DYNAMICS OF LANDSCAPE PATTERNS AND THEIR IMPACTS ON URBAN THERMAL AND BIOMASS ENVIRONMENTS IN THE kUNMING METROPOLITAN AREA

ACKNOWLEDGEMENTS First and foremost, I would like to express my earnest gratitude to my advisor, Dr. Yi-Chen Wang for her patient guidance as I gradually got into my academic track, helpful instructions when I was bewildered, gentle criticisms and supportive encouragements when I did something wrong or right, and role model in terms of the commitment and enthusiasm in scientific research and scrutiny in reading and writing that will continually influence me in my future study and career. It is her who taught me step by step from the basic academic writing, to critical thinking and conscientious working. I am sincerely grateful for her supervision in the past short but meaningful two years. In addition, I would like to thank A/P Xixi Lu for his constant concerns and great help in my entire graduate studies. Thanks also go to Dr. Chen-Chieh Feng for his advices from GIS perspective and constructive suggestions in my study. Also, I really appreciate A/P Alan Ziegler and Dr. Jun Zhang and other lecturers in geography department for their suggestion and guiding. I have also received many helps from the administrative staff and technical staff, including Mr. Lee Choon Yoong, Mr. Yong Sock Ming and Ms. Sakinah bte Yusof. I am particularly grateful to Pauline for the timely and detailed answers for all administrative matters and joys we had together outside campus. Also, tremendous thanks go to all my friends whom I share my laughters, confusions and excitements with and I learned from, especially I those lovely guys in the research clusters, Yu Liang, Yikang, Tzu-Yin, Yiqiong, Mr. Huan and Valerie; the post-graduate fellows, Rana, Lina, Deborah, Jianjun, Swe Hlaing, Lishan, Serene, Fred, and Aidan; and my NUS friends, Quchen, Haigang, Yuqi and others. Most of all, I would like to thank my parents for their infinite support and encouragement. Their continual help in every aspect is always accompanying me. Thanks for being with me in my journey to the end of my master study! Xiaolu Zhou May 2010 II TABLE OF CONTENTS Page ACKNOWLEDGEMENTS I TABLE OF CONTENTS III SUMMARY VIII LIST OF TABLES X LIST OF FIGURES XI LIST OF PLATES XIII LIST OF ACRONYMS XIV CHAPTER ONE: INTRODUCTION 1.1 Land use change in the process of urbanization 1 1.2 Implications of green space loss 2 1.3 Greening initiatives 3 1.4 Process of landscape pattern change 3 1.5 Scope and aims of this study 5 1.6 Outline of the thesis 7 CHAPTER TWO: LITERATURE REVIEW 2.1 Landscape pattern change analysis 8 2.1.1 Landscape ecology applied in the urban environment 8 2.1.2 Landscape metrics in sustainable development 9 III 2.1.3 Landscape gradient analysis 11 2.1.4 Change intensity analysis 12 2.2 Urban thermal environment analysis 13 2.2.1 Impact of land use change on urban thermal environment 14 2.2.2 Measurements of the thermal environment 14 2.2.3 Relationship between LST and land use change 15 2.2.4 Spatial variation in LST modeling 17 2.3 Green biomass analysis 18 2.3.1 Importance to measure green biomass 18 2.3.2 Approaches to measure biomass 20 2.4 Urban expansion in Chinese cities 22 CHAPTER THREE: THE STUDY AREA KUNMING, CHINA 3.1 Geography 25 3.2 Climate 26 3.3 Vegetation and wildlife 26 3.4 Economic growth 27 3.5 Urbanization and green space loss 27 3.6 Greening policies 28 3.7 Area of the research site 28 CHAPTER FOUR: MATERIALS AND METHODS 4.1 Data used in this study 29 IV 4.2 Data preparation 32 4.2.1 Preparations for planning maps and field survey data 32 4.2.2 Image pre-process 33 4.2.3 Land use classification system 34 4.2.4 Classification and accuracy assessment 35 4.3 Landscape pattern analysis 35 4.3.1 Landscape metrics analysis 36 4.3.2 Change intensity analysis 39 4.4 Thermal environment analysis 40 4.4.1 LST derivation 41 4.4.2 Derivation of remote sensing indices 42 4.4.3 Impact of pattern change on LST 43 4.4.4 Modeling LST based on land use indicators 44 4.5 Green biomass analysis 45 4.5.1 Calculation of VD 46 4.5.2 VD sampling 47 4.5.3 Linking vegetation indices with VD 48 4.5.4 VD change analysis in 2000, 2006 and 2009 49 V CHAPTER FIVE: LANDSCAPE PATTERN DYNAMICS IN RESPONSE TO RAPID URBANIZATION AND GREENING POLICIES 5.1 Landscape analysis and change intensity analysis 50 5.1.1 Synoptic landscape pattern 50 5.1.2 Concentric landscape pattern 52 5.1.3 Directional landscape pattern 57 5.1.4 Change intensity analysis 58 5.2 Discussion of the green space dynamics analysis 61 5.2.1 Landscape change in response to urbanization and greening policies 61 5.2.2 Concentric and directional landscape analyses and change intensity 62 5.2.3 Interpretation of landscape indices 63 CHAPTER SIX: THERMAL ENVIRONMENTAL IMPLICATIONS OF THE LANDSCAPE PATTERN DYNAMICS 6.1 The change of the urban thermal environment 66 6.1.1 Impacts of landscape pattern change on LST 66 6.1.2 Modeling LST based on remote sensing indices 72 6.2 Discussion of the thermal environment analysis 77 6.2.1 Process of the thermal environmental change 77 6.2.2 Effects of green policies on thermal environment 78 6.2.3 Localized statistic in LST modeling 79 VI CHAPTER SEVEN: GREEN BIOMASS IMPLICATIONS OF THE LANDSCAPE PATTERN DYNAMICS 7.1 Green biomass dynamics of the study area 80 7.1.1 Comparison of vegetation density and vegetation cover 80 7.1.2 Vegetation indices in modeling VD 81 7.1.3 Estimated VD in 2000, 2006 and 2009 82 7.1.4 Distribution of the areas with VD variation 86 7.2 Discussion of the green biomass analysis 87 7.2.1 Comparison of vegetation indices in VD regression 87 7.2.2 VD dynamics in 2000, 2006 and 2009 87 7.2.3 Effects of greening policies on VD 88 CHAPTER EIGHT: DISCUSSION 8.1 Landscape pattern and process 89 8.1.1 Impacts of the processes on landscape patterns 90 8.1.2 Impacts of landscape patterns on processes 91 8.2 Relationship between thermal environment and biomass amount 92 8.3 Environmental planning strategies 93 CHAPTER NINE: CONCLUSION AND FUTURE WORK 96 BIBLIOGRAPHY 99 VII SUMMARY Urban expansion is occurring at an unprecedented rate in most countries worldwide with no exception in China. The conversion of natural land into impervious areas has resulted in many environmental consequences. Having realized the important role of green space in urban ecosystems, many municipal governments in China have set out a series of policies to introduce green elements into urban areas. Insights into how urban landscape pattern changes in response to urbanization and greening policies and to what extent land use transformation affects local environment are essential for guiding sustainable urban development. This study investigated urban landscape pattern change in response to the rapid urbanization and greening policies in the Kunming metropolitan areas, China. Urban thermal environment and green biomass were investigated in the context of landscape pattern change. The concentric and directional landscape analyses along with landscape metrics were first used to characterize landscape patterns. Change intensities of the landscape patterns were then calculated for the study area as a whole, the concentric belts, and the directional transects to examine the variation of the green space change rate in the city. Next, the study used land surface temperature (LST), derived from remotely sensed images, to characterize the thermal environment of the study area and to associate the LST with the changing landscape patterns. Global and local models were performed to explore the impacts of different land use types on LST variations. Urban green biomass was represented by vegetation VIII density (VD) to evaluate the urban green space conditions and wildlife habitats. VD was derived from remote sensing indices and its spatial change was analyzed using Geographical Information Systems. Results revealed that both rapid urbanization and greening policies accounted for the process of landscape pattern change. Among different green space types, agriculture land was largely encroached and fragmented by urban sprawl, especially in the outer belts of the city. Forest land was also impacted but encountered a relatively moderate loss rate compared to agriculture land. Conversely, greening policies contributed to the recovery of grass land in the last decade. Land use