The Application of Airborne Lidar Data in the Modelling of 3D Urban Landscape Ecology [1 ed.] 9781443857604, 9781443899864

Citation preview

The Application of Airborne Lidar Data in the Modelling of 3D Urban Landscape Ecology

The Application of Airborne Lidar Data in the Modelling of 3D Urban Landscape Ecology By

Ziyue Chen

The Application of Airborne Lidar Data in the Modelling of 3D Urban Landscape Ecology By Ziyue Chen This book first published 2017 Cambridge Scholars Publishing Lady Stephenson Library, Newcastle upon Tyne, NE6 2PA, UK British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library Copyright © 2017 by Ziyue Chen All rights for this book reserved. No part of this book may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without the prior permission of the copyright owner. ISBN (10): 1-4438-9986-0 ISBN (13): 978-1-4438-9986-4

TABLE OF CONTENTS

Preface ........................................................................................................ vi Chapter One ................................................................................................. 1 Introduction Chapter Two .............................................................................................. 26 Urban DTM Generation Chapter Three ............................................................................................ 57 Urban Land Cover Classification Chapter Four .............................................................................................. 88 Urban Landscape Pattern Analysis Chapter Five ............................................................................................ 109 Urban Landscape Pattern Evaluation Chapter Six .............................................................................................. 147 Supporting Specific Research with 3D Urban Landscape Models: Visual Effect Analysis of Urban Trees Chapter Seven.......................................................................................... 177 Suggestions for Urban Landscape Planning and Improvement Chapter Eight ........................................................................................... 187 Conclusions and Future Work

PREFACE

Remote sensing may seem prohibitive to scholars from other backgrounds due to the complexity of its data format and processing algorithms. I felt the same way even though I majored in Geographical Information System, which is closely related to RS, before I came to Cambridge. So it is not difficult for you to imagine the look on my face when I was told that I should do some research on airborne Lidar, an emerging remote sensing technology. After a thorough review of preliminary Remote Sensing knowledge, I started to work on the processing of airborne Lidar data. To my surprise, the data format and processing methods of raw Lidar point clouds are quite different from traditional remote sensing images. Therefore, a strong background of remote sensing is dispensable for understanding airborne Lidar data, which significantly lowers the difficulty of employing this new technology for beginners. In my own case, I mastered basic methods for processing airborne Lidar data and started designing specific algorithms within one month. Airborne Lidar data is highly suitable for urban studies. The additional 3D positional information provided by airborne Lidar data effectively offsets the missing feature of traditional remote sensing images. Either employed solely or fused with other data sources, airborne Lidar data is an ideal source for establishing and applying 3D urban landscape models, which provides important decision support for a diversity of disciplines. In this case, it is of practical significance to introduce airborne Lidar data to scholars from different subjects (e.g. geography, ecology and urban planning), and I believe proper use of airborne Lidar data can significantly promote the development of other research fields. Different from previous books that mainly introduce its physical meaning and general principles, this book demonstrates several main aspects of Lidar data processing and applications by illustrating specific case studies. These methods proposed in each chapter are not only highly efficient, but also simply implementable, which is well suited to beginners. Following the introduction in this book, you can pick up your own experiments using

The Application of Airborne Lidar Data in the Modelling of 3D Urban Landscape Ecology

vii

airborne Lidar data. Even if you do not want to go through complicated programming to implement specific algorithms, you can still benefit significantly from this book, which explains in details the practicality and potential of airborne Lidar data. By knowing how Lidar data can be processed for a diversity of subjects, geographers, ecologists, planners and decision makers can simply design their research projects based on Lidar data, and seek for technical support from Lidar experts. It is a great pleasure to publish my work with Cambridge Scholars Publishing. Although some parts have been published with relevant Journals, presenting the book as a whole is of great value to demonstrate to readers, especially those without remote sensing background, a complete frame of processing and applying airborne Lidar data for different disciplines. Herein, I gratefully acknowledge my PhD supervisor professor Bob Haining and Dr Bernard Devereux, for introducing me to the field of airborne Lidar and preparing me a solid research background. My heartfelt gratitude goes to my parents, who supported all my big decisions without reservation. I would also like to express my deepest gratitude to my wife and my twin boys, who are my lifelong motivation for becoming a better scholar and person. Finally, my most sincere gratitude to you, readers of this book. Thanks so much for your attention to this book. I simply hope that this book can give you some understandable concepts, if not a deep understanding, of Lidar processing and applications, and inspire you to come up with feasible ideas and methods to better design and implement your research projects. —Ziyue Chen

CHAPTER ONE INTRODUCTION

1.1 Background Landscape ecology is poised to play an important role in tackling major conservation and land-use issues, and in developing responses to pressing problems that result from human-induced global changes (Hobbs, 1997). The rise of landscape ecology is mainly ascribed to the increasing recognition that many conservation and land-use issues can only be solved in a sensible way within a landscape framework (Saunders et al., 1991; Franklin, 1993). Based on the understanding of research problems, landscape ecology aims to deal with landscape patterns, functioning and dynamics (Wiens, 1999; Chen et al, 2002). Since the systematic framework of landscape ecology was proposed in Forman’s work (1986), landscape ecology has experienced rapid developments. In the past decades, landscape ecology has become one of the most promising subjects in geographic research and many studies have been conducted on the theory (Hall, 1991; Barbault, 1995; Selman and Doar, 1998; Li, 2000; Naveh, 2000; Makhzoumi, 2000; Antrop, 2001; Wu, 2008; Wang and Paul, 2009) and methodology (Ihse and Lindahl, 2000; Freeman and Ray, 2001; Mortberg et al., 2007; Silva et al., 2008; Rashed, 2008; Steiniger and Hay, 2009; Chen et al., 2012). In addition, the principle of landscape ecology has been applied to a diversity of research fields, such as the design and planning of green spaces (Yahner et al., 1995; Jim and Chen, 2003; Uy and Nakagoshi, 2008; Tagliafierro et al., 2013), the management of water sources (Aspinall and Pearson, 2000; Smith et al., 2002; Wiens, 2002), suburban and urban planning (Froment and Wildmann, 1987; Selman, 1993; Flores et al., 1998; Girvetz et al., 2008), the management of forests (Hansson, 1992; Bell, 2001; Lundquist and Klopfenstein, 2001; Venema et al, 2005) and so forth.

2

Chapter One

Landscape ecology serves as the bridge between landscape patterns and ecological processes. To examine the interactions between spatial patterns and ecological processes in the environment and ecosystems, it is essential to understand both aspects comprehensively and accurately. Among the two key factors, ecological processes (e.g. the frequency of forest fires, the distribution of vegetation and so forth) are more likely to be measured by a definite approach whilst spatial patterns can be understood from different perspectives. As a result, growing research emphasis is placed on designing appropriate methods to analyze and evaluate spatial patterns of different landscape types. Urban landscape ecology mainly focuses on the interaction between social-ecological issues and spatial arrangements of urban features (e.g. trees, buildings, green spaces, etc.), which are closely related to people’s daily life, mental and psychological health and aesthetic preferences. Therefore, urban landscape ecology is receiving growing research emphasis. In the diversity of research on urban planning (Sun et al., 2006; Long et al., 2012; etc.), land cover change (López et al. 2001; Du et al., 2010; He et al., 2011; etc.) and sustainable development (Käyhkö and Skånes, 2006; Termorshuizen et al., 2007; Renetzeder et al., 2010; Estoque and Murayama, 2013; etc.), one of the key tasks is to quantitatively analyze urban landscape patterns. Landscape patterns can be analyzed with words, statistics, graphics and landscape metrics, the last of which is the most widely used approach to quantify landscape patterns. In the past decades, designing and interpreting landscape metrics has developed into an important research topic in landscape ecology. More than 100 landscape metrics have been coined (Romme, 1982; Forman and Godron, 1986; Gardner et al., 1987; O’Neill et al., 1988; Gustafson and Parker, 1994; McGarigal and Marks, 1995; Riitters et al., 1995; Li and Archer, 1997; Ricotta, 2000; Ong, 2003; Ludwig et al., 2007; Parrott et al., 2008). These metrics have been applied to urban ecology (Wu et al., 2000; Luck and Wu, 2002; Dumas et al, 2008; Li et al., 2011; Ramachandra et al., 2012), landscape planning (Leitao and Ahern, 2002; Sundell-Turner and Rodewald, 2008; Frank et al, 2012), monitoring of landscape changes (Lausch and Herzog, 2002; Herold et al., 2003,2005; Narumalani et al, 2004; Ji et al. 2006; Solon, 2009), forest dynamics (Welsh Jr et al, 2008; Geri et al. 2010; Wang et al., 2012; Tang et al., 2012), ecological network planning (Sklenicka and Charvatova, 2003;

Introduction

3

Jim and Chen, 2003; Zhang and Wang, 2006; Zhang et al., 2009; Schaubroeck et al., 2012) and so forth. These studies have shown the practicality of connecting quantitative analysis to landscape ecological issues. In the meantime, some important landscape metrics, such as Patch Number, Mean Patch Area, Patch Density and Shannon Diversity Index, have been well accepted as the fundamental indicators of landscape configuration. However, limitations still exist in traditional 2D landscape models. As people may overlook, lacking quantitative information in the vertical direction can result in inaccurate or non-discriminatory description of landscape patterns. For instance, the land cover percentage of building areas in a town centre may equal that in a metropolitan area whilst the building structure, height in particular, can differ a lot in the two landscapes (Fig 1.1). The situation also occurs in urban forests, which may have similar tree cover area but different tree heights. In addition to the height of urban features, 2D landscape models cannot provide researchers with terrain information, which is an important factor in the study of ecological processes. According to these limitations, Chen et al. (2008) pointed out that understanding landscape models at multiple-dimensions was a challenging yet significant trend, for future landscape ecology research. Amongst the developments in relevant disciplines, airborne Lidar (also written for LIDAR or LiDAR, Light detection and ranging) data may be the most suitable source for adding height information to 2D landscape models. Airborne Lidar is an emerging technology that obtains elevation information of surface targets by calculating the time of flight taken for laser pulses to travel between a Lidar sensor and a target scene. Relying on the accuracy of GPS (Global Position System) and IMU (Inertial Measurement Unit) components in the system, Lidar can produce data of high resolution and accuracy in both horizontal and vertical directions. With the rapid development of this technology, the applications and processing methods of airborne Lidar data have been significantly broadened and improved. In addition, some mature Lidar processing software has been designed to assist researchers to process airborne Lidar data automatically. With the growing availability of airborne Lidar data, this data source can be adopted as an ideal tool for modelling 3D urban landscape ecology.

4

Chapter One

Fig 1.1 Different landscape patterns may appear the same in 2D landscape models due to the missing height information

1.2 Limitations of Current Urban Landscape Ecology Research For years, 2D landscape models provided landscape ecologists with a mature and efficient tool to examine landscape patterns. However, considering the growing need for more accurate landscape pattern analysis and more incisive understanding of the interactions between landscape patterns and ecological processes, limitations still exist in the 2D landscape models. These limitations are presented as follows:

1.2.1. Lacking a Systematic Methodology of Establishing 3D Urban Landscape Models In spite of great progress made in the subject of landscape ecology, traditional 2D landscape models can hardly meet the requirements of comprehensively describing landscape patterns. To better understand urban landscape patterns and provide more useful sources for specific research, urban landscape patterns should be analyzed from a 3D

Introduction

5

perspective. To date, the methodologies of Lidar data processing have experienced rapid developments, which provides the establishment of 3D landscape models with important theoretical backup. However, since most algorithms are only valid in certain landscape types or are only applicable with additional data sources, very few algorithms for Lidar data processing can be applied to urban areas. Therefore, it is of theoretical and practical significance to design an applicable and efficient methodology of establishing 3D landscape models (Chen et al., 2008) using airborne Lidar data.

1.2.2 Lacking a Mature Framework of 3D Landscape Pattern Analysis Based on the 3D urban landscape model, researchers can conduct 3D landscape pattern analysis, which is another challenging subject in the current research of landscape ecology (Chen et al., 2008). Since a 3D urban landscape model is not commonly available and applied, very limited research emphasis has been placed on designing 3D metrics for 3D landscape pattern analysis. Compared with a strict and comprehensive framework of 2D landscape metrics, very few 3D landscape metrics have been proposed. Some researchers (Cain et al., 2003; Mirzaei and Haghighat, 2010) designed several 3D indicators and these indicators work efficiently to describe some pattern characteristics. However, these indicators are usually designed for specific disciplines and can hardly be generally applied. Without a systematic set of 3D landscape metrics, researchers cannot make full use of the extra vertical information from 3D urban landscape models. In this case, more properly designed and generally applicable 3D metrics are urgently required for researchers to have a comprehensive understanding of urban landscape patterns.

1.2.3 Lacking Systematic and Applicable Methodologies of Urban Landscape Pattern Evaluation One main research purpose of landscape ecology is to analyze, evaluate and enhance current landscape patterns. Research on landscape pattern analysis can be conducted using proper metrics whilst the evaluation of landscape patterns in general, and urban landscape patterns in particular, is still challenging due to the lack of

6

Chapter One

generally applicable criteria. To efficiently evaluate landscape patterns, researchers should have a better understanding of the interactions between landscape patterns and the well-being and preferences of residents, the distribution and diversity of wildlife, the sustainable development of local environments and so forth, which are the main subject and difficulty of landscape ecology research. Although previous studies have pointed out the relationship between landscape patterns and some ecological processes, such as the diffusion of epidemics (Waller, 2000; Schærström, 2009; Dirk and Pfeiffer, 2011; etc.), the distribution of wildlife (Sorace and Visentin, 2007; Kretser et al, 2008; etc.), and so forth, it is still a substantial shortcoming that landscape patterns have not been explicitly linked with ecological processes (Turner, 1989; Turner et al., 2001; Thompson and McGarigal, 2002; Corry and Nassauer, 2005; etc.) . Chen et al. (2008) concluded that most landscape metrics came from statistics and geometry (Li et al., 2004) and have very limited socio-ecological meaning. For years, landscape pattern analysis has mainly focused on the depiction of landscape pattern characteristics and can hardly be used to indicate ecological processes, which causes great controversies (Chen et al., 2008). Without robust linkages between landscape pattern analysis and those social-ecological issues which are closely related to residents’ daily life, research on urban landscape pattern evaluation is weakened significantly through lacking general and applicable criteria. As a result, designing efficient and generally applicable methodologies, which explicitly integrate quantitative 3D landscape pattern analysis with specific social-ecological issues, is of great importance for urban landscape pattern evaluation.

1.3 Research Aims and Objectives 1.3.1 Research Aims To fill these above mentioned research gaps, this PhD book focuses on the creation and analysis of 3D urban landscape models using airborne Lidar data. This research aims to produce new tools and indices, based on 3D landscape information, which could assist urban landscape planning and management. The methodology proposed in this research can provide reference for other landscape ecologists that may not be

Introduction

7

familiar with the use of airborne Lidar data, so more scholars can be encouraged to apply the advanced tool to 3D urban landscape ecology, a new yet promising discipline.

1.3.2 Research Objectives In accordance with the research aim, the objectives of this research are to propose: 1.3.2.1 The Methodology of Establishing 3D Urban Landscape Models To establish 3D urban landscape models, researchers need to obtain urban DTMs and classified land cover types with height information. Since object height information can be acquired from urban DTMs, the two indispensable procedures of establishing 3D urban landscape models are urban DTM generation and urban land cover classification. DTMs have traditionally been produced through on-site survey, which takes much time and human resources. Due to its high efficiency and accuracy, airborne Lidar has become one of most widely used approaches for DTM generation and many algorithms of DTM generation have been designed to derive DTMs in different situations. However, due to the complexity of different urban features, most existing algorithms cannot work efficiently in urban terrain situations. For decades, urban land cover classification has experienced rapid developments and many studies have been conducted using multi-spectral remote sensing images and airborne photographs (Haack et al., 1987; Gastellu-Etchegorry, 1990; Eyton, 1991; Jensen, 1997; Zhang and Foody, 1998; Barr and Barnsley, 2000; Stefanov et al., 2001; David and Wang, 2002; Yang et al., 2003; Lu and Weng, 2006; Zhou et al., 2008, 2009; Myint et al., 2011; Zhu et al., 2012; etc). Since airborne Lidar data can provide urban land cover classification with additional elevation information, some researchers have integrated multi-spectral images with airborne Lidar data for better classification accuracy (Teo and Chen, 2004; Rottensteiner et al., 2005; Chen et al., 2009; etc.). Although the methodologies of urban land cover classification are mature and efficient, most algorithms involve the use of multi-spectral images, especially high-resolution data sources,

8

Chapter One

whilst very few studies have conducted urban land cover classification using airborne Lidar data only. As discussed, airborne Lidar data is the indispensable source for deriving urban DTMs and establishing 3D urban landscape models. If additional data sources, such as high-resolution images or airborne photographs, are employed for the procedure of urban land cover classification, extra cost will be added to research projects. Considering the high cost of airborne Lidar data, it is not always feasible for researchers to purchase another type of data sources for urban land cover classification. Therefore, there is a practical need for designing algorithms for urban land cover classification using airborne Lidar data. According to the research problems, one major objective of this research is to propose efficient and applicable methods of urban DTM generation and urban land cover classification and thus establish 3D urban landscape models using airborne Lidar data as the only source. 1.3.2.2 Theoretical Framework and Case Studies of Applying 3D Urban Landscape Models Since Forman (1986) proposed the patch-corridor-matrix model for landscape ecology, the majority of research on landscape ecology has been conducted to analyze landscape patterns and interactions between landscape patterns and ecological processes based on this model. The patch-corridor-matrix model is an important foundation for landscape ecology and the rapid development of landscape ecology research proved the practicality of this model. However, height information is not included in traditional 2D models (Chen et al., 2008) and the vertical structure of landscape features or patches cannot be analyzed. 3D urban landscape models provide researchers with a foundation to improve current research methods and expand the scope of landscape ecology. With limited time, resources and research experience, it is not feasible for the author to establish a very concrete system of 3D urban landscape ecology. Instead, this book suggests a framework for 3D urban landscape ecology with possible future directions. Next, this research explains these key factors using some specifically designed case studies. Similar to 2D landscape ecology research, 3D landscape pattern analysis and interactions between 3D patterns and socio-ecological

Introduction

9

issues are the focus of 3D urban landscape ecology. The main difficulties in applying 3D urban landscape models lie in appropriately utilizing the additional height attribute. 2D patch-based landscape models are easily applicable based on a well-designed and widely accepted system of 2D landscape metrics (Lausch and Herzog, 2002; Herold et al., 2005; Sundell-Turner and Rodewald, 2008). Nevertheless, since the 3D landscape model is not commonly used, the methodology of 3D landscape pattern analysis is very limited. To demonstrate the methodology and efficiency of 3D landscape pattern analysis, the author coins some 3D landscape metrics, which may be generally applied to common landscape pattern analysis, for a comparative study. Based on the 3D landscape models, urban landscape patterns can be further compared, evaluated and improved. Landscape patterns may be evaluated in terms of the well-being and preferences of residents, the distribution and diversity of wildlife, the sustainable development of local environments and so forth. However, since it is still a difficult task to link these social-ecological issues to quantitative 3D landscape pattern analysis, researchers can hardly design robust methodologies and applicable criteria for urban landscape pattern evaluation. This research introduces some potential criteria for landscape pattern evaluation and conducts a case study to demonstrate the methodology of obtaining efficient and robust criteria, which are closely related to quantitative landscape pattern analysis, for urban landscape evaluation. In addition to landscape pattern analysis and evaluation, 3D urban landscape models can also support specific research. This research suggests some potential areas that 3D urban landscape models can be applied to, and one case study is conducted to demonstrate that 3D urban landscape models can be an accurate and efficient tool to replace large-scale surveys for some specific research subjects. 1.3.2.3 Suggestions for Improving Urban Landscape Patterns The ultimate goal for urban landscape ecology is to propose useful decision support for landscape planners and policy makers. To date, growing research emphasis has been put on holistic and sustainable landscape planning. In addition, many international and regional eco-projects (e.g. UK National Ecosystem Assessment UK NEA project) have been conducted for better landscape planning. Based on

10

Chapter One

these studies and projects, some general principles (e.g. conducting landscape planning from a holistic and sustainable perspective, conducting landscape planning according to the characteristics and development potential of the study site) have been proposed for landscape planning. However, limitations still exist in the current strategy for urban landscape planning. Most strategies for landscape planning and improvement are of large-scale. Although these types of holistic principles are theoretically effective, they are not always feasible under different land use policies. For instance, the implementation of large-scale landscape planning projects may be constrained by those land sources that belong to individuals or some institutions. To provide alternative strategies for landscape planners when comprehensive and large-scale methodologies may not work, this research proposes some suggestions for improving urban landscapes from a micro perspective. In addition, suggestions for pre-planning survey and landscape change monitoring are also discussed.

1.4 Book Structure The framework of this research is described in Fig 1.2. Two main tasks of this research are to establish and apply 3D urban landscape models. Establishing 3D urban landscape models requires two fundamental sources, urban DTMs and classified urban land cover types; whilst 3D landscape models can be applied to three aspects: quantitatively analyzing 3D urban patterns, evaluating 3D urban landscape patterns and supporting specific research.

Introduction

11

Fig 1.2 The realization and applications of 3D urban landscape models

According to research objectives and the framework described in Fig 1.2, this book is structured in the following sections:

Urban DTM Generation (Chapter Two) Although a large body of algorithms (Kraus and Pfeifer, 1998, 2001; Elmqvist, 2002; Wack and Wimmer, 2002; Hu, 2003; etc.) has been designed, a generally applicable approach for generating DTMs, especially urban DTMs using airborne Lidar data is yet to be developed. Even if some advanced algorithms exist, the task of implementation causes extra difficulties. As a result, an easily applicable method, the upward-fusion algorithm is designed in this research to process raw airborne Lidar data for high quality urban DTMs. Algorithms of urban DTM generation need to be examined

12

Chapter One

using field-collected data of high accuracy and the availability of reliable reference data is the key factor for a successful case study. Cambridge is a typical urban landscape, which consists of a diversity of urban features with different sizes. In addition, the ground control points are available using high-accuracy GPS tools. As a result, Cambridge is a suitable site for the case study of urban DTM generation. Field-collected reference data are included in the study for the accuracy assessment. To further validate the accuracy and efficiency of this algorithm, the upward-fusion method is also compared with some leading Lidar processing software, Lasground, Terrascan and Tiffs.

Urban Land Cover Classification (Chapter Three) Many researchers employed airborne Lidar data for urban land cover classification. However, most of these studies have used airborne Lidar data as complementary sources to multi-layer images and very few researchers employed Lidar data as the sole source for urban land cover classification. To fill this research gap, the author proposes an object-based urban land cover classification method. Cambridge has a large proportion of trees and buildings that locate together in the city center, which causes common difficulties in urban land cover classification. Therefore, the algorithm of urban land cover classification may be generalized to other cities if this method works efficiently in Cambridge. Based on the output of urban DTM generation, a case study of urban land cover classification using airborne Lidar data was conducted. To evaluate the accuracy of this algorithm, accuracy assessment and comparison with Lidar processing software, Terrascan, is included in the case study.

Urban Landscape Pattern Analysis (Chapter Four) Based on the generated urban DTM and classified land cover types, 3D urban landscape models can be established. To demonstrate the methodology of landscape pattern analysis based on the 3D model, a comparative study is conducted. Central Cambridge and the residential area in Canvey Island have a similar landscape composition and structure from a 2D perspective.

Introduction

13

However, their landscape patterns in the vertical direction differ significantly. As a result, comparing landscape patterns between the two sites from both the 2D and 3D perspectives can efficiently examine the advantages of 3D landscape models and the practicality of a diversity of landscape metrics. This case study is conducted as follows. Firstly, a set of 2D pixel or object-based landscape metrics is employed to analyze and compare horizontal patterns of the two cities. Next, some 3D landscape metrics, which are suitable for the case study and have the potential to be generally applied, are designed and adopted to examine the vertical patterns of the two cities. By analyzing the difference between 2D and 3D landscape pattern analysis, the practicality of proposed 3D landscape metrics can be successfully examined.

Evaluating Urban Landscape Patterns (Chapter Five) Urban landscape patterns can be well measured using 2D and 3D landscape metrics. However, lacking systematic evaluation systems, it remains difficult to evaluate urban landscape patterns. Based on 3D landscape models, interactions between landscape patterns and social-ecological issues, especially the preferences and benefits of urban residents and wildlife, can be further examined. With these types of analysis, urban landscape patterns can be evaluated from different perspectives. One example is given in this section. The frequency and intensity of specific ecological processes, the diversity of wildlife and vegetation, sustainability of farm land use and so forth, which are important factors for rural or wild landscape pattern evaluation, are not generally practical for urban landscape pattern evaluation. Compared with suburban or rural areas, the evaluation of urban landscapes should place more emphasis on the needs of local residents. People’s preferences towards landscape patterns have been widely researched, yet the research on linking quantitative landscape pattern analysis (3D patterns in particular) to landscape preferences is very limited. As a result, the public’s preferences towards landscape preferences, integrated with landscape pattern analysis based on 3D landscape models, can be employed as a useful evaluation criterion for urban landscape evaluation. This part introduces a survey on the public’s preferences towards urban landscape patterns.

14

Chapter One

The public’s landscape preferences may vary in terms of geographical locations and cultures (Purcell et al., 1994; Rauwald and Moore, 2002; etc.). As a result, a comparative survey should be conducted in different geographical areas. Cambridge is a famous university town with a typical British urban landscape whilst Nanjing (China) has a typical Chinese urban landscape. As a result, responses from the two sites can be used to understand landscape preferences of different cultural groups. Integrated with landscape pattern analysis, the findings from this survey can be employed as feasible criteria for urban landscape pattern evaluation.

Supporting Specific Research with 3D Urban Landscape Models (Chapter Six) In addition to landscape pattern analysis and evaluation, 3D urban landscape models can provide accurate data sources for specific research of landscape ecology. Some potential applications are discussed in this chapter, and assessing tree green availability using 3D urban landscape models is introduced in this part as an instance. Tree green availability (how much green perception individual trees can provide for local residents), which is a very important factor in people’s perception of local environments, is usually calculated by time-consuming field work. In this section, the 3D landscape model is employed to assess tree green availability. Considering the diversity of tree species, sizes and shapes, Cambridge is selected as the study site. Firstly, the tree green availability of some tree samples is recorded and analyzed through a photography- based survey. Next, some variables concerning tree shapes and sizes are acquired from the 3D landscape model. Following this, this study employs these model-derived variables to simulate the true value of tree green availability by establishing a robust and applicable regression model. To prove the reliability of the proposed regression model, a cross-validation is conducted using a K-fold cross-validation1. (Kohavi, 1995)

1

In k-fold cross-validation, the entire data set is rstly partitioned into k equally (or nearly equally) sized folds. Subsequently, k iterations of training and validation are performed. Within each iteration a different fold of the data is chosen for validation whilst the remaining k - 1 folds are used for learning. As a result, each fold of the data is used for validation for exactly once.

Introduction

15

Suggestions for Improving Urban Landscape Patterns (Chapter Seven) Many sustainable and comprehensive strategies have been employed by some governments, institutions, and landscape planners to provide residents with better urban landscape patterns, yet some principles may not be generally applicable due to the limitation of different local land use regulations. Based on the findings from the present study, some practical approaches, which serve as complements to existing methods of urban landscape improvement, are suggested for urban planners and decision makers to design and improve urban landscapes. These specific suggestions are proposed in accordance with corresponding stages of urban landscape planning.

Conclusions and Future Work (Chapter Eight) The methodology and key findings of each chapter are concluded in this section. In addition, the research plan of future work in terms of different areas is discussed in accordance with some limitations of the present study.

References Antrop, M. (2001) The language of landscape ecologists and planners: A comparative content analysis of concepts used in landscape ecology. Landscape and Urban Planning. 55(3), 163-173 Aspinall, R., Pearson, D. (2000) Integrated geographical assessment of environmental condition in water catchments: Linking landscape ecology, environmental modelling and GIS. Journal of Environmental Management. 50(4), 299-319. Barbault, R. (1995) Biodiversity dynamics: from population and community ecology approaches to a landscape ecology point of view. Landscape and Urban Planning. 31(1–3), 89-98. Barr, S., Barnsley, M. (2000) Reducing structural clutter in land cover classifications of high spatial resolution remotely-sensed images for urban land use mapping. Computers & Geosciences. 26(4), 433-449. Bell, S. (2001) Landscape pattern, perception and visualisation in the visual management of forests. Landscape and Urban Planning.

16

Chapter One

54(1-4), 201-211. Cain, A.T., Tuovila, V.R., Hewitt, D.G., Tewes, M.E. (2003) Effects of a highway and mitigation projects on bobcats in Southern Texas. Biological Conservation. 114, 189-197. Chen, L.D., Liu, Y., Lv, Y.H., Feng , X.M., Fu, B.J. (2008) Pattern analysis in landscape ecology: progress, challenges and outlook. ACTA ECOLOGICA SINICA. 28(11), 5521-5531. Chen, L. Yang, Z.F., Chen, B. (2012) Landscape ecology planning of a scenery district based on a characteristic evaluation index system—a case study of the Wuyishan scenery district. Procedia Environmental Sciences. 13, 30-42. Chen ,W.B., Xiao, D.N., Li, X.Z. (2002) The characteristics and contents of landscape spatial analysis. Acta Ecologica Sinica. 22(7), 1135–1142. Chen, Y. H., Su, W., Li , J., Sun, Z.P. (2009) Hierarchical object oriented classification using very high resolution imagery and LIDAR data over urban areas. Advances in Space Research. 43, 1101–1110. Corry, R.C., Nassauer, J.I. (2005) Limitations of using landscape pattern indices to evaluate the ecological consequences of alternative plans and designs. Landscape and Urban Planning. 72, 265–280. David, C.H., Wang, X.Y. (2002) Urban land cover classification from high resolution multi-spectral IKONOS imagery. Geoscience and Remote Sensing Symposium, 2002. IGARSS. 2, 1204-1206. Dirk, S., Pfeiffer, U. (2011) Spatial modelling of disease using dataand knowledge-driven approaches. Spatial and Spatio-temporal Epidemiology. 2(3), 125-133. Du, P. J., LI, X. L., Cao, W., Luo, Y., Zhang, H. P. (2010) Monitoring urban land cover and vegetation change by multi-temporal remote sensing information. Mining Science and Technology (China). 20(6), 922-932. Dumas, E., Jappiot, M., Tatoni, T. (2008) Mediterranean urban-forest interface classification (MUFIC): A quantitative method combining SPOT5 imagery and landscape ecology indices. Landscape and Urban Planning. 84(3-4), 183-190. Elmqvist, M. (2002) Ground surface estimation from airborne laser scanner data using active shape models. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. XXXIV ( part 3/W3), 114–118.

