain system, a large scale increase in suitability is also predicted, which is rarely mentioned in the present study. According to the fourth giant panda survey report of Sichuan Province, compared with the third giant panda survey, the area of giant panda habitat in Liangshan Mountain system increased by 37.18%, ranking the third among the six mountain systems. At the same time, through the analysis of the spatial genetic structure of the population, it was found that the Liangshan Mountain system was an independent genetic cluster, which could be divided into five local populations: Lewu, Fengfengding, Lami, Jinpingshan and Wuzhishan. Le wu LAN, population is the biggest local population is composed of 92 wild pandas, the smallest is the mi, and local populations in Wuzhishan, respectively is composed of 3 wild pandas, Liangshan Mountains, a total of 124 wild pandas live, food is a Sichuan panda habitat in the Minshan Mountain range and the largest number of mountains in Qionglai Mountains besides wild pandas, Therefore, it is suggested that the relevant protection and management departments should strengthen the protection of the Liangshan Mountain system. (4) The suitability of Ya’an research area, the core area of giant panda habitat at a small scale, will increase a little under the background of climate change only, which is different from the results of the decrease of giant panda habitat predicted by most previous studies, but the overall increase is not large, around 10%. Detailed systematic analysis can be found in the research group’s published article (Jing et al. 2018). To sum up, when applying different models to assess the impact of climate change, the choice of scale in the study area is very important. Different scales may lead to different conclusions, which need special attentions.

7.5 Habitat Assessment of Giant Pandas Under the Combined Impact …

357

7.5 Habitat Assessment of Giant Pandas Under the Combined Impact of Human Disturbance and Climate Change 7.5.1 Assessment of Giant Panda Habitat Change in Sichuan Giant panda habitat by natural and man-made two aspects of interference, including natural interference earthquake, bamboo flowering, debris flow, landslide, fire, etc. Human interference includes logging, grazing, roads, tourist attractions, mines and so on. With the development of social economy, human disturbance has become the main disturbance of giant panda habitat. Human interference directly destroys the habitat vegetation (such as logging, medicine collection), affects the normal life of giant pandas (such as tourism, farming), and causes the shrinking and fragmentation of the giant panda habitat (such as roads, power transmission lines). In the fourth giant panda survey in Sichuan Province, a total of 17 types of disturbance factors of human activities were recorded in the habitat. According to the impact range of disturbance factors, the study divided them into general disturbance of production and operation and large disturbance such as facility construction. Generally speaking, large disturbance such as facility construction has a greater impact on the habitat, as shown in Sect. 7.2.6. For further detailed assessment of climate change and human disturbance under the influence of the giant panda habitat conditions, selected research infrastructure projects such as large interference of high rate of 5 class met in artificial interference factors to analysis, including traffic road (state roads, provincial roads, and highways), hydropower stations, quarries, tourist attractions, and transmission lines. The analysis function of this part is obtained by superposition analysis of human disturbance factors and habitat suitability of giant pandas in 2050 with the support of ArcGIS 10.2 software. First of all, high anthropogenic interference factor for 5 classes meet rate buffer analysis, the setting of buffer reference fourth report records of wild giant pandas in Sichuan Province trace point number and density with the change of distance from interference factors, research of highway buffer radius set to 3 km, the rest of the interference factor of the buffer radius are 2 km. Secondly, the buffer zone of human disturbance factors and the giant panda habitat suitability map predicted by 2050 were superimposed. As the suitable habitat of the giant panda in Sichuan as a whole was reduced, the influence of adding human disturbance on the suitability change area was analyzed. For the small scale Ya’an research area, because the predicted habitat range is slightly increased, so the study conducted a superposition analysis of suitability and human disturbance in the small scale, and obtained the suitable habitat for giant pandas after removing the human disturbance area.

358

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

Finally, the habitat conditions of giant pandas before and after the increase of human disturbance were compared and analyzed, and the corresponding opinions and suggestions were given based on the comparative analysis results. According to the above steps, the habitat suitability distribution diagram and change diagram of Sichuan giant panda in 2050 under the premise that the current five types of human disturbance factors remain unchanged are obtained, as shown in Fig. 7.18. Although only considering the impact of climate change, the suitable habitat area for giant pandas in Sichuan has not decreased significantly, with the total amount being about 10%, the comprehensive analysis of Fig. 7.18 shows that the potential suitable habitat for giant pandas in Sichuan will be greatly reduced and the fragmentation will be more serious after the addition of human disturbance. Especially, the suitability degree of the northern Minshan Mountain system and southeastern Qionglai Mountain system, where wild giant pandas are distributed in large numbers,

Fig. 7.18 Changes in habitat suitability under the combined influence of climate change and human disturbance in 2050

7.5 Habitat Assessment of Giant Pandas Under the Combined Impact …

359

will be significantly reduced. In addition, although it is predicted that the suitability will increase slightly in Liangshan Mountain system, the hydropower station and transmission network will affect the areas where the suitability may increase after the addition of human disturbance. Therefore, it is suggested that the infrastructure planning and construction in this region can comprehensively consider the possible future improvement of suitability.

7.5.2 Assessment of Giant Panda Habitat Change in Ya’an Research Area After considering the artificial disturbance factors, the area of potential suitable habitat in Ya’an study area will change from the overall increase trend to a significant decrease (Fig. 7.19), and the suitable habitat will decrease by 58.56% rapidly. The area of sub-suitable habitat is still decreasing, and the decrease will reach 62.29% (Table 7.7). At the same time, due to the segmentation of roads and transmission lines, habitat fragmentation is more serious.

Fig. 7.19 a Overlay analysis of human disturbance factors and b buffer analysis of human disturbance factors

360

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

Table 7.7 Change in the suitability of giant panda habitats in the study area considering human disturbance factors in 2050

Habitat type

Suitable

Moderately suitable

Present/km2

3038.00

3922.00

2050/km2

4057.00

3241.00

Buffer analysis of human disturbance/km2

1259.00

1479.00

−1779.00

−2443.00

−58.56

−62.29

Change/km2 Change rate/%

7.5.3 Comparative Study of Habitat Change Trends at Different Scales To sum up, if the comprehensive impact of climate change and human disturbance is considered comprehensively, the suitability of giant panda habitat will be significantly reduced at both large and regional scales, and the fragmentation will be more serious. At present, the degradation and fragmentation of giant panda habitat are two important factors affecting the number of giant panda population and habitat suitability. Existing protection system is based on more current habitat condition, therefore after research under different scales of the giant panda habitat conditions to carry out the forecast and analysis, the analysis of suitability should focus on increase and decrease area, and suggests relevant management department when making protection is the most suitable for the future may change.