transformation largely altered the local thermal environment and green biomass. A remarkable LST increase was detected in the urban fringe when natural land cover was replaced by impervious surface. Green space was confirmed to be important in mitigating urban heat. As for green biomass, low vegetated areas encountered a substantial biomass loss, mainly due to the rapid shrinkage of the agriculture land. Densely vegetated areas maintained a relatively stable biomass status, suggesting forest areas remained less impacted by human disturbances. This study unveiled the processes of landscape pattern change in the presence of two seemingly contradicting driving forces, i.e. urbanization and greening policies, providing insights into the mechanisms of urban land use change and the subsequent environmental implications. Based on the results, several planning strategies were put forward to ensure a sustainable urban development. Keywords: Green biomass; Kunming; Landscape pattern; Urban sustainable development; Urban thermal environment; IX LIST OF TABLES Page Table 4.1: Landscape metrics used in this study 38 Table 4.2: Weights of VD for different vegetation cover 47 Table 5.1: Average land use change rate in percent in two time periods 59 Table 5.2: Correlations between selected metrics 65 Table 6.1: Derived LST for each land use type 68 Table 6.2: Land use change and its effect on LST 70 Table 6.3: Correlation coefficients between LST and remote sensing 72 indices in 1992 and 2006 Table 6.4: Comparison of the diagnostics for OLS and GWR modeling 75 Table 6.5: Parameter coefficients of the multivariate models based on OLS and GWR methods 75 Table 7.1: Comparison of the regression results for different remote sensing indices 82 Table 7.2: VD changes in 2000, 2006 and 2009 85 X LIST OF FIGURES Page Figure 1.1: Process of landscape pattern change 5 Figure 3.1: Location of the study area 25 Figure 4.1: Flow chart of this study 29 Figure 4.2: Elaboration of data used in this study 30 Figure 4.3: The processes of data preparation 32 Figure 4.4: Flow chart of the landscape pattern analysis 36 Figure 4.5: The layout of concentric belts and directional cells 39 Figure 4.6: Flow chart of the thermal environment analysis 40 Figure 4.7: Flow chart of the green biomass analysis 46 Figure 5.1: Land use of the study area in 1992, 2000 and 2009 51 Figure 5.2: Synoptic landscape pattern change quantified by landscape metrics 52 Figure 5.3: Concentric landscape pattern change characterized by landscape metrics 54 Figure 5.4: Concentric land use diversity from 1992 to 2009 quantified 57 using SHDI Figure 5.5: Percentage of landscape along eight directional transects in 58 1992, 2000, and 2009 Figure 5.6: Land use change rate in seven concentric belts 60 Figure 5.7: Spatial patterns of land use change rate in eight directions 61 Figure 6.1: Derived LST in 1992 and 2006 66 Figure 6.2: Remote sensing indices of the study area in 1992 and 2006 67 Figure 6.3: Spatial correspondence between changes in land use and LST 71 XI Figure 6.4: Scatter plots of LST against remote sensing indices 73 Figure 6.5: Local coefficients of multivariate regression based on GWR analysis 77 Figure 7.1: Computation of VC and VD 81 Figure 7.2: VD spatial distribution in 2000, 2006 and 2009 83 Figure 7.3: VD change maps 86 Figure 8.1: Relationship between landscape patterns and processes reflected by this study 90 XII LIST OF PLATES Page Plate 5.1: Photos of newly constructed urban parks and golf courses in Kunming City 62 XIII LIST OF ACRONYMS AICc corrected Akaile Information Criterion BT Brightness Temperature DN Digital Number ETM+ Enhanced Thematic Mapper Plus FLAASH GDP Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes Gross Domestic Product GIS Geographical Information Systems GPS Global Positioning Systems GWR Geographically Weighted Regression LAI Leave Area Index LPI Largest Patch Index LSI Landscape Shape Index LST Land Surface Temperature LUCC Land Use/Cover Change MNDWI Modified Normalized Difference Water Index NDBI Normalized Difference Built-up Index NDVI Normalized Difference Vegetation Index NDWI Normalized Difference Water Index OLS Ordinary Least Square PD Patch Density PLAND Percentage of Landscape SHDI Shannon’s Diversity Index XIV SPOT Systeme Probatoire d’Observation dela Tarre TM Thematic Mapper UHI Urban Heat Island UTM Universal Transverse Mercator VC Vegetation Cover VD Vegetation Density XV Chapter 1: Introduction 1.1 Land use change in the process of urbanization In the past several decades, population has been growing worldwide. The explosion of population brings about urban expansion occurring at an unprecedented rate worldwide with 65% of the population expected to reside in urban areas by 2025 (Schell & Ulijaszek, 1999). The rate in some developing countries is more remarkable due to the pursuit of fast development (Li et al., 2009). In China, the launch of economic reforms in the late 1970s largely pushed forward the urbanization process in the past several decades (Luo & Wei, 2009). The “Open Door” Policy initiated in 1978 and the land reform regulation launched in 1987 markedly expedited the urbanization rate (Cheng and Masser, 2003; Luo and Wei, 2009). The urbanization level is predicted to reach 50% with 1.5 billion urban residents by the end of 2020 (Tian et al., 2005). In the process of urbanization, the most remarkable land use change is the loss of green space. The impervious land sprawls unrestrainedly, causing tremendous suburban green areas to be swallowed by the overwhelming urban growth. For example, in China, large areas of arable land were encroached by urban expansion (Tan et al., 2005). Forest land shrank tremendously in the suburban areas (Li et al,. 2006; Fan et al., 2007). Concomitant with green space loss in the suburbs is the reduction of the semi-natural green space within urban areas (Tan, 2006). Due to the intense competition for the limited 1 urban land, green elements in the most urbanized area are substituted for developments, further aggravating the rigidness of cities. 1.2 Implications of green space loss The loss of urban green space gives rise to many environmental problems. In the urban context, drastic reduction of green space tends to elongate green patch distance and decrease habitat size, which tremendously degrades the quality and lessens the quantity of habitats, causing a sharp decrease of wildlife (Bender et al., 2003). The removal of vegetated areas changes land surface properties, such as moisture and optical characteristics, largely modifies the urban thermal environment (Owen et al., 1998). The increasing emission of greenhouse gas in urban areas and the lowered photosynthesis process induce the heat island effect in cities and increase the potential risk of global warming (Nowak & Crane, 2002; Guan & Chen, 2003). In addition, urban green space can purify air and contain water. The reduction of green space aggravates air pollution, lessens the recharging of groundwater and results in polluted surface runoff (Chen & Jim, 2008). The degradation of urban green space also significantly affects the energy flow and the nutrient cycling of the ecosystem, causing the degradation of the functions of a life-supporting environment and lead to unsustainable development (Grimm, 2000; Whitford et al., 2001; Yeh & Huang, 2009). Moreover, since urban green space provides citizens recreational opportunities (Fleischer & Tsur, 2003), renders aesthetic enjoyments (Chen & Jim, 2008), promotes physical health (Hartig, 2003) and adjusts psychological well being (Milligan et al., 2 2004), the loss of urban green space substantially degrades the citizens’ quality of life. 1.3 Greening initiatives Having realized the important role played by green space in the urban sustainable development and the severe reduction of green space in cities, there is a growing attention to urban greening worldwide. For example, conservation and restoration of remnant green space, establishing habitats with good quality surrounding urban areas were stressed by municipal government in Australia (Low et al,. 