Introduction

17

Estoque, R.C., Murayama, Y. (2013) Landscape pattern and ecosystem service value changes: Implications for environmental sustainability planning for the rapidly urbanizing summer capital of the Philippines. Landscape and Urban Planning. 116, 60-72. Eyton, J.R. (1991) Urban land use classification and modelling using cover-type frequencies. Applied Geography. 13(2), 111-121. Flaspohler, D.J., Giardina, C.P., Asner, G.P., Hart, P., Price, J., Lyons, C.K., Castaneda,X. (2010). Long-term effects of fragmentation and fragment properties on bird species richness in Hawaiian forests. Biological Conservation. 143, 280-288. Flores, A., Pickett, S.T.A., Zipperer, W.C., Pouyat, R.V., Pirani, R. (1998) Adopting a modern ecological view of the metropolitan landscape: the case of a greenspace system for the New York City region. Landscape and Urban Planning. 39, 295–308. Forman, R.T.T., Gordon, M. (1986). Landscape Ecology, Wiley, New York. Franklin, J.F. (1993). Preserving biodiversity: species, ecosystems or landscapes? Ecological Applications. 3, 202-205. Frank, S., Fürst, C., Koschke, L., Makeschin, F., (2012). A contribution towards a transfer of the ecosystem service concept to landscape planning using landscape metrics. Ecological Indicators. 21, 30-38. Freeman, R. E., Ray, R. O (2001) Landscape ecology practice by small scale river conservation groups. Landscape and Urban Planning. 56(3–4), 171-184. Froment, A., Wildmann, B. (1987) Landscape ecology and rural restructuring in Belgium. Landscape and Urban Planning. 14, 415-426. Gardner, R.H., Milne, B.T., Turner, M.G., O’Neill, R.V. (1987). Neutral models for the analysis of broad-scale landscape pattern. Landscape Ecology. 1, 19–28. Gastellu-Etchegorry, J.P (1990) An assessment of SPOT XS and Landsat MSS data for digital classification of near-urban land cover. International Journal of Remote Sensing. 11(2), 225-235. Geri, F., Rocchinib, D., Chiarucci. A. (2010). Landscape metrics and topographical determinants of large-scale forest dynamics in a Mediterranean landscape. Landscape and Urban Planning. 95, 46–53. Girvetz, E. H., Thorne, J. H., Berry, A. M., Jaegera, J.A.G. (2008) Integration of landscape fragmentation analysis into regional planning: A statewide multi-scale case study from California, USA.

18

Chapter One

Landscape and Urban Planning. 86, 205–218. Gustafson, E.G., Parker, G.R. (1994) Using an index of habitat patch proximity for landscape design. Landscape and Urban Planning. 29, 117-130. Haack, B., Bryant, N., Adams, S. (1987) An assessment of landsat MSS and TM data for urban and near-urban land-cover digital classification. Remote Sensing of Environment. 21(2), 201-213. Hall, D. L. (1991) Landscape planning: functionalism as a motivating concept from landscape ecology and human ecology. Landscape and Urban Planning. 21(1-2), 13-19. Hansson, L. (1992) Landscape ecology of boreal forests. Trends in Ecology & Evolution. 7(9), 299-302. He, C., Wei, A.N., Shi, P.J., Zhang, Q.F., Zhao, Y.Y. (2011) Detecting land-use/land-cover change in rural–urban fringe areas using extended change-vector analysis. International Journal of Applied Earth Observation and Geoinformation. 13(4), 572-585. Herold, M., Goldstein, N.C., Clarke, K.C. (2003) The spatiotemporal form of urban growth: measurement, analysis and modeling. Remote Sensing of the Environment. 86, 286–302. Herold, M., Couclelis, H., Clarke K.C. (2005) The role of spatial metrics in the analysis and modelling of urban land use change. Computers, Environment and Urban Systems. 29, 369-399. Hobbs, R. (1997) Future landscapes and the future of landscape ecology. Landscape and Urban Planning. 37, 1-9. Hu, Y. (2003) Automated extraction of digital terrain models, roads and buildings using airborne lidar data. PhD Thesis. University of Calgary, Canada. Ihse, M., Lindahl, C. (2000) A holistic model for landscape ecology in practice: the Swedish survey and management of ancient meadows and pastures. Landscape and Urban Planning. 50(1-3), 59-84. Lausch, A., Herzog, F. (2002) Applicability of landscape metrics for the monitoring of landscape change: issues of scale, resolution and interpretability. Ecological Indicators. 2, 3–15. Leitao,A.B., Ahern,J. (2002) Applying landscape ecological concepts and metrics in sustainable landscape planning. Landscape and Urban Planning. 59, 65-93. Levick, S., Rogers, K. (2008) Patch and species specific responses of savanna woody vegetation to browser exclusion. Biological Conservation. 141, 489-498. Li, B. L. (2000) Why is the holistic approach becoming so important in

Introduction

19

landscape ecology? Landscape and Urban Planning. 50(1-3), 27-41. Li, B. L., Archer, S. (1997) Weighed mean patch size: a robust index for quantifying landscape structure. Ecological Modelling. 102, 353-361. Li, J. X., Song, C., Cao, L., Zhu, F., Meng, X. L., Wu, J. G. (2011) Impacts of landscape structure on surface urban heat islands: A case study of Shanghai, China. Remote Sensing of Environment. 115(12), 3249-3263. Luck, M., Wu, J. (2002) A gradient analysis of urban landscape pattern: a case study from the Phoenix metropolitan region, Arizona, USA. Landscape Ecology. 17, 327–339. Ludwig, J.A., Bastin, G.N., Chewings, V.H., Eager, R.W., Liedloff, A.C. (2007). Leakiness: A new index for monitoring the health of arid and semiarid landscapes using remotely sensed vegetation cover and elevation data. Ecological Indicators. 7, 442–454. Lundquist, J.E., Klopfenstein, N.B. (2001) Integrating concepts of landscape ecology with the molecular biology of forest pathogens. Forest Ecology and Management. 150(3), 213-222. Li, X.Z., Bu, R.C., Chang, Y. (2004) The response of landscape metrics against pattern scenarios. Acta Ecologica Sinica. 24(1), 123-134. Long, Y., Gu, Y. Z., Han, H. Y. (2012) Spatiotemporal heterogeneity of urban planning implementation effectiveness: Evidence from five urban master plans of Beijing. Landscape and Urban Planning. 2-4, 103-111. López, E., Bocco, G., Mendoza, M., Duhau, E. (2001) Predicting land-cover and land-use change in the urban fringe: A case in Morelia city, Mexico. Landscape and Urban Planning. 55(4), 271-285. Lu, D.S., Weng, Q.H. (2006) Use of impervious surface in urban land-use classification. Remote Sensing of Environment. 102(1-2), 146-160. Jensen, L. M. (1997) Classification of urban land cover based on expert systems, object models and texture. Computers, Environment and Urban Systems. 21(3-4), 291-302. Ji, W., Ma, J., Twibell, R.W., Underhill, K. (2006) Characterizing urban sprawl using multi-stage remote sensing images and landscape metrics. Computers, Environment and Urban Systems. 30, 861–879. Jim, C.Y., Chen, S.S. (2003) Comprehensive greens pace planning based on landscape ecology principles in compact Nanjing city,

20

Chapter One

China. Landscape and Urban Planning. 65, 95–116. Käyhkö, N., Skånes, H. (2006) Change trajectories and key biotopes—Assessing landscape dynamics and sustainability. Landscape and Urban Planning. 75 (3–4), 300-321. Kohavi, R (1995). "A study of cross-validation and bootstrap for accuracy estimation and model selection". Proceedings of the Fourteenth International Joint Conference on Artificial Intelligence 2 (12): 1137–1143. Kraus, K., Pfeifer, N. (1998) Determination of terrain models in wooded areas with airborne laser scanner data. ISPRS Journal of Photogrammetry and Remote Sensing. 53, 193–203. Kraus, K., Pfeifer, N. (2001) Advanced DTM generation from LIDAR data. International Archives of Photogrammetry and Remote Sensing, Annapolis, MD, 22-24 Oct. 2001, XXXIV (part 3/W4), 23–30. Kretser , H.E., Sullivan , P.J., Knuth, B.A. (2008) Housing density as an indicator of spatial patterns of reported human–wildlife interactions in Northern New York. 84, 282–292. Makhzoumi, J. M. (2000) Landscape ecology as a foundation for landscape architecture: application in Malta. Landscape and Urban Planning. 50(1-3), 167-177. McGarigal, K., Marks, B.J. (1995). Fragstats: spatial pattern analysis program for quantifying landscape structure. Program documentation. Oregon State University, Corvallis. Mirzaei, P.A., Haghighat,F. (2010) A novel approach to enhance outdoor air quality: Pedestrian ventilation system. Building and Environment. 45, 1582-1593. Mortberg, U.M., Balfors, B., Knol, W.C. (2007) Landscape ecological assessment: A tool for integrating biodiversity issues in strategic environmental assessment and planning. Journal of Environmental Management. 82, 457–470. Narumalani, S., Mishra, D. R., Rothwell, R. G. (2004) Change detection and landscape metrics for inferring anthropogenic processes in the greater EFMO area. Remote Sensing of Environment. 91(3-4), 478-489 Naveh, Z. (2000) What is holistic landscape ecology? A conceptual introduction. Landscape and Urban Planning. 50 (1-3), 7-26. O’Neill, R.V., Krummel, J.R., Gardner, R.H., Sugihara, G., Jackson, B., DeAngelis, D.L., Milne, B.T., Turner, M.G., Zygmunt, B., Christensen, S.W., Dale, V.H., Graham, R.L. (1988) Indices of

Introduction

21

landscape pattern. Landscape Ecology. 1, 153–162. Ong, B.L. (2003) Green plot ratio: an ecological measure for architecture and urban planning. Landscape and Urban Planning. 63, 197–211. Parrott, L., Proulx1, R., Thibert-Plante, X. (2008) Three-dimensional metrics for the analysis of spatiotemporal data in ecology. Ecological Matics. 3, 343-353. Purcell, A.T., Lamb, R.J., Peron, E.M., Falchero, S. (1994) Preference or preferences for landscape? Journal of Environmental Psychology. 14, 195–209. Ramachandra, T.V., Aithal, B. H., Sanna, D.D. (2012) Insights to urban dynamics through landscape spatial pattern analysis. International Journal of Applied Earth Observation and Geoinformation. 18, 329-343. Rashed, T. (2008) Remote sensing of within-class change in urban neighborhood structures. Computers, Environment and Urban Systems. 32, 343–354. Rauwald, K.S., Moore, C.F. (2002) Environmental attitudes as predictors of policy support across three countries. Environment and Behavior. 34 (6), 709–739. Renetzeder, C., Schindler, S., Peterseil, J., Prinz, M.A., Mücher, S., Wrbka, T (2010) Can we measure ecological sustainability? Landscape pattern as an indicator for naturalness and land use intensity at regional, national and European level. Ecological Indicators. 10(1), 39-48. Ricotta, C. (2000) From theoretical ecology to statistical physics and back: self-similar landscape metrics as a synthesis of ecological diversity and geometrical complexity. Ecological Modelling. 125, 245–253. Riitters, K.H., O’Neill, R.V, Hunsaker, C.T., Wickham, J.D., Yankee, D.H., Timmins, S.P., Jones, K.B., Jackson, B.L (1995) A factor analysis of landscape pattern and structure metrics. Landscape Ecology. 10 (1), 23–39. Romme, W.H. (1982) Fire and landscape diversity in subalpine forests of Yellowstone National Park. Ecological Monographs. 52, 199–221. Rottensteiner, F., Trinder, J., Clode, S., Kubik, K. (2005) Using the Dempster–Shafer method for the fusion of LIDAR data and multi-spectral images for building detection. Information Fusion. 6, 283–300.

22

Chapter One

Schaubroeck, T., Staelens, J., Verheyen, K., Muys, B., Dewulf, J. (2012) Improved ecological network analysis for environmental sustainability assessment; a case study on a forest ecosystem. Ecological Modelling. 247, 144-156. Schærström, A. (2009) Disease Diffusion. International Encyclopedia of Human Geography. 2009, 222-233. Selman, P. (1993) Landscape ecology and countryside planning: Vision, theory and practice. Journal of Rural Studies. 9(1), 1-21. Selman, P., Doar, N. (1992) An investigation of the potential for landscape ecology to act as a basis for rural land use plans. Journal of Environmental Management. 35(4), 281-299. Silva, E.A. Ahern, J., Wileden, J. (2008) Strategies for landscape ecology: An application using cellular automata models. Progress in Planning. 70(4), 133-177. Sklenicka, P., Charvatova, E. (2003) Stand continuity—a useful parameter for ecological networks in post-mining landscapes. Ecological Engineering. 20(4), 287-296. Smith, S.V., Renwick, W.H., Bartley, J.D., Buddemeier, R.W(2002) Distribution and significance of small, artificial water bodies across the United States landscape. Science of The Total Environment. 299, 21-36. Solon, J. (2009) Spatial context of urbanization: Landscape pattern and changes between 1950 and 1990 in the Warsaw metropolitan area, Poland. Landscape and Urban Planning. 93(3-4), 250-261. Sorace, A., Visentin, M. (2007) Avian diversity on golf courses and surrounding landscapes in Italy. Landscape and Urban Planning. 81, 81–90. Stefanov, W.L., Ramsey, M.S., Christensen, P.R. (2001) Monitoring urban land cover change: An expert system approach to land cover classification of semiarid to arid urban centers. Remote Sensing of Environment. 77(2), 173-185. Steiniger, S., Hay, G. J. (2009) Free and open source geographic information tools for landscape ecology. Ecological Informatics. 4(4), 183-195. Sun, J., Xia, H.P., Lan, C.Y., Xin, K. (2006) A gradient analysis based on the buffer zones of urban landscape pattern of the constructed area in Guigang City, Guangxi, China. Acta Ecologica Sinica. 3, 655-662. Sundell-Turner, N.M. Rodewald, A.D. (2008) A comparison of landscape metrics for conservation planning. Landscape and Urban

Introduction

23

Planning. 86, 219–225. Tagliafierro, C., Longo, A., Eetvelde, V.V., Antrop, M., Hutchinson, W.G. (2013) Landscape economic valuation by integrating landscape ecology into landscape economics. Environmental Science & Policy. http://dx.doi.org/10.1016/j.envsci.2012.12.001 Tang, J. M., Bu, K., Yang, J. C., Zhang, S. W., Chang, L. P. (2012) Multitemporal analysis of forest fragmentation in the upstream region of the Nenjiang River Basin, Northeast China. Ecological Indicators. 23, 597-607. Teo, T.A., Chen, L.C. (2004) Object-based building detection from LiDAR data and high resolution satellite imagery, In: Proceedings of Asian Conference on Remote Sensing, November 22–26. Ching-Mai, Thailand, 2004. Termorshuizen, J.W., Opdam, P., Brink, A. (2007) Incorporating ecological sustainability into landscape planning. Landscape and Urban Planning. Volume 79(3-4), 374-384. Thompson, C.M., McGarigal, K. (2002) The influence of research scale on bald eagle habitat selection along the lower Hudson River, New York (USA). Landscape Ecology. 17 (6), 569–586. Turner, M.G. (1989). Landscape ecology: the effect of pattern on process. Annual Review of Ecology and Systematics. 20, 171–197. Turner, M.G., Gardner, R.H., O’Neill, R.V. (2001) Landscape Ecology in Theory and Practice: Pattern and Process. Springer-Verlag, New York. Umapathy, G., Kumar, A. (2000) The occurrence of arboreal mammals in the rain forest fragments in the Anamalai Hills, south India. Biological Conservation. 92, 311-319. Uy, P.D., Nakagoshi, N. (2008) Application of land suitability analysis and landscape ecology to urban greenspace planning in Hanoi, Vietnam. Urban Forestry & Urban Greening. 7(1), 25-40. Venema, H.D., Calamai, P.H., Fieguth, P (2005) Forest structure optimization using evolutionary programming and landscape ecology metrics. European Journal of Operational Research. 164(2), 423-439. Wang, L. Y., Paul, F.J.E ( 2009) Some theoretical considerations: From landscape ecology to waterscape ecology. Acta Ecologica Sinica. 29(3), 176-181. Wang, X. L., Hamann, A., Cumming, S. G. (2012) Measuring boreal forest fragmentation after fire: Which configuration metrics are best? Ecological Indicators. 13(1), 189-195.

24

Chapter One

Waller, L.A. (2000) A Civil Action and Statistical Assessments of the Spatial Pattern of Disease: Do We Have a Cluster? Regulatory Toxicology and Pharmacology. 32(2), 174-183. Wack, R., Wimmer, A. (2002) Digital terrain models from airborne laser scanner data- a grid based approach. International Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences. 34(part 3B), 293–296. Welsh Jr., H.H., Pope, K. L., Wheeler, C. A. (2008) Using multiple metrics to assess the effects of forest succession on population status: A comparative study of two terrestrial salamanders in the US Pacific Northwest. Biological Conservation, 141, 4, 1149-1160. Wiens J A. (1999) Landscape ecology: the science and the action. Landscape Ecology. 14: 103. —. (2002) Riverine landscapes: taking landscape ecology into the water. Freshwater Biology. 47(4), 501-515. Wu, J. (2000) Landscape Ecology: Patterns, Process, Scale and Hierarchy. Higher Education Press, Beijing. —. (2008) Landscape ecology. Encyclopedia of Ecology. 2103-2108. Yahner, T.G., Korostoff, N., Johnson, T.P., Battaglia, A. M., Jones, D.R. (1995) Cultural landscapes and landscape ecology in contemporary greenway planning, design and management: a case study. Landscape and Urban Planning. 33(1-3), 295-316. Yang, L. M., Xian, G., Klaver, J.M., Deal, B. (2003) Urban Land-Cover Change Detection through Sub-Pixel Imperviousness Mapping Using Remotely Sensed Data. Photogrammetric Engineering & Remote Sensing. 69(9), 1003–1010. Zhang, J., Foody, G.M. (1998) A fuzzy classification of sub-urban land cover from remotely sensed imagery. International Journal of Remote Sensing. 19(14), 2721-2738. Zhang, L.Q., Wang, H.Z. (2006) Planning an ecological network of Xiamen Island (China) using landscape. Metrics and network analysis. Landscape and Urban Planning. 78,449–456. Zhang, Y., Yang, Z.F., Yu, X.Y. (2009) Ecological network and emergy analysis of urban metabolic systems: Model development, and a case study of four Chinese cities. Ecological Modelling. 220(11), 1431-1442. Zhou, W.Q., Huang, G.L., Troy, A., Cadenasso, M.L. (2009) Object-based land cover classification of shaded areas in high spatial resolution imagery of urban areas: A comparison study. Remote Sensing of Environment. 113(8), 1769-1777.

Introduction

25

Zhou ,W.Q., Troy, A., Grove, M. (2008) Object-based Land Cover Classification and Change Analysis in the Baltimore Metropolitan Area Using Multitemporal High Resolution Remote Sensing Data. Sensors. 8(3), 1613-1636. Zhu, Z., Woodcock, C.E., Rogan, J., Kellndorfer, J. (2012) Assessment of spectral, polarimetric, temporal, and spatial dimensions for urban and peri-urban land cover classification using Landsat and SAR data. Remote Sensing of Environment. 117, 72-82.

CHAPTER TWO URBAN DTM GENERATION

2.1 Background Lidar is an active remote sensing technique that provides range measurements between the laser scanner and surface targets by emitting the pulses and calculating the time of flight taken for laser pulses to travel between the sensor and the targets. Combined with some georeferencing processes, such distance measurements can be turned into 3D Lidar point clouds. Relying on the accuracy of GPS and IMU components in the system, Lidar can produce data of high resolution and accuracy in both horizontal and vertical directions. Within the travel path of a laser pulse, there may be objects of different ranges that generate a series of backscattered echoes. Therefore, traditional Lidar systems record the first echo of the incoming signal (first return) and the last echo (last return) (Chauve et al., 2007). In this research, airborne Lidar data contains only the first and last return, whilst some advanced Lidar systems are able to measure up to six returns. In addition to the elevation information, most Lidar systems also record the intensity (sometimes referred to as the amplitude) of each received return, which is another useful attribute for the applications of Lidar data. Lidar systems can be carried on ground-based equipment, vehicles, and aircraft. This book mainly discusses the principles and applications of airborne Lidar data. Based on the 3D information extracted from airborne Lidar data, researchers can analyze tree volumes and building structures, and create 3D urban landscape models. Although applications of Lidar data vary, all these modelling tasks are built around one indispensable procedure: the generation of Digital Terrain Models (DTMs) using raw Lidar point clouds.

Urban DTM Generation

27

A Lidar point cloud is a set of points associated with x, y, z positional information. These points include ground points (signals returned from the terrain) and non-ground points (signals returned from objects such as trees, buildings, etc.). The elevation information from the whole point cloud (including both ground points and non-ground points) forms a Digital Surface Model (DSM) whilst elevation information from ground points forms a DTM. Since ground points are mixed with non-ground points in the point cloud, processing algorithms are needed to generate DTMs from DSMs. The use of local minima is the basis for one of the most widely used DTM generation approaches. According to this method, the lowest point in a moving window is assumed to be a ground point. By moving this window across the study area, one ground point can be selected in every cell and a DTM can be established with position information from these minimum points. (Fig 2.1)

Fig 2.1 Moving windows and the local minima

The local-minimum method works well in flat terrain with few trees and buildings, but struggles to achieve the balance between the fine resolution (requiring a small window size) and little noise (requiring a large window size to filter large buildings). This problem has attracted significant attention and many attempts have been made to generate more accurate DTMs. Kraus and Pfeifer (1998, 2001) designed an algorithm, which has been widely accepted by researchers, based on

28

Chapter Two

robust linear prediction Firstly, a rough approximation of the surface is computed using some local lowest points. Next, the residuals (oriented distances from the surface to the measured points) are calculated and each point is given a weight according to its residual. Points with a large weight will attract the surface whilst points with a small weight will have little influence on the form of the surface. The process of weight iteration continues until a stable surface is acquired or the maximum number of iterations is reached. Some other methods have also been coined following the principle of refining the output DTM step by step (Pfeifer and Stadler, 2001; Elmqvist, 2002; Wack and Wimmer, 2002; Hu, 2003). Vosselman (2000) proposed a filtering method based on the mathematical morphology and preserved terrain information by analyzing elevation differences between neighbouring points. This method was varied by Sithole (2001) and Roggero (2001a). To reduce the influence of terrain relief, Sithole (2001) introduced a local operator which can alter parameters as a function of the slope of the terrain whilst Roggero (2001a) considered local morphology by setting the values of certain terrain parameters. Another emerging research topic is to derive a DTM with the help of a triangulated irregular network (TIN). Axelsson (2000) established a TIN with local minimum points and analyzed the relationship between residual points and the TIN. If a residual point meets certain criteria, the point is included in the TIN to refine it. Following the procedure, all ground points can be added to the final TIN. Sohn and Dowman (2002) adopted a ‘downward and upward divide-and-conquer triangulation’ strategy to refine DTM iteratively. Firstly, a coarse TIN surface is established with some pre-selected ground points. Then ‘Downward divide-and-conquer’ is conducted to find points lower than the trend surface and update the TIN model using these new ground points. Next, ‘Upward divide-and-conquer’ is conducted to examine the spatial relationship between the rest of the points and the TIN. A model is adopted to find candidate ground points which can be added into the TIN surface and divide the local area into more planar terrain surfaces. When more than one candidate point exist for a planar terrain surface, a Minimum Description Length Criterion (MDL) is used to decide the most reliable points. This iteration continues until no new ground points can be added into the TIN model. Segmentation and classification are also important tools for the derivation of DTMs (Roggero, 2001b, 2002; Lohmann, 2002; Roggero, 2002; Nardinocchi et al., 2003; Brovelli et al., 2004; Tovari and Pfeifer, 2005; Sithole and

Urban DTM Generation

29

Vosselman, 2005). By putting points from the original point cloud into different categories according to classification rules, ground points can be extracted. Bartels et al. (2006a, 2006b, 2010) designed an unsupervised Lidar filtering algorithmˈSkewness Balancing. This method only calculates the skewness value of the remaining point cloud and removes the highest point if the skewness value is greater than 0. As a result, Skewness Balancing is a threshold-free algorithm, unlike most known filtering methods. Yao et al. (2008) and Bao et al. (2008) also employed Skewness Balancing and further improved this algorithm. Since many algorithms have been developed to extract DTMs from point clouds, Sithole and Vosselman (2003) designed a comparative study to evaluate the performance of these methods under different circumstances. The sample data and results of this study are available for researchers’ reference at: http://www.itc.nl/isprswgIII-3/filtertest/ index.html. With the development of Lidar application, some specific software has also been developed to assist researchers to process Lidar data. Toolbox for LiDAR data Filtering and Forests Studies (Tiffs), Terrascan, Lasground, SCOP++ and so forth, are designed to filter point clouds, generate DTMs, and extract useful information from airborne Lidar data. By classifying ground and non-ground points, this type of software can produce high quality DTMs automatically. Generating DTMs from Lidar data has thus been widely researched. However, few methods concerning urban DTM generation were conducted. Many DTM generating algorithms work efficiently in forest or suburban areas, yet fail to work well in urban areas due to the existence of particularly large and flat-topped buildings, which are usually recognized as flat and large ground areas. In addition, the requirements for output DTMs vary significantly for different applications. For instance, landscape ecologists may need Lidar-generated DTMs to assist their research whilst they have no access to commercial Lidar processing software and do not necessarily want to implement complicated algorithms (very few DTM generating methods have been integrated with common GIS software). Therefore, it remains important to design simple and effective methods for generating reliable urban DTMs.

30

Chapter Two

To meet these requirements, this book introduces an upward-fusion method (Chen et al., 2012) specifically designed for generating urban DTMs using airborne Lidar data. This method generates accurate DTMs efficiently and can easily be implemented using standard GIS tools.

2.2 Methodology Because of its simplicity and efficiency in terms of processing time, the local minimum method is widely used by researchers who would like to use DTMs as research sources but do not have access to complex and often expensive DTM generation software. However, when applied to urban areas, this approach has its limitations. If the moving window size in the method is not big enough, the lowest point in a cell will have a significant probability of being returned from a large building or tree, whose area can cover several cells. Nevertheless, if the moving window size is set very large (e.g. 50m) to minimize the influence of non-ground points, the resolution of the DTM will be relatively low with excessive smoothing and will not provide sufficiently accurate information for further studies, such as classification and modelling. To solve the problem, Zhang et al (2003) filtered non-ground points using gradually-increasing window sizes and a constant slope operator that was decided by comparing the filtered and unfiltered data iteratively. Chen et al (2007) adopted increasing window sizes and a building mask to iteratively update the elevation of each point in the point cloud. Both studies achieved satisfactory results. Nevertheless, due to the existence of various urban features, the strict setting of the slope operator and the building mask to filter very large buildings may result in some small features remaining in the output DTM. In addition, the required programming work to implement these algorithms can cause difficulties for some researchers. To easily obtain a satisfactory urban DTM, a raster-based upward-fusion algorithm is proposed to achieve accuracy as well as simplicity and efficiency. The principle of this method can be described as follows:

Urban DTM Generation

31

2.2.1 Generation of preliminary DTMs The local minimum method generates qualified DTMs with high efficiency. However, some issues exist that may lead to errors in the process of DTM generation. Generally, the main inaccuracy in the process is caused by points that are much lower than their surrounding points. For instance, Lidar system errors can result in records with abnormal elevations (Fig 2.2), which add uncertainty to the process of deriving preliminary DTMs. Although most DTM-generating algorithms include iterative refining procedure, these errors caused by low outliers can hardly be corrected through further refining. As a result, it is necessary to remove outliers in the raw point cloud before it is used for DTM generation.