References Anderson R, Martinez-Meyer PE (2004) Modeling species’ geographic distributions for preliminary conservation assessments: an implementation with the spiny pocket mice (Heteromys) of Ecuador. Biol Cons 116(2):167–179 Brambilla MG, Ficetola F (2012) Species distribution models as a tool to estimate reproductive parameters: a case study with a passerine bird species. J Anim Ecol 81(4):781–787 Chang C, Wang X, Yang R et al (2015) Quantitative delineation of alpine vegetation zones based on DEM-NDVI. Geogr Res (11):2113–2123 (in Chinese) Chen I, Hill JK, Ohlemüller R et al (2011) Rapid range shifts of species associated with high levels of climate warming. Science 333(6045):1024 Dawson TP, Jackson ST, House JI et al (2011) Beyond predictions: biodiversity conservation in a changing climate. Science 332(6025):53–58 Della Pietra S, Della Pietra V, Lafferty J (1997) Inducing features of random fields. J IEEE Transactions Pattern Analysis and Machine Intelligence, 19(4): 380–393 Ekins P, Speck S (2014) The fiscal implications of climate change and policy responses. Mitig Adapt Strat Glob Change 19(3):355–374 Elith JS, Phillip SJ, Hastie M et al (2011) A statistical explanation of MaxEnt for ecologists. Divers Distrib 17(1):43–57 Fan J, Li J, Xia R et al (2014) Assessing the impact of climate change on the habitat distribution of the giant panda in the Qinling Mountains of China. Ecol Model 274:12–20

References

361

Gong M, Guan T, Hou M et al (2017) Hopes and challenges for giant panda conservation under climate change in the Qinling Mountains of China. Ecol Evol 7(2):596–605 Grinnell J (1917) The niche-relationships of the California Thrasher. Auk 34(4):427–433 Harald P, Michael G, Karl R et al (2007) Signals of range expansions and contractions of vascular plants in the high Alps: observations (1994–2004) at the GLORIA* master site Schrankogel, Tyrol Austria. Glob Change Biol 13(1):147–156 Hijmans RJ, Cameron SE, Parra JL et al (2005) Very high resolution interpolated climate surfaces for global land areas. Int J Climatol 25(15):1965–1978 Howard C, Stephens PA, Pearce-Higgins JW et al (2014) Improving species distribution models: the value of data on abundance. Methods Ecol Evol 5(6):506–513 Hu JC (1986a) Biology of the giant panda. Science (3):181–191, 239 (in Chinese) Hu JC (1986b) The giant pandas of Qionglai Mountain. J Nanchong Teach College (Nat Sci Edn) 1:21–28 (in Chinese) Hu J, Deng Q, Yu Z et al (1980) Ecological biology of giant pandas, golden monkeys and other rare animals. J Nanchong Teach College (Nat Sci Edn) 2:3–41, 127–134 (in Chinese) Hu Jinchu GB, Schaller GK, Johnson G (1990) Foraging ecology of giant pandas in Tangjiahe Nature Reserve. J West China Norm Univ (Nat Sci Edn) 1:182–194 (in Chinese) Hu J, Zhang Z, Wei F (2011) History, status and future of the development of giant panda reserves in China. J Vet Sci (1):10–14 (in Chinese) Hull V, Zhang JD, Huang JY et al (2016) Habitat use and selection by giant pandas. PLoS ONE 11(9):e0162266 Intergovernmental Panel on Climate Change (IPCC) (2014) Climate Change 2014: synthesis report. IPCC, Geneva, Switzerland Jian J, Jiang H, Jiang Z et al (2014) Predicting giant panda habitat with climate data and calculated habitat suitability index (HSI) map. Meteorol Appl 21(2):210–217 Jing Z, Wang X, Meng Q et al (2018) Fine-Scale Evaluation of Giant Panda Habitats and Countermeasures against the Future Impacts of Climate Change and Human Disturbance (2015–2050): A Case Study in Ya’an, China[J]. Sustainability, 10(4):1081 Kang DW, Li JQ (2016) Connect the fragmented habitat patches for giant panda. Environ Sci Pollut Res 23(12):11507–11508 Legault A, Theuerkauf J, Chartendrault V et al (2013) Using ecological niche models to infer the distribution and population size of parakeets in New Caledonia. Biol Cons 167(3):149–160 Li J, Liu F, Xue YD et al (2017) Assessing vulnerability of giant pandas to climate change in the Qinling Mountains of China. Ecol Evol 7(11):4003–4015 Li R, Xu M, Wong MHG et al (2015) Climate change threatens giant panda protection in the 21st century. Biol Cons 182:93–101 Liao Y (2016) Fine observation and quantitative delineation of vertical zones of subalpine vegetation in Wanglang Nature Reserve. Master’s thesis. University of Chinese Academy of Sciences (Institute of Remote Sensing and Digital Earth Research, Chinese Academy of Sciences), Beijing (in Chinese) Liu G, Guan T, Dai Q et al (2016) Impacts of temperature on giant panda habitat in the north Minshan Mountains. Ecol Evolut 6(4):987–996 Oliver TH, Gillings S, Girardello M et al (2012) Population density but not stability can be predicted from species distribution models. J Appl Ecol 49(3):581–590 Ouyang ZY, Liu JG, Xiao H et al (2001) Habitat evaluation of giant pandas in Wolong Nature Reserve. J Ecol 21(11):1869–1874 (in Chinese) Parmesan C, Yohe G (2003) A globally coherent fingerprint of climate change impacts across natural systems. Nature 421(6918):37–42 Peng S, Li Q, Ren H (2002) Impacts of global climate change on wildlife. J Ecol 22(7):1153–1159 (in Chinese) Petit JR, Jouzel J, Raynaud D et al (1999) Climate and atmospheric history of the past 420000 years from the Vostok ice core Antarctica. Nature 399(6735):429–436 Phillips SJ, Anderson RP, Schapire RE (2006) Maximum entropy modeling of species geographic distributions. Ecol Model 190(3/4):231–259