2005). Despite the severe land use competition for developments of high economic yields, Singapore embarked upon the implementation of an island-wide network of greenways to link urban parks to natural areas. This policy significantly ameliorated the urban environment, distinguishing Singapore as a famous garden city worldwide (Tan, 2006). Cities in the United States witnessed an upsurge in “greening”, such as green roofs, new parks and tree plantings (Daniels, 2008). In China, several greening policies were also implemented recently, such as the comprehensive green space planning strategies for promoting sustainable development in the Beijing metropolitan area (Li et al., 2005). Other examples included a greening plan based on landscape ecology in Nanjing (Jim & Chen, 2003) and a greenway augment plan in Xiamen Island (Zhang & Wang, 2006). 1.4 Process of landscape pattern change Understanding the process of landscape pattern change requires the examinations of the driving forces and the subsequent implications. There are 3 complicated interactions among different driving forces, such as climate change, socioeconomic development and municipal regulations, which jointly determine the direction of the landscape pattern change (Verburg et al., 2004). In the urban context, urbanization and urban greening are two conflicting driving forces that simultaneously modify the green space pattern. It is thus important to explore the change of urban landscape patterns under these two driving forces collectively. As mentioned in Section 1.2, the change of landscape patterns will give rise to many environmental consequences. In the urban areas, two environmental implications are closely related to land use change, i.e. urban thermal environment and green biomass, which were also used by prior studies to evaluate the urban environment qualities (Nichol & Wong, 2005). In this study, these two implications were thus investigated under the rapid urbanization and greening policies. Figure 1.1 illustrates the framework of this study. 4 Figure 1.1: Process of landscape pattern change. Urbanization and greening policies serve as driving forces to modify landscape patterns. The change of landscape patterns results in a variety of implications, such as thermal environment and green biomass dynamics. 1.5 Scope and aims of this study This study investigated the dynamics of landscape patterns and the associated environmental implications in a city with rapid urbanization, Kunming, China. In Kunming City, rapid urban growth encroached much green space on the urban fringe and suburbs, while the intensive urban development took place in many semi-natural places within the urban areas in the past several decades. To ameliorate the urban environment and to enhance citizens’ quality of life, the municipal government has implemented several urban greening policies since 2000. To examine the effects of driving forces on landscape pattern change and the subsequent environmental implications, this study consisted of three analyses. The first part focused on the landscape dynamics in response to the rapid urbanization and greening policies. Integrated approaches were used to 5 characterize the changing patterns and intensities of green space change in Kunming. Spatial variations of landscape patterns were derived through concentric and directional landscape analyses integrated with landscape metrics. Change intensities were calculated for the study area as a whole, the concentric belts, and the directional transects to examine the variation of the green space change rate in the city. The second and the third analyses focused on the environmental implications of the changing landscape patterns. The second part concerned with urban thermal environment change. Integrated methods of Geographical Information Systems (GIS) and remote sensing were used to investigate the impact of land use change on the dynamics of land surface temperature (LST). Remote sensing indices were used to quantify land use types and also as explanatory variables in LST modeling. The third part explored the urban green biomass in Kunming City in the past decade. Vegetation density (VD) was used as an indicator of green biomass. VD, derived from remote sensing indices, reflected the vegetation condition and the relative habitat abundance in the study area. Therefore, the aim of this study is to address the following research questions: (1) How does the landscape pattern change in response to the rapid urbanization process and the greening policies? (2) To what extent do landscape pattern changes influence the local thermal environment? (3) How does the green biomass vary in the changing landscape pattern? 6 1. 6 Outline of the thesis Following this brief introduction on the general rationale of the research, Chapter Two reviews a series of prior studies conducted in each of the three aspects of this study. Chapter Three describes the Kunming City in terms of the geography, climate, vegetation, economy and the urbanization process. Chapter Four unveils the research methods employed in this study, comprising the concentric and directional landscape metric analysis and change intensity analysis, derivation of LST and its relationship with land use change, and the measurement of urban green biomass. Chapter Five presents the results and discusses the dynamics of the urban green space pattern in response to the urbanization and greening policies. Chapter Six displays and discusses findings of the thermal environment change due to landscape alterations. Chapter Seven reveals the results of the amount of green biomass derived from VD and discusses its temporal change. Chapter Eight discusses the relationship between landscape patterns and processes as well as provides suggestions for the potential environmental planning strategies for the study area. Chapter Nine concludes this study by summarizing the major findings and illustrating possible directions for future research. 7 Chapter 2: Literature Review 2.1 Landscape pattern change analysis Urban sprawl has caused tremendous land use alteration. This trend raises many alerts and a number of studies have been carried out to investigate the changing landscape patterns. Landscape ecology is a science which provides concepts and methods to study land use/cover change (LUCC). The following part reviews the state of the art research in landscape ecology and its applications. 2.1.1 Landscape ecology applied in the urban environment Landscape ecology is a science contributing to improve the understanding of the relationship between spatial patterns and ecological processes on different landscape scales and organizational levels (Forman & Godron, 1986; Wu, 2006). It has been widely used in investigating LUCC, inferring ecological flows and guiding sustainable planning (Wu & Hobbs, 2002). While landscape ecology emerged in Europe half a century ago (Naveh & Lieberman 1984), the past two decades witnessed a rapid development due to the growing awareness of the environmental problems (Wu & Hobbs, 2002). Landscape ecology has been introduced into urban land use study for at least three reasons. First, since landscape ecology is capable to look simultaneously at a broad scale (the scale of the entire landscape) and a local scale (the scale of neighborhood), it is particularly useful in the urban 8 environment to depict overall urban morphology and complicated heterogeneous patches (Low et al., 2005). Second, landscape ecology seeks to explain the fundamental landscape structure based on three spatial concepts, i.e. matrix, patch and corridor, which fit the elements in the urban ecological context (Jim & Chen, 2003). Patches correspond to urban parks, grass land, and open space with different sizes and shapes, clustering, dispersion or isolation in urban areas. Corridors can be reckoned as linearly arranged elements, such as riparian green belts, green avenues with different length, width and shapes. Matrix is associated with uniform and dominant areas over space, such as urbanized areas, with different heterogeneous extents. By depicting the urban environment using these basic structural elements, complicated urban landscape can be better understood (Dramstad et al., 1996). Third, Landscape ecology examines landscape composition and configuration which can provide insights into urban ecological process. Concepts of landscape ecology can be incorporated into sustainable urban planning, maximizing the ecological functions of the limited green elements in the urban areas (Opdam et al., 2002; Leitão et al., 2006; Roy et al., 2009). 