Fig 2.2. A few abnormal, low-elevation points may cause inaccuracy when using the local minimum method to generate preliminary surfaces

Two approaches for filtering outliers and generating preliminary DTMs are suggested as follows: 2.2.1.1 Application of GIS tools Some algorithms have been designed for outlier detection. Outliers can be detected using distribution-based approaches (Silven-Cardenas and Wang, 2006; Meng et al, 2009), mathematical morphology methods (Kobler et al, 2007; Chen et al, 2007) and density-based methods (Breunig et al, 2000; Sotoodeh, 2007). Silven-Cardenas and Wang (2006) and Meng et al (2009) used an elevation histogram to remove points with extreme high or low elevation and a Delaunay triangulation to examine remaining outliers. For each point, Kobler et al (2007) computed the vertical difference D between the elevation of each point and the average elevation of its neighbouring points. Next, all the

32

Chapter Two

points were ranked according to D, and P percent of points with the largest negative D values were discarded as outliers. Chen et al (2007) assumed that outliers were scattered and metres lower than their neighbouring points. So those points that had a small neighbouring area and were obviously lower than their neighbouring points were deleted as outliers. Breunig et al (2000) proposed a local outlier factor (LOF) algorithm and simply calculated the local density for each object in the Lidar data set. Those points with a large LOF value were easily removed as outliers. To filter outliers that may appear at various scales, Sotoodeh (2007) employed a global and then a local outlier detection using Delaunay Triangulation (Delaunay, 1934; a Delaunay Triangulation for a set P of points in a plane is a triangulation DT such that no point in P is inside the circumcircle of any triangle in the generated graph DT), Euclidean Minimum Spanning Tree ( an EMST connects a set of points using lines such that the total length of all lines is minimized and any point can be reached from any other by following these lines) generation and Gabriel Graph (Gabriel and Sokal, 1969; a Gabriel Graph of a set of points in the Euclidean plane expresses the notion of proximity or nearness of those points) generation. Researchers can choose appropriate methods according to different terrain situations. In addition, most GIS software includes a module for estimating distribution statistics that can be used for automatic outlier detection. For instance, since these unusual low-elevation points only take up a small proportion in the whole Lidar data set, researchers can adopt a module “Quantile classification” (a classification method that distributes a set of values into groups that contain an equal number of values) integrated in some GIS software to remove 3% to 5% of the lowest points. Following this process, researchers can obtain a point cloud with limited noises. Since airborne Lidar data is not distributed regularly in a grid system, the processed Lidar point cloud needs to be interpolated to a regular grid DSM and the grid size is decided according to the resolution of raw point clouds. Although many experiments have been carried out to test a diversity of interpolation methods, Fisher and Tate (2006) pointed out that there seemed to be no preferable interpolation algorithm for all terrain situations. Therefore, researchers can choose suitable interpolation method (such as Inverse Distance weighed)

Urban DTM Generation

33

according to terrain characteristics. Based on the interpolated DSM, the local minimum method can be used to generate several preliminary DTMs of variable grid sizes (For instance, DTM_50m, DTM_30m, DTM_10m and DTM_2m) for the following upward-fusion. 2.2.1.2 Extended local minimum method The function that supports distribution statistics in GIS tools enables researchers to filter most outliers automatically. However, this method works efficiently to detect global extreme values, but some points with a local extreme value may be retained. To better detect local outliers, morphology-based or density-based operators can be adopted. In addition to algorithms introduced above, this chapter introduces an extended local minimum method to generate preliminary DTMs as well as filter abnormal points. Since the standard local minimum method uses the lowest point in a cell to decide the elevation of its local area, extremely low outliers cause serious errors in a trend surface. Due to the large difference between the erroneous trend surface and the real terrain, this type of errors can hardly be corrected through further process. To reduce the influence of outliers in the trend surface, DTMs of large grid sizes (e.g. DTM_50m, DTM_30m) that serve as trend surfaces can be generated using an extended local minimum method. (The preliminary DTMs of medium and fine scale can still be generated using the standard local minimum method, as outliers retained in these preliminary DTMs can be corrected using trend surfaces). In the extended version, several low-lying points, instead of only the lowest point in a cell, are selected and analyzed to better describe the local terrain. By analyzing the positional relationship between these lowest points, problematic, unusual terrain situations can be detected. Then a proper elevation value can be assigned for every cell according to some specific rules. Take one possible rule for instance. Researchers can analyze the elevation difference between the lowest point and the second lowest point. If this difference is smaller than a given threshold Ȝ, then the elevation of the lowest point can be used to set the value of this cell. If the difference is larger than Ȝ, then the lowest point may be an outlier. In this case, the lowest point is discarded and the second lowest point will be set as the candidate point and compared with the

34

Chapter Two

third lowest point following the same procedure (Fig 2.3). Iteration continues until an appropriate point is found and the value of the cell is decided using the elevation of this point. By adopting this rule, above-mentioned noises can be significantly reduced.

Fig 2.3 Analyzing the relationship between several lowest points can reduce the error caused by abnormal low-elevation points.

According to local terrain, the resolution of Lidar data set, the structure and density of housings and trees, the approach of analyzing the relationship between the lowest points and then calculating a proper value for the output DTM can be flexible. Choosing an appropriate approach to generate preliminary DTMs can greatly improve the accuracy of following upward-fusion. For instance, in a flat terrain, analyzing the difference between the lowest points and selecting the appropriate point will work efficiently to reduce noises from limiting cases. In some undulating terrain where laser returns may not be dense enough, considering only one low-lying point may cause excessive smoothing and the loss of terrain details (which cannot be refined if occurs in large-scale preliminary DTMs). In this case, employing the mean value of these low-lying points in a large-scale grid may have a better effect than the sole use of one chosen point. The extended local minimum method also has the potential to generate qualified DTMs for complicated terrain situations. The landscape of some cities in the UK and other countries can be very hilly. Hence, if it is difficult to use only one value to feature the elevation of a large cell, some procedures, such as re-interpolation using multiple low-lying points, can be conducted for better preliminary DTMs.

Urban DTM Generation

35

In addition to analyzing the points in a single cell, the elevation of neighbouring cells can also be examined in the extended local minimum method to reduce serious noises. As discussed, since a preliminary DTM of large grid size will be used as the trend surface for further processing, the grid size can neither be set too small (or the result will be influenced by buildings and other non-ground features) nor too large (or the result will not be accurate enough). Generally, preliminary DTMs can be generated with satisfactory accuracy using the grid size 40m×40m or 50m×50m. However, in some metropolitan areas, there may be a few large buildings that cover more than one complete cell. In this case, the elevation value of this cell features the height of the building, instead of the terrain elevation. By analyzing the difference between neighbouring cell values, obvious noises can be detected and the value of this exceptional cell can be set using the mean value of its unaffected neighbouring cells.

2.2.2 Upward-fusion After several preliminary DTMs have been generated using GIS tools or the extended local minimum method, upward-fusion can be started using the preliminary DTM with the largest grid size. The resolution of preliminary DTMs is flexible in accordance with the terrain situation. In this case study, the resolution of preliminary DTMs is set as 50m, 30m, 10m and 2m. The procedure of upward-fusion is as follows: a. The analysis starts with DTM_50m and DTM_30m; the elevation of a point (x, y) on a refined DTM_30m+ can be decided by analyzing the elevation difference between DTM_50m and DTM_30m at the location (x, y): If the difference ǻh=DTM_30m(x, y)-DTM_50m(x, y) < ĭ1 (a given threshold aiming to filter points from large buildings), then this difference is most likely to reflect more accurate terrain details from a fine resolution. Thus, DTM_30m+(x, y) = DTM_30m(x, y); If h ı ž1, the difference may result from non-ground objects and the refined DTM should retain the elevation value from the coarse surface. Then DTM_30m+(x, y) = DTM_50m(x, y).

36

Chapter Two

By conducting image fusion following this principle, a refined DTM_30m+, whose resolution is 30m, is acquired (Fig 2.4a). b. By analogy, upward-fusion is continued with DTM_30m+ and DTM_10m. If the difference ǻh=DTM_10m(x, y)-DTM_30m+(x, y) < ĭ2 (a given threshold aiming to filter points from buildings and large trees, smaller than ĭ1), then DTM_10m+(x, y)=DTM_10m(x, y); If h ı ĭ2, DTM_10m+(x, y)= DTM_30m+(x, y). A further refined DTM_10m+, whose resolution is 10m, can be acquired using upward-fusion. (Fig 2.4b) c. The fusion can be processed further with DTM_10m+ and DTM_2m. If the difference ǻh=DTM_2m(x, y)-DTM_10m+(x, y) < ĭ3 (a given threshold aiming to filter points from cars, trees and other urban features, smaller than ĭ1 and ĭ2) , then DTM_2m+(x, y)=DTM_2m (x, y); If h ı ĭ3, DTM_2m+(x, y)= DTM_10m+(x, y). Finally, DTM_2m+, with the resolution of 2m, is established. (Fig 2.4c) This procedure of upward-fusion is shown in Fig 2.4 and the flowchart of the upward-fusion DTM generation method is shown in Fig 2.5.

Urban DTM Generation

a. The fusion of coarse DTM and large DTM

b. The fusion of DTM+ and M_DTM

c. The fusion of DTM++ and F_DTM Fig 2.4 Upward-fusion of preliminary DTMs

37

38

Chapter Two

Fig 2.5 The flowchart of the upward-fusion method

This method has several advantages, making it an efficient algorithm for urban DTM generation. a. Accuracy. Following the principles of other methods, the upward-fusion method uses a large-scale, coarse resolution DTM to

Urban DTM Generation

39

minimize the influence from buildings and trees in urban areas and form an approximate terrain surface. Successively finer DTMs are compared to this surface. Thus, a series of increasingly accurate DTMs is established by updating qualified elevation values from the finer DTMs, and retaining the values from the original DTMs when the elevation value from the finer DTM is beyond a threshold. Iterative refinement continues until a satisfactory resolution is achieved. Theoretically, the best resolution DTM can approximate the horizontal resolution of the original Lidar point cloud. The high accuracy of this method is assessed in the following section with field-collected data. b. Flexibility. Since terrain in urban areas varies significantly, this methodology enables researchers to decide a series of parameters to fit particular study sites by conducting experiments and fieldwork. These conditions include the number and resolution of preliminary DTMs, the values of fusion thresholds and the resolution of the final, refined DTM. c. Simplicity and Efficiency. Interpolation and neighbourhood (moving window) analysis functions are available in most GIS software, so it is not difficult for researchers to acquire the necessary, preliminary DTMs of different grid sizes. Furthermore, since the fusion procedures do not include complicated analysis such as matrix calculation, this algorithm is very fast to execute. In the experiment described below, it took only minutes to conduct fusion analysis using a dataset of more than 10 million points.

2.3 Study site and data preparation The accuracy of the output DTM is the most crucial criterion for evaluating DTM generation algorithms. The availability of high-quality airborne Lidar data, as well as ground reference data, is an important factor for the choice of the study site. Considering the data availability, the typicality of the terrain situation, and the possibility of collecting reliable reference points, Cambridge, UK was selected for the case study of urban DTM generation using airborne Lidar data. Cambridge is a world-famous university town, which lies in East Anglia about 50 miles north-by-east of London, UK. The terrain of Cambridge is generally flat although there are obvious elevation

40

Chapter Two

changes in some parts of the city. Cambridge has a very high density of buildings, trees and shrubs, which is a typical urban situation and can cause common difficulties in generating urban DTMs. Therefore, if the upward-fusion method works well in Cambridge, it has the potential to be tranferred to most urban terrains. The airborne Lidar data used in this case study were from an ALTM-3033 Lidar System carried on a Piper Chieftain aircraft. The survey was carried out in central Cambridge in June, 2009 and the average height of this flight was 800m. The Lidar point cloud included about 10 million records (a comparatively small data set in airborne Lidar terms) and the horizontal resolution was 2 points/ m2 on average. The main tasks of data preparation were to integrate separate tiles of Lidar records into a complete file, and to convert the original data format (such as .txt) to other formats (such as Shapefile, Grid and so on) for the convenience of data processing.

2.4 Results Unusual terrain situations and system errors may cause points with comparatively low elevation to bias the generation of preliminary DTMs, especially DTMs of large grid size. Global outliers can be easily filtered using a distribution statistics module, but some local outliers may be retained. As displayed in Figure 2.6a, these abnormal points lead to a cell with a much lower elevation value than its neighbouring cells (so the color of the cell is obviously darker on the map). Since these unusual low-elevation points take up only a small proportion of the whole Lidar data set, this type of error can be modified easily through manual editing. In addition to manual editing, the extended local minimum method can filter these outliers automatically and smoothly. The main improvement with the extended version is that several lowest points in every cell are analyzed following certain rules to calculate an appropriate value, instead of the elevation of the lowest point, to decide the elevation of the cell. In this case study, the rule was set as follows: Ten lowest points were selected in each cell and an appropriate elevation value was calculated for every cell. Next, another process to reduce noises was conducted.

Urban DTM Generation

41

The elevation values of every four neighbouring cells were compared. If one cell had an elevation that differed significantly with three neighbors’, the elevation of this cell would be replaced by the average elevation value of the three neighbouring cells. Following these processes, advanced preliminary DTMs were acquired. As Figure 2.6b shows, errors related to the low points were removed efficiently. As a result, the extended local minimum method was an ideal way to avoid errors in the output DTM.

a. Errors in preliminary DTMs generated using standard local minimum method

b. Errors removed in preliminary DTMs generated using extended local minimum method

Fig 2.6 Errors removed using extended local minimum method

Through the local minimum method, four preliminary DTMs (50m, 30m, 10m, 2m) were obtained (as Fig 2.7 shows) for upward-fusion. The upward-fusion thresholds ĭ1, ĭ2, ĭ3 were set by taking experiments and finding the optimal parameters according to the visual effect of generated DTMs. Due to comparatively flat terrain in central Cambridge, the thresholds were chosen as follows: ĭ1=1.8m, ĭ2= 0.5m, ĭ3=0.1m Then the upward fusion process was conducted and a refined DTM_2m+ with the resolution of 2m was created (Fig 2.8).

Chapter Two

42

Preliminary DTM_50m

Preliminary DTM_30m

Preliminary DTM_10m

Preliminary DTM_2m

Fig 2.7 Preliminary DTMs of different grid sizes generated using the local minimum method

Urban DTM Generation

43

Hillshade image of DSM

Hillshade image of output DTM Fig 2.8 Output DTM generated using upward-fusion method

As shown in Fig 2.8, the zoomed-in area is the city center of Cambridge, which has many tall trees and buildings. By comparing the images of the DSM and generated DTM, one can see that upward-fusion successfully removes the influence from these non-ground objects and retains most terrain details.

Chapter Two

44

2.5 Accuracy assessment Some Lidar processing software with advanced DTM generating algorithms is designed to generate DTMs using Lidar point clouds automatically. To compare the accuracy and efficiency of the upward-fusion method with these existing tools, this case study also processed the same data set using some leading Lidar software: Tiffs, Terrascan and Lasground. All these three types of software can automatically produce DTMs from raw Lidar points and the only key parameter that users need to set is the largest building size in the data set. To decide the suitable value of this parameter for each tool, one can gradually increase this value and check when the traces of the largest buildings are completely removed. In this research, this value was set as 60m for all software. The results of DTM generation were shown in Fig 2.9. To compare the accuracy of output DTMs generated using the upward-fusion method and other tools, 100 reference points were employed for quantitative accuracy assessment. By analyzing the terrain difference between the real elevation value (acquired from the ground-collected data) and the estimated value acquired from these generated DTMs (Fig 2.9), the accuracy of different DTM generation algorithms can be examined.

Grid image of DSM

Hillshade image of DSM

Urban DTM Generation

Tiffs generated DTM

Lasground generated DTM

Terrascan generated DTM

Upward-fusion generated DTM

45

Fig 2.9 DTMs generated with upward-fusion method and some leading software

The ground control points for the accuracy assessment were mainly collected using a Leica RTK (Real - Time Kinematic) GPS tool. With the real-time correction from UK Smartnet, the vertical and horizontal error in the RTK GPS can be constrained to within 0.5cm. Since RTK GPS fails to measure the elevation of some sites, such as the edge of large buildings or trees due to limited available satellites, some manually filtered points (from the raw point cloud) were also used as control points. All these reference points were selected across the study site, including sufficient sites close to buildings, trees and undulating terrain, where DTM generating algorithms were most likely to perform

46

Chapter Two

badly. Fig 2.10 and Fig 2.11 shows the error distribution of output DTMs generated using the upward-fusion method and other tools. Table 2.1 gives the detailed statistics of the accuracy assessment, which includes the mean error, the largest error, the number of reference points that show large error and the filtering efficiency. Lasground and Terrascan produced smooth and satisfactory DTMs, but noises still existed. Terrascan efficiently removed the influence from large buildings and trees whilst some small non-ground objects were retained. Lasground adopted a fine filtering scale and filtered most non-ground objects. Therefore, the proportion of points with satisfactory accuracy was high. However, its disadvantage was that some traces of large buildings were retained in the generated DTM, resulting in several areas with serious bias. An experiment was carried out based on two generated DTMs. When upward-fusion was conducted between DTMs generated using Lasground and Terrascan (Fig 2.12.a), 5 out of 9 points with large error ( absolute deviation >0.5m) were filtered and the mean bias of Lasground was improved from 0.254 m to 0. 114 m Tiffs did not produce a smooth DTM as many small non-ground objects were included. When the value of the parameter “the largest building size” was adjusted to 40m, Tiffs removed most small non-ground objects, yet retained some traces of large buildings, which were reduced efficiently when the value of the largest building size was set as 60m. To improve the performance of Tiffs, upward-fusion was carried out between DTMs generated using Tiffs with different parameter values. The result showed that 16 out of 24 points with large error (absolute bias larger than 0.5m) were filtered and the mean bias was improved from 0.755 m to 0.206 m. (Fig 2.12.b)

Urban DTM Generation

47

Ground control points

Tiffs generated DTM

Terrascan generated DTM

Lasground generated DTM

Upward-fusion generated DTM

Fig 2.10 Accuracy comparisons between DTMs generated using different methods

0

10

20

30

40

50

60

70

(-2,-1)

Fig 2.11 Error distribution of DTM generation methods [-0.5,-0.2)

[-0.2,0)

no bias

(0,0.2]

(0.2,0.5]

(0.5,1]

Elevation bias between model estimated value and real value (m)

[-1,-0.5)

(1, 8]

Miffs Lasground Terra Upward-fusion

48 Chapter Two

Frequency

Urban DTM Generation

49

Table 2.1. Error statistics of DTMs generated using different methods 0-0.2m

0.2-0.5m

0.5-1m

Error >1m

Largest Error

Tiffs

63

13

6

18

8m

Lasground

88

3

5

4

7.5m

Terrascan

82

5

9

4

2.4m

Upward-fusion

89

4

6

1

1.4m

Mean Error

Adjusted (exclude error>1m) mean error

Adjusted Standard deviation

Filtering time

Tiffs

0.755m

0.289m

0.85m

98s

Lasground

0.254m

0.14m

0.30m

78s

Terrascan

0.211m

0.12m

0.17m

73s

Upward-fusion

0.115m

0.093m

0.16m

71s

a. Upward-fusion between the DTMs generated using Lasground and Terrascan

b. Upward-fusion between DTMs generated using Tiffs with different parameter values

Fig 2.12. Upward-fusion can be used to improve DTMs generated using other methods

50

Chapter Two

Since the upward-fusion method used different scales to filter the DSM, it successfully removed large buildings as well as small non-ground objects and produced satisfactory DTMs. The mean error and the proportion of satisfactory records proved the reliability of this method. Different from other algorithms, the upward-fusion method is more likely to underestimate rather than overestimate terrain elevation. The ratio of overestimated points to underestimated points for upward-fusion was 39:46, compared with 77:12 (Tiffs), 72:18 (Lasground) and 66:22 (Terrascan). The mean bias of underestimation in DTMs generated by the upward fusion method was 0.12m. Amongst those overestimated points created by upward-fusion, the mean bias was 0.14m, compared with 0.89m (Tiffs), 0.30m (Lasground) and 0.23m (Terrascan). The upward-fusion method works well to filter urban features as well as retain ground surface morphology information. Nevertheless, compared with Terrascan and Lasground, this method is more likely to cause error on undulating terrain. An effective approach to reduce this type of error is to include more preliminary DTMs for upward-fusion and choose more appropriate upward-fusion parameters. Besides, since interpolation is conducted at the first stage in the upward-fusion method whilst interpolation is the last process in most algorithms, some parts of the DTMs generated using upward-fusion may not be as smooth as DTMs generated using other methods. Not all software has an integrated function for interpolation and experiments proved that it took equal time for these methods to interpolate Lidar points to a DSM (upward fusion) or a DTM (Tiffs, Terrascan, Lasground) using the same algorithm. As a result, the time efficiency can be analyzed using filtering time. These methods were all highly time-efficient and it took less than two minutes to filter a set of airborne Lidar data with more than 3 million points. Therefore, Tiffs, Lasground, Terrascan and upward-fusion are all ideal DTM generating tools for large projects.

2.6 Discussion Based on an upward-fusion methodology, this research conducted urban DTM generation using airborne Lidar data. The results indicated that upward-fusion was an effective approach for deriving DTMs

Urban DTM Generation

51

because of its simplicity, efficiency and accuracy. No additional programming work was required to adopt the methodology because all processes could be realized in existing GIS software (If researchers choose the extended local minimum method or other methods for the outlier detection, some simple programming is necessary). This makes it very convenient for a broad spectrum of users. In addition, since this methodology did not rely on complex filtering algorithms, it took much less processing time, which is a key factor in DTM generation from Lidar data. Although the study area is a comparatively small one, the dataset still consisted of 10 million points. It was possible to generate a DTM using the upward-fusion algorithm within an elapsed time of one hour. The accuracy of the output DTM generated using the upward-fusion method was satisfactory. Compared with some leading Lidar processing software, the upward-fusion method generated DTMs with the least mean bias and the least points with large bias. In addition, the upward-fusion method can also be conducted with DTMs generated using other DTM generating methods. Experiments suggested that several types of errors were produced when the value of some key parameters was varied in different DTM generation algorithms or software. When upward-fusion was conducted between DTMs generated using different methods or different parameter values, the number of points with large bias and the mean bias were both reduced significantly. The research on upward-fusion urban DTM generation is on-going. The optimization of different parameters (the number and grid size of preliminary DTMs, thresholds for DTM fusion and so on), the choice of rules in the extended local minimum method to better suit different terrain situations and some other processes in this approach are all important issues. In future studies, further experiments can be carried out to improve the current method, including: a. Experiments can be conducted to assess the methodology in urban areas with intensively changing terrains, and to analyze how parameters should be adjusted in the process of upward-fusion or how rules should be set in the extended local minimum method to achieve optimal urban DTMs. b. Threshold free upward-fusion may be feasible with more specific statistics included. For instance, if the average height of buildings,

52

Chapter Two

trees, shrubs and other non-ground objects can be calculated through statistical analysis in the process of upward fusion, the thresholds can be set according to these results automatically. In this case, a threshold free upward-fusion method may become feasible. c. Some DTM generation methodologies achieve satisfactory results on undulating terrain whilst the upward-fusion method has its limitation in such situations. As a result, selecting appropriate DTM generation methods, instead of the local minimum method, for producing preliminary DTMs, can be an efficient way to improve the accuracy of output DTMs. d. Although the noise from large buildings can be reduced effectively by adopting large scale preliminary DTMs, some errors may also be caused in the following procedure of upward-fusion. To better remove the noise from large non-ground features without losing terrain details, preliminary DTMs with large grid sizes may be replaced by some other materials, such as building masks. Masks for buildings, especially large buildings, can be acquired using appropriate image segmentation or classification methodologies. By employing this type of building masks to filter small or medium scale preliminary DTMs, the influences from buildings or other large non-ground features can be reduced whilst terrain details can be better reserved.

2.7 Conclusions Despite its simplicity, the upward-fusion method can be used to generate urban DTMs with very high accuracy and efficiency. Algorithms of DTM generation commonly blunder by failing to identify particularly large, flat-topped buildings, but in the upward-fusion method, these types of error can be efficiently removed by adopting a preliminary DTM with a large grid size as the coarse surface. By fusing a large scale preliminary DTM with a fine scale DTM, upward-fusion also successfully avoids errors from small non-ground objects. The results of accuracy assessment showed that the bias in the experimental DTMs was, on average, less than 12cm. Considering the high density of buildings and trees in the study site, which is a common landscape situation in urban areas, this case study was very typical. The result of this case study suggested that the

Urban DTM Generation

53

upward-fusion DTM generation methodology has the potential to be applied to other urban areas. In addition, experiments proved that the upward-fusion method could be used to improve the accuracy of other DTM generating algorithms. This upward-fusion method provides researchers with an effective approach to generate high-quality urban DTMs, based on which the vertical structure of urban features can be further extracted for establishing 3D landscape models.

References Axelsson, P. (2000) DEM generation form laser scanner data using adaptive TIN models. International Archives of Photogrammetry and Remote Sensing. XXXIII (part B4), 110-117. Bao, Y.F., Li, G.P., Cao, C.X., Li, X.W., Zhang, H., He, Q.S, Bai, L.Y., Chang, C.Y. (2008) Classification of LIDAR point cloud and generation of DTM from LIDAR height and intensity data in forested area. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. XXXVII (part B/3b), 313–318. Bartels, M., Wei, H. (2006a) Segmentation of LIDAR data using measures of distribution. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. XXXVI (part 7), 426–431. Bartels, M., Wei, H., Mason, D. C. (2006b) DTM generation from LIDAR data using Skewness Balancing. In: 18th International Conference on Pattern Recognition, 20-24 August 2006, Hong Kong, China, I, pp, 566–569. Bartels, M., Wei, H. (2010) Threshold-free object and ground point separation in LIDAR data, Pattern Recognition Letters. 31(10), 1089-1099. Breunig, M. M., Kriegel, H.P., Ng, R.T., Sander J. (2000) LOF: Identifying density-based local outliers. In: Proceedings of the ACM SIGMOD International Conference on Management of Data, 2000. pp, 93–104. Brovelli, M. A., Cannata, M., Longoni, U. (2004) Lidar data filtering and dtm interpolation within grass. Transactions in GIS. 8(2), 155–174. Chauve, A., Mallet, C., Bretar1, F., Durrieu, S., Deseilligny, M.P.,

54

Chapter Two

Puech, W. (2007) Processing full-waveform Lidar Data: Modelling Raw signals. ISPRS Workshop on Laser Scanning 2007 and SilviLaser 2007, Espoo, September 12-14, 2007, Finland. Chen, Q., Gong, P., Baldocchi, D., Xie, G. (2007) Filtering airborne laser scanning data with morphological methods, Photogrammetric Engineering & Remote Sensing, 73(2), 175-185. Chen, Z.Y., Devereux, B., Gao, B.B., Amable, G (2012) Upward-fusion urban DTM generating method using airborne Lidar data. ISPRS Journal of Photogrammetry and Remote Sensing. 72, 121-130. Delaunay, B (1934): Sur la sphère vide, Izvestia Akademii Nauk SSSR, Otdelenie Matematicheskikh i Estestvennykh Nauk. 7, 793–800. Elmqvist, M. (2002) Ground surface estimation from airborne laser scanner data using active shape models. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. XXXIV ( part 3/W3), 114–118. Fisher, P. F., Tate, N. J. (2006) Causes and consequences of error in digital elevation models. Progress in Physical Geography. 30, 467–489. Gabriel, K. R., Sokal, R. R. (1969), A new statistical approach to geographic variation analysis. Systematic Zoology (Society of Systematic Biologists). 18(3): 259–270. Hu, Y. (2003) Automated extraction of digital terrain models, roads and buildings using airborne lidar data. PhD Thesis. University of Calgary, Canada. Kobler, A., Pfeifer, N., Ogrinc, P., Todorovski, L., Ostir, K., Dzeroski, S. (2007) Repetitive interpolation: a robust algorithm for DTM generation from aerial laser scanner data in forested terrain. Remote Sensing of Environment. 108 (1), 9-23. Kraus, K., Pfeifer, N. (1998) Determination of terrain models in wooded areas with airborne laser scanner data. ISPRS Journal of Photogrammetry and Remote Sensing. 53, 193–203. Kraus, K., Pfeifer, N. (2001) Advanced DTM generation from LIDAR data. International Archives of Photogrammetry and Remote Sensing, Annapolis, MD, 22-24 Oct. 2001, XXXIV (part 3/W4): 23–30. Lohmann, P. (2002) Segmentation and filtering of laser scanner digital surface models. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. 34 (part XXX), 311–315.

Urban DTM Generation

55

Myint, S. W., Gober, P., Brazel, A., Clarke, S.G., Weng, Q.H. (2011) Per-pixel vs. object-based classification of urban land cover extraction using high spatial resolution imagery. Remote Sensing of Environment. 115(5), 1145-1161. Meng, X., Wang, L., Silvan-Cardenas, J.L., Currit, N. (2009) A multidirectional ground filtering algorithm for airborne Lidar. ISPRS Journal of Photogrammetry and Remote Sensing. 64(1), 117-124. Nardinocchi, C., Forlani, G., Zingaretti, P. (2003) Classification and filtering of laser data. International Archives of Photogrammetry and Remote Sensing. XXXIV (part 3/W13). Pfeifer, N., Stadler, P., Briese, C. (2001) Derivation of Digital Terrain Models in the SCOP ++ Environment. In Proceedings of the OEEPE Workshop on Airborne Laser scanning and Interferometric. SAR for Digital Elevation Models, Stockholm, Sweden. Powell, R.L., Robertsa, D.A., Dennison, P.E., Hess, L.L. (2007) Sub-pixel mapping of urban land cover using multiple endmember spectral mixture analysis: Manaus, Brazil. Remote Sensing of Environment. 106(2), 253-267. Roggero M. (2001a) Dense DTM from laser scanner data, Proceedings of OEPEE Workshop on airborne laser scanning and Interferometric SAR. Stockholm Roggero, M. (2001b) Airborne laser scanning: clustering in raw data, International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, XXXIV (part 3/W4), 227-232. Roggero, M. (2002) Object segmentation with region growing and principal component analysis. ISPRS Commission III Symposium. Saunders, D,A.. Hobbs, R.J., Margules, C.R. (1991) Biological consequences of ecosystem fragmentation: a review. Conservation Biology. 5, 18-32. Silvan-Cardenas, J. L., Wang, L (2006) A multi-resolution approach for filtering Lidar altimetry data. ISPRS Journal of Photogrammetry and Remote Sensing. 61(1), 11-22 Sithole, G. (2001) Filtering of laser altimetry data using a slope adaptive filter. International Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences. XXXIV (part 3/W4), 203–210. Sithole, G., Vosselman, G. (2003) Comparison of filtering algorithms. International Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences 34 (Part 3/W13), 71–78.

56

Chapter Two

Sithole, G., Vosselman, G. (2005) Filtering of airborne laser scanner data based on segmented point clouds. Workshop Laser scanning 2005, pp, 66–71. Sohn, G., Dowman, I. (2002) Terrain surface reconstruction by the use of tetrahedron model with the mdl criterion. International Archives of Photogrammetry and Remote Sensing. XXXIV (part 3A), 336–344. Sotoodeh, S. (2007) Hierarchical clustered outlier detection in laser scanner point clouds. International Archive of Photogrammetry and Remote Sensing. 35 (part 3/W52), 383-388. Tovari, D., Pfeifer, N. (2005) Segmentation based robust interpolation-a new approach to laser data filtering. Workshop Laser scanning. 2005, pp, 79–84. Vosselman, G. (2000) Slope based filtering of laser altimetry data. International Archives of Photogrammetry and Remote Sensing. XXXIII (part B3/2), 935–942. Wack, R., Wimmer, A. (2002) Digital terrain models from airborne laser scanner data- a grid based approach. International Archives of Photogrammetry, Remote Sensing and Spatial Information Sciences. 34(part 3B), 293–296. Yao, W., Hinz, S., Stilla, U. (2008) Automatic vehicle extraction from airborne LiDAR data of urban areas using morphological reconstruction. In: 5th IAPRS Workshop on Pattern Recognition in Remote Sensing (PRRS 2008). p. 4. Zhang, K.Q., Chen, S.C., Whitman, D., Shyu, M.L., Yan, J.H., Zhang, C.C. (2003) A progressive morphological filter for removing nonground measurements from airborne LIDAR data, IEEE Transactions on Geoscience and Remote Sensing. 41(4): 872–882.