362

7 Detailed Evaluation of Giant Panda Habitats and Countermeasures …

Phillips SJ, Schapire RE (2004) A maximum entropy approach to species distribution modeling[C]. International Conference on Machine Learning :83 Phillips S, Dudík JM (2008) Modeling of species distributions with Maxent: new extensions and a comprehensive evaluation. Ecography 31(2):161–175 Pounds JA, Mpl JF, Campbell H (1999) Biological response to climate change on a tropical mountain. Nature 398(398):611–615 Qi D, Hu Y, Gu X et al (2012) Quantifying landscape linkages among giant panda subpopulations in regional scale conservation. Integr Zool 7(2):165–174 Root TL, Price JT, Hall KR et al (2003) Fingerprints of global warming on wild animals and plants. Nature 421(6918):57–60 Shen GZ, Pimm SL, Feng CY et al (2015) Climate change challenges the current conservation strategy for the giant panda. Biol Cons 190:43–50 Sichuan Forestry Department (2015) Giant pandas in Sichuan: report of the fourth survey of giant pandas in Sichuan province. Sichuan Science and Technology Press, Chengdu (in Chinese) Songer M, Delion M, Biggs A et al (2012) Modeling impacts of climate change on giant panda habitat. Int J Ecol 2012:108752 State Forestry Administration (2015) The State Forestry Administration held a press conference on the results of the fourth national giant panda survey. http://www.scio.gov.cn/xwfbh/gbwxwfbh/ fbh/Document/1395514/1395514.Htm [2015-02-28] (in Chinese) Swaisgood R, Wang DX, Weif F (2016) Ailuropoda melanoleuca. (errata version published in 2016) the IUCN red list of threatened species 2016: e.T712A102080907 Swets JA (1988) Measuring the accuracy of diagnostic systems[J]. Science 240(4857):1285-1293 Tanner EP, Papes M, Elmore RD et al (2017) Incorporating abundance information and guiding variable selection for climate-based ensemble forecasting of species’ distributional shifts. PLoS ONE 12(9):e0184316 Thomas CD, Cameron A, Green RE et al (2004) Extinction risk from climate change. Nature 427(6970):145–148 Tuanmu M, Jetz NW (2014) A global 1 km consensus land-cover product for biodiversity and ecosystem modelling. Glob Ecol Biogeogr 23(9):1031–1045 Walther GR, Post E, Convey P et al (2002) Ecological responses to recent climate change. Nature 416(6879):389–395 Wang YS, Xie BY, Wan FH et al (2007) Application of ROC curve analysis in evaluating the performance of alien species’ potential distribution models. J Biodiversity Science 15(4):365-372 (in Chinese) Wang R, Fan X, Liu Q et al (2010) Impacts of climate change on giant panda habitat in Sichuan. Highl Mt Meteorol Res 30(4):57–60 (in Chinese) Wei F, Feng J, Wang Z (1999) Habitat selection by giant pandas and red pandas in the Xiangling mountain system. J Zool (1):57–63 (in Chinese) Winkler JA (2016) Embracing complexity and uncertainty. Ann Am Assoc Geogr 106(6):1418–1433 Wu J, Lv J, Ai L (2009) Impacts of climate change on biodiversity: vulnerability and adaptation. J Ecol Environ 18(2):693–703 (in Chinese) Xu ZL, Peng HH, Peng SZ et al (2015) The development and evaluation of species distribution models. Acta Ecol Sin 35(2):557–567 Zeng J (2018) Fine-scale assessment and response to future climate change impacts on giant panda habitat. Ph.D. thesis. University of Chinese Academy of Sciences (Institute of Remote Sensing and Digital Earth Research, Chinese Academy of Sciences), Beijing (in Chinese) Zhang ZJ, Hu JC (2000) A study on the Giant Panda’s habitat selection. J Journal of West China Normal University (Natural Science Edition) 1:18–21 (in Chinese) Zhen J, Wang X, Meng Q et al (2018) Fine-Scale Evaluation of Giant Panda Habitats and Countermeasures against the Future Impacts of Climate Change and Human Disturbance (2015–2050): A Case Study in Ya’an, China[J]. Sustainability 10(4):1081 Zhou J (2008) The Study on the Quality Evaluation System of Giant Panda Habitat[D]. Ph.D Dissertation of Beijing Forestry University (In Chinese) Zhu LF, Hu YB, Zhang ZJ et al (2013) Effect of China’s rapid development on its iconic giant panda. Chin Sci Bull 58(18):2134–2139

Chapter 8

Suggestions on Sustainable Development of Giant Panda Habitat

8.1 Conclusion of the Impact of Climate Change on Giant Panda Habitat Climate change has posed a severe challenge to global biodiversity. With the rapid warming of the earth and the change of precipitation pattern, many species have to adjust their distribution range, change their ecological characteristics, reproductive law and other ways to adapt to climate change. As a flagship species of global biodiversity conservation, giant pandas have attracted a large number of domestic and foreign scholars to carry out research on the impact of climate change on their habitat. Based on remote sensing, geographic information system and other spatial technologies, this book uses the maximum entropy model, which is widely used at present, to evaluate and forecast the giant panda habitat at both macro and regional scales. The research results of this book are of great significance for the effective protection of the current and future panda habitat, as well as the coordinated development of the ecological protection of the habitat and the local economy. The results show that: (1) The impact of climate change on giant panda habitat is different at different research scales. In terms of the giant panda habitat in Sichuan, no matter which global climate model and typical concentration path are adopted, the suitable habitat area of giant panda will decrease by 2050. Under the RCP4.5 emission scenario, the reduction of suitable habitat for giant pandas is less than 10%. (2) The impact of climate change on different mountain systems is also different. The Liangshan Mountain system in the south and Minshan Mountain system in the north will have a small increase in the northeast, while the Daxiangling Mountain system and Xiaoxiangling Mountain system in the south of the center will have a large decrease in the northeast. The Daxiangling Mountain system, in particular, is experiencing a loss of suitable habitat for almost the entire region, and conservation efforts in this area should be focused on. © Science Press 2023 X. Wang et al., Spatial Observation of Giant Panda Habitat, https://doi.org/10.1007/978-981-19-8794-6_8