2.1.2 Landscape metrics in sustainable development Investigating urban landscape space often requires researchers to first quantify the urban landscape composition and configuration so that a sound understanding of the distribution of urban landscape pattern can be attained (Forman & Godron, 1986). One of the milestones of landscape ecology was the extensive use of landscape metrics in spatial pattern analysis due to the increased availability of spatial data (Gustafson, 1998; Turner et al., 2001; Li 9 & Wu, 2004). Most of the landscape metrics were originally derived from percolation theory, fractal theory and information theory (Cardille & Turner, 2002), devoted to investigating landscape characteristics, such as dominance, fragmentation, connectivity, diversity, and complexity (McGarigal et al., 2002). Landscape metrics were widely used in natural landscape analysis and monitoring (Herzog et al., 2001). Landscape metrics have been used to guide conservation and restoration planning. Metrics were proved to be an effective approach when performing conservation planning (Sundell-Turner & Rodewald, 2008). These metrics also can be used to model alternative planning and management scenarios and thereafter guide ecologically sound planning (Lasanta et al., 2006). Biodiversity can be reflected or monitored by landscape metrics (Bailey et al., 2007). To maintain biodiversity and monitor animal migration, animal movement was attempted to be modeled by different patch isolation metrics (Bender et al., 2003). To curb deforestation, forest pattern was also monitored and analyzed by landscape metrics (Frohn & Hao, 2006). In addition, there has been a growing awareness of the usefulness of landscape metrics in metropolitan environmental planning (Jim & Chen, 2003; Li et al., 2005), such as using them to optimize the benefit of green space in urban sustainable development (Uy & Nakagoshi, 2008). Landscape metrics can also provide foundations for assessing, evaluating and visualizing biodiversity in response to urbanization (Mörtberg et al., 2007). Moreover, landscape metrics were used to examine the landscape connectivity and green 10 networks, especially in urban areas where large areas of green patches are not allowed to be developed (Zhang & Wang, 2006). While landscape metrics provide many insights into environmental processes, interpreting landscape metrics requires some cautions in order to decipher the true ecological processes. Three aspects require special attention when applying landscape metrics, i.e. conceptual design in landscape pattern analysis, appropriate selection of landscape indices, and inherent limitations of landscape indices (Li & Wu, 2004). In addition, since landscape metrics take on a variety of forms, selecting representative and understandable metrics shows great significance (McGarigal et al., 2002). 2.1.3 Landscape gradient analysis Quantifying gradient using landscape metrics is an important breakthrough because it attempts to reflect the spatial landscape variations instead of describing only the overall patterns (McDonnell & Hahs, 2008). The most intensive urban development generally takes place in the city centre, while developments are less densely distributed in the urban fringe. This variation cannot be revealed by the conventional synoptic metrics analysis. The advent of “urban-rural” gradient largely unveils spatial heterogeneity, bringing insights into the directional landscape variations (Luck & Wu, 2002). The layout of spatial gradients commonly originates from the city centre and radiates to different orientations. Prior studies used transects in different directions, such as two directions (Luck & Wu, 2002), four directions (Hahs & McDonnell, 2006; Yu & Ng, 2007; Yeh & Huang, 2009), and eight directions (Kong & Nakagoshi, 2006). 11 The spatial patterns and variations obtained from the gradient analysis in conjunction with landscape metrics, however, often depend on the positions and the orientations of transects (Hahs & McDonnell, 2006; Yeh & Huang, 2009). Linear gradient analysis examines patterns along a predefined direction which may be limited in capturing the spatial variation of land use patterns. Urban morphology suggests that there are several urban development forms, such as linear, concentric, sector and multiple nuclei forms (Yu & Ng, 2007). For cities of which development patterns are other than linear form, linear gradient analysis may be limited in capturing the landscape patterns (Yeh & Huang, 2009). Scant research, however, examines the spatial variation of green space in cities whose developments conform to a concentric form, which is also a common development model in many Chinese cities (Jim & Chen, 2003; Tian et al., 2010). This study thus proposes concentric and directional landscape analyses with landscape pattern metrics to characterize the changing land use patterns. The proposed method compensates the deficiency of the linear transect method in capturing the landscape feature in different concentric belts. 2.1.4 Change intensity analysis Apart from understanding the spatial variation of land use patterns, insights into the rate of land use change is desirable because such information is vital for land use monitoring and prediction (Xiao et al., 2006). The availability of multi-temporal remotely sensed data and their increased spatial resolution and coverage have enabled more extensive monitoring of land use change (Yang et al., 2003). Timely and accurate change detection of land use 12 can build up concrete premise for a better understanding of land surface features and is also useful in quantifying the land use change intensity (Lu et al., 2004). Change intensity is an important indicator to evaluate the urban land use dynamics (Yu & Ng, 2007). Areas which urbanize at a fast rate are mostly involved with an intensive loss of natural land and severe environmental degradations. This indicator can be used directly as planning guide to monitor and control urban expansion. In addition, change intensity is a critical factor for predicting land use conditions (Liu et al., 2003). Many prior studies identified the transitions among different land use types as an overall change rate (Chen et al., 2005; Xiao et al., 2006; Yu & Ng, 2007). Localized change intensity analysis to reveal the spatially varied land use change intensities would therefore be desirable. Therefore, this study calculates the change intensity of the study area as a whole, the concentric belts, and the directional transects to examine the local variation of land use change rate in the city. 2.2 Urban thermal environment analysis Incessant urbanization process causes a major alteration to the earth surface, with much of the natural land cover being replaced by impervious materials. Both the increase of impervious surface and the loss of green space bring about a series of negative environmental impacts. Of these impacts, urban heat island (UHI), a phenomenon whereby surface and air temperature in the urban core is higher than that in the suburban areas (Voogt & Oke, 2003), is commonly associated with cities (Weng, 2001). This part reviews 13 prior studies on the urban thermal environmental dynamics caused by the change of land use. 2.2.1 Impact of land use change on urban thermal environment Since each land use type has its unique thermal, moisture, and optical spectral properties, the change of land use affects local thermal environment (Oke, 1982). The expansion of impervious surface alters the heat capacity and radiative properties of land surface (Streutker, 2002). The removal of vegetated areas alters the surface energy balance, largely increasing the sensible heat flux at the cost of the latent heat flux (Owen et al., 1998). The evapotranspiration of vegetation is substantially reduced. The warming thermal environment may induce many negative influences, such as the rise of urban pollution and the modification of precipitation patterns (Yuan & Bauer, 2007). Intense heat in the urban environment will bring much discomfort, and in most severe cases even cause mortality (Chang et al., 2007; Johnson et al., 2009). Hence, there is a growing interest concerning the urban thermal environment and its driving forces (Nichol, 2005; Buyantuyev & Wu, 2010). 