CHAPTER THREE URBAN LAND COVER CLASSIFICATION

3.1 Background An image of land cover classification can be directly employed for 2D landscape pattern analysis using specific methods or tools (e.g. Fragstats). Integrated with the vertical pattern information, classified land cover types can be further adopted for 3D landscape pattern analysis. Since pattern analysis is a significant procedure for landscape ecology research, the methodology of land cover classification is of great importance for modelling 3D landscape ecology.

3.1.1 Land cover classification using airborne Lidar data Airborne Lidar data is traditionally used to classify ground and non-ground points (Cobby et al., 2001; Vosselman, 2000; Lohmann et al., 2000) and some methods have been designed to detect vegetation or building features from Lidar point clouds using topological reasoning (Heuel et al., 2000; Lohmann, 2002), morphology-based classification (Nardinocchi et al., 2003; Meng et al, 2009), robust interpolation (Tovari and Pfeifer, 2005) and specific indicators (Miliaresis and Kokkas, 2007; Yu et al, 2010). As well as elevation information, most Lidar systems also record the intensity of each received return. Since these returns were generated from different surface types within the travel path of laser pulses, the intensity of these returns serves as an important variable to classify different land cover types. The intensity of laser pulses indicates the reflectance characteristics of the surface targets in the near infrared spectra between wavelengths of 800 nm and 1550 nm (Bao et al, 2008). Since different land cover features usually have a diversity of reflectance, Flood (2001) and Song et al (2002) pointed out that the intensity attribute of airborne Lidar data could also provide important information for land cover classification. In the past decade, many

58

Chapter Three

researchers employed the intensity for tree species classification (Holmgren and Persson, 2003; Moffiet et al., 2005; Ørka et al., 2007; Donoghue et al. 2007; etc.). Since elevation information from airborne Lidar data can offset the missing attribute of remote sensing and airborne photographic images, researchers are placing growing emphasis on conducting land cover classification by integrating airborne Lidar data with other data sources. Multi-spectral images (Rottensteiner et al., 2005; Bork and Su, 2007; Koetz et al., 2008), hyperspectral data (Csatho et al. 2003; Dalponte et al., 2012), high resolution imagery (Lee and Shan, 2003; Teo and Chen, 2004; Syed et al., 2005; Chen et al, 2009; Cook et al., 2009; Ke et al, 2010; Arroyo et al., 2010), and airborne photography (Hill and Thomson, 2005; Mesas-Carrascosa et al., 2012) are fused with airborne Lidar data to achieve better classification accuracy.

3.1.2 Object-based classification Although pixel-based land cover classification has developed for decades and both its theory and applications have experienced significant progress, shortcomings still exist. For instance, different land cover types may have the same spectrum whilst the reflection from the same land cover type may appear with different spectra under changing situations. Even if an ideal classification model can be established to distinguish all land cover types, it is still very difficult, if not impossible, to classify every pixel into a definite type due to the inevitable existence of mixed pixels. Although algorithms such as minimum distance and maximum likelihood (Curran, 1985; Campbell, 1987; Richards, 1993) may be used to decide the types of some mixed pixels, these methods may result in serious salt and pepper noise (salt and pepper noise, which describes a tendency to produce randomly occurring white and black pixels) in land cover classification. Object-based land cover classification attempts to conduct classification from a macro perspective. The first step is to group similar neighbouring pixels into multi-pixel objects and then classify these objects into different types according to their spectral and spatial characteristics (Aplin et al., 1999; Baatz and Schape, 2000). Since the classification procedure is based on objects that are a group of neighbouring pixels, this method can efficiently minimize the salt and

Urban Land Cover Classification

59

pepper noise and obtain more homogeneous classification results. Theoretically, the useful attributes from airborne Lidar data for land cover classification are only elevation and one band (as compared to tens, even hundreds of bands in multi-spectral remote sensing images) of intensity. As a result, it is not practical to conduct pixel-based land cover classification using airborne Lidar data. Nevertheless, owing to its high resolution in the horizontal direction, the neighbouring points in a Lidar point cloud are highly correlated in terms of both the elevation and intensity, which makes Lidar data an ideal data source for object-based classification. More and more studies have been conducted using the object-based method. Based on the characteristics of building structures, specific approaches have been coined to extract building features from Lidar point clouds. Teo and Chen (2004) fused Lidar data and QuickBird orthoimage to detect buildings using an object-based method. High resolution imagery (Syed et al. 2005; Chen et al. 2009) has also been integrated with Lidar data to carry out the object-based land cover classification in urban areas. Ke et al (2010) employed object-based forest species classification using Lidar and QuickBird multispectral imagery. However, since useful attributes of airborne Lidar data are limited, few researchers used airborne Lidar data as the sole source for comprehensive land cover classification. Johansen et al. (2010) monitored environmental conditions of riparian zones using airborne Lidar data by assessing some riparian condition indicators. The empirical modelling results proved the high accuracy of object-based classification. Antonarakis et al. (2008) integrated vegetation height models, percentage canopy hit models, intensity models and skewness and kurtosis models for object-based land cover classification in forest areas using airborne Lidar data. Johansen and Antonarakis’ research presented efficient approaches for object-based land cover classification, but the study sites were not in urban areas. In this case, these classification methods did not include any criteria to differentiate trees and buildings, a key issue in urban land cover classification. Brennan and Webster (2006) employed multiple Lidar derived surfaces to classify 10 land cover types based on object-oriented classification. In this study, they successfully classified the land cover type of non-tree structures using the attribute of multiple echoes. Nevertheless, since the research was conducted in a coastal area with only a few

60

Chapter Three

scattered buildings, the classification rules could hardly be efficiently applied to urban areas with many more clusters of mixed trees and buildings. Hence, more specific classification methodologies are required. Classified land cover types are required for landscape pattern analysis and landscape ecology research. Since airborne Lidar data is indispensable for establishing 3D landscape models whilst other high-resolution data sources may not be always available, it is of practical importance to propose a methodology for urban land cover classification using airborne Lidar data solely. At the object level, this study analyzes the differences, which may not be detected at the pixel level, between a diversity of land cover types in airborne Lidar data and explores the possibility of establishing an efficient and applicable methodology for urban land cover classification based on these unique characteristics.

3.2 Methodology The object-based urban land cover classification method (Chen and Gao, 2014) introduced in this research is to be carried out in two steps. Firstly, several thematic images are produced using those attributes that can distinguish different land cover types. Following this, these images are employed as hierarchical layers for image segmentation. Next, specific classification rules for different land cover types are set and all objects are classified into corresponding land cover types.

3.2.1 Thematic images for segmentation Since Lidar data can offer only two main attributes, many researchers (Teo and Chen, 2004; Syed et al., 2005; Chen et al., 2009; Ke et al., 2010; etc.) fused Lidar data with multi-spectral images for better classification results. However, information from the Lidar point cloud has not been fully utilized and other useful attributes can also be extracted. Thus, satisfactory classification results may be achieved using airborne Lidar data as the sole source when other high resolution data are not available. In this research, several thematic images are produced using useful attributes for segmentation, and these images are explained below.

Urban Land Cover Classification

61

3.2.1.1 Height image Different land cover types have a variety of height ranges, so height information is an important factor for land cover classification. However, the elevation attribute from Lidar points cannot be used as height information without processing. As introduced above, a Lidar point cloud includes ground points and non-ground points. Elevation information from the entire point cloud, both ground points and non-ground points, forms the DSM. The DSM cannot be used for classification, because elevation is the height above a fixed reference point, not the height above the real terrain. To avoid the influence from terrain differences, the height information (namely nDSM, normalized DSM information), instead of the elevation information, is more appropriate for land cover classification. The difference between the DSM image and the image of height information is shown in Fig 3.1.

Image of DSM

Image of the object height

Fig 3.1 An image of the DSM and the corresponding nDSM image

The height information of urban features can be calculated simply by subtracting the DTM from the DSM. The DSM can be acquired directly through interpolation on all Lidar points whilst the DTM can be generated using a variant of above-mentioned DTM generating methods, as well as Lidar processing tools. With the height information, preliminary classification results that include low land cover types (such as lawns, bare ground, etc.), medium land cover types (such as shrubs, short trees, etc.) and high land cover types (such as tall trees and buildings) can be achieved.

62

Chapter Three

3.2.1.2 Intensity image The intensity information is of great importance for land cover classification (Flood, 2001; Song et al, 2002), but this attribute cannot provide researchers with a classification rule, based on which every land cover type can be placed into a non-overlapping intensity range. For instance, since both trees and buildings occupy a very wide and overlapping range of intensity values, it is very hard to differentiate trees from buildings in terms of either height or intensity information. Clearly, classifying trees and buildings is a significant, yet challenging task in urban land cover classification. Different methodologies have been designed to solve the problem. Some researchers have utilized the spectral information from multi-spectral images whilst some have designed building feature models or tree feature models to detect the outline of buildings or trees from a Lidar point cloud. The modelling methods proved efficient in some research. However, when it applies to some urban areas where trees and buildings of similar heights are located closely, clusters of buildings and trees may cause inaccuracy in the processes of image segmentation and classification. More information is required to obtain satisfactory classification results. Therefore, some overlooked characteristics from Lidar data can be used to assist the classification procedures. 3.2.1.3 Elevation difference image Since the laser pulse has the ability of penetration, every record of Lidar data includes the first and last return information. As we know, different objects have different transmittances and it is much easier for the laser pulse to penetrate leaves than building roofs. When the laser pulse successfully penetrates leaves, the elevation value of the last return differs from that of the first return in the same record. Since the laser pulse can hardly penetrate through the building roof, the elevation value of the last return remains the value of the first return. Therefore, this type of elevation difference only occurs on the edge of buildings. Based on this conclusion, analyzing the elevation difference between the first return and the last return is a useful way to classify trees and buildings.

Urban Land Cover Classification

63

A field survey was conducted in Cambridge and several buildings and their neighbouring trees were manually positioned on the map to examine the elevation difference of Lidar points reflected from different objects. The sizes of buildings are more than 600 m2 and the sizes of trees are between 23 m2 and 50 m2. Fig 3.2 shows the proportion of elevation edge points within different objects (because the largest elevation difference usually occurs on the edge of objects, a point with elevation difference can be named as an elevation_edge point).

Trees (45 out of 188, 23.9%)

Buildings (91 out of 3514, 2.59%)

Fig 3.2 Elevation_edge points within trees and buildings (Blue-outlined points are elevation_edge points)

Although some researchers (Lee and Lucas, 2007; Kato et al., 2009; etc.) have mentioned the elevation difference between the first and last return, they only used this information to distinguish ground or non-ground points rather than to classify different land cover types. Fig 3.2 illustrates the difficulty in applying the elevation difference information for pixel-based land cover classification. As shown in this figure, there are some Lidar points that show no elevation difference, so we cannot classify all individual points using this attribute. In addition, since the range of the elevation difference is very wide, it is very hard to distinguish whether the point is from a tree or a building. However, researchers can make the best use of the elevation difference information through object-based land cover classification. According

64

Chapter Three

to the statistics in Fig 3.2, trees have a much larger proportion of elevation_edge points than buildings do. As mentioned above, the first step of the object-based classification is to segment pixel-based images into multi-pixel objects. If this procedure is conducted successfully, points from the same tree or the same building will be grouped into one unclassified object. And the second step is to set some rules to classify these unclassified objects into different land cover types. Since the proportion of elevation_edge points differs greatly between trees and buildings, the elevation difference can be a promising tool for the object-based urban land cover classification. 3.2.1.4 Intensity difference image In addition to the elevation difference, the intensity difference between the first and the last return can also be utilized as a valuable attribute for land cover classification. Nevertheless, due to similar limitations, the attribute of the intensity difference is not suitable for urban land cover classification at the pixel level. Following the same procedure of analyzing elevation difference, the field test was also used to find out unique characteristics of trees and buildings in terms of this attribute. Due to different transmittances, tree objects have a much larger proportion than building objects of intensity_edge points (because those points with intensity difference are mainly on the edge of objects, they can be named as intensity_edge points) (this is shown in Fig 3.3). Due to the distinctive difference in the proportion of intensity_edge points at the object level, the intensity difference can be an efficient feature in object-based classification to distinguish trees and buildings.

Urban Land Cover Classification

Trees (54 out of 188, 28.7%)

65

Buildings (141 out of 3514, 4.01%)

Fig 3.3. Intensity_edge points within trees and buildings

3.2.2 Image segmentation and classification Image segmentation, which aims to completely partition an image into non-overlapping segments (Schiewe, 2002), is one key issue for object-based land cover classification. For years, image segmentation has received growing emphasis and many studies have been conducted based on the algorithms of image segmentation (Tseng and Lai, 1999; Abkar et al., 2000; Li and Peng, 2004; Kurnaz et al., 2005; Lhermitte et al., 2008; Fan et al., 2009; Wang et al., 2010; Wang et al., 2012; Zhang et al., 2013; etc). In addition, some software has been developed to conduct image segmentation automatically. Ecognition, one of the leading and most commonly used object-based image analysis tools, can be used for image segmentation and classification. Ecognition applies a region-growing algorithm to image segmentation. Firstly, a seed pixel is selected and then neighbouring pixels with similar attribute values can be merged into the region where the seeding pixel lies. When the seed pixel has consolidated all qualified neighbouring pixels, new seed pixels are selected to conduct the same procedure of region-growing. The segmentation task finishes when the entire image has been partitioned into non-overlapping regions (Garbay et al., 1986; Chang and Li, 1994).

66

Chapter Three

Ecognition makes hierarchically-weighted segmentation possible, so thematic images of different attributes can be analyzed comprehensively to achieve optimum segmentation results. Image layers can be weighted depending on their importance or suitability for the segmentation result. The larger the weight assigned to a thematic image, the more weight will be assigned to the image's pixel information during the segmentation process, assuming that the segmentation uses the pixel information. (Definiens Developer - User Guide, http://www.definiens.com) In addition, the segmentation scale, which refers to the heterogeneity within one object, is flexible and researchers can adjust this parameter according to characteristics of different object types to obtain better segmentation results. For instance, when the research needs to classify trees, the segmentation scale should be small whilst a large segmentation scale is more efficient to classify large buildings. When successful segmentation has been conducted, the image space can be partitioned into non-overlapping unclassified objects with attributes from thematic images. For instance, if the height and intensity images are used for segmentation, each segmented object has a height and intensity value. Researchers can choose some reference areas and analyze characteristics of different land cover types. Based on the sample analysis, classification rules can be established and every land cover type can be explained with specific ranges of different attributes. Next, all objects in the image space can be put into corresponding land cover types and preliminary classification results can be achieved. The results of image segmentation and classification can be further optimized by carrying out experiments and fieldwork for more accurate parameters.

3.3 Study site Based on the output of urban DTM generation, a case study of urban land cover classification was also conducted in central Cambridge using the same Lidar data set. Like other cities, there are dense buildings in the city center. On the other hand, trees and other urban vegetation occupy a large proportion of the whole urban landscape. Cambridge is a typical study site to test the methodology of object-based urban land cover classification for the reason as follows:

Urban Land Cover Classification

67

Trees and buildings of similar heights are frequently distributed together in central Cambridge. As discussed, one of the most difficult processes in urban land cover classification is to distinguish trees from buildings, which have overlapping height and intensity ranges. If trees are located close to buildings, the tasks of image segmentation and classification become more difficult. Therefore, if the classification methodology works efficiently in the challenging site of Cambridge, this method can also be applied to most urban landscapes. There are very few studies on urban land cover classification using solely airborne Lidar data. Chen et al (2009) classified water, shadow, vegetation, shrub, grassland, high building, low building, road, and vacant land type areas using airborne Lidar data and high resolution imagery. Inspired from previous studies and characteristics of the local landscape, there are eight main and distinguishable land cover types in central Cambridge: lawn, grass and crop, road and bare ground, building, tree, shrub, public infrastructure and water. In this work, shrubs are defined as trees that are shorter than 1.5 m.

3.4 Results Thematic images for image segmentation and classification can be generated directly through interpolation using raw Lidar data whilst the process of image segmentation and classification requires appropriate setting of parameters. The detailed setting and results of the thematic maps, image segmentation and image classification are presented as follows:

3.4.1 Thematic images With the DTM generated using the process discussed in Chapter 2, the nDSM (DSM- DTM) was obtained as well as the intensity image, elevation difference image and intensity difference image. Therefore, all the thematic images required for image segmentation are ready and are illustrated in Fig 3.4.

Chapter Three

68

a. Height Image

c. Elevation Difference Image

b. Intensity Image

d. Intensity Difference Image

To better observe the characteristics of different landscape types, these pictures are zooming in on the city center, instead of the whole study area. Fig3.4 Thematic images for image segmentation and classification

3.4.2 Image segmentation These four images were analyzed as hierarchical layers with different weights in Ecognition. As introduced, Ecognition enables researchers to adjust the weight of different layer images for the satisfactory abstraction and shaping result. The experiment for deciding segmentation parameters was conducted in some sample areas with densely distributed buildings and trees, and some trees and buildings were manually segmented for reference. By comparing the segmentation results (generated using different segmentation parameters) with the reference data, suitable segmentation parameters were decided. Through experiments, the optimum weight for Height, Intensity, Elevation Difference and Intensity Difference was set as 4:4:1:1. Next, by setting appropriate segmentation parameters (such as scale

Urban Land Cover Classification

69

parameters and homogeneity criteria), researchers can acquire a satisfactory segmentation result (most individual trees and buildings are one-to-one corresponding to a segmented object). Two strategies may be employed for multi-solution segmentation. a. Using similar scale parameters to segment all land cover types. Thus, trees can be segmented efficiently yet large buildings may be divided into several objects, which adds some difficulties to the process of classification. To solve this problem, researchers can firstly classify these land cover types with more specific rules and then merge neighbouring objects that are classified into the same types. b. Using different scale parameters (e.g. Image resolution: 0.5 m. Scale parameter: 35. Shape: 0.5. Compactness: 0.8 for trees, shrubs, small buildings and so forth and Image resolution: 0.5 m. Scale parameter: 80. Shape: 0.5. Compactness: 0.8 for large buildings) for different target types. Thus, it is much easier to establish classification rules for land cover types. However, this approach may result in several neighbouring trees being segmented into one object. In this study, the first strategy proved to be more efficient to segment individual tree objects; divided building objects can then be merged after the process of image classification. Therefore, the optimal segmentation result for this research was achieved by analyzing different thematic images as hierarchical layers with different weights (the weight for Height, Intensity, Elevation Difference and Intensity Difference was set as 4:4:1:1) and applying the following segmentation parameters (Image resolution: 0.5 m. Scale parameter: 35. Shape: 0.5. Compactness: 0.8). Through image segmentation, the entire image was segmented into 110306 objects and the average area of these objects was 112 m2.

3.4.3 Image classification Eight land cover types in the present research were classified following some general principles that can also be applied to other urban areas, although the setting of some classification parameters may vary in different study sites.

70

Chapter Three

Lawn, grass and crop, road and bare ground These three land cover types can be firstly distinguished from other types because the height values of these types are the lowest. Next, the obvious difference in intensity values enables researchers to set clear thresholds to differentiate them. Among the three types, the lawn has the highest intensity value whilst the road and bare ground has the lowest intensity value. Following these rules, the lowest objects can be put into the category of lawn, grass and crop or road and bare ground. Shrubs and Public infrastructure Shrubs (defined as trees shorter than 1.5m in this study) have medium height values and comparatively high intensity values whilst public infrastructures (includes benches and other infrastructure in public spaces) have medium height and very high intensity values (larger than that of shrubs). Water It is difficult to classify water sources using airborne Lidar data, because when Lidar pulses reach the surface of water, most of them will be absorbed and only very few returns can be received. As a result, even if some pixels can be classified as water in a pixel-based classification, the outline of the entire water area cannot be developed using these scattered pixels due to too many missing pulses. However, at the object level, the classification of water areas is feasible. In the process of image segmentation, all similar neighbouring pixels, even without the intensity or height information, can be grouped into a multi-pixel object. Therefore, the entire water area can be segmented into one or several objects with the lowest height value and a very small intensity value (much smaller than that of the road and bare ground and is close to 0). Through these procedures, water areas can be classified with a satisfactory result. Buildings and trees As discussed, trees and buildings share an overlapping range in both height and intensity values, which makes it very difficult to set clear classification rules to distinguish them at the pixel level. However, at the object level, some parameters can be better utilized to classify trees

Urban Land Cover Classification

71

and buildings. Area: Some buildings cover a large area that cannot be occupied by a single tree. Therefore, if the area of an object exceeds a certain threshold, this object can be classified as a building. Elevation difference and intensity difference: Since trees have a much larger proportion of elevation_edge and intensity_edge points than buildings do, objects with a large elevation_edge or intensity_edge value are more likely to be classified as trees. These two attributes can work efficiently to classify most urban trees and buildings. Exclusive height and intensity value: Although trees and buildings have an overlapping range in terms of the intensity and height, there are some exclusive values that can be used for classification. For instance, objects with a large height value (e.g. 30 meters), which goes beyond the range of urban trees, can be classified as buildings. Similarly, some objects with comparatively high intensity values, which exceed the range of buildings, can be classified as trees. Shape index: On an orthophoto map, individual trees tend to have a near-circle outline whilst individual buildings tend to have a rectangular or irregular outline. As a result, tree objects usually have a smaller value than that of building objects in terms of shape index ( Perimeter / 2 S * Area ), roundness, and so forth. Nevertheless, the accuracy of this parameter partly depends on the efficiency of image segmentation. If several trees or clusters of trees and buildings are segmented into one object, the shape index may not provide a reliable reference. The characteristics of different land cover classes that can be used for classification are also summarized in Table 3.1.

Chapter Three

72

Table 3.1 Useful attributes for land cover classification Class

Height Intensity

Intensity

Elevation and difference

Lawn

low

high

very small

Grass and crop

low

medium

very small

Road and bare ground

low

low

very small

Water

low

extremely low

very small

Shrub

medium

medium

small

Public

medium

high

small

high

medium -high

large

medium -high

medium

Other useful attributes

infrastructure Tree

Building

high

exclusive intensity small shape index exclusive area large shape index

According to Table 3.1, researchers can find unique characteristics for each land cover class. Nevertheless, due to errors that exist in image segmentation and the diversity of buildings and trees, it is not practical to use just one feature to classify all trees and buildings. All useful attributes should be considered comprehensively and the sequence for classification should be optimized by conducting experiments repeatedly. According to these principles, all these urban land cover types can be classified. By selecting training plots and analyzing value ranges of height, intensity, intensity difference and elevation difference for different land cover types, the setting of some key parameters was decided and land cover classification in the central Cambridge was conducted. Most land cover types can be easily classified using

Urban Land Cover Classification

73

specific intensity and height ranges, but the classification of trees and buildings requires more detailed design. Since there are thousands of trees and buildings of different shapes and sizes in the study site, it is very difficult to use a fixed threshold and principle to classify all tree and building objects. In addition, as the classified tree and building objects will be used in the following landscape pattern analysis, the completeness of the classified objects is another important issue. As discussed, the image segmentation process may cause some divided buildings, and thus the classification of tree and building objects also involves the process of merging divided objects. Therefore, this research conducted the classification of trees and buildings gradually. Ecognition allows users to gradually adjust the thresholds of classification and check the classification results. Different attributes were decided following the principle that misclassified trees and buildings should be controlled and the completeness of tree and building objects should be retained. According to the decision tree described in Fig 3.5, a land cover classification map of Cambridge is shown in Fig 3.6. According to Fig 3.5, one can see that most land cover types can be simply classified using one or two rules whilst tree and building objects need to be classified using a diversity of attributes. Although many features were employed for classifying trees from buildings, the elevation difference and intensity difference proved to be the most efficient approach. In the case study, more than 81% of tree objects and 85% of building objects were classified using the two variables, whilst other attributes serve as complementary approaches to classify a small proportion of the remaining trees and buildings.

74

Chapter Three

I_Edge: value of intensity difference. E_Edge: value of elevation difference. The range of intensity, I_Edge and E_edge have been normalized to 0-255 for image segmentation whilst height range remains unchanged. Fig 3.5 Decision tree for urban land cover classification

Urban Land Cover Classification

Fig 3.6 Land cover classification image of central Cambridge

75

76

Chapter Three

3.5 Accuracy assessment To conduct accuracy assessment, more than 500 reference points (each class included no less than 10 points) were generated randomly across the study site using the module “Accuracy Assessment” of the image processing tool, Erdas. Next, Each reference position was given an estimate value of land cover type according to the results of land cover classification. In addition to the model-estimated value, a real value of the land cover type was required for the accuracy assessment. Airborne photography data acquired during the same period (June, 2009) as airborne Lidar data and field collected data (to cover areas that are not available from airborne photography data) were employed to obtain a real value of the land cover type for each reference point. By comparing the real value and the estimated value of the land cover type, the accuracy assessment of urban land cover classification was conducted. The confusion matrix and accuracy assessment of object-based classification method are shown in Table3.2 and Table 3.3. Table 3.2 Confusion matrix of object- based land cover classification Reference Data Lawn Grass Shrub Road Water Tree Public Building and crop and bare ground infrastructure Lawn 95 1 0 0 0 0 0 Grass and 0 21 0 0 0 0 2 crop Shrub 0 0 0 0 0 17 0 Road and 0 1 38 0 0 0 0 bare ground Water 0 0 1 0 11 0 0 Tree 7 0 4 0 0 0 207 Public 0 0 0 14 0 0 0 infrastructure Building 0 0 1 0 0 0 10

0 0 0 1 0 6 0 97

Urban Land Cover Classification

77

Table 3.3 Accuracy assessment of object-based landscape classification Class

Reference Classified Number Producers Users Total Total Correct Accuracy Accuracy

Lawn 102 Grass 23 and crop Road and 44 bare ground Public 14 infrastructure Water 11 Shrub 17 Tree 219 Building 104 Total 534 Overall Accuracy

96 23

95 21

93.14% 91.30%

98.96% 91.30%

40

38

86.36%

95%

14

14

100.00%

100.00%

12 17 224 108 534

11 17 207 97 500

100.00% 100.00% 94.52% 93.27%

100.00% 100.00% 92.41% 89.81%

93.63%

In this study, the overall accuracy of land cover classification was 93.6% and the classification accuracy for most land cover types achieved 90%. The types of Lawn, grass and crop, road and bare ground, shrub, public infrastructure were classified efficiently using intensity and height attributes and limited misclassifications mainly resulted from overlapping intensity values (e.g. lawn and grass and crop). By adopting the elevation and intensity difference as well as some other attributes, this method classified most trees and buildings, although some misclassification remained. Water areas were developed into complete objects through image segmentation and were thus classified with a high accuracy at the object-based level. In line with the applications of Lidar technology, some specific Lidar processing software, such as Terrascan, Scop++, Tiffs, Lidar Analyst and so forth, has been developed to classify land cover types from Lidar data automatically. To compare the performance of the object-based methodology with these pixel-based tools, the same data set was also tested using one of the leading Lidar processing tools, Terrascan. Since Terrascan is a tool designed to process raw Lidar point clouds automatically, only several parameters need to be decided

78

Chapter Three

upon. To avoid the influence from an inappropriate setting of parameters, experiments were conducted using different values for some key parameters. In this study, pulses from trees that are close to buildings are more likely to be classified as buildings. To achieve a better result, loose parameters were adopted for tree detection to include as many qualified tree points as possible whilst strict parameters were used for building detection to reduce the influence from clusters of neighbouring buildings and trees. As a result, the optimal results for land cover classification using Terrascan were achieved by applying such settings: ground point searching (maximum building size 60 m), tree detection (more trees) and building detection (strict rules, minimum building size 10 m). Like most Lidar processing tools, Terrascan can only detect ground points, trees and buildings automatically. Therefore, such land cover types as water and public infrastructure were not included in the final result (Fig 3.7c). Ground points classified using Terrascan include lawn, grass and crop, and road and bare ground, which can also be classified using the object-based classification method. The comparison between classification images produced using the object-based method and Terrascan is shown in Fig 3.7. In addition, the same set of reference points was also used to evaluate the classification accuracy of Terrascan (Table 3.4).

Urban Land Cover Classification

79

a. Hillshade image of DSM

b. Classification results using object-based method

c. Classification results using Terrascan Fig 3.7 Comparison between classification images using object-based method and Terrascan

Chapter Three

80

Table 3.4 Accuracy assessment of land cover classification using Terrascan Class Reference Classified Number Producers Users Total Total Correct Accuracy Accuracy Tree 219 193 172 78.54% 89.12% Building

104

136

85

81.73%

62.5%

Ground

169

163

158

93.49%

96.93%

Total

492

492

415

Overall Classification Accuracy

84.35%

According to the visual observation from Fig 3.7 and the overall accuracy of 84.35%, urban land cover classification using Terrascan achieved satisfactory results. Considering the complexity of the landscape structure in the study site, Lidar processing software proved its ability to classify trees and buildings automatically. Nevertheless, since Terrascan is a pixel-based classification, the classified image is more fragmented than the image produced using the object-based approach. Additionally, this type of tool is more likely to produce scattered building pixels within the bare ground or scattered tree pixels within large buildings, which reduces the classification accuracy of buildings and trees. Compared with the existing software, the object-based classification method can efficiently filter scattered abnormal pixels and produce more accurate and homogeneous classification images, which can be more useful data sources for such disciplines as landscape ecology, environmental modelling and so forth. In addition, the classification of different ground types or specific land cover types can hardly be realized automatically in Lidar tools whilst such land cover types as water can only be extracted using the object-based classification method. As a result, despite the existence of highly automatic Lidar processing software, the algorithm of object-based urban land cover classification is still of practical significance.