363

364

8 Suggestions on Sustainable Development of Giant Panda Habitat

(3) The habitat change trend of Ya’an study area in the middle Qionglai Mountain system is different from that of the whole Sichuan giant panda habitat in the context of climate change. Specifically, under the background of climate change only, the area of suitable habitat in this region will increase by 2050, which is different from most previous predictions that the area will decrease due to temperature rise. The newly increased suitable habitat for giant pandas showed an expansion trend to the west, north and east, Baoxing County in the northwest, Tianquan County in the west, Lushan County in the northeast, Yingjing County in the south have increased, this conclusion is consistent with the change trend of China’s fourth panda habitat actual survey data; although the suitable habitat shows an increasing trend under the background of climate change, the area of suitable habitat will decrease significantly (about 58.56%) and the fragmentation will be more obvious after considering the human disturbance factors such as roads, power transmission lines, mining and hydropower stations. The three protected areas in the current study area are currently far apart from each other and do not cover potentially suitable habitats that will emerge from future climate change. Necessary adjustments should be made. This book further analyzes and predicts the habitat suitability of giant pandas due to climate change after increasing human disturbance. Although the suitable habitat of the giant panda habitat in Sichuan is not significantly reduced under the influence of climate change only, and the total amount is less than 10% under the typical concentration path of RCP 4.5, the potential suitable habitat of the giant panda habitat in Sichuan will be significantly reduced and the fragmentation will be more serious after the addition of human disturbance. Especially, the suitability degree of the northern Minshan Mountain system and southeastern Qionglai Mountain system, where wild giant pandas are distributed in large numbers, will be significantly reduced. In addition, although it is predicted that the suitability will increase a little in Liangshan Mountain system, the hydropower station and transmission network will affect the areas where the suitability may increase with the addition of human disturbance. After considering artificial disturbance factors, the area of potential suitable habitat in Ya’an study area will change from the overall increase trend to a significant decrease, and the suitable habitat will decrease by 58.56%. The area of sub-suitable habitat will be reduced by 62.29%. At the same time, due to the segmentation of roads and transmission lines, habitat fragmentation is more serious.

8.2 Proposed Protection Measures for the Sustainable Development of Giant Panda Habitat This book uses the niche principle, using maximum entropy model in the whole Sichuan giant panda habitat and the core area of giant panda habitat—Ya’an research area two levels of climate change, as well as climate change and anthropogenic

8.2 Proposed Protection Measures for the Sustainable Development …

365

influence under the two conditions of suitability for accurate analysis. Based on the findings, the following recommendations are made: Recommendation 1: for the Minshan Mountains which have the largest number of wild giant pandas, there are potentially suitable habitat in northwest, so it can strengthen the management work of protection of the region, in addition, in the north of Minshan Mountains, hydroelectric power station and the traveling scenic area is relatively dense, in the development of the economy at the same time pay attention to the protection of giant pandas, moderately reduce the effects of human disturbance on habitat. Recommendation 2: It is predicted that there will be a small increase in suitability in the Liangshan Mountain system, but with the addition of human disturbance, hydropower stations and power transmission networks will affect the areas where suitability may increase. Therefore, it is suggested that the potential suitability areas of giant panda habitat should be considered comprehensively in infrastructure planning and construction in this region. As for Ya’an Research Area, the core area of giant panda habitat, the three nature reserves in the research area are scattered at present, which is not conducive to the communication between individual giant pandas in different areas of the same mountain system, as well as the stability and growth of giant panda population (Hu et al. 2011), so necessary adjustments need to be made. Recommendation 3: Based on the Fengtongzhai National Nature Reserve, we can consider extending the reserve to the east. Under the background of future climate change, there are large areas of suitable habitat patches in the northwest and east of Fengtongzhai Nature Reserve. In addition, according to the fourth giant panda survey report, only in this protected area, the habitat area of giant pandas in the fourth survey compared with the third survey increased from 30,936 to 32,899 hm2 , with a change rate of 6.35%. It is recommended to expand the Fengtongzhai Protection Area to the east based on the Fengtongzhai Protection Area, as shown in solid blue line in Fig. 8.1, with an area of 12,784.2 hm2 . Recommendation 4: Extend the new north-south Link reserve to the west of the Trumpet River Provincial Reserve to cover suitable habitat that may be added later. Model prediction shows that there will be more intensive new suitable habitat for giant pandas in the west and northwest of the Trumpet River Reserve in the future. At the same time, according to the fourth giant panda survey report, the number of wild giant pandas in the Trumpet River Reserve has increased significantly, from 11 to 20, with a change rate of 81.82%. It is suggested that the protected area should be focused on and the new protected area of the north-south link should be expanded in the west to cover suitable habitats that may be added in the future, as shown in solid blue line B in Fig. 8.1, with an area of 54,826.8 hm2 . Recommendation 5: Off site conservations can be considered in Daxiangling Provincial Reserve in the future. According to the prediction of the model, the Daxiangling Provincial Reserve located in Yingjing in the south will have a large area of habitat degradation, which is consistent with the research results of Zhao et al. (2016). Before the adjustment of the boundary with Qionglai Mountain system, Daxiangling Mountain system had the smallest population and habitat area of giant pandas. There

366

8 Suggestions on Sustainable Development of Giant Panda Habitat

Fig. 8.1 Status of nature reserves in the study area and distribution map of proposed new nature reserves. Zone A: the proposed protected area based on the extension of Fengtongzhai National Nature Reserve; Zone B: the proposed new protected area based on the Labahe Provincial Nature Reserve

were only 16 giant pandas in the mountain system at the time of the third survey. The fourth survey showed a total of seven giant pandas in the Daxiangling Reserve. Although Yingjing plans to build an important Niba Mountain giant panda corridor, which covers the Daxiangling nature Reserve, the corridor is heavily divided by the 108 national highway (from Chengdu to Kunming). At the same time, the distribution of mines and hydropower stations in this corridor is relatively dense. In conclusion, the habitat degradation and fragmentation of giant pandas in this region will be very serious in the future, and gene exchange between populations will be very difficult.