2.2.2 Measurements of the thermal environment The thermal environment can be represented by atmospheric heat and surface heat (Yuan & Bauer, 2007). Atmospheric heat can be measured in the urban canopy layer, the layer extending from the earth surface upwards to approximately mean building height, and the urban boundary layer, the layer above the urban canopy layer, which is influenced by the earth surface (Voogt 14 & Oke, 2003). Previously, studies of urban thermal environment are usually based on the air temperature obtained from in-situ weather stations or automobile transects (Li et al., 2009). Although in-situ data accurately record the local temperature, they suffer from some limitations. They are normally costly and subject to governmental restrictions when conducting regional scale field studies (Owen et al., 1998). Furthermore, data collected from several meteorological stations are discrete points in a continuous space, which can hardly reflect the spatial temperature variation caused by different land use types. An alternative to measuring urban thermal environment is land surface temperature (LST) because it is able to modulate the air temperature of the layer immediately above the earth surface and is a major parameter that closely associates with surface radiation and energy exchange (Voogt & Oke, 1998; Weng, 2009). With the advent of thermal airborne sensors, satellite and aircraft platforms provide considerable opportunities to observe surface temperature with high spatial detail and temporal frequency (Jensen & Cowen, 1999). A variety of sensors can be used to retrieve thermal infrared data, such as Landsat TM/ETM+, MODIS, ASTER and AVHRR, facilitating the monitoring and investigation of surface thermal properties (Weng, 2009). 2.2.3 Relationship between LST and land use change Many studies have been conducted to investigate the LST pattern and its relationship to land use change (Chen et al., 2006; Xiao & Weng, 2007). Prior studies on this subject mainly used two approaches: comparison between 15 LST in different land use conditions and modeling of LST based on remote sensing indices. When comparing LST in different land use conditions, multispectral techniques were frequently used since land use and thermal information can be obtained simultaneously by a single sensor (Voogt & Oke, 2003). To combine the thermal and land use information, GIS was incorporated into the analytical process to address the relationship between LST and surface characteristics (Weng, 2001). GIS functions, such as spatial overlay, buffering and image differencing, coupled with biophysical parameters derived from remotely sensed images, such as surface temperature and emissivity, effectively unveiled the impacts of land use alteration on thermal environment change. The revealed contributions of different land use types on surface temperature can be exploited to predict LST in the future (Chen et al., 2006). In order to quantify the relationship between LST and LUCC, some remote sensing indices have been used to model LST instead of using the categorical land use types (Dousset & Gourmelon, 2003). A series of remote sensing indices have been used to reflect the relationship between land use types and surface temperature. Vegetation indices, such as the normalized difference vegetation index (NDVI), were used to validate the important role played by green space in mitigating UHI (Yuan & Bauer, 2007). Other indices, such as the normalized difference built-up index (NDBI) (Zha et al., 2003) and the normalized difference water index (NDWI) (Gao, 1996), can be also used to represent water and urban areas quantitatively. Prior studies have adopted these indices to model LST (Chen et al. 2006), but few of them compared the modeling results among different stages of urbanization. This study thus 16 compares the LST in two different stages of urbanization, aiming at investigating the impacts of urban expansion on the local thermal environment. 2.2.4 Spatial variation in LST modeling Prior studies used different global models, such as linear, nonlinear, multivariate models to estimate LST based on different explanatory variables (Price, 1990; Weng et al., 2004; Chen et al., 2006). These global regressions assumed that relationship between LST and explanatory variables were spatially constant. In other words, the estimated parameters remained as constants across space (Bagheri et al., 2009). However, in most cases, the relationship might vary spatially. Since different land use types are not evenly distributed across the space and the impact intensity of land use on LST may differ from place to place, the conventional models may overlook the spatial variation in LST modeling. Localized modeling, on the other hand, allows the relationship between predictor and explanatory variables to alter over space, assessing the local influences on the dependent variable (Fotheringham et al., 2002). Localized modeling techniques have been used in spatial pattern modeling, such as the analysis of accessibility to facilities (Bagheri et al., 2009), driving forces of urban growth (Luo & Wei, 2009), and population pattern estimation (Luo & Wei, 2006). However, few studies documented the application of localized statistical models in analyzing spatially varied impacts of land use on LST patterns. In this study, a localized modeling based on geographically weighted regression (GWR) will thus be applied to reveal the spatial variations. 17 2.3 Green biomass analysis Another implication which comes along with the landscape pattern change is associated with the biophysical environment (Brovkin et al., 1999). Under the pressure of population growth and incessant urbanization, many anthropogenic activities have resulted in high green space loss rates worldwide. It was estimated that 30% to 50% of land surface has been transformed by human actions and a large portion of them is associated with deforestation (Brovkin et al., 2004). Biophysical impacts from urban expansion include the disorder in the interaction between the biosphere and atmosphere (Anaya et al., 2009) and the crisis in biodiversity conservation (Marsh & Grossa, 1996). It is imperative to investigate the biophysical impacts of rapid urbanization. Aboveground green biomass is an important indicator for evaluating ecosystem functions and understanding carbon equilibrium, which has been used to represent biophysical characteristics (Li et al., 2009). 2.3.1 Importance to measure green biomass Green space is the primary source of carbon conservation. Green plants absorb solar radiation to manufacture organic subsistence which is the fundamental premise of vegetation. Although less than 1% of the solar energy is utilized in this process, it is essential to the entire system of life on the earth (Marsh & Grossa, 1996). Among all the green space, forest conserves more carbon than other terrestrial ecosystems, accounting for 90% of the annual carbon flux between the atmosphere and the earth surface (Winjum et al., 1993). The stored carbon in the forests emits to atmosphere through 18 deforestation and forest fires, and this carbon emission is one of the most important sources of global warming (Lu, 2006). To monitor the forest carbon storage, particularly in support of a sustainable forest management and carbon accounting, biomass estimation and mapping become critical (Labrecque et al., 2006). Estimation of green biomass is an effective way to assess forest ecosystem productivity (Scott et al., 2010). Delineation of green biomass distribution helps in reducing the uncertainty of carbon sequestration and understanding its role in environmental processes and sustainability (Foody, 2003). Mapping temporal biomass change can detect areas where forests are endangered and areas where afforestation are needed. Another negative impact of the biomass loss is associated with the reduction in biodiversity. Habitat loss has been regarded as one of the most important factors in the global biodiversity crisis (Sala et al., 2000) and as a significant predicator of the amount of the threatened species in certain hotspots (Brooks et al., 2002). Land use change often makes the green space become more fragmented, separating species in isolated patches. This trend further impacts the local biodiversity, because species with small populations are prone to extinct since the relatively few individuals are more vulnerable to predators, and the population may be under the critical threshold for breeding (Marsh & Grossa, 1996). More species tend to be found in habitats with dense vegetation cover where there may be fewer human disturbances (Leitão et al., 2006). Therefore, green space with different VD might have varied capacities in terms of biodiversity conservation. Indeed, VD has been used to represent biomass (Nichol & Wong, 2005), an important indicator of biodiversity. 