Urban Land Cover Classification

81

3.6 Discussion Antonarakis et al (2008) adopted object-based classification using airborne Lidar data and successfully classified nine land cover types with an accuracy of 95% in three forest areas. Brennan and Webster (2006) applied the object-oriented methodology using airborne Lidar data to classify ten land types with an overall accuracy of 94%. Considering the complexity of urban landscapes, especially the indistinguishable spatial structure of mixed trees and buildings in the study site, the overall 93.6% accuracy of this study is satisfactory. Nevertheless, the classification accuracy for some land cover types may be further improved. The classification accuracy for most land cover types achieved 90% except for the road and bare ground type, the produced accuracy of which was only 86.36%. From the accuracy assessment table, we can see that the road and bare ground class had less classified pixels than reference pixels and these misclassified pixels were mainly put into the tree or building class. Since there were clear differences in classification rules between these land cover classes and misclassified pixels usually occurred on the edge of tree or building objects, this type of errors was mainly caused during the process of image segmentation when these ground points were grouped into neighbouring non-ground objects. For a similar reason, some pixels from lawns were also misclassified as trees or buildings. To reduce this type of error and improve the classification accuracy, the weight of the height information in the hierarchical image may be adjusted for a better segmentation result. Object-based urban land cover classification can be further explored using an airborne Lidar system with multiple returns. Theoretically, every laser pulse emitted to the surface target has a series of returns. Nevertheless, the number of returns that can be recorded differs in terms of Lidar systems. Airborne Lidar data used in this research only contained the first and last return information. However, there are some advanced Lidar systems that can record several returns. In this case, the correlation between different returns can be examined and more useful information for land cover classification can be extracted. Brennan and Webster (2006) used the number of echoes as an important indicator and successfully classified building structure. Although a two-return Lidar system can also be used for urban land

82

Chapter Three

cover classification, multiple-return Lidar systems are more likely to achieve better classification accuracy.

3.7 Conclusions Based on an object-based methodology, object-based land cover classification was conducted using airborne Lidar data and an overall classification accuracy of 93.6% was achieved. Compared with two previous studies, one study in forest areas with an accuracy of 95% and the other study in a coastal area with an accuracy of 94%, this research has proved that discrete Lidar data could also be used as the only source, instead of the complementary source to multi-spectral or airborne photography images, for land cover classification in urban areas. By analyzing the height information and intensity information comprehensively, researchers can efficiently classify the class of lawn, crop and grass, road and bare ground, shrub, public infrastructure and other urban land cover types, which have a unique combination of height and intensity ranges. An entire water area can hardly be classified using Lidar data at the pixel level, but in object-based classification, it can be segmented into complete objects and identified according to extremely low height and intensity values. The most challenging task in urban land cover classification is to distinguish trees from buildings because they share an overlapping range in terms of height and intensity values. This research suggests a possible solution. Although it has been applied in previous research to classify ground and non-ground points, the elevation difference and intensity difference cannot be used to classify specific land cover types at the pixel level. However, the two featyres become very useful attributes at the object level. Since trees have a much larger proportion of elevation_edge and intensity_edge points, most tree and building objects can be successfully distinguished by setting thresholds on the values of the elevation difference and intensity difference. In line with additional information such as the area, unique height and intensity range, shape index and so forth, the classification of urban trees and buildings can achieve satisfactory accuracy. In addition to typical urban landscape, clusters of neighbouring trees and buildings of similar heights in the study site cause extra difficulties in the process of image segmentation and classification. As a result, the methodology of object-based classification has the potential to be generally applicable.

Urban Land Cover Classification

83

Classified urban land cover types serve as important sources for establishing 3D urban landscape models. Since airborne Lidar data are also required for generating urban DTMs, which are indispensable for 3D landscape models, the object-based urban land cover classification method, as well as the urban DTM generation method, provides researchers with an efficient approach for establishing 3D urban landscape models using airborne Lidar data as the sole source. The Lidar-only methodology will be of practical significance for 3D landscape ecology research when additional data sources are not available.

References Abkar, A.A., Sharifi, M.A., Mulder, N.J. (2000) Likelihood-based image segmentation and classification: a framework for the integration of expert knowledge in image classification procedures. International Journal of Applied Earth Observation and Geoinformation. 2(2), 104-119. Antonarakis, A.S., Richards, K.S., Brasington, J. (2008). Object-based land cover classification using airborne LiDAR. Remote Sensing of Environment. 112, 2988–2998. Aplin, P., Atkinson, P., Curran, P. Per-field. (1999) classification of land use using the forthcoming very fine resolution satellite sensors: problems and potential solutions. Advances in Remote Sensing and GIS Analysis. Wiley and Son, Chichester, pp. 219–239, 1999. Arroyo, L. A., Johansen, K., Armston, J., Phinn, S. (2010) Integration of LiDAR and QuickBird imagery for mapping riparian biophysical parameters and land cover types in Australian tropical savannas. Forest Ecology and Management. 259(3), 598-606 Baatz, M., Schape, A. (2000) Multiresolution Segmentation: an optimization approach for high quality multi-scale image segmentation, Proceedings of Angewandte Geo. Informationsverarbeitung XII, in: Strobl, J., Blaschke, T. (Eds.). Wichmann, Heidelberg, 12̢23. Bao, Y.F., Li, G.P., Cao, C.X., Li, X.W., Zhang, H., H, Q.S., Bai, L.Y., Chang, C.Y. (2008) Classification of Lidar point cloud and generation of DTM from Lidar height and intensity data. The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. XXXVII (B3b). Beijing 2008. Bork, E.W., Su, J.G. (2007) Integrating LIDAR data and multispectral

84

Chapter Three

imagery for enhanced classification of rangeland vegetation: A meta analysis. Remote Sensing of Environment. 111, 11–24. Brennan, R., Webster, T. L. (2006) Object-oriented land cover classification of lidar derived surfaces. Canadian Journal of Remote Sensing. 32(2), 162í172. Campbell, B.J. (1987) Introduction to Remote Sensing. Guilford Press, New York. Chen, Y. H., Su, W., Li , J., Sun, Z.P. (2009) Hierarchical object oriented classification using very high resolution imagery and LIDAR data over urban areas. Advances in Space Research. 43, 1101–1110. Chen, Z. Y. Gao, B.B. (2014) An Object-Based Method for Urban Land Cover Classification Using Airborne Lidar Data. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing. 7(10):4243-4254 Cobby, D.M., Mason, D.C., Davenport, I.J. (2001) Image processing of airborne scanning laser altimetry data for improved river flood modelling. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. 56(2), 121-138. Cook, B.D., Bolstad, P.V., Næsset, E., Anderson, R.S., Garrigues, S., Morisette, J.T., Nickeson, J., Davis, K.J. (2009) Using LiDAR and quickbird data to model plant production and quantify uncertainties associated with wetland detection and land cover generalizations. Remote Sensing of Environment. 113(11), 2366-2379. Csatho, B., Schenk, T., Shin, S., Seo, S. (2003) Spectral interpretation based on multisensor fusion for urban mapping, In: Process of the 2nd GRSS/ISPRS Joint Workshop on Data fusion and remote sensing over urban areas, Berlin, May, pp. 8–11, 2003. Curran, P.J. (1985) Principles of Remote Sensing. Longman, New York. Dalponte, M., Bruzzone, L., Gianelle, D. (2012) Tree species classification in the Southern Alps based on the fusion of very high geometrical resolution multispectral/hyperspectral images and LiDAR data. Remote Sensing of Environment. 123, 258-270. Donoghue, D.N.M., Watt, P.J., Cox, N.J., Wilson, J. (2007) Remote sensing of species mixtures in conifer plantations using LIDAR height and intensity data. Remote Sensing of Environment. 110(4), 509–522. Fan, J.C., Han, M., Wang, J. (2009) Single point iterative weighted fuzzy C-means clustering algorithm for remote sensing image

Urban Land Cover Classification

85

segmentation. Pattern Recognition. 42(11), 2527-2540. Flood, M (2001) LIDAR activities and research priorities in the commercial sector. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. XXXIV (part 3/W4), 3í7. Heuel, S. Forstner, W., Lang, F. (2000) Topological and geometrical reasoning in 3D grouping for reconstructing polyhedral surfaces. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. XXIII (part 3), 397-404. Hill, R.A., Thomson, A.G. (2005) Mapping woodland species composition and structure using airborne spectral and LiDAR data. International Journal of Remote Sensing. 26, 3763–3779. Holmgren, J., Persson, Å. (2003) Identifying species of individual trees using airborne laser scanning. Remote Sensing of Environment. 90, 415í423. Johansen, K., Arroyo, L. A., Armston, J., Phinn, S., Witte., C. (2010) Mapping riparian condition indicators in a sub-tropical savanna environment from discrete return lidar data using object-based image analysis, Ecological Indicators. 10, 796-807. Kato ,A., Moskal , L.M., Schiess, P., Swanson, M.E., Calhoun, D., Stuetzle, W. (2009) Capturing tree crown formation through implicit surface reconstruction using airborne lidar data. Remote Sensing of Environment. 113, 1148–1162. Ke, Y. H., Quackenbush , L.J., Im, J. (2010) Synergistic use of QuickBird multispectral imagery and LIDAR data for object-based forest species classification. Remote Sensing of Environment. 114(6), 1141-1154. Koetz, B., Morsdorf, F., van der Linden, S., Curt, T., Allgöwer, B. (2008) Multi-source land cover classification for forest fire management based on imaging spectrometry and LiDAR data. 256(3), 263-271. Kurnaz, M.N., Dokur, Z., Ölmez, T (2005) Segmentation of remote-sensing images by incremental neural network. Pattern Recognition Letters. 26(8), 1096-1104. Lee, A. C., Lucas, R. M. (2007) A LiDAR-derived canopy density model for tree stem and crown mapping in Australian forests. Remote Sensing of Environment. 111, 493–518. Lee, D. S., Shan, J (2003) Combining LIDAR elevation data and IKONOS multispectral imagery for coastal classification mapping. Marine Geodesy. 26, 117í127.

86

Chapter Three

Lhermitte, S., Verbesselt, J., Jonckheere, I., Nackaerts, K., van Aardt, J.A.N., Verstraeten, W.W., Coppin, P. (2008) Hierarchical image segmentation based on similarity of NDVI time series Remote Sensing of Environment. 112(2), 506-521. Li, F., Peng, J.X. (2004) Double random field models for remote sensing image segmentation. Pattern Recognition Letters. 25(1), 129-139. Lohmann, P. (2002) Segmentation and filtering of laser scanner digital surface models. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. 34 (part XXX), 311–315. Lohmann, P., Koch, A., Schaeffer, M. (2000) Approaches to the filtering of laser scanner data. International Archives of Photogrammetry and Remote Sensing. XXXIII(B3), 540-547. Mesas-Carrascosa, F. L., Castillejo-González, I. L., Orden, M. S., Porras, A. G. (2012) Combining LiDAR intensity with aerial camera data to discriminate agricultural land uses. Computers and Electronics in Agriculture. 84, 36-46. Meng, X.L., Currit, N., Wang,L. (2009) Morphology-based building detection from airborne LiDA data. Photogrammetric Engineering & Remote Sensing. 75 (4), 437-442. Miliaresis, G., Kokkas, N. (2007) Segmentation and object-based classification for the extraction of the building class from LIDAR DEMs. Computers & Geosciences. 33(8), 1076-1087. Moffiet, T., Mengersen, K., Witte, C., King, R., Denham, R. (2005) Airborne laser scanning: Exploratory data analysis indicates potential variables for classification of individual trees or forest stands according to species. ISPRS Journal of Photogrammetry and Remote Sensing. 59, 289í309. Nardinocchi, C., Forlani, G., Zingaretti, P. (2003) Classification and filtering of laser data. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Science. XXXIV, (3/W13), 79-86. Ørka, H.O., Næsset, E., Bollandsås, O.M., 2007. Utilizing airborne laser intensity for tree species classification. International Archives of Photogrammetry and Remote Sensing. XXXVI (3/W52), 300–304. Richards, J.A. (1993) Remote Sensing Digital Image Analysis: An Introduction, 2nd ed Springer-Verlag, Berlin. Rottensteiner, F., Trinder, J., Clode, S., Kubik, K. (2005) Using the

Urban Land Cover Classification

87

Dempster–Shafer method for the fusion of LIDAR data and multi-spectral images for building detection. Information Fusion. 6, 283–300. Schiewe, J. (2002) Segmentation of high-resolution remotely sensed data concepts, applications and problems. Symposium on Geospatial Theory, Processing and Applications, 2002. Song, J. H., Han, S. H., Yu, K., Kim, Y. I. (2002) Assessing the possibility of land-cover classification using LIDAR intensity data. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. 34 (3B), 259–262. Syed, S., Dare, P., Jones, S. (2005) Automatic classification of land cover features with high resolution imagery and LIDAR data: an object-oriented approach. http://www.ecognition.com/sites/default/files/266_0185.pdf. Teo, T.A., Chen, L.C. (2004) Object-based building detection from LiDAR data and high resolution satellite imagery, In: Proceedings of Asian Conference on Remote Sensing, November 22–26. Ching-Mai, Thailand, 2004. Tovari, D., Pfeifer, N. (2005) Segmentation based robust interpolation-a new approach to laser data filtering. Laser scanning 2005, 79–84. Tseng, D.C., Lai, C.C. (1999) A genetic algorithm for MRF-based segmentation of multi-spectral textured images. Pattern Recognition Letters. 20(14), 1499-1510. Vosselman, G. (2000) Slope based filtering of laser altimetry data. International Archives of Photogrammetry and Remote Sensing. XXXIII (part B3/2), 935–942. Wang, Q., Zhang, Q.P., Zhou, W. (2012) Study on Remote Sensing Image Segmentation Based on ACA–FCM. Physics Procedia. 33, 1286-1291. Wang, Z.W., Jensen, J.R., Im, J. (2010) An automatic region-based image segmentation algorithm for remote sensing applications. Environmental Modelling & Software. 25(10), 1149-1165. Yu, B.L, Liu, H.X., Wu, J.P, Hu, Y.J., Zhang, L. (2010) Automated derivation of urban building density information using airborne Lidar data and object-based method, Landscape and Urban Planning. 98, 210-219. Zhang, X.L., Xiao, P.F., Song, X.Q., She, J.F. (2013) Boundary-constrained multi-scale segmentation method for remote sensing images. ISPRS Journal of Photogrammetry and Remote Sensing. 78, 15-25.

CHAPTER FOUR URBAN LANDSCAPE PATTERN ANALYSIS

4.1 Introduction One main task of landscape ecology is to understand the interactions between landscape patterns and ecological processes. As a result, the development of landscape ecology research is based on a comprehensive and accurate understanding of both landscape patterns and ecological processes. Amongst the two subjects, ecological processes can usually be analyzed using some standard criteria (e.g. the frequency, density, intensity, diversity, etc.), yet the analysis of landscape patterns may be conducted from different perspectives. Therefore, growing research priorities are placed on understanding landscape patterns. 2D landscape pattern analysis has experienced rapid developments in the past decades. More than 100 metrics have been designed to analyze general landscape patterns and specific pattern characteristics of different landscapes. These landscape metrics usually include patch level metrics (e.g. Patch Area, Patch Perimeter, Perimeter-Area Ratio, Core Area, Euclidean Nearest Neighbor Distance, Edge Contrast Index, etc.), class level metrics (e.g. Total (Class) Area, Percentage of Landscape, Core Area Distribution, Patch Cohesion Index, Proximity Index Distribution, etc.) and landscape level metrics (Total Area, Largest Patch Index, Landscape Shape Index, Patch Density, Patch Richness, Shannon’s Diversity Index, etc.) In addition, some research (Lausch and Herzog, 2002; Herold et al., 2005; Sundell-Turner and Rodewald, 2008) has been conducted to examine the performance of a diversity of metrics under different situations. Therefore, researchers can easily analyze landscape patterns in the horizontal direction with a systematic and mature methodology, including a set of well-accepted metrics and principles of selecting appropriate metrics.

Urban Landscape Pattern Analysis

89

Although great progress has been made in 2D landscape ecology, lacking vertical information of landscape features limits people’s comprehensive understanding of landscape patterns, and of interactions between landscape patterns and ecological processes. Chen et al. (2008) pointed out that 3D landscape pattern analysis was an important trend of future landscape ecology research, as the vertical pattern was closely correlated with some ecological processes. More and more researchers have realized the significance of vertical patterns in landscape ecology and the height information has been included in many studies as an important predictor of ecological processes. Amongst these studies, the tree and canopy height was correlated with such processes as the abundance and diversity of birds (Kirk and Hobson, 2001; Santos et al, 2006; Hodgkison et al, 2007; Flaspohler et al, 2010), occurrence and abundance of specific species (Timoney et al, 1997; Umapathy and Kumar, 2000; Meulebrouck et al, 2007; Levick and Rogers,2008; Stewart et al, 2009), total species richness (Reeder et al, 2005; Estrada et al, 2006), nest success (Shochat et al, 2005), rates of CO2 exchange (Petrone et al, 2008), SOC (soil organic carbon) stock (Li et al. 2010), WTP2 (Nielsen et al, 2007) and suitable habit conditions (Payer and Harrison, 2003). The plant, grass and general vegetation height also proved to be important influence factors of the perception of naturalness (Lamb and Purcell, 1990), bird community (Berg, 2002; Roth et al, 2005; Filloy and Bellocq, 2007) and abundance (Perkins et al, 2000; Rahman et al, 2012), vegetation dynamics (Koniak and Noy-Meir, 2009), soil erosion (Botterweg et al,1998), nutrient enrichment (Craft et al, 2007) and ecological networks (Bazelet and Samways, 2011). In addition to the vegetation height, researchers also examined the correlation between the building height and pollutant dispersion (Gerdes and Olivari, 1999), viewshed property (Sander and Manson, 2007), the intensity of urban heat island effects (Mirzaei and Haghighat, 2010) and environmental aesthetics (Ameel and Tani, 2011). Corridors play a significant role in landscape modelling and the correlation between heights of corridors and ecological processes receives growing research emphasis. According to findings of these studies, the height of corridors correlates with the intensity of turbulence (Prueger et al, 1996), corridor use 2

WTP, Willingness To Pay. WTP usually refers to Willingness To Pay for the management of trees and green spaces in landscape ecology.

90

Chapter Four

(Laurance.S.G and Laurance.W.F. 1999; Cain et al, 2003) and the occurrence of bat species (Hein et al, 2009). To efficiently utilize the height information, this attribute is better to be understood as landscape patterns in the vertical dimension, which works together with horizontal patterns to form the comprehensive landscape structure. Cain et al (2003) pointed out that the culvert use of bob cats was correlated with openness ratio (width×height/length) of culverts. Mirzaei and Haghighat (2010) found the positive correlation between the aspect ratio (the height of building to street breadth) and urban heat island effects. However, without more accurate data sources, most researchers can only acquire and apply the mean height, instead of the detailed vertical structure (e.g. height heterogeneity within landscape features), for landscape pattern analysis. Under such circumstances, complicated spatial patterns and correlations in the vertical direction may be over-simplified or omitted. Additionally, although much research has been conducted to add vertical information to 2D landscape ecology, these studies are mainly focusing on establishing the correlation between the height information and specific issues. Since methods from these scattered studies can hardly be applied to other research fields, a systematic methodology of analyzing 3D landscape patterns is yet developed. To propose a feasible framework of 3D landscape pattern analysis, some 3D landscape metrics, which have the potential to be generally applied to other urban landscapes, are designed as examples and landscape patterns of two urban areas are analyzed and compared from both 2D and 3D perspectives.

4.2 Methodology With the pre-processed DTM and classified land cover types, a 3D urban landscape model can be easily established using the ArcGIS tool ArcScene. Based on the model, the 3D structure for each urban feature can be built. Next, a set of 2D metrics, including some generally used metrics and some specific metrics designed for this study, is adopted to analyze landscape composition and structure at both the pixel and object level. In addition, some 3D landscape metrics (Chen et al., 2014) are proposed to analyze and compare landscape pattern differences in the vertical direction, which cannot be revealed in 2D landscape models.

Urban Landscape Pattern Analysis

91

4.3 Study sites and data preparation A comparative study was conducted in two areas, central Cambridge, UK and the residential area in Canvey Island, UK. The two sites are typical urban landscapes, with a large proportion of buildings, trees and green spaces. Through preliminary analysis, the total cover area of trees in the Cambridge is very close to that of Canvey. In addition, the total building cover area of buildings in Cambridge is also very close to that of Canvey. Thus, from a 2D perspective, landscape patterns of the two areas seem to have many similarities. However, since there is a diversity of trees and buildings in Cambridge whilst the spatial structure of trees and buildings in Canvey is more unified, vertical landscape patterns in the two areas are notably different. As a result, this case study can efficiently examine the performance of 3D landscape metrics in revealing pattern differences that would otherwise be omitted in 2D models. The airborne Lidar data used for the Cambridge study has been introduced in previous chapters. The data set employed for Canvey Island was obtained in March, 2010 using the same ALTM-3033 Lidar System, also with a horizontal resolution of 0.5 m. The main task for data preparation was to integrate separate Lidar point clouds and convert the raw data format to other formats for the convenience of further analysis. Since the entire point cloud of Canvey survey includes some suburban and coastal areas, the residential area was clipped from the original data set.

4.4 Results Using the similar methodology adopted in central Cambridge, DTM generation and land cover classification were conducted in the residential area of Canvey Island. As the methodology has been discussed in details above, only the results are described here. Land cover classification images of Cambridge and Canvey are shown in Fig 4.1.

92

Chapter Four

a. Central Cambridge

b. Residential area in Canvey Fig 4.1 Land cover classification images of Cambridge and Canvey

According to the land cover classification images, landscape composition and structure in the two sites can be analyzed and compared. Since the methodology of 2D landscape pattern analysis is highly mature, this study did not include detailed pattern analysis of all land cover types. To illustrate the utility of the height information, two important land cover types, buildings and trees in the two areas, were selected for spatial pattern analysis from both 2D (Table 4.1) and 3D perspectives. The perception of naturalness is closely related to urban residents’ preferences towards the local environment (Lamb and Purcell, 1990; Ode et al. 2009; etc.). In addition to urban trees, green

Urban Landscape Pattern Analysis

93

spaces are the main source of urban ‘naturalness’. Considering the significance of green spaces (in the land cover image of Cambridge, green spaces include lawns and grass and crops) and the requirements of following analysis, the land cover type of green spaces was also analyzed using some 2D metrics. Table 4.1 Spatial pattern analysis using 2D landscape metrics 2D Landscape Metrics ( pixel level)

Cambridge 2

Canvey

Total Building Cover Total Tree Cover Tree Cover/Building Cover Total Area of Green Spaces

1522467 m 3360707 m2 2.21 5787688 m2

1443617 m2 3289053 m2 2.28 3685078 m2

2D Landscape Metrics ( object level)

Cambridge

Canvey

Total Building Number Mean Building Area Total Tree Number Mean Tree Area Total Tree Patch Number Number of Large Tree Patches (area>500m2) Mean Area of Tree Patches Mean Distance between A Building and A Nearest Tree Mean Number of Trees Surrounding A Building (500m2) Mean Distance Between a Building and Nearest Green Space (area>2500m2)

2464 617.9m2 23064 145.7 m2 3019 722

17048 84.7m2 54144 60.4 m2 9512 952

1113 m2 12.3m

346 m2 5.74m

6

15

4720 1226 51m

2433 1515 162m

152m

254m

94

Chapter Four

Based on the visual observation from Fig 4.1 and spatial pattern analysis at the pixel level, landscape patterns can be compared between the two sites. Except for patterns of green spaces, Cambridge and Canvey have a similar spatial composition in terms of buildings and trees, as the value of total building cover and total tree cover are very close in the two sites. Nevertheless, according to the pattern analysis at the object level, structural differences between the two areas were revealed in terms of the tree size, building size, tree patch number and so forth. Besides, the spatial correlation between individual trees and buildings can also be analyzed using specific landscape metrics. As examples of this, two object-based 2D metrics, Mean Distance between A Building and A Nearest Tree, and Mean Number of Trees Surrounding A Building proved the differences in tree-building patterns in Cambridge and Canvey. Although the total building and tree cover are very close in the two areas, the mean cover area of trees and buildings in Canvey is much smaller than that in Cambridge and trees and buildings in Canvey are more closely distributed. Although landscape compositions and structures in the horizontal direction were understood well from an object-based perspective, spatial patterns in the vertical direction require further analysis. To add the vertical attribute, airborne Lidar data were used to generate the nDSM image for urban features, which described detailed spatial structures in the vertical direction (DSM-DTM, shown as Fig 4.2). Next, the height image was fused with the classification image and a 3D structure was then modeled for each classified object (Fig 4.3).

Urban Landscape Pattern Analysis

a. Spatial structure of urban features in height image

b. Overview of Cambridge in height image

c. Overview of Canvey in height image Fig 4.2 Height images of Cambridge and Canvey

95

96

Chapter Four

a. Overview of Cambridge in 3D landscape model

b. Overview of Canvey in 3D landscape model Fig 4.3 Classified urban features with 3D structure

Based on the 3D landscape model, the height heterogeneity within landscape features can be well observed and some 3D landscape metrics can be designed (Table 4.2). All 3D landscape metrics were implemented and calculated through programming using the software package ArcGIS Engine and C#.

Urban Landscape Pattern Analysis

97

Table 4.2 Spatial pattern analysis using 3D landscape metrics Landscape Metrics

Cambridge

Canvey

Mean Tree Height

11.6m

4.0m

Area Weighted Mean Tree Height

11.3m

3.9m

Mean Building Height

10.2m

5.3m

Area Weighted Mean Building Height

13.2m

6.2m

Area Weighted Mean Height Deviation Within Individual Trees1

1.90m

0.72m

Area Weighted Mean Height Deviation Within Tree Patches2

3.95m

1.42m

Height Deviation Between Trees

4.93m

1.97m

Height Deviation Between Tree Patches3

4.83m

3.70m

Area Weighted Mean Height Range of Individual Trees4

0.89

0.75

Area Weighted Mean Height Deviation Within Individual Buildings5

2.32m

1.07m

Mean Building Structure Index6

42.6

13.7

Height Deviation Between Buildings

3.97m

1.93m

Mean Ratio of Building Height to Mean Height of Its Surrounding Trees7

1.172

1.302

1

Height Deviation Within Individual Trees: The standard deviation of the heights of all pixels within tree objects. 2 Height Deviation Within Tree Patches: The standard deviation of the heights of all trees within individual tree patches. 3 Height Deviation Between Tree Patches: The standard deviation of the heights of

98

Chapter Four

all tree patches. The height of tree patches refer to the area weighted mean tree height within individual tree patches. 4 Height Range of Individual Trees = (Hmax – Hmin) / Hmax. 5 Height Deviation Within Individual Buildings: The standard deviation of the heights of all pixels within building objects. 6 Building Structure Index = Building Area/ Building Height. 7 Ratio of Building Height to Mean Height of Its Surrounding Trees: The ratio of the height of a building to the mean height of all trees that are surrounding this building. “surrounding” means that the distance between these trees and the building is less than a certain value.

Table 4.2 illustrates some examples of 3D landscape metrics. Mean Building Height and Mean Tree height: The two metrics were used to explain the general structure of buildings and trees in the vertical direction. Due to the existence of large college buildings in Cambridge and the large proportion of single unit houses in Canvey, the mean height of buildings in Cambridge is much higher than that in Canvey. In terms of trees, the mean tree height is close to the mean building height in both Cambridge and Canvey. Mean Height Deviation Within Individual Trees and Mean Height Range of Individual Trees: The height deviation and height range were calculated for each tree using GIS tools and the mean height deviation and mean height range can be used to compare the general tree shape in different areas. Since there are many freely growing trees in Cambridge (Fig 4.4c) and regularly shaped trees in Canvey (Fig 4.4b), trees in Cambridge show more height heterogeneity and understory structure, which can be revealed using the two metrics. In addition, in urban forests with one unified tree type, the value of the two metrics can be used to assist the classification of tree species. Height Deviation Within Tree Patches, Height Deviation Between Trees and Height Deviation Between Tree Patches: Tree patches can be acquired by merging neighbouring trees with GIS or Remote Sensing software. The three metrics were used to examine the diversity of tree heights, which is an important factor of people’s aesthetical preferences towards local environments (Nielsen, et al., 2007; Gundersen and Frivold, 2008; Zheng et al., 2011). Since there are several types of trees with different heights in Cambridge, whilst most trees in Canvey are of unified types and heights, the value of these

Urban Landscape Pattern Analysis

99

metrics in Cambridge is larger than that in Canvey. Mean Height Deviation Within Individual Buildings: The height deviation within each building is an indicator of the roof type and thus the mean height deviation can be used to understand general roof types in the study site. As there are some buildings with pointed arches (Fig 4.4a) or gable roofs in Cambridge whilst there are many buildings with comparatively flat gable roofs in Canvey (Fig 4.4b), the value of this metric in Cambridge is much larger than that in Canvey. Mean Building Structure Index: The ratio of building area to building height was used to analyze general building structure in different landscape types. For instance, although some buildings in Cambridge cover a large area, these buildings are not similar to metropolitan high-rise buildings, which have many more layers and thus a much smaller value of building structure index. By comparison, a large proportion of single unit buildings with small area in Canvey resulted in a small value of this metric. Height Deviation Between Buildings: This metric can be used to examine the height diversity of buildings, which can be related to people’s aesthetical perception of local landscapes. Since Cambridge is full of buildings with different heights and roof types whilst a large proportion of buildings are designed with a similar height and structure in Canvey, the value of this metric is much larger in Cambridge than that in Canvey. Mean Ratio of Building Height to Mean Height of Its Surrounding Trees: Through programming, the ratio of building height to mean height of its surrounding trees was calculated for each building and the mean ratio in an area was designed to examine the spatial relationship between buildings and surrounding trees. People’s preference (which may be acquired through questionnaires; see Chapter 5) on the value of this metric can be used as evaluation reference for urban planners.