8.2 Proposed Protection Measures for the Sustainable Development …

367

The survival prospects of giant pandas distributed in this region are worrying. In view of this situation, if habitat improvement measures are difficult to implement, off site conservation measures shoule be considered. Recommendation 6: There is still a risk of food shortages in 2050 due to widespread flowering and dying of bamboo, the staple food for pandas. Giant pandas are exclusively carnivorous and 99% of their food comes from bamboo. In Sichuan Province, there are 32 species of 7 genera of giant pandas staple food bamboo, with a distribution area of 1.9255 million hm2 , accounting for 94.88% of the total habitat area of giant pandas in the province. Among them, the largest distribution area was Bashania fangiana, followed by Fargesia denudata, Yushania brevipaniculata and Chimonobambusa szechuanensis. According to Sichuan Province staple bamboo flowering history records show that in the mid-1970s to the mid-1980s, the research area within the Tianquan County, Baoxing County, Lushan County, Yingjing County four counties have occurred staple bamboo flowering and dying phenomenon, and according to historical documents and local interviews, the flowering cycle of bamboo has three records: 50, 60, and 90–100 years. If we calculate it by 60 years, a new round of bamboo flowering and dying will come around the 2040s. Meanwhile, studies have shown (Parmesan and Yohe 2003) that climate change may lead to the advance of plant phenology, so the rejuvenation of bamboo and the abundance of bamboo species in this region need to be considered in the future. The phenomenon of large areas of bamboo flowering and dying has a long cycle (60–120 years), but it has a great impact on giant pandas. In view of this situation, some studies suggest that: artificial planting of bamboo, increasing the species diversity of bamboo, and protecting some older forests (Li et al. 2015). For Ya’an region, the core research area of giant panda habitat, we superimposed the protected areas of A and B to the GF-1 image of the region in March 2017 (Fig. 8.2) and DEM of the study area (Fig. 8.3). The vegetation type of zone A is coniferous forest or mixed coniferous and broadleaved forest. The altitude ranges from 965 to 2229 m, with an average altitude of 1561 m. In zone B, the existing Trumpet River Reserve will be extended to the north by a complete branch, and most of the area is woodland and grassland, with an altitude of 1500–3400 m. According to the analysis of the habitat utilization of the existing giant pandas, the proposed areas A and B are feasible to build new habitats for giant pandas. In addition, according to research, the minimum nesting range of a giant panda is 400 hm2 (Shen et al. 2008; Hu 1990), and the fourth survey showed that there were seven giant pandas in daxiangling Nature Reserve. According to our calculation, the protected areas A and B will be added in the north with an area of 67611 hm2 , and the new protected areas of Fengtongzhai and Labahe will be moved to the north with an expanded capacity to carry the habitat range required by the current giant pandas in the Daxiangling Mountain system, which is also conducive to enhancing the communication between different populations (Songer et al. 2012).

368

8 Suggestions on Sustainable Development of Giant Panda Habitat

Fig. 8.2 Superposition analysis of proposed protected area expansion and GF-1 image in March 2017

8.3 Prospect of Space Observation of Giant Panda Habitat According to the data of the four giant panda surveys, although the number and habitat area of giant pandas in Sichuan Province have continued to recover, habitat fragmentation and population segmentation have further intensified, so the overall situation is not optimistic, and the main threats are reflected in the following four points.

8.3 Prospect of Space Observation of Giant Panda Habitat

369

Fig. 8.3 DEM data overlay analysis of the proposed extension of protected area and the study area

1) Challenges from the biology of the species itself The endangerment of any species is the result of the interaction between its own biological characteristics and the surrounding environment. The giant panda has a single diet, and more than 99% of its food comes from bamboo. The bamboo has the phenomenon of periodic flowering and dying, and the rejuvenation of the bamboo after flowering takes 5–10 years. Therefore, if there is a single bamboo species in the giant panda habitat, once a large area of flowering events may lead to a shortage of food for the giant panda, resulting in physical decline and even death of the giant panda. In the mid-1970s, a bamboo bloom in the Minshan region killed nearly 150

370

8 Suggestions on Sustainable Development of Giant Panda Habitat

pandas. Combined with the low reproduction rate of wild pandas, this has largely hindered the recovery of the population. 2) Habitat fragmentation Giant panda habitat fragmentation situation is severe, the main reason is the construction and crossing of national road, provincial road and highway, as well as the existence of mountains with snow all year round, wide rivers with rapid water flow, scattered rural settlements and deep river valleys with cliffs. The analysis of giant panda population viability shows that the extinction risk of local small species will increase with the passage of time. Even if its habitat is well protected and human hunting and other factors are minimized, it is difficult to guarantee its long-term survival. Therefore, how to prevent the small population of giant panda from extinction in the fragmented habitat in the future is a difficult challenge for giant panda conservation. 3) Giant panda protection conflicts with the needs of regional economic development The results of the fourth national giant panda survey show that the top four categories of general human disturbance to the giant panda habitat in Sichuan Province are grazing, transportation, other collection and medicine collection. In the long run, it is necessary to protect pandas and their habitats by providing incentives to nearby residents or providing educational opportunities so that young people living in panda range can find jobs and settle down in cities. Alternatives that are economically affordable, socially acceptable, ecologically sound and sustainable in the long term need to be accelerated. In addition, with the rise of ecological tourism in western China, the development of tourism resources has led to the disappearance of giant panda habitats. Further efforts are needed to coordinate the development of ecological tourism and habitat protection. 4) Geological disasters, bamboo flowering and other natural disasters Most of the giant panda habitat is located within the impact of the Longmenshan fault zone. The Wenchuan earthquake in 2008 and the Ya’an earthquake in 2013 both caused some damage to the giant panda habitat. The earthquake will not only change the living environment of giant pandas, but also induce the flowering of bamboo, a staple food, which seriously threatens the health and food safety of giant pandas. Natural disasters, such as earthquake, fire, bamboo flowering of the giant panda and its habitat influence also to be reckoned with, in addition to using satellite remote sensing etc. New technology advantage to carry out a wide range of monitoring, to strengthen the field inspection and other measures, in these natural disasters affect suitability evaluation should also be taken into consideration, also is the important content of the follow-up study. According to scholars’ prediction of panda’s suitable habitat, habitat will shift, that is, some currently suitable habitat will be sub-suitable or even unsuitable habitat in the future, and vice versa. However, giant pandas may not be able to adapt well to habitat changes due to their slow movement and weak reproductive ability. In