19 2.3.2 Approaches to measure biomass Three approaches are commonly used to measure biomass (Lu, 2006). First, conventional techniques based on field measurement are generally the most accurate approach to derive green biomass (Brown, 2002; Lu, 2006). To derive green biomass, field measurements or spatial inventory data are combined with allometric equations to develop a function to estimate tree height and above ground biomass (Parresol, 1999; Fournier et al., 2003; Labrecque et al., 2006). The number of field sites and measurements must attain to a sufficient amount to achieve a high accuracy. To meet this requirement, especially at a regional scale, it is normally time consuming and labor intensive. Some remote places are even difficult to measure (Labrecque et al., 2006; Lu, 2006). In addition, in-situ biomass measurement carried out at a certain time may become useless if the landscape undergoes a major change (Hall et al., 2006). The second approach is GIS based green biomass estimation. Spatial interpolation and extrapolation are performed based on the known biomass samples and vegetation types to model the biomass in the entire study area (Houghton et al., 2001). Different geostatistical methods, such as Kriging, are widely used to model the green biomass (Sales et al., 2007). However, the accuracy of the interpolation method depends on the acquired data quality and it will become less effective where green space distribution is largely heterogeneous. GIS based biomass estimation also employs some ancillary data (e.g. elevation, soil, slope and precipitation) combined with sampled green biomass to estimate regional green biomass (Brown & Gaston, 1995). 20 This approach has its own limitation as well because high quality ancillary data are required and more importantly, the indirect relationship between biomass and the ancillary data weakens the reliability of the estimated results (Lu, 2006). The third approach is remote sensing based biomass estimation. Several characteristics, such as the multi-temporal imageries, fine resolution and high correlations with vegetation parameters, make remote sensing the best way to estimate green biomass at a regional scale when field data are relatively scarce (Anaya et al., 2009). It provides efficient and timely estimation of green biomass based on repetitive, comprehensive observations at different scales (Patenaude et al, 2005; Scott et al., 2010). Remote sensing based regional biomass estimation have been used in, for example, India (Roy & Ravan, 1996), Malaysia (Phua & Saito, 2003), and Wisconsin, USA (Zheng et al., 2004). In addition, a variety of optical sensors with different spatial resolutions have been used (Lu, 2006), such as the fine spatial resolution IKONOS and Quick Bird images (Lévesque & King, 2003), the medium spatial resolution TM/ETM+ images (Labrecque et al., 2006), and the coarse resolution AVHRR (Barbosa et al., 1999). For remote sensing based biomass estimation, three pathways are generally used to derive green biomass. The first is based on the direct correlations between different spectral values and biomass (Asner & Heidebrecht, 2002). The spectral values can be satellite radiance, reflectance and vegetation indices (Labrecque et al., 2006). Other methods such as spectral mixture analysis and canopy reflectance are also employed to estimate green biomass (Asner et al., 2003). 21 The second pathway is using biomass conversion tables to link with land use classification, and then calculate the biomass for each land use type (Labrecque et al., 2006). The level of accuracy depends on how detail the land use classification and the corresponding conversion tables are. The third pathway is to use other indicators to represent green biomass, such as canopy texture index and VD. Canopy texture index was proposed by Couteron et al. (2005) as an alternative to measure physical attributes of tree crowns. In some cases relative abundance of greenness may suffice, and the accurate biomass value is not required. This is particularly true in the urban areas. In this context, the absolute biomass value is represented by some other indicators with explicit meanings. VD is one indicator which is devised to coarsely represent the amount of urban green biomass and explicitly describe the complicated vegetation structures (Nichol & Lee, 2005). Due to its simplicity and representativeness, VD is calculated to represent urban green biomass and model urban environmental quality (Nichol & Wong, 2005). However, only a few studies used this indicator to reveal the temporal differences in urban green biomass. Thus, this study examines the biomass dynamics in response to urbanization and greening policies using VD, which is an appropriate substitute of absolute biomass value for urban vegetation measurement. 2.4 Urban expansion in Chinese cities A number of studies have been conducted in Chinese cities to examine the process of land use change in recent decades. Cities like Hangzhou, 22 Jiaxing and Huzhou exhibited a remarkable urban expansion from 1994 to 2003 (Su et al., 2010). Driven by economic growth and population increase, urbanization was accelerated, resulting in the degradation, isolation and fragmentation of agricultural land in many parts of China (Su et al., 2010). For example, the built-up area in Guangzhou City expanded 325.5 km2 from 1979 to 2002, replacing much of the vegetation areas (Ma & Xu, 2010). In addition, many coastal cities in the eastern China face the decline and destruction of farmland and wetland. For example, wetland in Lianyungang was found to shrink 54.4 km2 from 2000 to 2006 (Li et al., 2010). Cities in central China also face huge challenges of urban expansion. The encroachment of green space and arable land in Wuhan city from 1993 to 2000 is documented (Cheng & Masser, 2003). Apart from these, 13 Chinese mega cities were reported to be experiencing rapid urbanization, largely driven by demographical change, economic growth, and land use regulations (Liu et al., 2005). Almost all the urban expansion in those mega cities accompanies a reduction of natural space. The process of urbanization has posed many problems to the landscape and affected ecosystem functioning of the cities and their surrounding areas (Xian et al., 2007). Finding certain countermeasures to curb the rate of urban expansion and evaluating the influence of those countermeasures are of great significance. Additionally, most case studies are conducted in economically developed cities, such as Beijing, Shanghai, Guangzhou, and Hangzhou, while cities in southwestern China are somewhat neglected (Cheng & Masser, 2003). Therefore, this study uses Kunming City in the southwest China as an empirical case to investigate the process of urban expansion. Moreover, this study attempts to evaluate the effect of greening policies as a countermeasure 23 to urban expansion in Kunming City. The following chapter will detail the study area and describe the greening policies. 24 Chapter 3: The Study Area Kunming, China 3.1 Geography Kunming is located between 102°10’ to 103°40’E and 24°23’ to 26°33’N, in the north-central Yunnan province, China (Figure 3.1). The topography of Kunming is characterized by a plateau with a decrease of altitude from north to south. Elevation in Kunming ranges from 1500m to 2800m and the metropolitan area is located in the Dianchi basin with an elevation of 1890m, surrounded by mountains on three sides. Kunming has the sixth largest fresh water lake in China, the Dianchi Lake, with an area of approximate 340 km2. This lake plays an important role in climate regulation and city beautification, and is thus entitled “the Pearl of the Plateau”. Figure 3.1: Location of the study area. (a) Yunnan Province in China; (b) Kunming City in Yunnan Province; (c) Kunming metropolitan area with newly constructed urban parks and ecological wedges labeled. 