Chapter Four

100

a. Buildings with pointed arches in Cambridge

b. Buildings with gable roof in Canvey

c. Freely grown large trees in Cambridge

d. Regularly shaped medium-sized trees in Canvey

Fig 4.4 Comparison of building and tree structure in Cambridge and Canvey

Based on the 3D landscape model and a set of 3D landscape metrics, the vertical pattern differences, which cannot be analyzed in 2D landscape models, were examined between the two sites. With the results of 3D pattern analysis, researchers can have a comprehensive understanding of the spatial structure of urban features. Furthermore, researchers may relate 3D landscape metrics to people’s aesthetical preferences, ecological benefits from urban forests, wildlife’s diversity and other ecological processes according to their specialized fields. In addition to the object height, terrain information (although not employed in the present study) as shown in Fig 4.5, can also be integrated with 3D urban feature models for specific metrics (Chen et al., 2008), which may be used for such studies as pollutant diffusion, flood control and so forth.

Urban Landscape Pattern Analysis

101

a. DTM of Cambridge

b. DTM of Canvey Fig 4.5 Terrain information from 3D landscape models

4.5 Discussion In addition to the extra vertical attribute, object-based landscape models are more likely to provide reliable data for landscape pattern analysis. To compare the differences between object-based and pixel-based landscape pattern analysis, the object-based vector format was converted into pixel-based raster format and conducted 2D landscape pattern analysis using one pattern analysis tool, Fragstats, which can analyze pixel-based landscape patterns automatically and is thus widely adopted by many researchers. Compared with the

102

Chapter Four

object-based landscape model, the pixel-based landscape model is more fragmented and the result of spatial pattern analysis at the pixel level needs further revision. For instance, the tree patch number calculated using the pixel-based model was 6088, compared with 3019 using the object-based landscape model. By manually setting a minimum-area threshold for pixel-based tree patch detecting, the number of tree patches was reduced significantly. Based on the comparative experiments, one can see that the pixel-based landscape pattern analysis is probably influenced by the fragmentation effects. On the other side, since each object in 3D urban landscape models corresponds to an individual feature in the real landscape, object-based landscape pattern analysis can provide researchers with more accurate and practical results. In line with the progress of RS and GIS, it is not difficult for researchers to implement 3D landscape models. Nevertheless, the development of 3D pattern analysis falls behind that of 3D observation technology. From the 2D perspective, researchers can easily understand landscape composition, fragmentation and diversity using corresponding landscape metrics. In contrast, lacking systematic methodology and generally applicable 3D landscape metrics, researchers can hardly extract all 3D pattern characteristics from 3D landscape models and thus it is difficult to analyze, compare and evaluate vertical patterns between different sites. In the case study, some 3D landscape metrics that have the potential to be generally applied for pattern analysis of trees and buildings were proposed as instances. However, due to limited knowledge and experience, the amount and scope of 3D landscape metrics presented here can hardly compete with that of 100 well-developed 2D metrics. In addition, lacking time and resources, this research did not include the terrain-based 3D metrics. Therefore, the main purpose of this case study is to illustrate the design and utility of 3D landscape metrics and encourage more researchers to contribute towards a comprehensive and systematic framework for 3D landscape pattern analysis. By designing more general or specific 3D metrics, researchers can make full use of 3D landscape models to understand landscape patterns. Furthermore, based on the development of 3D landscape metrics, scholars from different disciplines can explore the possibility of examining the correlations between 3D landscape patterns and a

Urban Landscape Pattern Analysis

103

diversity of ecological processes. Although landscape patterns of trees and buildings were understood well based on the landscape model established in the case study, this model has its limitations. Due to its overhead perspective, airborne Lidar can only obtain height information without detailed understory structure of trees and buildings. Therefore, a landscape model established using airborne Lidar data is not sufficient for some specific studies concerning the patterns of underpasses or the understory structure of trees. This problem may be solved by ground or vehicle based Lidar tools, which can scan the structure of landscape features from different perspectives. However, establishing such a 3D landscape model is highly time and resource consuming. Hence, future research of 3D landscape modelling can be conducted on the methodology of integrating large-scale airborne data with small-scale ground or vehicle based Lidar data to meet the requirements of specific studies (Bremer and Sass, 2012).

4.6 Conclusions 3D landscape models enable researchers to distinguish the structural differences between different areas by designing and employing proper 3D landscape metrics. A case study was conducted to compare landscape patterns between two sites, central Cambridge and the residential area in Canvey. Firstly, some pixel-based 2D landscape metrics were employed to understand the general composition of different land cover types. The results indicated that the overall cover area of trees and buildings in Cambridge was very similar to that in Canvey. Following the pixel-based 2D landscape pattern analysis, a set of object-based 2D metrics was employed to understand the spatial structure of individual features. By comparison, the amount of trees and buildings in Canvey was much larger than that in Cambridge and the buildings and their neighbouring trees were generally located more closely in Canvey. By adopting a diversity of object-based and pixel-based 2D metrics, the spatial pattern of individual urban features and overall land cover types in the horizontal direction were analyzed comprehensively. However, 2D landscape pattern analysis fails to reveal pattern characteristics of the two sites in the vertical direction. As a result, 3D pattern analysis was conducted to further compare vertical landscape patterns between the two sites. A set of 3D

104

Chapter Four

landscape metrics was designed and employed. According to the result of 3D pattern analysis, the general height of trees and buildings, the height deviation within individual trees and the height deviation between trees and buildings in Cambridge are much larger than those in Canvey. Since the two sites have some similar patterns in the horizontal direction, the obvious vertical pattern differences revealed in this case study proved that landscape patterns were better understood and compared using properly designed 3D landscape metrics. 3D landscape pattern analysis based on 3D urban landscape models provides urban planners with an efficient approach for understanding the general landscape patterns in the study site. Instead of large-scale on-site surveys, researchers can quantitatively understand the spatial configuration of trees, buildings, green spaces and other land cover types. For instance, it is not feasible for researchers to visit every corner of the study site to analyze the structure of all individual trees and tree patches. Instead, by using proper metrics (e.g. area-weighted mean height of trees, mean height deviation within individual trees, height deviation between trees, height deviation between tree patches. etc.), urban planners can have a comprehensive and accurate understanding of tree and building heights, tree shapes, roof types, the diversity of tree heights, and many other pattern characteristics. Based on the 3D pattern analysis, urban planners can further evaluate urban landscape patterns and propose feasible approaches of enhancing current landscape patterns to better meet people’s needs (an example will be introduced in the following sections). Since landscapes in both Cambridge and Canvey are typical urban landscapes, which include a large number of trees, buildings and green spaces, 2D and 3D landscape metrics proposed in this case study may also be generalized to other urban areas. In addition to urban planning, quantitative 3D landscape pattern analysis can promote the development of landscape ecology. Since the key issue of landscape ecology is to examine the interactions between landscape patterns and ecological processes, a comprehensive and better understanding of 3D landscape patterns can provide more accurate results and an expanded scope for landscape ecology research (some potential applications of 3D landscape pattern analysis will be discussed in the following parts). To this end, researchers from different disciplines should work together to establish a generally

Urban Landscape Pattern Analysis

105

applicable system of 3D landscape metrics. In this case, the design and application of 3D landscape metrics demonstrated in this study serves as an instance for other landscape ecologists. Chen et al. (2008) discussed that one main challenge of landscape ecology was the applications of landscape metrics. As many metrics were designed without social-ecological meaning, they can hardly be linked to ecological processes or applied for landscape evaluation. On the other hand, 3D landscape metrics proposed in this section may be correlated with some social-ecological issues and have the potential to be employed for landscape pattern evaluation, which will be discussed in the following chapter.

References Ameel, L., Tani, S. (2011) Everyday aesthetics in action: Parkour eyes and the beauty of concrete walls Emotion. Space and Society. http://dx.doi.org/10.1016/j.emospa.2011.09.003. Bazelet,C.S., Samways, M.J. (2011) Identifying grasshopper bioindicators for habitat quality assessment of ecological networks. Ecological Indicators. 11, 1259-1269 Berg, A. (2002) Breeding birds in short-rotation coppices on farmland in central Sweden—the importance of Salix height and adjacent habitats. Agriculture, Ecosystems and Environment. 90, 265–276. Botterweg, P., Leek, R., Romstad, E.,Vatn, A. (1998) The EUROSEM-GRIDSEM modeling system for erosion analyses under different natural and economic conditions. Ecological Modelling. 108, 115-129 Bremer, M., Sass, O. (2012) Combining airborne and terrestrial laser scanning for quantifying erosion and deposition by a debris flow event. Geomorphology. 138(1), 49-60. Cain, A.T., Tuovila, V.R., Hewitt, D.G., Tewes, M.E. (2003) Effects of a highway and mitigation projects on bobcats in Southern Texas. Biological Conservation. 114, 189-197. Chen, L.D., Liu, Y., Lv, Y.H., Feng , X.M., Fu, B.J. (2008) Pattern analysis in landscape ecology: progress, challenges and outlook. ACTA ECOLOGICA SINICA. 28(11), 5521-5531. Chen Z.Y., Xu, B., Devereux, B. (2014) Urban landscape pattern analysis based on 3D landscape models. Applied Geography. 55, 82-91.

106

Chapter Four

Craft, C., Krull, K., Graham,S. (2007) Ecological indicators of nutrient enrichment, freshwater wetlands, Midwestern United States ( U.S.) . Ecological Indicators. 7, 733-750 Estrada, C.G., Damon, A., Hernández, C.S., Pinto, L.S., Núñez, G.I. (2006) Bat diversity in montane rainforest and shaded coffee under different management regimes in southeastern Chiapas, Mexico. Biological Conservation. 132, 351-361 Filloy, J., Bellocq, M.I. (2007) Patterns of bird abundance along the agricultural gradient of the Pampean region. Agriculture, Ecosystems and Environment. 120, 291–298 Gerdes, F., Olivari, D. (1999) Analysis of pollutant dispersion in an urban street canyon. Journal of Wind Engineering and Industrial Aerodynamics. 82, 105-124. Gundersen, V.S., Frivold, L.H. (2008) Public preferences for forest structures: A review of quantitative surveys from Finland, Norway and Sweden. Urban Forestry & Urban Greening. 7, 241–258. Hein, C.D., Castleberry, S.B., Miller, K.V. (2009) Site-occupancy of bats in relation to forested corridors. Forest Ecology and Management. 257, 1200-1207. Herold, M., Couclelis, H., Clarke K.C. (2005) The role of spatial metrics in the analysis and modelling of urban land use change. Computers, Environment and Urban Systems. 29, 369-399. Hodgkison, S., Hero, J.M., Warnken,J. (2007) The efficacy of small-scale conservation efforts, as assessed on Australian golf courses. Biological Conservation. 135, 576-586 Kirk, D.A., Hobson, K.A. (2001) Bird–habitat relationships in jack pine boreal forests. Forest Ecology and Management. 147, 217-243. Koniak, G., Noy-Meir, I. (2009) A hierarchical, multi-scale, management-responsive model of Mediterranean vegetation dynamics. Ecological Modelling. 24, 1148-1158 Lamb, R.J., Purcell A.T. (1990) Perception of naturalness in landscape and its relationship to vegetation structure. Landscape and Urban Planning. 19, 333-352 Laurance, S.G., Laurance, W.F. (1999) Tropical wildlife corridors: use of linear rainforest remnants by arboreal mammals. Biological Conservation. 91, 231-239 Lausch, A., Herzog, F. (2002) Applicability of landscape metrics for the monitoring of landscape change: issues of scale, resolution and interpretability. Ecological Indicators. 2, 3–15. Levick, S., Rogers, K. (2008) Patch and species specific responses of

Urban Landscape Pattern Analysis

107

savanna woody vegetation to browser exclusion. Biological Conservation. 141, 489-498 Li, P. H., Wang, Q., Endo, T., Zhao, X., Kakubari.Y. (2010) Soil organic carbon stock is closely related to aboveground vegetation properties in cold-temperate mountainous forests. Geoderma. 154, 407-415 Meulebrouck, K., Ameloot, E., Verheyen, K., Hermy, M. (2007) Local and regional factors affecting the distribution of the endangered holoparasite Cuscuta epithymum in heathlands. Biological Conservation. 140, 8-18 Mirzaei, P.A., Haghighat,F. (2010) A novel approach to enhance outdoor air quality: Pedestrian ventilation system. Building and Environment. 45, 1582-1593. Nielsen, A.B., Olsen, S. B., Lundhede, T. (2007) An economic valuation of the recreational benefits associated with nature-based forest management practices. Landscape and Urban Planning. 80, 63–71. Payer, D.C., Harrison,D.J. (2003) Influence of forest structure on habitat use by American marten in an industrial forest. Forest Ecology and Management. 179, 145-156 Perkins, A.J., Whittingham, M.J., Bradbury, R.B., Wilson, J.D., Morris, A.J., Barnett.P.R. (2000) Habitat characteristics affecting use of lowland agricultural grassland by birds in winter. Biological Conservation. 95, 279-294. Petrone, R.M., Chahil, P., Macrae, M.L., English, M.C. (2008) Spatial variability of CO2 exchange for riparian and open grasslands within a first-order agricultural basin in Southern Ontario. Agriculture, Ecosystems & Environment. 125, 137-147. Prueger, J.H., Hipps, L.E., Cooper, D.I. (1996) Evaporation and the development of the local boundary layer over an irrigated surface in an arid region. Agricultural and Forest Meteorology. 78, 223-237. Rahman, M.L., Tarrant, S., McCollin, D., Ollerton, J. (2012) Influence of habitat quality, landscape structure and food resources on breeding skylark Alauda arvensis territory distribution on restored landfill sites. Landscape and Urban Planning. 105, 281-287 Reeder, K.F., Debinski , D.M., Danielson, B.J. (2005) Factors affecting butterfly use of filter strips in Midwestern USA. Agriculture, Ecosystems and Environment. 109, 40–47. Roth, A.M., Sample, D.W., Ribic, C.A., Paine, L., Undersander, D.J., Bartelt, G.A. (2005) Grassland bird response to harvesting

108

Chapter Four

switchgrass as a biomass energy crop. Biomass and Bioenergy. 28, 490-498. Sander, H.A., Manson,S.M. (2007) Heights and locations of artificial structures in viewshed calculation: How close is close enough? Landscape and Urban Planning. 82, 257–270. Santos, T., Telleria, J.L., Diaz, M., Carbonell, R. (2006) Evaluating the benefits of CAP reforms: Can afforestations restore bird diversity in Mediterranean Spain? Basic and Applied Ecology. 7, 483—495. Shochat, E., Wolfe, D. H., Patten, M. A., Reinking, D.L., Sherrod,S.K. (2005) Tallgrass prairie management and bird nest success along roadsides. Biological Conservation. 121, 399-407 Stewart, G.H., Meurk, C.D., Ignatieva, M.E., Buckley, H.L., Magueur, A., Case, B.S., Hudson, M., Parker, M. (2009) URban Biotopes of Aotearoa New Zealand URBANZ II: Floristics, biodiversity and conservation values of urban residential and public woodlands. Urban Forestry & Urban Greening. 8, 149-162 Sundell-Turner, N.M. Rodewald, A.D. (2008) A comparison of landscape metrics for conservation planning. Landscape and Urban Planning. 86, 219–225. Timoney, K.P., Peterson, G., Wein, R. (1997) Vegetation development of boreal riparian plant communities after flooding, fire, and logging, Peace River, Canada. Forest Ecology and Management. 93,101-120. Umapathy, G., Kumar, A. (2000) The occurrence of arboreal mammals in the rain forest fragments in the Anamalai Hills, south India. Biological Conservation. 92, 311-319. Zheng, B., Zhang, Y.Q., Chen, J.Q. (2011) Preference to home landscape: wildness or neatness? Landscape and Urban Planning. 99, 1–8.

CHAPTER FIVE URBAN LANDSCAPE PATTERN EVALUATION

5.1 Introduction Landscape patterns can be analyzed and described quantitatively using landscape metrics and other methods, yet the evaluation of landscape patterns is still challenging due to the lack of generalizable criteria. To solve this problem, one efficient approach is to examine the correlations between landscape patterns and socio-ecological issues, and then establish evaluation proper standards according to the preferences of people, the well-being of wildlife and the sustainable development of environment. Based on 3D urban landscape models, correlations between social-ecological issues and 3D landscape patterns can be further examined. Following this, the findings from these studies can be used as applicable principles for evaluating 3D urban landscape patterns. Interactions between landscape patterns and social-ecological issues can be examined from different perspectives. With limited time and sources, this book will not explore all potential aspects of this broad topic. Instead, an example is given to demonstrate how a set of evaluation criteria may be established and how the evaluation criteria can be integrated with landscape pattern analysis (landscape metrics) for urban landscape pattern evaluation.

5.1.1 Interactions between landscape patterns and social-ecological issues Many studies have been conducted to relate landscape patterns to people’s preferences, the distribution of wild life and other social-ecological issues. Many researchers (Schroeder and Cannon, 1983; Buhyoff, 1984; Ulrich, 1986; Schroeder, 1988; Smardon, 1988; Murray, 1996; Dwyer et al., 2003; Van Herzele et al., 2005; Arnberger, 2006; Bernath and Roschewitz, 2008) have pointed out the positive

110

Chapter Five

correlation between the existence of vegetation and people’s aesthetical and recreational preferences. In addition, the correlation between landscape patterns and the diffusion of diseases has become a hot research field in landscape ecology (Waller, 2000; Banerjee, 2007; Schærström, 2009; Dirk and Pfeiffer, 2011). In terms of wildlife, such landscape factors as the proportion of forested area (Sorace and Visentin, 2007), housing density (Kretser et al, 2008) and bird–habitat preference indices (Fuller et al, 2005) proved to be directly related to the richness of wildlife. Few studies, however, have included the consideration of 3D landscape patterns. Due to the differences in landscape composition, management and functions, some factors (e.g. the frequency and intensity of specific ecological processes, the diversity of wildlife and vegetation, sustainability of farm land use, etc.) that can be common and useful criteria for rural or wild landscape pattern evaluation, are not generally practical for urban landscape pattern evaluation. Since the main role of cities is to serve urban residents, the evaluation of urban landscape patterns should place more emphasis on people’s needs. Amongst these needs, people’s aesthetic preferences are closely related to the landscape patterns. One case study (Chen et al., 2016) is therefore conducted to explore the public’s aesthetic preferences towards some urban landscape patterns. Integrated with the landscape pattern analysis, people’s preferences concluded in this study can be employed as criteria for evaluating urban landscape patterns.

5.1.2 The public’s aesthetic preferences towards urban landscape patterns To date, landscape aesthetics has proven to be closely linked to stress relief (Ulrich et al., 1991), mental and physical benefits (Kaplan et al., 1998; Kaplan and Kaplan, 1989; Hartig et al., 1991; Ode and Fry, 2002; Bhatti and Church, 2003) and attraction to migrants (Waltert and Schlapfer, 2010). As a result, growing research emphasis and policy priorities have been put on evaluating and optimizing local landscapes, and a large body of studies has been conducted to understand the public’s landscape preferences. Ulrich (1993) concluded that people

Urban Landscape Pattern Evaluation

111

from various cultures like natural landscapes in preference to human-influenced landscapes. In spite of the general consensus on the public’s preferences towards natural landscapes, some research pointed out that substantial group differences existed in landscape preferences. Landscape preferences have been found to differ with age (Balling and Falk, 1982; Lyons, 1983; Zube et al.,1983; Howley, 2011; Howley et al., 2012), ethnical groups (Herzog et al., 2000; Ho et al., 2005; Buijs et al., 2009; Suckall et al., 2009), educational level (Yu, 1995; Canas et al.,2009), cultures and geographic locations (Purcell et al., 1994; Rauwald and Moore, 2002), living environments (Yu, 1995; Van den Berg and Koole, 2006) and occupational groups ( Yu, 1995; Van Den Berg et al. 1998; Brush et al., 2000), However, Benjamin et al (2007) pointed out that some commonly used variables such as age and educational level may have weak influences on landscape perception. The appearance of the landscape can affect people’s perceptions of their everyday environment (Ode et al., 2009) and theories of landscape aesthetics have suggested that landscape patterns could be used to predict landscape preferences (Kaplan and Kaplan, 1989; Tveit et al., 2006). Tveit et al (2006) identified nine key concepts describing visual characters, including stewardship, coherence, disturbance, historicity, visual scale, imageability, complexity, naturalness and ephemera, However, there are few systematic studies that test for relationships between visual indicators and landscape preferences (Ode et al., 2009). Aesthetic preferences are usually analyzed using such research methods as questionnaires (Ozguner and Kendle, 2006; Sayadi et al., 2009; Buijs et al., 2009; Canas et al.,2009; Tveit, 2009; Sevenant and Antrop, 2009; Banski and Wesoáowsk, 2010; Howley, 2011; Zheng et al., 2011; Howley et al., 2012; Jin et al., 2013), online survey (Wherrett, 1999; Roth, 2006; Lee and Kozar, 2009; Ode et al., 2009; Smith et al, 2012) and so forth. Furthermore, with the development of remote sensing and GIS (Geographic Information System) techniques, researchers can establish more accurate and detailed landscape models. Based on the improved quantitative landscape pattern analysis, some specific visual indicators (Dramstad et al., 2006; Tveit, 2009; Ode et al., 2009) have been designed to relate comprehensive landscape patterns to people’s aesthetic preferences. Therefore, the public’s aesthetic preferences towards comprehensive landscape configuration have been well examined.

112

Chapter Five

However, large-scale landscape planning is subject to land use policies and the enhancement of landscape patterns based on people’s landscape preferences is not always feasible. Therefore, adjusting the spatial patterns of individual urban features may be an alternative approach to improve people’s perception of local landscapes. The aim of this case study is to assess the public’s aesthetic preferences towards urban landscape patterns. In contrast to some previous studies focusing on the correlation between landscape preferences and comprehensive landscape patterns (general visual indicators), this research mainly explores people’s preferences towards the composition and structure of individual urban features. Hence, the methodology and results have the potential to provide landscape planners and policy makers with decision support from a micro-scale perspective whilst previous studies mainly focus on the general configuration of local landscapes. In addition to general landscape preferences, group differences in the landscape aesthetics and the correlations between landscape preferences towards different urban features are discussed in this chapter.

5.2 Methodology and study sites This case study employs a well-structured questionnaire to gain some insight into people’s aesthetic preferences towards the composition and structure of urban features. Previous research on people’s aesthetic preferences has commonly been criticized for being subjective and lacking in standardization and replicability (Bruns and Green, 2001; Daniel, 2001; Terkenli, 2001; Ndubisi, 2002), and not being feasible to integrate with quantitative pattern analysis. To acquire more applicable landscape evaluation criteria, the questionnaire should be designed to link landscape metrics to people’s aesthetic preferences. In this case, the findings from this research may also be applied to other areas.

5.2.1 Questionnaire design The aim of this questionnaire-based survey is to gain some insight into the public’s general landscape preferences. Integrated with the results of landscape pattern analysis (described in Chapter 4), these preferences can be employed as criteria for urban landscape pattern

Urban Landscape Pattern Evaluation

113

evaluation. To better correlate with 3D urban landscape pattern analysis, questions included in this questionnaire are specifically designed according to landscape metrics proposed above. Since comprehensive landscape patterns and the structure of individual urban features can be accurately measured, landscape patterns may thus be evaluated, if the public’s landscape preferences towards some urban landscape patterns can be acquired. Therefore, questions included in the survey aimed to explore people’s preferences towards such urban features as buildings, trees and green spaces, which can be directly correlated with specific landscape metrics and employed for landscape pattern evaluation. The method of linking corresponding landscape metrics to the design of the questionnaire is explained as follows. Take the tree shape as an example. Since the general shape of trees can be understood using the metric “Mean Height Deviation Within Individual Trees”, then the value of this metric can be used to compare the general tree shape between different sites. However, without evaluation criteria, researchers cannot decide what tree shape is preferable. To fill this gap, such a question as “your preference towards tree shapes” is included in the questionnaire. Once enough qualified responses are received, researchers can know whether the freely-growing shape or the pruned shape is preferred. A large value of “Mean Height Deviation Within Individual Trees” indicates a large proportion of freely growing trees whilst a small value of this metric usually indicates a large proportion of regularly shaped trees. Correlated with responses to the question “your preference towards tree shapes”, the value of this metric “Mean Height Deviation Within Individual Trees” can thus be used for comparing and evaluating landscape patterns. Following the same principle, other questions concerning preferences towards building shapes, tree-building patterns and so forth, were also designed specifically to correlate with corresponding landscape metrics provided in Chapter 4 (correlations between landscape preferences and landscape metrics will be further discussed in the following section). Since the survey is not an on-site interview (an on-site interview means that respondents are brought to the actual landscapes described in the questionnaire), photograph materials are used in the survey. Although photographic images are widely employed due to the time-efficiency and low cost they offer (García et al., 2006), this approach has been criticized since it does not entirely reflect the complex reality of landscape, which includes such factors as temperature, smell and so

114

Chapter Five

forth. To improve the reliability of the questionnaire, questions proposed in this survey only concern the preferences towards spatial patterns of individual urban features, not the comprehensive feeling of different landscape types. The photos used in the research were taken in some cities within UK. Only photos describing some generalized urban feature types were selected to eliminate possible effects from unique local landscape patterns. All these photos were taken under similar weather conditions to reduce influences from such environmental factors as cloud cover or visibility. These photos were taken with focal lengths that framed their subjects in order to reduce the influences from neighbouring features. In addition, brief text description was provided that to assist respondents’ understanding of the presented landscape patterns. This questionnaire consists of two parts, socio-demographic information and aesthetic preferences towards landscape patterns. The former includes such social factors as age, gender, education and so forth. The latter includes the preference towards comprehensive setting of landscape patterns (e.g. tree-building relationship, Fig 5.1) as well as the pattern of individual urban features (e.g. lawns, Fig 5.2; trees, Fig 5.3; buildings, Fig 5.4; roofs, Fig 5.5; urban forests, Fig 5.6). Due to the limitations of interview-based survey (e.g. respondents are usually not willing to answer many questions), not all urban features are included in the survey. Instead, this questionnaire mainly focuses on the structure of urban features that are closely related to people’s daily life. In these days, people are very concerned about their environments. Trees, buildings and green spaces are indispensable parts of residential neighborhood, and their structure has significant influences on residents’ aesthetic perception. Therefore, this questionnaire mainly focused on these urban features. Other important features, such as water, which are not available in all landscape scenarios and has been discussed by previous studies (Yu, 1995; Kaltenborn and Bjerke., 2002; Canas et al., 2009; etc.), were not included in this study. Considering the research aim, limitations and available landscape metrics, eleven questions: “your attitude towards landscape patterns”, “your attitude towards vertical landscape patterns”, “your preference towards the distance between your home and the nearest green space”, “your preference towards the design of green spaces”, “your preference towards the size of green spaces”, “your preferences towards the shape of trees”, “your preference towards the tree-building pattern”, “your

Urban Landscape Pattern Evaluation

115

preference towards the height diversity of trees”, “your preference towards building types”, “your preference towards the roof types” and “your preference towards spatial patterns of urban buildings” were included in the questionnaire (more details of these questions are presented in the following section). A five-point scale, ranging from 1 ‘Dislike’ to 5 ‘Like’, was applied for those questions concerning the evaluation of people’s preferences towards different structures.

c. a. Trees higher than neighbouring buildings

b. Trees and buildings with similar height

c.Trees shorter than neighbouring buildings

Fig 5.1. Tree-building patterns

a. Lawns only Fig 5.2 Structure of lawns

b. Mixture of lawns and trees

Chapter Five

116

a. Rectangular shape

d. Vase Shape

b. Round Shape

c. Cone shape

e. Freely Growing Shape

Fig 5.3 Structure of trees

a. High rise buildings

b. Multi-layer buildings

c. Individual houses

d. Single unit flats

Fig 5.4 Structure of buildings

Urban Landscape Pattern Evaluation

a. Flat roof

b. Gable roof

c. Curved roof

d. Gothic roof

117

Fig 5.5 Structure of roofs

a. Trees of different height

b. Trees of similar height

Fig 5.6 Patterns of urban forests

5.2.2 Study sites and survey procedure In addition to general preferences towards landscape patterns, a comparative research strategy was applied to examine the cultural influence on the public’s aesthetic preferences. As pointed out by some researchers (Purcell et al., 1994; Rauwald and Moore, 2002), landscape preferences may differ significantly between cultures and geographic locations. If comparative surveys are conducted in cities in the same country (e.g. the UK), the results cannot prove the existence of some

118

Chapter Five

shared landscape preferences. In this case, the methodologies and findings of these surveys may not be generalized to a series of areas across different geographical locations and cultures. Therefore, a cross-culture study is required to enhance understanding of what landscape preferences can be generally applied to different areas and cultures. Cities in China and the UK are usually of different landscape patterns, and residents in the two countries are from different cultural backgrounds. Hence, if some landscape preferences are shared in both countries, these preferences are more likely to be generally applicable. The questionnaire-based survey was conducted in the city of Cambridge, UK and the city of Nanjing, China. Residents in both sites enjoy significant area of urban greenery. The green ratio in Nanjing is 45%, ranking the first amongst Chinese cities whilst central Cambridge also has a very large proportion of green spaces and the total cover of urban vegetation is over 60%. Cambridge is a famous university city with a typical British urban landscape that has a large proportion of individual houses whilst Nanjing has a typical Chinese urban landscape that is characterized with a large proportion of high rise buildings. As a result, responses from the two sites can be used to compare the aesthetic preferences of different cultural groups. The interview-based survey was mainly carried out in some green spaces (e.g. Jesus Green, Parkside Green at Cambridge and Xuanwu Lake Park at Nanjing) where people are more likely to participate in a detailed survey with four pages and more than 10 questions. A random sampling strategy, which was widely used in previous studies (Kaltenborn and Bjerke, 2002; Ozguner and Kendle, 2006; Canas et al., 2009; etc.), was employed for the interview. To have an unbiased sample in terms of the age, gender, education, status and so forth, there was no pre-selection of respondents. Therefore, in the process of the interview, this researcher randomly invited people (including students and local residents) who were in the sample area (green spaces introduced above) to take part in the survey. As an interview survey, the researcher endeavored to explain details to respondents when confusion occurred, and to make sure that most questionnaires were finished with respondents’ serious thoughts. The average time spent for each respondent to fill in the questionnaire was around 10 minutes. In the survey, young people were more willing to

Urban Landscape Pattern Evaluation

119

be interviewed than the elderly, which was reflected in the imbalanced distribution of different age groups.