8.3 Prospect of Space Observation of Giant Panda Habitat

371

addition, some of the transferred habitats are far from the existing habitats and the habitat fragmentation is serious. It is suggested that some captive giant pandas be relocated artificially and corridors should be added. The establishment of connected communication corridors between protected areas can also increase the flow between isolated populations, which is an effective measure to increase genetic diversity and help mitigate the negative effects of habitat fragmentation, since the contribution of small populations to genetic diversity cannot be ignored. In addition to the problems faced by natural reasons, the impact of human factors on giant pandas is also great. At present, there are protection red line management, to reduce human factors to the giant panda habitat destruction and disturbance. However, in addition to directly banning human activities in some areas, the government should also pay attention to the living standards of people around the giant panda habitat to indirectly reduce the damage caused by human activities to the local ecological environment: In some areas with low living standards and low employment rates, people often take illegal hunting and picking as income sources, which will undoubtedly cause disturbance to pandas. It is suggested that the local government increase employment opportunities. At present, there are some confusion in the management system of nature reserves, scenic spots and other places in the distribution area of giant pandas. Moreover, due to the topography of the distribution area of giant pandas, the distribution of protection, monitoring and data centers, such as meteorological centers, is still sparse, resulting in the acquisition of high-quality data and the delay in taking measures. Therefore, it is suggested that meteorological observation stations should be set up in protected areas or near habitats in order to better understand the climate change situation, formulate protection plans and take protection measures in time. In short, this book based on remote sensing and geographic information system technology advantage, using ecological niche model in large scale and key areas of the giant panda habitat suitability under the background of climate change and possible trends are discussed in this paper, accumulated some preliminary research results, but with climate data the spatial resolution and limitations on the inversion temperature, precipitation gradient change, as well as the limitation of environmental factors involved in the model calculation (for example, only considering the impact of planar bamboo distribution on the habitat suitability of giant pandas, the change trend of specific bamboo species under the background of future climate change has not been further analyzed), the main problems to be solved in the follow-up research methods include the following three aspects. (1) Rely on remote sensing technology to obtain high-resolution bioclimatic data to improve the prediction accuracy of niche model. For climate data application, usually is the study of large scale under the condition of macroscopic overall trend, and meteorological data of the measurement site also uneven, may be more densely populated, developed economic regions and remote economic less-developed areas are often doesn’t even have a meteorological observation sites, and different parts of the world have observed data of different starting period, The observation content and data format are also different, which leads

372

8 Suggestions on Sustainable Development of Giant Panda Habitat

to the accuracy of climate data interpolation in dozens or even hundreds of kilometers, and most of them are expressed as 5°, 2.5°, 1° or 30' of latitude and longitude. In 2005, Hijmans et al. collected and processed the data of meteorological observation stations around the world, and obtained the climate data with the highest resolution of 30 s, which was more than 400 times higher than the previous climate data resolution, equivalent to 1 km2 horizontal resolution in the equatorial region. It is also the highest resolution climate data available on WorldClim. However, in terms of ecological applications, such as niche model suitability evaluation, if combined with climate data, there will be problems in matching with remote sensing data with a resolution of about 30 m that we commonly use. The spatial resolution of baseline climate data is low, so it is urgent to solve the problem of how to match and use it comprehensively with remote sensing thematic data with high resolution. At present, most of the large-scale studies are to sample high resolution remote sensing data into low resolution climate data, which will lose many details of remote sensing data itself. Or the climate data can be subdivided and sampled into 30 m resolution to match remote sensing data. Both methods have application examples. At present, the former method is used for processing, and at the same time, we are communicating with the relevant technical personnel of the meteorological department, in order to find a more practical and effective way to combine the two data. On the other hand, with the rapid development of remote sensing technology in recent years, both spatial and temporal resolution have been greatly improved, and there are more and more global or large-scale standardized remote sensing data sets with high resolution. For example, Deblauwe et al. (2016) obtained temperature and precipitation data estimated by remote sensing through comprehensive processing of remote sensing data. Its resolution is 0.05° (about 6 km), and its temperature data span from 2001 to 2013, and precipitation data span from 1981 to 2013. The author used WorldClim climate data set to compare with the global climate data set obtained by 0.05° (about 6 km) remote sensing (Zeng 2018). Although the latter was not adopted due to its low spatial resolution, it is a feasible research direction to improve the prediction accuracy of the model. (2) Assessment and prediction of habitat suitability of typical plants and animals under the background of climate change at a more detailed scale. Through the comparison of the prediction results of the model at two scales in this book, it can be seen that the same model and method, with different research scales, may have a great difference in research results. Climate change or comprehensive assessment of climate change and anthropogenic impacts on a macro scale is still insufficient to support policy making. Research the technical route and research steps can be used as a reference with other species or reserve, used for regional level or a specific area to carry out targeted careful evaluation, conclusions and recommendations can get more direct operation, better provide data and information for the construction of giant pandas national park, which is the direction of follow-up study.