25 3.2 Climate Situated in a lake basin, surrounded by mountains on three sides, and seated at an elevation of 1890m on the plateau, the city is controlled by a subtropical highland climate, enjoying a mild climate in most months in a year and is known as the “City of Eternal Spring”. Precipitation is around 1000mm/year, with about 2,250 hours of sun exposure in one year. In Kunming, summer is warm and humid while winter is cold and dry. Spring and autumn are relatively longer than summer and winter. Weather is characterized by intense sunlight in the daytime and a slight chill at night. 3.3 Vegetation and wildlife Due to the mild climate and intense sunlight, vegetation in Kunming enjoys a benign natural environment. Products such as grain, potato and a variety of fruits are exported to other cities in China. Kunming is also famous for its flowers. The abundance of camellias, orchids, and azaleas rejuvenates the city. Because of these characteristics, Kunming held the International Horticulture Exposition in 1999, which helped attract more attention to Kunming and expedited the urban development. In addition, the physical environment of Kunming with mountains on three sides offers extensive forest areas. These forest areas are important habitats for most wildlife in Kunming. In addition, urban environment provides habitats for wildlife. Because of the moderate temperature in winter, lots of seagulls fly from Siberia to Kunming every winter, which brings much attraction to this city. 26 3.4 Economic growth Before 1999, Kunming had been mostly overlooked in the rapid development of China because its economic growth rate was lower than most of the eastern Chinese cites. Kunming received much more attention after it held the International Horticulture Exposition in 1999. As the main passage from China to most of the countries in Southeast Asia, such as Vietnam, Laos, Thailand, and Cambodia, and the traffic hubs of the Mekong regional economic cooperation circle, international trade developed by leaps and bounds. Kunming has become an important city in the southwest part of China. The gross domestic product (GDP) increased from 16.9 billion Chinese Yuan in 1992 to 56.2 billons in 1998. From 1998 to 2008, the GDP was nearly tripled, attaining to 160.5 billion Chinese Yuan. Concomitant with the fast economy growth was the population increase, which grew from 4.4 million to 6.2 million from 1992 to 2008 (Kunming Statistical Year Book, 2009). 3.5 Urbanization and green space loss Fast economic growth was accompanied with a rapid urbanization process. The urbanized area increased from 184.42 km2 in 1992 to 257.83 km2 in 2005 (Cai, 2007). Urban expansion in Kunming largely conformed to the concentric model, with varied developing intensities in different concentric belts (Zhou, 2009). Many agriculture lands on the urban fringe were encroached during the past two decades. Forest lands were also endangered in the process of urban expansion, some of which were converted into built-up areas and agriculture lands. Within the city, intensive urban development deprived many semi-natural elements, reducing many urban amenities. 27 3.6 Greening policies Having realized the important role of green space in urban ecosystems and the losing trend of urban green areas, the Kunming Municipal Government launched a series of greening policies. To rejuvenate the city with natural elements, in 2000, the Kunming Landscape Bureau carried out several greening projects to construct more urban parks and community gardens. In 2005, the municipal government set the goal to develop Kunming into one of the National Garden Cities in China in three years. A series of campaigns and activities were therefore implemented from 2006 to 2009 to allot more green areas and corridors in the urban areas, aiming to meet the standard of the National Garden City. 3.7 Area of the research site Urban regions comprise cities and their interdependent suburban areas (Yeh & Huang, 2009). To examine the interaction between the urbanized area and its surroundings, the study area is composed of three areas: urban area (dominated by impervious land), urban fringe (mixture of impervious land and agriculture land), and suburban area (dominated by forest land). The total area for these three parts is 538.89km2 in this study (Figure 3.1c). 28 Chapter 4: Materials and Methods In this chapter, Section 4.1 introduces the data used in the thesis. Section 4.2 describes the process of data preparation. Then, methods for the following three analyses are elaborated (Figure 4.1). The first analysis on landscape patterns in response to urbanization process and greening policies from 1992 to 2009 is given in Section 4.3. The spatial variations of landscape pattern and change intensities were characterized by the proposed concentric and directional landscape analyses with landscape pattern metrics. Section 4.4 introduces the analysis of the urban thermal environment. The procedure of green biomass analysis is presented in Section 4.5. Figure 4.1: General flow chart of this study. Each box will be further illustrated in the following sections. 4.1 Data used in this study Data used in this study included remotely sensed images, field survey data and green space maps (Figure 4.2). For landscape pattern analysis, the following cloud-free remotely sensed images were analyzed: Landsat 5 Thematic Mapper (TM) image acquired on 16 August 1992, Landsat 7 Enhanced Thematic Mapper Plus (ETM+) image acquired on 11 November 29 2000, and Systeme Probatoire d’Observation dela Tarre (SPOT 4) image acquired on 20 September 2009. Figure 4.2: Data used in this study. “+” indicates data used in landscape pattern analysis; “○” indicates data used in thermal environment analysis; and “□” indicates data used in green biomass analysis. For surface thermal environment study, since SPOT image does not include any thermal infrared band to derive temperature information, Landsat image observed on 19 May 2006 was selected to derive the surface temperature. In addition, the Landsat ETM+ image observed on 11 November 2000 was discarded because the surface temperature in early winter is much lower than summer and early autumn, although most evergreen vegetation still can be detected from the image. Therefore, in surface thermal analysis, Landsat TM/ETM+ images in 1992 and 2006 were used. These two images were able to represent the general change tendency of surface thermal environment from the beginning of the 1990s to the end of the 2000s. 30 For green biomass analysis, Landsat TM/ETM+ images in 2000 and 2006 and SPOT image in 2009 were used. Landsat TM image in 1992 was not used because measuring VD relies on fine resolution data as sampling sources, but such data for 1992 were not available. The green space map was prepared by the Municipal Planning Bureau in 2000, the Quickbird images were observed on 3 December 2006 and 20 September 2007, and the current Google Earth image in Kunming were served as sampling maps to derive VD for 2000, 2006 and 2009. In spite of no 1992 data for the green biomass information, the rest of the images still can reflect the green biomass dynamics in the context of landscape pattern change. In terms of spatial resolution of the remotely sensed images used, Landsat images (bands 1-5 and 7) have a spatial resolution of 30m. ETM+ image provides an additional panchromatic band with a spatial resolution of 15 meters. Both TM and ETM+ are loaded with a thermal sensor, with a spatial resolution of 120m and of 60m respectively. SPOT image has a resolution of 20m in multispectral bands and 10m in panchromatic band. Quickbird image has four multi-spectral bands with 2.4m spatial resolution. Panchromatic band of Quickbird provides a fine resolution at 0.6m. Apart from remotely sensed images, green space map in 2000 and green space planning map in 2010 were used to show the green space change tendency in Kunming. Field survey data included the ground control points collected using Global Positioning Systems (GPS) for geo-referencing satellite images and photographs for assisting in image interpretation. 31 4.2 Data preparation Data preparation was done for remotely sensed data, planning maps and field survey data respectively (Figure 4.3). Figure 4.3: The processes of data preparation. 