5.3 Results The survey was conducted in Cambridge and Nanjing in July and August, 2012. Following a validation check, unqualified responses (e.g. respondents failed to answer some questions or used the same score for all evaluation questions) were removed and thus 182 respondents from Cambridge and 180 respondents from Nanjing were included in the statistics and analysis. The socio-demographic information of this survey is shown in Table 5.1. Table 5.1 Socio-demographic information of the research population

Gender:

Male Female Age: 50 Highest education: Primary school Secondary school Undergraduate Master PhD Major1: Liberal arts Science and engineering Status: Student Resident Tourist from other countries Length of Living in the UK2:

1

Less than 1 year 1-3 years 3-5 years More than 5 years

Nanjing

Cambridge

112 68 95 38 18 12 17 1 17 75 71 16 47 117 99 81

102 80 110 39 13 7 13 3 41 71 35 32 81 77 114 55 13 Cambridge 28 17 12 125

Optional question. Respondents can choose whether or not to answer this

120

Chapter Five

question. Some respondents did not go to college and thus had no fixed major in their educational career. But they may also choose to answer this question according to their interest in different majors. In the survey, when respondents were not sure about their major, a unified suggestion was given for their reference, e.g. economics and law were regarded as “Liberal arts” in the present research. 2

International students in Cambridge and the UK take up a large proportion of total students whilst there are very limited overseas students (none in the survey) in China. To reduce the biased response from international students who just arrive in the UK as well as foreign tourists, the term “Length of living in the UK” was included in the UK survey.

From Table 5.1, one can see that there is no significant difference between the Chinese survey and the UK survey in terms of social-demographic characteristics, except for the category of “Major”. This difference is consistent with the general setting of college courses in the two countries. In China, many more students choose courses related to “science and engineering” than those who choose courses concerning “liberal arts” whilst the distribution of majors is more balanced in the UK. Additionally, since Cambridge is a university-based area, the proportion of students is larger than most cities.

5.3.1 General aesthetic preferences Since there are no significant differences in social and demographic characteristics between the two samples, responses from the two groups of respondents can generally be compared with little bias from the composition of samples. The comparison of general aesthetic preferences towards landscape patterns is presented in Table 5.2.

Nanjing

121

Cambridge ( adjusted1) 1 Your attitude towards local landscape planning (the composition and patterns of trees, buildings, green space, etc) is 5 very concerned 1 not interested Mean (STD) 3.89 (1.03) 3.51 (1.06) 2 Your attitude towards local landscape planning at the vertical direction (building height, tree shapes, heterogeneity of tree height) is 5 very concerned 1 not interested Mean (STD) 3.48 (1.17) 3.31 (1.04) 81.6 % 77.9 % 3 Your preference towards the distance between your home and the 0.1 Green=2.4169*(distance)-1.0284 0.054 0.192 15 Mean Largest Relative error relative error1 relative error > 50% Green=2.4169*(distance)-1.0284 45.37% 277.02% 20 Relative error: Abs(estimated value - observed value)/ observed value*100%

This type of large bias mainly resulted from lacking variables that described the tree size. As the 93 sample trees included in the present study were of different sizes and large trees can provide much more green visibility than small trees from the same distance away, this regression model was more likely to produce large bias when analyzing large or small trees. To enhance the reliability, more variables concerning the tree size need to be added to the regression model as well as Distance. According to the result of correlation analysis, there were no other direct variables correlated with tree green visibility. Therefore, composite variables proposed using Distance and other variables were required for better regression models. Yang et al. (2009) indicated that such parameters as Average height of viewable trees/shrubs normalized by their distances to the photographer (AHT) were correlated with tree green visibility. Inspired by previous research, several composite variables were designed. Since Area, Perimeter, Mean and Max were direct indicators of the tree size and strongly correlated with each other, some composite variables were designed using Distance and these variables respectively. The statistics and comparison of different regression models are shown in Fig 6.6 and Table 6.3

2 0.575

Green=0.210*(Area/ Distance2)0.522

Green=1.081*(Mean/ Distance )

2 0.556

Green=0.897*(Max/ Distance2)0.574

Green=0.409*(Perimeter/Distance )

Green=0.210*(Area/ Distance )

2 0.522

Green=1.081*(Mean/ Distance2)0.556

Green=0.897*(Max/ Distance )

2 0.574

Green=0.409*(Perimeter/Distance2)0.575

.765

.811

.845

.834

R2

Largest

.000

.000

.000

.000

Sig

38.77%

36.69%

34.39%

34.20%

240.02%

152.51%

132.66%

136.07%

.0489

.0478

.0464

Mean error .0425

.211

.162

.188

> 50%

25

21

18

17

Larges t error .153

Relative error

relative error relative error

Mean

296.3

390.5

496.6

457.0

F

Table 6.3 Statistics of regression models established using different composite variables

Supporting Specific Research with 3D Urban Landscape Models

13

13

9

Error > 0.1 7

161

Chapter Six

Fig 6.6 Regression models of tree green visibility using composite variables

162

Supporting Specific Research with 3D Urban Landscape Models

163

According to Table 6.3 and Fig 6.6, all these composite variables were strongly correlated with tree green visibility. The regression models established using these variables achieved better results than the regression model using the only variable Distance in terms of R2, mean error and so forth. Among the four composite variables, the regression model built using “Area/ Distance2” had the smallest R2 and the most large errors. The possible reason may be explained as Fig 6.7 shows. In the horizontal direction, tree green visibility is directly related to the crown diameter of the trees. On an orthophoto map, the outline of ordinary trees is near-circular in shape. In this case, the map derived variable Perimeter § ʌ * Diameter and the variable Area §ʌ * (Diameter/2)2 is . Therefore, the variable Perimeter is linearly correlated with available tree size in the horizontal direction whilst the variable Area is not.

Fig 6.7 Tree green visibility from viewers’ perspective

Although the R2 and mean bias of regression models with composite variables were satisfactory, a large amount of sample trees with obvious error still existed. As a result, the factor of tree shapes, as well as tree sizes, should be added to these models. To further improve regression models, another variable, Range was included in composite variables. When the size of a tree object and the distance between the tree and the viewer are fixed, a larger Range value is more likely to produce larger green visibility. Therefore, several advanced composite variables were designed, such as Perimeter*Range/Distance2, Max*Range/Distance2, Mean*Range/Distance2 and Area*Range/Distance2. Based on these variables, regression models were established and compared (Table 6.4 and Fig 6.8).

Chapter Six

235.0

389.9

439.1

Largest

.000

.000

.000

.000

Sig

36.76% 42.78%

Green=0.223*(Area*Range/ Distance2)0.477

32.91%

Green=0.920*(Max*Range/Distance2)0.545 2 0.539

16.73%

334.21%

173.30%

121.36%

53.73%

.182

.175

.174

> 50%

9

1

12

13

11

4

> 0.1

Error

23

20

Relative error

.0479

.0471

.0490

.125

error

error .0413

Largest

Mean

relative error relative error

Green=0.451*(Perimeter*Range/Distance2)0.559

Green=1.143*(Mean*Range/Distance )

F

468.8

Mean

.721

Green=0.223*(Area*Range/ Distance )

.811

2 0.477

.828

.837

Green=1.143*(Mean*Range/Distance2)0.539

Green=0.920*(Max*Range/Distance )

2 0.545

Green=0.451*(Perimeter*Range/Distance2)0.559

R2

Table 6.4 Statistics of regression models established using advanced composite variables

164

Fig 6.8 Regression model of tree green visibility using advanced composite variables

Supporting Specific Research with 3D Urban Landscape Models 165

166

Chapter Six

In accordance with Table 6.4 and Fig 6.8, the accuracy of regression models using composite variables Perimeter*Range/Distance2 and Max*Range/Distance2 was significantly improved from previous models in terms of the number of sample units with large bias. Both models had small mean error and fewer than 10% of sample trees had large relative error. The model Green=0.451*(Perimeter*Range/Distance2)0.559 achieved the optimum regression result. Through simulation, only 4 sample trees had large absolute bias and 1 tree had large comparative bias. In addition, the largest bias occurred in this model was the smallest amongst all regression models. The reason why the combination of Perimeter and Range had a better regression result may be that Perimeter and Range feature the characteristics of trees in different directions whilst both Max and Range feature the size and shape of trees in the vertical direction.

6.5 Cross-validation To examine the validity of the optimum model Green=a*(Perimeter*Range/Distance2)b K-fold cross-validation was adopted for accuracy assessment. According to the size of the sample, 93 sample units were randomly divided into 10 groups: 7 groups with 9 units and 3 groups with 10 units. Each time, 9 groups were used for training and 1 group was used for validation. Iteratively, each sample unit was used for the accuracy assessment once. The result of cross-validation is shown in Table 6.5.

0.840 0.843 0.844 0.865 0.854 0.841 0.839 0.825 0.768 0.840

Green=0.447*(Perimeter*Range/Distance2)0.561

Green=0.459*(Perimeter*Range/Distance2)0.563

Green=0.444*(Perimeter*Range/Distance2)0.560

Green=0.450*(Perimeter*Range/Distance2)0.565

Green=0.458*(Perimeter*Range/Distance2)0.572

Green=0.459*(Perimeter*Range/Distance2)0.568

Green=0.458*(Perimeter*Range/Distance2)0.561

Green=0.449*(Perimeter*Range/Distance2)0.559

Green=0.412*(Perimeter*Range/Distance2)0.504

Green=0.455*(Perimeter*Range/Distance2)0.558

1

2

3

4

5

6

7

8

9

10

Mean error and total large error

R2

Regression Model

ID

Table 6.5 The result of cross-validation

0.041

0.033

0.040

0.034

0.030

0.016

0.035

0.063

0.035

0.063

0.064

Mean Error

0.127

0.066

0.078

0.094

0.060

0.032

0.074

0.127

0.064

0.102

0.117

Largest Error

Supporting Specific Research with 3D Urban Landscape Models

4

0

0

0

0

0

0

2

0

1

1

Error >0.1

167

168

Chapter Six

Table 6.5 proved the validity of this regression model. Each test obtained similar regression parameters with satisfactory R2 and small mean error. In addition, the amount of sample units with large error was controlled efficiently. Therefore, this regression model has the potential to be generalized to other study sites.

6.6 Discussion Most trees in the study site have a freely growing shape. As a result, some trees included in the sample have irregular crown shapes (Fig 6.9). According to error statistics, many large errors in the regression model resulted from these sample trees. As a result, regression models proposed in this research may be more effective when applied to some sites with unified and regular tree shapes.

Fig 6.9 Some sample trees with abnormal ratio of height to perimeter

The regression models of tree green visibility can be applied for specific purposes. By calculating the distance between buildings and their surrounding trees, as well as characteristics of tree structures, researchers can calculate the total tree green visibility for each building in the study site and the general affluence of tree greenery in a study site. For instance, through programming and spatial data analysis, the mean tree green visibility around one building in Cambridge is 0.46 whilst the mean tree green visibility around one building in Canvey is 0.81 (without considering overlapping effects between neighbouring trees). Although the total tree cover and building cover in Cambridge is similar to that in Canvey, and trees in Cambridge are much taller than those in Canvey, the value of this metric in Cambridge is much smaller

Supporting Specific Research with 3D Urban Landscape Models

169

than that in Canvey due to the spatial distribution of trees and buildings. Therefore, tree green visibility can be an efficient metric to examine the tree-building pattern and to analyze residents’ perception of urban green presence. Instead of large-scale field photograph collection, all the process of tree green visibility analysis can be conducted automatically with the aid of GIS tools. Based on the analysis of tree green visibility, urban planners may make some adjustments on local landscape patterns to enable people from most buildings to enjoy tree green presence. Based on the urban land cover classification images and nDSM information, variables concerning tree shapes were obtained for the regression model. In addition to the comprehensive urban land cover classification, a large body of studies (Nelson et al., 1984, 1988; Næsset 1997a, 1997b; Persson and Holmgren, 2002; Brandtberg et al., 2003; Leckie et al., 2003; Popescu et al., 2003; Holmgren and Persson, 2004; Morsdorf et al., 2004; Peuhkurinen et al., 2007; Yu et al., 2011; Vastaranta et al., 2012; etc.) have been conducted to design specific algorithms for tree structure extraction and modelling. To examine the performance of different methods, Kaartinen et al. (2012) conducted a comparative study that included more than 10 algorithms. According to previous studies, future researchers may also employ suitable tree exaction methods to obtain necessary parameters of tree shapes and the accuracy of the model may be further improved. Research that assesses tree green visibility can be further improved. In this study, all the photographs were taken from the sides of streets or edges of buildings. In future studies, photographs can be taken from different floors of buildings (Fig 6.10). Therefore, by adding the vertical position of viewers to a regression model, people’s perception of tree green affluence from different floors may be quantified.

Ground floor

First floor

Fig 6.10 Tree green visibility from different floors

Second floor

170

Chapter Six

All photographs for this study were taken in the summer, which excluded the differences between evergreen types and deciduous types. If the on-site survey and regression analysis can be conducted in different seasons, a refined model with a seasonal factor may be established. In addition, the factor of tree species has significant impacts on tree shapes and their understory structure, which are closely related to tree green visibility. Researchers can analyze the characteristics of specific tree types and adjust the regression model for individual tree types. In this case, the regression model can achieve a better accuracy in some urban forests that consist of unified tree species. The view-blocking effects from buildings and the overlapping effects from neighbouring trees, which are excluded in the present case study, can be further examined in future studies.

6.7 Conclusions 3D urban landscape models can support specific research with fine and reliable data sources and herein, the analysis of tree green visibility based on 3D landscape models was conducted as an example. The visual effects of urban greenery are closely related to people’s perceptions of local environments, yet very limited research has been conducted on the design of research methods that involve quantitative analysis. Based on the established 3D urban landscape model and photographs collected through on-site survey, this research has explored the correlations between green visibility from individual trees and variables that can be acquired from the 3D landscape model. As proposed by previous studies, average height of trees/ crowns normalized by their distances to the photographer was correlated with visual greenery. This study also proved that some composite variables, Perimeter*Range/Distance2, such as Perimeter/Distance2, 2 2 Max/Distance , and Mean/Distance were strongly correlated with tree green visibility. In addition, some regression models were designed using variables derived from the 3D landscape model to quantitatively estimate tree green visibility and achieved satisfactory effects. Amongst these models, the model form Green=a*(Perimeter*Range/Distance2)b

Supporting Specific Research with 3D Urban Landscape Models

171

proved to be the optimum model. This model had small mean error and the amount of sample units with large error was controlled efficiently. Considering the diversity of sample trees and the results of cross-validation, this model has the potential to be applied to other areas. Compared with on-site photography, it is more feasible for researchers to conduct a large-scale project for assessing tree green visibility using 3D urban landscape models. In line with the public’s growing emphasis on local environment and natural perception, tree green visibility is becoming an important factor related to people’s landscape preferences. As a result, this research provides urban planners and decision makers an approach to analyze and improve local landscape patterns by providing people with more available tree green. Reliable position and pattern information of different urban features from 3D landscape models have the potential to replace time-consuming on-site survey in a diversity of research subjects. In addition to the UFORE model and tree green visibility analysis, researchers from different disciplines may explore the applications of 3D urban landscape models in the broad scope of landscape ecology.

References Alonso, R., Vivanco, M. G., González-Fernández, I., Bermejo, V., Palomino, I., Garrido, Juan., Elvira, S., Salvador, P., Artíñano, B. (2011) Modelling the influence of peri-urban trees in the air quality of Madrid region (Spain). Environmental Pollution. 159(8–9), 2138-2147. Aoki, Y. (1991) Evaluation methods for landscapes with greenery. Landscape Research. 16, 3–6. Arnberger, A. (2006) Recreation use of urban forests: An inter-area comparison. Urban Forestry & Urban Greening. 4, 135–144. Baumgardner, D., Varela, S., Escobedo, F.J., Chacalo, A., Ochoa, C. (2012) The role of a peri-urban forest on air quality improvement in the Mexico City megalopolis. Environmental Pollution. 163(10), 174-183. Beckett, K.P., Freer-Smith, P.H., Taylor, G. (1998) Urban woodlands: their role in reducing the effects of particulate pollution. Environmental Pollution. 9, 347–360.

172

Chapter Six

Bernath, K., Roschewitz, A. (2008) Recreational benefits of urban forests: Explaining visitors’ willingness to pay in the context of the theory of planned behavior. Journal of Environmental Management. 89, 155–166. Bhatti, M., Church, A. (2003) Home, the culture of nature and meanings of gardens in late modernity. Housing Studies. 19 (1), 37–51. Brandtberg, T., Warner, T., Landenberger, R., McGraw, J. (2003) Detection and analysis of individual leaf-off tree crowns in small footprint, high sampling density lidar data from the eastern deciduous forest in North America. Remote Sensing of Environment. 85, 290-303. Buhyoff, G.J., Gauthier, L.J., Wellman, J.D. (1984) Predicting scenic quality for urban forests using vegetation measurements. Forest Science. 30, 71–82. Chen, Z.Y., Xu, B, Gao, B.B. (2015) Assessing visual green effects of individual urban trees using airborne Lidar data. Science of the Total Environment. 536, 232-244 Escobedo, F.J., Wagner, J.E., Nowak, D.J., De la Maza, C.L., Rodriguez, M., Crane, D.E. (2008) Analyzing the cost effectiveness of Santiago, Chile’s policy of using urban forests to improve air quality. Journal of Environmental Management. 86, 148–157. Hartig, T., Mang, M., Evans, G.W. (1991) Restorative effects of natural environment experiences. Environment and Behavior. 23 (1), 3–26. Holmgren, J., Persson, Å. (2004) Identifying species of individual trees using airborne laser scanning. Remote Sensing of Environment. 90, 415-423. Hörnsten, L., Fredman, P. (2000) On the distance to recreational forests in Sweden. Landscape and Urban Planning. 51(1), 1-10. Jim, C.Y., Chen, W.Y. (2008) Assessing the ecosystem service of air pollutant removal by urban trees in Guangzhou (China). Journal of Environmental Management. 88(4), 665-676. Jim, C.Y., Chen, W.Y. (2009) Ecosystem services and valuation of urban forests in China. Cities. 26,187–194. Kaartinen, H., Hyyppä, J., Yu, X., Vastaranta, M., Hyyppä, H., Kukko, A., et al. 2012. An international comparison of individual tree detection and extraction using airborne laser scanning. Remote Sensing. 4(4), 950-974. Kaplan, R., Kaplan, S. (1989) The Experience of Nature. Cambridge University Press, Cambridge.

Supporting Specific Research with 3D Urban Landscape Models

173

Kaplan, R., Kaplan, S., Ryan, R.L. (1998) With People in Mind. Island Press, Washington, DC. Kenney,W.A. (2000) Leaf area density as an urban forestry planning and management tool. The Forestry Chronicle. 76, 235–239. Kraus, K., Pfeifer, N. (1998) Determination of terrain models in wooded areas with airborne laser scanner data. ISPRS Journal of Photogrammetry and Remote Sensing. 53, 193–203. Kraus, K., Pfeifer, N. (2001) Advanced DTM generation from LIDAR data. International Archives of Photogrammetry and Remote Sensing, Annapolis, MD, 22-24 Oct. 2001, XXXIV (part 3/W4): 23–30. Leckie, D., Gougeon, F., Hill, D., Quinn, R. (2003) Armstrong, L.; Shreenan, R. Combined high-density lidar and multispectral imagery for individual tree crown analysis. Canadian Journal of Remote Sensing. 29, 633-649. Lohmann, P. (2002) Segmentation and filtering of laser scanner digital surface models. International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. 34 (part XXX), 311–315. McPherson, E.G. (1998) Atmospheric carbon dioxide reduction by Sacramento’s urban forest. Journal of Arboriculture. 24, 215–223. McPherson, E.G., Simpson, J.R. (2000) Carbon dioxide Reductions through Urban Forestry: Guidelines for Professional and Volunteer Tree Planters. PSWGTR-171. USDA Forest Service, Pacific Southwest Research Station, Albany, CA. McPherson, E.G., Simpson, J.R., Pepper, P.J., Xiao, Q. (1999) Benefit cost analysis of Modesto’s municipal urban forest. Journal of Arboriculture. 25 (5), 235–248. Morsdorf, F., Meier, E., Kötz, B., Itten, K. I., Dobbertin, M., Allgöwer, B. (2004) LIDAR-based geometric reconstruction of boreal type forest stands at single tree level for forest and wildland fire management. Remote Sensing of Environment. 3, 353-362. Næsset, E. (1997a) Determination of mean tree height of forest stands using airborne laser scanner data. ISPRS Journal of Photogrammetry and Remote Sensing. 52, 49-56. Næsset, E. (1997b) Estimating timber volume of forest stands using airborne laser scanner data. Remote Sensing of Environment. 61, 246-253. Nardinocchi, C., Forlani, G., Zingaretti, P. (2003) Classification and filtering of laser data. International Archives of Photogrammetry

174

Chapter Six

and Remote Sensing. XXXIV (part 3/W13). Nelson, R., Krabill, W., Maclean, G. (1984) Determining forest canopy characteristics using airborne laser data. Remote Sensing of Environment. 15, 201-212. Nelson, R., Krabill, W., Tonelli, J. (1988) Estimating forest biomass and volume using airborne laser data. Remote Sensing of Environment. 24, 247-267. Nowak, D.J. (2006) Institutionalizing urban forestry as a “biotechnology” to improve environmental quality. Urban Forest & Urban Greening. 5, 93–100. Nowak, D.J., Crane, D.E. (2000) The Urban Forest Effects (UFORE) Model: quantifying urban forest structure and functions. In: Hansen, M. and T. Burk (Eds.) Integrated Tools for Natural Resources Inventories in the 21st Century. Proc. Of the IUFRO Conference. USDA Forest Service General Technical Report NC-212. North Central Research Station, St. Paul, MN. pp. 714-720. See also http://www.ufore.org. Nowak, D.J., Crane, D.E. (2002) Carbon storage and sequestration by urban trees in the USA. Environment Pollution. 116, 381–389. Ode, Å., Fry, G. (2002) Visual aspects in urban woodland management. Urban Forestry & Urban Greening. 1, 15–24. Ong, B.L. (2003) Green plot ratio: an ecological measure for architecture and urban planning. Landscape and Urban Planning. 63, 197–211. Persson, Å., Holmgren, J., Söderman, U. (2002) Detecting and measuring individual trees using an airborne laser scanner. Photogrammetric Engineering & Remote Sensing. 68, 925-932. Peuhkurinen, J., Maltamo, M., Malinen, J., Pitkänen, J., Packalén, P. (2007) Preharvest measurement of marked stands using airborne laser scanning. Forest Science. 53, 653-661. Popescu, S., Wynne, R., Nelson, R. (2003) Measuring individual tree crown diameter with lidar and assessing its influence on estimating forest volume and biomass. Canadian Journal of Remote Sensing. 29, 564-577. Rosenberger, R.S., Needham, M.D., Morzillo, A.T., Moehrke, C. (2012) Attitudes, willingness to pay, and stated values for recreation use fees at an urban proximate forest. Journal of Forest Economics. 18( 4), 271-281 Rowntree, R.A. (1984) Ecology of the urban forest-introduction to Part I. Urban Ecology. 8, 1- 11.

Supporting Specific Research with 3D Urban Landscape Models

175

Rowntree, R.A. (1986) Ecology of the urban forest-introduction to Part II. Urban Ecology. 9, 229-243. Sanders, R.A. (1986) Urban vegetation impacts on the hydrology of Dayton, Ohio. Urban Ecology. 9, 361–376. Schroeder, H.W. (1988) Visual impact of hillside development: comparison of measurements derived fromaerial and ground-level photographs. Landscape and Urban Planning. 15, 119–126. Schroeder, H.W., Cannon Jr., W.N. (1983) The esthetic contribution of trees to residential streets in Ohio towns. Journal of Arboriculture. 9, 237–243. Scott, K.I., McPherson, E.G., Simpson, J.R. (1998) Air pollutant uptake by Sacramento’s urban forest. Journal of Arboriculture. 24 (4), 224–233. Thayer, R.L., Atwood,W.G. (1978) Plants, complexity and pleasure in urban and suburban environment. Environmental psychology and nonverbal behavior. 3, 67–76. Tovari, D., Pfeifer, N. (2005) Segmentation based robust interpolation-a new approach to laser data filtering. Workshop Laser scanning. 2005, pp, 79–84. Ulrich, R.S. (1986) Human responses to vegetation and landscapes. Landscape and Urban Planning. 13, 29–44. Ulrich, R.S., Simons, R.F., Losito, B.D., Fiorito, E., Miles, M.A., Zelson, M. (1991) Stress recovery during exposure to natural and urban environments. Journal of Environmental Psychology. 11, 201–230. Van Herzele, A., De Clercq, E.M., Wiedemann, T. (2005) Strategic planning for new woodlands in the urban periphery: through the lens of social inclusiveness. Urban Forestry & Urban Greening. 3, 177–188. Vastaranta, M., Kankare, V., Holopainen, M., Yu, X., Hyyppä, J., Hyyppä, H. (2012) Combination of individual tree detection and area-based approach in imputation of forest variables using airborne laser data. ISPRS Journal of Photogrammetry and Remote Sensing. 67, 73-79. Waltert, F., Schlapfer, F. (2010) Landscape amenities and local development: a review of migration, regional economic and hedonic pricing studies. Ecological Economics. 70, 141–152. Xiao, Q.F., McPherson, E.G., Simpson, J.R. (1998) Rainfall interception by Sacramento’s urban forest. Journal of Arboriculture. 24, 235–244.

176

Chapter Six

Xiao, Q.F., McPherson, E.G., Ustin, S.L., Grismer,M.E. (2000) A new approach to modeling tree rainfall interception. Journal of Geophysical Research. 16, 29173–29188. Yang, J., McBride, J., Zhou, J., Sun, Z. (2005) The urban forest in Beijing and its role in air pollution reduction. Urban Forestry & Urban Greening. 3, 65–78. Yang, J., Zhao, L.S., Mcbride, J., Gong, P. (2009) Can you see green? Assessing the visibility of urban forests in cities. Landscape and Urban Planning. 91, 97–104. Yu, X., Hyyppä, J., Vastaranta, M., Holopainen, M. Viitala, R. (2011) Predicting individual tree attributes from airborne laser point clouds based on random forests technique. ISPRS Journal of Photogrammetry and Remote Sensing. 66, 28-37. Zhao, M., Kong , Z.H., Escobedo , F.J., Gao, J. (2010) Impacts of urban forests on offsetting carbon emissions from industrial energy use in Hangzhou, China. Journal of Environmental Management. 91(4), 807-813.

CHAPTER SEVEN SUGGESTIONS FOR URBAN LANDSCAPE PLANNING AND IMPROVEMENT

Although sub-disciplines in urban landscape ecology include a wide range of research fields, such as landscape pattern modelling and interpretation, urban ecology analysis and the correlation between urban landscape patterns and urban social-ecological issues, all these research fields contribute to one ultimate aim: providing decision support for a better landscape that meets the requirement of urban residents, wildlife and environments. The case studies introduced in previous chapters seem scattered, yet closely connected to this primary purpose of landscape ecology. Based on the outputs acquired from the above-discussed case studies, some suggestions for landscape planners are proposed as follows. These methodologies and policies are suggested in accordance with different stages of urban landscape planning.

7.1 Comprehensive survey before planning Holism is considered as a significant characteristic of landscape ecology (Naveh and Lieberman, 1994; Antrop, 1996) and the need for including environmental and other components in landscape planning is widely accepted (Hersperger, 1994; Forman, 1995; Dramstad et al., 1996; Ndubisi, 1997; Nassauer and Corry, 2004; Swaffield and Primdahl, 2004; Tippett et al., 2007; White and Ellis, 2007; Geneletti, 2008). As a result, a comprehensive pre-survey should be conducted to inform a better landscape planning. The content, aim and methodology of the survey should be determined by the details of the specific planning and characteristics of local landscapes. Firstly, the preferences and benefits of local residents should be understood and considered. Previous studies proved that some specific landscape patterns are either preferred by a majority of people or beneficial to people’s mental and psychological health. Some principles from a

178

Chapter Seven

diversity of research can be generalized to other sites (e.g. the correlation between landscape patterns and disease, pollutants or noise diffusion) and thus be directly used as guidelines for planning. Before the implementation of urban landscape planning, valuable conclusions from previous studies should be thoroughly examined and taken into account. In spite of the existence of generally applicable guidelines, a specific pre-survey is still required to examine the characteristics of target sites. Take the public’s landscape preferences for instance. Although most people share similar preferences towards urban landscape patterns, group differences still exist in terms of culture, geographic location, age, educational background and so forth. Therefore, planners may take into account the public’s general landscape preferences according to existing findings. Meanwhile, small scale interview-based surveys can be undertaken to meet specific preferences of local residents or different groups. In addition to people’s needs, the well-being of wildlife should also be considered. As suggested by previous studies, perceptions of nature, which includes contact with vegetation and wildlife, benefit urban residents’ mental health. Therefore, appropriate design of urban forests, wetlands and green spaces (Johnson, 1988; Loss et al., 2009; Fontana et al., 2011; Shanahan et al., 2011; Patón et al., 2012) does not only meet the preferences of urban avian and other wildlife (Vallejo Jr et al., 2009; Nichol et al., 2010; Bried et al., 2011; McKinney et al. 2011; Oliver et al., 2011; Zhou et al., 2012), but also provide people with an ideal place to experience wildlife-oriented recreation in the urbanized environment. Hence, a pre-study on the distribution, abundance, diversity and preferences of local urban wildlife needs to be conducted firstly. Based on this pre-study, the planning of urban forests, wetlands and green spaces can be better implemented. As well as the preferences and well-being of local residents and wildlife, urban landscapes are closely related to the local environment and ecological processes (Cook, 2002; Bryant, 2006; Ishikawa and Fukushige, 2012; Ramachandra et al., 2012). In line with people’s growing emphasis on a better living environment, the correlations between landscape patterns and urban environments and ecology should also be well analyzed before landscape planning. A comprehensive and well-designed survey on the interactions between landscape patterns and socio-ecological issues will lead to a

Suggestions for Urban Landscape Planning and Improvement

179

more practical landscape planning in the future. Since it is not realistic for landscape planners to master the knowledge of all related subjects, the participation of experts from different disciplines will be a key issue for a balanced planning. To date, more and more countries and governments are making efforts towards the comprehensive assessment of eco-systems, which can be used as important pre-studies for landscape planning. For example, in the UK, a National Ecosystem Assessment (UK NEA, http://uknea.unep-wcmc.org/Resources/tabid/ 82/Default.aspx ) was carried out between mid-2009 to early 2011. This project analyzes the UK natural environment in terms of the benefits it provides to the society and continuing prosperity. Hundreds of scholars from geography, ecology, environment, economics and other disciplines worked together to establish a comprehensive assessment framework and thousands of volunteers contributed to the on-site survey. Although great time and resources were spent, this project achieved a lot. The findings of this study mainly focus on discovering the current situations and development potential of different areas in terms of the society, economy and ecology. The UK NEA was part of a sub-global assessment network and the design of this project was based on a preliminary assessment framework worked out by researchers all over the world. So far, more than 90 countries have conducted their own NEA, which serves as a solid foundation for comprehensive landscape planning. Outputs of these large-scale projects, as well as possible regional and specific surveys, provide landscape planners and policy makers with valuable decision support. In the future, with the growing popularity of national or regional Ecosystem Assessment, urban landscape planners can conduct more balanced and sustainable planning with decision support from a comprehensive report. In addition, efficiently organized large-scale comprehensive assessment can reduce the redundancy of local surveys.