8.4 Recommendations for Achieving the Goals of SDG11.4 …

373

(3) Future impacts of climate change on specific bamboo species need to be further studied. Giant pandas are exclusively carnivorous and rely on bamboo for more than 99% of their food sources. The current study only incorporates the data of bamboo species with planar distribution into the model operation, and does not thoroughly analyze the impact of future climate change on specific bamboo species. Subsequent will contact for the panda’s staple food bamboo mountain systems (data, combined with the existing models and methods, at the same time to specific bamboo in corresponding pattern and magnitude of future climate change scenarios are studied, and comprehensive analysis of the existing research results, in order to get a more accurate distribution of the giant panda habitat suitability in the future. In addition, model prediction and field survey results, as well as the ecological characteristics of research species, must be closely combined and comprehensively analyzed in order to improve the feasibility and operability of the suggestions. The prediction using niche model is based on “static niche”, that is, under the premise that the mapping relationship of the model trained by sampling remains unchanged, the occurrence probability of new species under the condition of changing time or re projection of some environmental variables is obtained. However, the survival and reproduction of species are the result of long-term evolution and continuous adaptation to the surrounding environment. The potential distribution of species depends not only on the change of time and space, but also on their own adaptability and migration ability. For example, in order to adapt to the environment, giant pandas change from carnivores to 99% of their food comes from bamboo, and in order to minimize human interference to themselves, their distribution altitude is getting higher and higher. All these show that the niche of species is a “dynamic niche”. Therefore, the adaptive capacity and migration capacity of species should also be taken into account in the follow-up study of the impact of climate change on species. In short, this book has made some useful explorations in the assessment of the impact of climate change on China’s national treasure giant panda and its habitat, in order to provide reference for similar studies and contribute to the protection of biodiversity and the construction of ecological civilization in China.

8.4 Recommendations for Achieving the Goals of SDG11.4 for the Protection and Defense of World Heritage Sites World natural and cultural heritage, including giant panda habitat, is the common wealth of all mankind and has outstanding universal value (OUV). Ensuring the authenticity and integrity of world heritage is the primary condition for heritage protection. World heritage is the “material evidence” to understand the evolution history of our living earth, the evolution and development of human beings, and the customs and cultures of different nationalities. They have the significance and role of knowledge education, civilization inheritance, spiritual incentive and image

374

8 Suggestions on Sustainable Development of Giant Panda Habitat

solicitation, and can make unique contributions to world peace and security. World heritage is not only an honor, or a golden signboard for tourism, but also a solemn commitment to heritage protection. For this reason, the United Nations put forward the requirements and calls for “further efforts to protect and defend the world cultural and natural heritage” in the 2030 agenda for sustainable development. 1. Contents of sustainable development goal SDG11.4 proposed by the United Nations (UN) In 1972, the United Nations Educational, scientific and Cultural Organization (UNESCO) adopted the Convention for the protection of the world cultural and natural heritage (hereinafter referred to as the Convention) at the 17th general conference held at its headquarters in Paris. The Convention aims to recognize, protect, preserve, display and pass on cultural and natural heritage of outstanding universal value from generation to generation. The Convention is the most intelligent cry of mankind in the 20th century. It has received positive responses from all over the world. At present, 195 countries and regions have become contracting members. The Convention is of great significance for carrying forward the outstanding universal value of world heritage and protecting its authenticity and integrity. Since the adoption of the Convention in 1972, the international community has fully accepted the concept of “sustainable development”. The protection and preservation of natural and cultural heritage is a great contribution to sustainable development. On September 25, 2015, 193 Member States formally adopted the 2030 agenda for sustainable development at the United Nations Summit on sustainable development. It is determined to achieve 17 sustainable development goals in the next 15 years (sustainable development goals, SDGs) and 169 sub goals to completely solve the development problems in the three dimensions of society, economy and environment in a comprehensive way and turn to the path of sustainable development. Sustainable development is the result of human reflection on the process of industrial civilization. It is a new concept of development, morality and civilization. It is a development theory and strategy based on the protection of natural resources and environment, under the condition of stimulating economic development, and aiming at improving and improving the quality of human life. It involves many aspects such as nature, environment, society, economy, science and technology, politics and so on. It requires that development should be coordinated and adapted to resources, environment and population. Among the 17 sustainable development goals (SDGs), goal 11 is “building inclusive, safe, disaster resilient and sustainable cities and human settlements”. The goal includes seven direct sub goals and three indirect sub goals. Sub goal SDG11.4 proposes “further efforts to protect and defend the world cultural and natural heritage”, including one indicator (Table 8.1). However, this indicator actually has no data and no method, and belongs to tier III. For this indicator without data and actual methods, how to protect and defend the world heritage needs our in-depth research to find a solution. In particular, find

8.4 Recommendations for Achieving the Goals of SDG11.4 …

375

Table 8.1 Sub goal 11.4 further efforts to protect and defend world cultural and natural heritage Targets

Indicator

11.4 Strengthen efforts to protect and safeguard 11.4.1 To conserve, protect and maintain the total per capita expenditure (public and private) the world’s cultural and natural heritage for all the cultural and natural heritage, listed according to heritage type (cultural, natural, mixed and designated by the World Heritage Center), government level (national, regional and local/municipal), expenditure type (operational expenditure/investment) and private funding type (in-kind donations, private non-profit sector, sponsorship)

a more comprehensive and operable description of the goal by making full use of contemporary advanced science and technology. According to “11.4.1” given by UN, the total per capita expenditure (public and private) for the preservation, protection and conservation of all cultural and natural heritage, by heritage type (cultural, natural, mixed, designated by the World Heritage Center), government level (national, regional and local/city), expenditure type (operational expenditure/investment) and private funding type (in kind donation, private non-profit sector and sponsorship) “breakdown” actually measures the “strength” of “strengthening efforts” according to “total per capita expenditure (public and private)”. However, since world heritage exists in different countries and regions, how to evaluate it under different cultural backgrounds and different development levels “Further efforts to protect and defend world cultural and natural heritage” has become a difficulty. In fact, the size of a country’s total per capita expenditure is at least related to the following factors: (i) The number and total area of all cultural and natural heritage in the country; (ii) The amount of funds invested by the country in each cultural and natural heritage; (iii) The population of the country. Through the comprehensive consideration of the research and interpretation of the evaluation objective system, the convenient acquisition of reliable data and the formulation of practical measurement, we propose that the “measurement capital investment”, especially for natural heritage and mixed heritage, can be calculated from the investment per unit area of the reserve, That is, the index of input cost per unit area = total cost/Heritage site area (km2 or hm2 ) is used to measure the situation of “increasing capital investment”. Through the research and comparison of Chinese case sites, it is found that “input cost per unit area” is higher than “total per capita expenditure (public and private)” More reasonable. Therefore, recommendation 1: integrate the current United Nations Sustainable Development SDG11.4. The indicators given are summarized as 11.4.1 “increase capital investment per unit area to protect and defend world cultural and natural heritage”. According to the relevant provisions and contents of the Convention and the operation guide for the implementation of the World Heritage Convention, we believe that: 11.4.1 given by UN 1 the indicators do not fully meet the responsibility of “further efforts to protect and defend the world cultural and natural heritage”. Therefore,