7.2 Regular monitoring after planning Even if successful urban landscape planning is put into use, regular monitoring of land cover and landscape change is indispensable. When significant changes are detected, re-evaluation of the landscape should be conducted to examine whether the updated landscape patterns are still beneficial and appreciated by local residents.

180

Chapter Seven

As introduced as a main part of this book, regular surveys using airborne Lidar data represent the most suitable tool for establishing 3D urban landscape models and monitoring large-scale landscape changes. By conducting airborne Lidar based surveys, urban landscape patterns can be analyzed and evaluated on a regular basis. Even if airborne Lidar data is not available due to policy or financial limitations, it is still possible to establish 3D landscape models and monitor landscape changes with some alternative approaches, such as ground-based or vehicle-based Lidar, on-site survey, the integration of different data sources and so forth. 7.2.1 Ground-based or vehicle-based Lidar Ground-based and vehicle-based Lidar surveys acquire the elevation and location information of target objects based on the similar working principle as airborne Lidar. Compared with airborne Lidar, ground-based Lidar has its pros and cons. Since the equipment of ground-based Lidar is much cheaper and more convenient than that of airborne Lidar, it is more feasible for researchers to conduct small-scale data-collecting with ground-based Lidar. In addition, ground-based Lidar is more capable of being manually adjusted in the process of scanning, so it performs better than airborne Lidar if researchers want to build detailed 3D models of individual buildings or trees, especially the understory structure of urban features. The disadvantage of ground-based Lidar is that it takes much time to fix the equipment and move to another location, so it is not suitable for study sites of large area. Vehicle-based Lidar collects data automatically and efficiently. A vehicle equipped with a Lidar system can collect high-resolution data at the speed of more than 40km/h. However, this type of Lidar system is limited by road systems and can hardly scan all the districts of study sites. According to its characteristics, vehicle-based Lidar is an ideal choice for monitoring such 3D features as high-ways and rail routes. 7.2.2 On-site survey Without the support from Lidar data, researchers can monitor landscape changes through on-site survey. Levelling instruments,

Suggestions for Urban Landscape Planning and Improvement

181

portable GPS and other equipment enable researchers to accurately measure the elevation of terrain, trees and buildings. Nevertheless, the efficiency of on-site data collecting is comparatively low. To collect height information of buildings more efficiently, another approach is employed in some studies and projects. Instead of measuring the height of buildings directly, researchers just count the number of floors for each building and then estimate the building height by multiplying the average floor height by the number of floors. Through this method, researchers can acquire acceptable accuracy in much less time. A simplified approach for collecting vertical information of trees is also used in some projects (e.g. the production of 3D urban maps). If trees in an urban area are of a similar structure and height, only the locations of trees need to be recorded and a unified tree model can be established for each spot. The approach of on-site survey has its own disadvantages. Due to its low efficiency, on-site data collection is highly time-consuming and not appropriate when applied to a large sample area. In addition, the accuracy of on-site collected data can hardly compete with that of the data acquired using Lidar equipment. 7.2.3 Integration of different data sources Even if researchers do not have access to Lidar data or on-site collected data, they may establish 3D landscape models using other data sources. Although one type of data can hardly provide researchers with all required information, this limitation can be compensated by integrating different data sources. One indispensable and generally available data source is the DTM. Since academic institutions and corporations produce regional, national and global DTMs, researchers can easily apply for or purchase these types of data from such organizations. Another important factor for 3D landscape modelling is classified land cover types, which can be acquired by processing remote sensing images. According to the accuracy requirements, researchers can choose to download remote sensing images of medium resolution for free (http://glcf.umiacs.umd.edu/data/landsat/) or purchase remote sensing images of high resolution from local data distributors. The data source of nDSM is not directly available under most situations. However, some files, such as thematic maps, reports of

182

Chapter Seven

urban planning and reports of urban greenery, from local governments or corresponding divisions, may include the height information of individual buildings and the average height of different tree categories. If these data sources are accessible, researchers can manually add the vertical attribute to 2D landscape models. In summary, airborne Lidar data is recommended as an efficient tool for landscape pattern analysis and monitoring whilst several alternative approaches also work for researches with limited data sources to establish 3D landscape models. Since these alternative approaches all have advantages and limitations, researchers are suggested to combine these methods properly to achieve optimum results.

7.3 Enhancement of current urban landscapes The preferences of local residents, the habitats of local wildlife and the mechanism of local ecosystem change from time to time. Hence, urban landscapes planned in the past may fail to meet current requirements. As a result, it is of great importance to evaluate urban landscapes from different aspects on a regular basis (e.g. every ten years) and to make the necessary improvements to current landscapes according to the evaluation results. To achieve satisfactory urban landscape planning, appropriate land use policies are required. To date, a large body of studies has been conducted on the evaluation and improvement of large-scale or local landscape-related policies (Scholtens, 2001; Greaker, 2003; Brennan, 2006; Dwyer, 2011; Chang and Wang, 2012;). Although great progress has been made in suggesting more sustainable land use policies, unified and generally accepted standards of land use policy have yet been developed. Due to the great differences in land use policies and laws, feasible measures to improve urban landscapes differ significantly between countries. Take the situation in China and the UK as an example. In China, the land resources belong to the state, but local governments are usually allowed to make large-scale changes to current landscape designs. Therefore, researchers are given much freedom to conduct urban landscape planning without much concern of land use constraints. At the macro level, the distribution and size of green open spaces, built up

Suggestions for Urban Landscape Planning and Improvement

183

areas, urban forests and so forth can be decided based upon a comprehensive consideration. At the micro level, the height and spatial structure of individual buildings, trees and other urban features may also be adjusted according to the public’s preferences. The ownership of land resources is different in the UK, as a large proportion of land resources belong to individuals or institutions, which has its own advantages and problems. On one hand, preferable landscape features may be retained for a long period. For instance, some large lawns (which are generally favored by local residents and tourists) owned by some colleges in Cambridge, have remained unchanged for centuries. On the other hand, this type of land ownership may cause difficulties in implementing large-scale landscape planning, as land use change on private lands requires the consent from a diversity of residents and organizations. In this case, it is not feasible for landscape planners to conduct urban landscape planning fully according to their ideal plans. (For instance, an urban planner may believe that a large green space in the city center would be preferred by the public, but he cannot conduct such planning due to the land ownership). However, some enhancements of current landscapes may still be carried out to meet the public’s requirements. Landscape planners may find it difficult to make significant changes in terms of land use types, so transforming a large built area into a green space is not always possible. Nevertheless, landscape planners can find alternative ways to meet the public’s requirements for green spaces. According to people’s feedback, most respondents prefer a mixed design of lawns and trees, which can provide people with shadow in hot weather. As a result, just by adding some trees to the current green spaces, if most of them are pure lawns, landscape patterns can be enhanced to some extent. Urban trees are an important factor that affects people’s perception of urban landscapes. It may be difficult to change a building area into an urban forest, yet researchers can make efforts on the choice of tree species, shape, size and diversity to update an existing urban treescape according to people’s needs. Additionally, the findings from this research have suggested correlations between landscape preferences towards different urban features. By comprehensively considering the design of different features, such as trees, buildings and roofs, landscape planners can improve landscape patterns without making many changes.

184

Chapter Seven

Generally, landscape planners in the UK have to consider a more complicated scenario: what is the optimal landscape that could be achieved under current policies? If the enhancement can be conducted at the macro level, the distribution of different land cover types can be appropriately arranged according to some theoretical criteria (e.g. people’s landscape preference, the relationship between the presence of green spaces and people’s health, etc.) concluded in previous studies. Whilst a large-scale landscape enhancement may not be feasible, researchers can also improve urban landscapes by adjusting the structure of individual urban features (trees, buildings, green spaces, etc.) according to general landscape preferences. In conclusion, the survey and research on urban landscape enhancements should be conducted comprehensively and thoroughly whilst the optimization of urban landscapes is inevitably constrained by land use laws and regulations. Nevertheless, as long as the needs from different perspectives are well examined and a balanced outcome is considered, a better urban landscape can be achieved.

References Antrop, M. (1996) Ecological awareness and holism. In: Miklos, L. (Ed.), Methodological Problems of the Education of Ecological Awareness. BanskaÁ Stiavnica, Unesco Chair Workshop, 24-25 October 1996, pp. 1-6. Brennan, T.J. (2006) “Green” preferences as regulatory policy instrument. Ecological Economics. 56, 144– 154. Bried, J.T., Langwig, K.E., Dewan, A. A., Gifford, N.A. (2011) Habitat associations and survey effort for shrubland birds in an urban pine barrens preserve. Landscape and Urban Planning. 99, 218–225. Bryant, M. M. (2006) Urban landscape conservation and the role of ecological greenways at local and metropolitan scales. Landscape and Urban Planning. 76( 1-4), 23-44. Chang, C. C., Wang, C. M. (2012) Evaluating the effects of green port policy: Case study of Kaohsiung harbor in Taiwan. Transportation Research Part D. 17, 185–189. Cook, E.A. (2002) Landscape structure indices for assessing urban ecological networks. Landscape and Urban Planning. 58(2–4), 269-280. Dramstad, W., Olson, J., Forman, R. (1996). Landscape ecology

Suggestions for Urban Landscape Planning and Improvement

185

principles in landscape architecture and land-use planning. Washington, DC: Island Press. Dwyer, J. (2011) UK Land Use Futures: Policy influence and challenges for the coming decades. Land Use Policy. 28, 674–683. Fontana, S., Sattler, T., Bontadina, F., Moretti, M. (2011) How to manage the urban green to improve bird diversity and community structure. Landscape and Urban Planning. 101, 278–285. Forman, R. (1995). Some general principles of landscape ecology and regional ecology. Landscape Ecology. 10(3), 133–142. Geneletti, D. (2008). Incorporating biodiversity assets in spatial planning: Methodological proposal and development of a planning support system. Landscape and Urban Planning. 84(3–4), 252–265. Greaker, M. (2003) Strategic environmental policy; eco-dumping or a green strategy? Journal of Environmental Economics and Management. 45, 692–707. Hersperger, A. M. (1994). Landscape ecology and its potential application to planning. Journal of the Planning Literature. 9(1), 14–29. Imai, H., Nakashizuka, T. (2010) Environmental factors affecting the composition and diversity of avian community in mid- to late breeding season in urban parks and green spaces. Landscape and Urban Planning. 96, 183–194. Ishikawa, N., Fukushige, M. (2012) Effects of street landscape planting and urban public parks on dwelling environment evaluation in Japan. Urban Forestry & Urban Greening. 11(4), 390-395. Johnson, C. W. (1988) Planning for avian wildlife in urbanizing areas in American desert/mountain valley environments. Landscape and Urban Planning. 16, 245-252. Loss, S.R., Ruiz, M. O., Brawn, J.D. (2009) Relationships between avian diversity, neighborhood age, income, and environmental characteristics of an urban landscape. Biological Conservation. 142, 2578–2585. McKinney, R.A., Raposa, K.B., Cournoyer, R.M. (2011) Wetlands as habitat in urbanizing landscapes: Patterns of bird abundance and occupancy. Landscape and Urban Planning. 100, 144–152. Ndubisi, F. (1997). Landscape ecological planning. In G. F. Thompson & F. R. Steiner (Eds.), Ecological design and planning (pp. 9–44). New York: John Wiley and Sons. Nichol, J.E., Wong, M. S., Corlett, E., Nichol, J.E. (2010) Assessing avian habitat fragmentation in urban areas of Hong Kong (Kowloon)

186

Chapter Seven

at high spatial resolution using spectral unmixing. Landscape and Urban Planning. 95, 54–60. Nassauer, J., Corry, R. (2004). Using normative scenarios in landscape ecology. Landscape Ecology. 19, 343–356. Naveh, Z., Lieberman, A. (1994). Landscape ecology: theory and application. Springer. New York. Oliver, A. J., Jodi Devonshire, C.H., Olea, K.R., Rivas, G.F., Gahl, M.K. (2011) Avifauna richness enhanced in large, isolated urban parks. Landscape and Urban Planning. 102, 215– 225. Patón, D., Romero, F., Cuenca, J., Escudero, J.C. (2012) Tolerance to noise in 91 bird species from 27 urban gardens of Iberian Peninsula. Landscape and Urban Planning. 104, 1-8; Scholtens, B. (2001) Borrowing green: economic and environmental effects of green fiscal policy in The Netherlands. Ecological Economics. 39, 425–435. Shanahan, D.F., Miller, C., Possingham, H.P., Fuller, R.A. (2011) The influence of patch area and connectivity on avian communities in urban revegetation. Biological Conservation. 144, 722–729. Swaffield, S., Primdahl, J. (2004). Spatial concepts in landscape analysis and policy: Some implications of globalization. Landscape Ecology. 20, 657–673. Tippett, J., Handley, J., Ravetz, J. (2007). Meeting the challenges of sustainable development: A conceptual appraisal of a new methodology for participatory ecological planning. Progress in Planning. 67(1), 9–98. Ramachandra, T.V., , Aithal, B. H., Sanna, D.D. (2012) Insights to urban dynamics through landscape spatial pattern analysis. International Journal of Applied Earth Observation and Geoinformation. 18, 329-343. Vallejo Jr, B.M., Aloya, A.B., Ong, P.S. (2009) The distribution, abundance and diversity of birds in Manila’s last greenspaces. Landscape and Urban Planning. 89, 75–85. White, S. S., Ellis, C. (2007). Sustainability, the environment, and new urbanism: An assessment and agenda for research. Journal of Architectural and Planning Research. 24(2), 125–142. Zhou, D. Q., Fung, T., Chu, L.M. (2012) Avian community structure of urban parks in developed and new growth areas: A landscape-scale study in Southeast Asia. Landscape and Urban Planning. 108, 91– 102.

CHAPTER EIGHT CONCLUSIONS AND FUTURE WORK

8.1 Conclusions Due to the lack of vertical information, traditional 2D landscape models have limitations in representing comprehensive landscape patterns and understanding the interactions between landscape patterns and socio-ecological issues. To add the extra vertical pattern to 2D landscape models, this book proposes a methodology of establishing 3D urban landscape models using airborne Lidar data. In addition, this book also explores a framework of applying 3D urban landscape models to urban landscape ecology. Chapter 1 generally introduces the research background, objectives and the organization of this book whilst the following chapters give concrete discussion of corresponding methodologies and case studies. Urban DTMs and classified urban features are indispensable for the establishment of 3D urban landscape models. To obtain necessary data sources, two models, the urban DTM generation model (Chapter 2) and urban land cover classification model (Chapter 3) are introduced in the book. 3D landscape models are then employed for urban landscape pattern analysis (Chapter 4), landscape pattern evaluation (Chapter 5) and specific research (Chapter 6). Based on the findings from these case studies, some suggestions are proposed for landscape planning and enhancement (Chapter 7). The detailed contents of each chapter are summarized as follows: 8.1.1 The establishment of 3D urban landscape models The upward-fusion Urban DTM generation method (Chapter 2) This book aimed to propose a methodology for establishing and applying 3D urban landscape models, and urban DTM generation is

188

Chapter Eight

indispensable for establishing 3D urban landscape models. Although many studies have been conducted in generating DTMs using airborne Lidar data, urban DTM generation is still a challenging research area. In this case, the objectives of this chapter were to develop a methodology for urban DTM generation, and to evaluate this method with field-survey data and leading Lidar processing software. This research designed an upward-fusion methodology to generate urban DTMs using airborne Lidar data. Firstly, several preliminary DTMs of different resolution were obtained using a local minimum method. Next, upward fusion was conducted between these DTMs. This process began with the DTM of the largest grid size, which was treated as a trend surface. A finer DTM was compared with this large scale DTM. By setting appropriate thresholds, a new DTM was achieved by selecting qualified elevation values from the finer DTM and retaining the value of the trend surface when the value from the finer DTM was beyond a threshold. This process continued iteratively until all preliminary DTMs had been included in the upward fusion processing and a refined DTM of high resolution was achieved. To further reduce possible errors in the resulting DTM, an extended local minimum method was proposed for filtering outliers and generating preliminary DTMs. The upward-fusion method was experimented in central Cambridge, and the time efficiency, results of the accuracy assessments and comparison with leading software proved the success of the case study, and indicated that upward-fusion was an effective method for urban DTM generation. Considering the accuracy, simplicity and time efficiency, the upward-fusion method is a suitable tool for landscape ecologists to generate urban DTMs, based on which 3D urban landscape models can be established. The object-based urban land cover classification model (Chapter 3) Classified urban land cover types are also required for 3D urban landscape models. Integrated with multi-spectral images, hyperspectral images or airborne photographs, airborne Lidar data have been widely used for land cover classification. However, few researchers have conducted urban land cover classification using airborne Lidar data as

Conclusions and Future Work

189

the sole source. When additional data sources are not available, a method of urban land cover classification using airborne Lidar data solely is required for the establishment of 3D urban landscape models. The objectives of this chapter were therefore to develop an object-based land cover classification method using Lidar data as the sole source, and to evaluate this method with field-survey data and leading Lidar software. This research explored the possibility of applying airborne Lidar data to land cover classification in urban areas. The elevation difference and intensity difference between the first and last return, which may not work efficiently in pixel-based classification, were employed as two key attributes for object-based classification. Since tree objects have a much larger proportion of returns that show the elevation and intensity difference, the two indicators were used to classify the most distinguishable urban land cover types, buildings and trees. Besides, the height and intensity attributes were employed to classify other land cover types. This methodology was experimented in central Cambridge and eight urban land cover types were classified with an overall accuracy of 93.6%. Each land cover type was classified with an accuracy of between 80% and 100% and among these types, the accuracy of more than 90% for trees and buildings was satisfactory. To compare this method with existing Lidar processing software, one leading Lidar tool (Terrascan) was employed for experiments. The comparison result proved that the object-based classification method was more likely to produce accurate and homogeneous classified images. The object-based urban land cover classification method proposed in this chapter enables researchers to acquire classified object-based urban land cover types using Lidar data only. Integrated with the upward-fusion DTM generation method, this book provides scholars, who have no easy access to specific Lidar software, with a simple but efficient approach for establishing 3D urban landscape models.

190

Chapter Eight

8.1.2 Applications of 3D urban landscape models Urban landscape pattern analysis (Chapter 4) Landscape pattern analysis is an indispensable task for landscape ecology research, yet traditional 2D models have limitations in analyzing spatial patterns comprehensively and distinguishing the structural differences in the vertical direction. As a result, it is of great importance to conduct landscape pattern analysis based on 3D landscape models. The objectives of this chapter were therefore to analyze 3D urban landscape models in two different sites by exploring 2D and 3D landscape metrics, and demonstrate the advantages of using 3D landscape metrics. With the methodology of urban DTM generation and urban land cover classification, the 3D urban landscape models of central Cambridge and the residential area in Canvey were established respectively. To comprehensively analyze and compare landscape patterns between the two areas, both 2D and 3D landscape metrics were adopted for pattern analysis. Based on the pixel and object-based 2D pattern analysis, the spatial composition and structure of trees and buildings in the horizontal direction were compared whilst the differences in their vertical patterns were not revealed. A set of 3D landscape metrics was thus designed and applied to 3D pattern analysis. The results proved that properly designed and selected 3D landscape metrics enable researchers to better understand and compare comprehensive landscape patterns. The methodology and practicality of designing appropriate 3D metrics serves as an example for other landscape ecologists to conduct 3D landscape pattern analysis, and some of the 3D landscape metrics proposed in this research have the potential to be generalized to other urban sites. Evaluating urban landscape patterns in terms of the public’s aesthetic preferences (Chapter 5) Evaluating landscape patterns is an important application of 3D landscape models, yet the integration of landscape pattern analysis and evaluation criteria remains a challenging subject due to the lack of

Conclusions and Future Work

191

general criteria. Landscape patterns can be evaluated from the perspective of urban residents, wildlife and local environments. In this research, the public’s aesthetic preferences towards urban landscape patterns were explored as potential evaluation criteria. The objectives of this chapter were thus to explore the methodology of integrating 3D landscape metrics with people’s landscape preferences for evaluating different landscape patterns. This research explored people’s preference towards the composition and structure, especially the vertical structure of some specific urban features. A questionnaire-based survey was conducted in two cities: Cambridge, UK and Nanjing, China and more than 180 responses were collected in each site. In spite of some differences, respondents from both sites showed similar preferences towards freely growing trees, individual houses, gable roofs, mixed design of green spaces and so forth. Group differences in gender, age, education, major, status, etc. were analyzed and some specific preferences were explained. In addition, correlations between people’s preferences towards different urban features were examined. Although some group differences existed, this survey concluded that most people shared similar preferences towards the patterns of some urban features. Correlated with the general landscape preferences concluded in the survey, the value of some 2D and 3D landscape metrics can be employed to evaluate urban landscape patterns. This case study serves as an example for other landscape ecologists to demonstrate the methodology of designing feasible evaluation criteria according to designed 3D landscape metrics. Supporting research with 3D urban landscape models: visual effect analysis of urban trees (Chapter 6) In addition to pattern analysis, 3D urban landscape models can support such studies as environmental modelling and retrieval analysis with reliable data sources. In this research, assessing ‘tree green availability’ was introduced as an example. The objective of this chapter was to propose a methodology for the analysis of ‘ tree green availability’ using variables from a 3D urban landscape model. Urban trees benefit people’s daily life in terms of air quality, local

192

Chapter Eight

climate, recreation, ecology and aesthetics. Among these functions, a growing number of studies have been conducted to analyze the relationship between residents’ perception of local environments and visual effects of urban greenery. However, except for on-site photography, there are few quantitative methods to calculate green visibility, especially tree green visibility, from viewers’ perspectives. Photograph-based survey was conducted in Cambridge to estimate the real value of tree green visibility. In addition, the established 3D landscape model was employed in this study to measure the size and shape of individual trees. Next, correlations between the value of tree green visibility and some tree parameters extracted from the 3D model were examined. Through experiments, a model of the form Green=a*(Perimeter*Range/Distance2)b proved to be the optimum regression model. Considering the diversity of sample trees and the satisfactory result of cross-validation, this model has the potential to be applied to other study sites. This study of tree green visibility analysis provides urban planners and policy makers with an efficient method to analyze and evaluate landscape patterns in terms of the presence, density and character of urban greenery. This research demonstrates the methodology of applying 3D urban landscape models to some new research areas related to landscape planning and management. Suggestions for urban landscape planning and improvement (Chapter 7) The objective of this chapter was to briefly propose some suggestions for urban landscape planning and improvement. At the pre-planning stage, comprehensive surveys should be conducted to examine the correlation between landscape patterns and the preferences of local residents, the well-being of urban wildlife and the sustainable development of the environment. Following a successful implementation of landscape planning, some alternative approaches, such as ground or vehicle-based Lidar and on-site survey, as well as airborne Lidar, can be employed to monitor the changes of landscape patterns. The enhancement of urban landscape patterns may be limited by local land use policies or regulations and large-scale land cover

Conclusions and Future Work

193

change is not always feasible. A possible solution is to improve landscape patterns from a micro perspective. When the current landscape patterns needs to be adjusted to better meet people’s requirements and the adjustment of large-scale landscape patterns is not possible, the improvement of urban landscape patterns may be achieved by adjusting the spatial structure of individual urban features. This chapter provides landscape planners with suggestions for planning and improving landscape patterns in different stages and scenarios. These suggestions are proposed according to the diverse findings of this book, and the methodology of designing proper case studies, which lead to these suggestions, also provides other landscape ecologists and planners with a reference, pointing for future work.

8.2 Future work Due to the lack of time and resources, this book has not included detailed studies on all potential applications of 3D urban landscape models. However, the framework and case studies designed in this book provide a foundation for future studies to build upon and the researcher’s work on 3D urban landscape ecology is on-going. 8.2.1 Integrating airborne Lidar data with other data sources Although the airborne Lidar data employed in this research is of fine resolution (0.5m in the horizontal direction) for urban DTM generation and urban land cover classification, the resolution is not fine enough for modelling the detailed roof structure of specific buildings or tree crown structures. Airborne Lidar data of very high resolution (e.g. 0.25m or 0.1m) and multiple returns have the potential to produce 3D urban landscape models with enhanced details and accuracy. Nevertheless, with improved resolution, the size of the entire point cloud and the difficulties of data processing will also be increased significantly. Therefore, specific research emphasis should be placed on revising current methodology of Lidar data processing for high resolution data. Airborne Lidar data can be employed as the sole source for land cover classification using appropriate algorithms. However, due to its limited attributes, airborne Lidar data can hardly be used to classify as many

194

Chapter Eight

land cover types as multi-spectral remote sensing images or airborne photographs. In this case, the accuracy of DTM generation can also be improved by including multiple data sources. By analogy, some data sources such as city maps serve as building masks for urban DTM generation. To further improve the visual effect and accuracy of 3D landscape models, efforts should be made on properly integrating airborne Lidar data with relevant additional data sources. As pointed out in previous chapters, airborne Lidar has its limitations for describing the understory structure of trees and buildings whilst ground-based or vehicle-based Lidar can build more accurate and detailed vertical patterns. On the other hand, compared with airborne Lidar, ground-based or vehicle-based Lidar can rarely be used to cover a large-scale area. According to the research problem, the integration of large-scale airborne Lidar data and small-scale ground-based Lidar data (Bremer and Sass, 2012) will be a practical, yet challenging subject. The main difficulty of this task lies in designing proper methodologies for analyzing data sources from different (overhead and horizontal) perspectives. In conclusion, adopting more advanced airborne Lidar technology, as well as the combination of airborne Lidar data and other data sources, for better 3D landscape modelling will be one focus of future work. 8.2.2 Designing 3D landscape metrics and examining the correlation between 3D landscape patterns and urban social-ecological issues One key issue of urban landscape ecology is to analyze the composition and structure of different urban features. As a result, a set of rigid and efficient 3D landscape metrics is required for 3D pattern analysis. Some examples of 3D landscape metrics for urban areas were proposed in this book and achieved satisfactory results. Nevertheless, the number and scope of these 3D metrics cannot meet the requirements of a diversity of disciplines in landscape ecology. Future research should be conducted on the design of different landscape metrics for general pattern analysis and specific research topics by cooperating with scholars from different backgrounds. Terrain information, which is an important indicator of ecological processes,

Conclusions and Future Work

195

was not included in the 3D landscape metrics employed in this book. Therefore, growing research priorities should be placed on designing 3D landscape metrics using the terrain attributes as well. In addition, some studies (Herold et al., 2003; Seto and Fragkias, 2005; Parrott et al., 2008) have been conducted to examine spatio-temporal metrics for landscape change detection. By adding time series to 3D spatial landscape metrics, the efficiency of landscape pattern analysis and comparison can be further improved. Hence, four-dimensional metrics will also be a promising research field. Without necessary data sources, this book did not include the analysis on the interactions between 3D landscape patterns and specific ecological processes, which will be explored in future research. Take the correlation between urban avian fauna and 3D landscape patterns as an example. As mentioned, the correlations between the distribution of urban avian and specific 3D landscape indicators can be a very important subject. By analyzing the correlation between 3D landscape patterns (more specifically, a set of 2D and 3D landscape metrics) and the occurrence of birds, we can provide more suitable patterns according to avian preferences, as avian are preferred by local residents. However, avian influenza causes serious health threats to residents in East Asia. If there is a strong linkage between the occurrence of urban birds and the burst of H5N1 (H7N9), we can suggest some urban landscape patterns that are not preferred by birds. In this case, we can reduce the possibility of contacting avian influenza and reduce the danger of getting infected. A mature framework of 2D landscape ecology was achieved through a large body of studies. 3D urban landscape ecology is a new, yet promising subject. As a result, the future development of 3D landscape ecology also requires the efforts from ecologists and scholars from related disciplines.

References Herold, M., Goldstein, N., Clarke, K. (2003) The spatio-temporal form of urban growth: measurement, analysis and modeling. Remote Sensing of Environment. 85, 95-105. Parrott, L., Proulx1, R., Thibert-Plante, X. (2008) Three-dimensional metrics for the analysis of spatiotemporal data in ecology.

196

Chapter Eight

Ecological Matics. 3, 343-353. Seto, K.C., Fragkias, M. (2005) Quantifying spatiotemporal patterns of urban land-use change in four cities of China with time series landscape metrics. Landscape Ecology. 20, 871-888.