376

8 Suggestions on Sustainable Development of Giant Panda Habitat

Fig. 8.4 Decomposition map of SDG11 4. target

the indicators given are summarized as 11.4. In addition to “increasing capital investment to protect and defend world cultural and natural heritage”, we suggest 2: add 11.4.2 “increase investment in science and technology to protect and defend world cultural and natural heritage”, and 11.4.3 “strengthen education and publicity to protect and defend world cultural and natural heritage”. 11.4.1,11.4. 2 and 11.4. The three indicators can comprehensively reflect the role of education, science and culture in “further efforts to protect and defend the world cultural and natural heritage”. The target decomposition is shown in Fig. 8.4. 2. Increase investment in science, technology and education to achieve the UN sustainable development goal SDG11.4 As mentioned above, for 11.4 “further efforts to protect and defend world cultural and natural heritage”, it can be divided into three indicators to measure and describe. Only by the organic combination of capital investment, science and technology investment, education and publicity investment can we truly realize the effective protection and sustainable development of world heritage. Only by bringing science, technology and education, including spatial information, into full play, can we solve some urgent problems faced by natural heritage protection. In terms of natural heritage protection, first, we should increase the investment in the application of scientific research and technology to the preservation, protection, conservation, restoration and restoration of natural heritage. In particular, we should give full play to the macro, rapid and accurate advantages of spatial information technology and strengthen the monitoring of the impact of global changes and human activities and the identification and evaluation of potential threat risks. Find ways to eliminate threats to heritage through scientific and technological research. Second, make full use of spatial information technology to scientifically establish the boundary of natural heritage sites. Scientifically assess

8.4 Recommendations for Achieving the Goals of SDG11.4 …

377

whether the current boundary of the heritage site reflects the spatial requirements of habitats, species, processes or phenomena that become the basic conditions of the world heritage site, so that the boundary of the heritage site includes a large enough area close to the prominent universal value, so as to protect its heritage value from being damaged by human direct erosion and resource development outside the area. In addition, the boundary of the current heritage site may coincide with one or more existing or proposed protected areas, such as national parks or nature reserves, biosphere reserves, cultural or historical reserves, etc. Although the protected areas may contain several management zones, only individual zones may meet the requirements of world heritage. Therefore, we should pay close attention to distinguishing different protection requirements, which is of great significance to the countries (regions) where the world heritage sites are located. Third, digital natural heritage is realized based on spatial information technology. Digital Natural Heritage here refers to the resources or products from existing entity or analog natural heritage into digital form, including the generation, recording, preservation, protection and processing of digital products related to natural heritage (research), dissemination and presentation of all dynamic or static digital information. Digital natural heritage can realize the effective recording and preservation of natural heritage highlighting universal value, authenticity and integrity, realize the digitization and panoramic virtual display of natural heritage, and maximize the function of serving public education, scientific research and culture. It will play a fundamental role in protecting and defending world heritage and strengthening education and publicity. On the one hand, through education and publicity, we should enhance people’s appreciation and respect for world heritage sites, realize extensive understanding and understanding of world heritage sites, achieve the public’s deep understanding of the threats faced by heritage sites, promote everyone’s participation in scientific actions for heritage protection, and turn education and publicity into effective and targeted capacity-building activities, Support the implementation of the sustainable development goals of world heritage. On the other hand, through education and publicity to improve the scientific understanding of the relationship between heritage protection and utilization. Actively promote world heritage sites and contribute to the growth of employment and welfare for local communities. In particular, it is necessary to study and make a reasonable balance between heritage protection and tourism utilization, so as to truly achieve the organic unity and benign interaction between community development and conscious protection of world heritage, so as to achieve the development goals of the United Nations 2030 agenda for sustainable development.

378

8 Suggestions on Sustainable Development of Giant Panda Habitat

References Deblauwe V, Droissart V, Bose R et al (2016) Remotely sensed temperature and precipitation data improve species distribution modelling in the tropics. Glob Ecol Biogeogr 25(4):443–454 Hijmans RJ, Cameron SE, Parra JL et al (2005) Very high resolution interpolated climate surfaces for global land areas. Int J Climatol 25(15):1965–1978 Hu J, Zhang Z, Wei F (2011) History, status and future of the development of giant panda reserves in China. J Vet Sci 1:10–14 (in Chinese) Hu J (1990) Foraging ecology of black bears in Tangjiahe nature reserve. J Xihua Normal Univ (Nat Sci Edn) 3:182–194 (in Chinese) Li R, Xu M, Wong MHG et al (2015) Climate change threatens giant panda protection in the 21st century. Biol Cons 182:93–101 Parmesan C, Yohe G (2003) A globally coherent fingerprint of climate change impacts across natural systems. Nature 421(6918):37–42 Shen G, Feng C, Xie Z et al (2008) Proposed conservation landscape for giant pandas in the Minshan Mountains China. Conserv Biol 22(5):1144–1153 Songer M, Delion M, Biggs A et al (2012) Modeling impacts of climate change on giant panda habitat. Int J Ecol 20(12):1–12 Zeng J (2018) Fine assessment and response to future climate change impacts on giant panda habitat. University of Chinese Academy of Sciences (Institute of Remote Sensing and Digital Earth Research, Chinese Academy of Sciences), Beijing. Ph.D. thesis (in Chinese) Zhao C, Yue B, Ran J et al (2016) Relationship between human disturbance and endangered giant panda Ailuropoda melanoleuca habitat use in the Daxiangling Mountains. Oryx 51(1):146–152