4.2.1 Preparations for planning maps and field survey data For planning maps, image clipping was first carried out to extract land use information within the study area. Geo-rectification was then performed to impose spatial information onto each map which thereafter bore the spatial coordinates. In-situ collected data were sorted out and arranged for the subsequent image geo-referencing and interpretation. 32 4.2.2 Image pre-process Through the preliminary inspection of the selected satellite images of the study area, slight cloud cover was detected on ETM+ images on 19 May 2006. To remove the effect of cloud on land use classification and LST derivation, these clouds and the shade below them were masked out before processing images. Pre-processing of the satellite images involved the following steps (Figure 4.3). First, the images were geo-referenced to the Universal Transverse Mercator (UTM) coordinate system (Spheroid WGS 84, Datum WGS 84, Zone 48) using ERDAS IMAGINE 8.7 software package. Fifteen spatial evenly distributed ground control points were selected for geometric correction using a second-order-polynomial transformation and the nearest-neighborhood resampling method. Root mean square error was limited within 0.5 pixels. Because of the multi-temporal and multi-sensor images analyzed in this study, atmospheric correction was required and the Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes (FLAASH) atmospheric correction module in ENVI 4.6 was used. Although Kunming is surrounded by mountains, the study area is almost located within the basin without much topographic change. Topographic correction was thus not performed in this study. In order to match the spatial resolution, the SPOT image was resampled to 30m, the same resolution as the Landsat images. The thermal band of the TM image was resampled to 60m, the same as the ETM+ image. 33 4.2.3 Land use classification system The selection of the land use classification system is important because it influences the results and subsequent interpretations (Lu & Weng, 2007). For landscape pattern analysis, land use of the study area was classified into green space, built-up land (impervious land including residential, commercial and industrial land together with public transportation land), water (streams, canals, lakes and reservoirs), and barren land (bare exposed rocks and derelict lands). Prior work has further suggested three main types of green space in the cities, namely forests, farmlands and parks (Li et al., 2005). Accordingly, green space in this study was subdivided into forest land (deciduous, evergreen and mixed forest), agriculture land (croplands and orchards), and grass land (gardens, parks with low tree density, and golf courses). This classification system largely conformed to the Anderson classification standard (Jensen, 2005) and the Chinese land use system (Chinese Current Land Use Classification, 2007). For thermal environment change analysis, when imposing land use classification layers onto the LST layers, the average LST of grass lands was similar to that of forest lands but with a large variation in the pilot trials. Since the spatial resolution of thermal band on Landsat images is limited (60m for ETM+ and 120m for TM) and grass land in the study area is small, the digital value of a singular pixel might comprise a mixture of thermal radiation from grass and other land use types. Besides, since the average surface temperature of grass land was similar to that of forest land in the pilot trials, in thermal environment analysis, grass land was removed from the classification system for the thermal analysis. 34 4.2.4 Classification and accuracy assessment The maximum likelihood classifier was used in land use classification. Statistics of training samples for each image was plotted to ensure that different land use classes could be separated. Accuracy assessment was performed using the stratified random sampling method. Field survey assisted by historical photos and knowledge of the area were used as the references to operate the assessment. Images of 1992, 2000 and 2009 were classified based on the classification system for landscape pattern analysis while images observed in 1992 and 2006 were classified according to the classification system for thermal environment analysis (i.e. without grassland). After classification, two hundred reference points, derived from the stratified random sampling method, were selected for each image to assess the classification accuracy. The overall accuracy for all the images was above 81%. The Kappa coefficient was greater than 0.82. Minor adjustments were done to correct the misclassified classes through field work verification. 4.3 Landscape pattern analysis After preparing and classifying images, landscape metrics and change intensity analyses were carried out in the landscape pattern analysis (Figure 4.4). 35 Figure 4.4: Flow chart of the landscape pattern analysis. Image preparation has been described in section 4.2.2. 4.3.1 Landscape metrics analysis The classification results from the 1992, 2000, and 2009 satellite images were converted into GRID format for the calculation of landscape metrics in FRAGSTATS 3.3 (McGarigal et al., 2002). A total of five landscape metrics, including percentage of landscape (PLAND), patch density (PD), largest patch index (LPI), landscape shape index (LSI) and Shannon’s diversity index (SHDI), were used to quantify the spatial patterns of urban green space in 1992, 2000, and 2009. PLAND is a general index which depicts the relative abundance of each land use type; PD is sensitive to landscape fragmentation; LPI can be used to illustrate landscape dominance; LSI is associated with landscape complexity; SHDI represents landscape diversity (Table 4.1). These landscape metrics were used in three analyses (Figure 4.4). First, synoptic analysis was carried out to quantify overall spatial patterns. 36 Second, the study area was divided into seven concentric belts (Figure 4.5a), each of which spanned two kilometers, approximately the average distance between the adjacent ring roads in Kunming City. Concentric analysis was then conducted to measure the landscape patterns in different belts radiated from the city center to the fringe. Third, following the eight directions undertaken in Kong and Nakagoshi (2006), directional analysis of PLAND was performed to examine the green space distribution by transects in eight directions (Figure 4.5b). Each transect consisted of a number of 1.8 km x 1.8 km cells. The 1.8 km length was slightly modified from the averaged distance between the adjacent ring roads so as to allow the eight transacts with cells to fit into the study area completely. Both concentric belts and directional cells are created in ArcGIS. These belts and cells were exported into FRAGSTATS to derive the five landscape metrics in Table 4.1. 37 Table 4.1: Landscape metrics used in this study (McGarigal et al., 2002). Metrics Description Percentage of Landscape (PLAND) The proportion of the area of certain land use class to the entire landscape area Percent 0< PLAND ≤ 100 General index Patch Density (PD) The number of patches in certain class divided by the entire landscape area Number per 100 hectares PD > 0 Index of fragmentation Largest The largest patch of Patch Index certain class divided by (LPI) the entire landscape area Percent 0 < LPI ≤ 100 Index of fragmentation and dominance Landscape Patch perimeter divided Shape by the minimum Index (LSI) perimeter possible for a maximally compact patch of the corresponding patch area None LSI ≥ 1 Index of shape Shannon’s Diversity Index (SHDI) None SHDI ≥ 0 Minus the sum, across all patch types, of the proportional abundance of each patch type multiplied by that proportion Unit Range Justification Index of diversity 38 Figure 4.5: The layout of concentric belts and directional cells. (a) Seven concentric belts radiated from the city center; (b) Forty-four sampling cells in transects of eight directions. 4.3.2 Change intensity analysis Land use change rate, computed as percent change per year, was used to reflect the change intensity (Xiao et al., 2006). The change rate was calculated as below: Ci = Ai (T 1) − Ai (T 2 ) T1 −T 2 (Equation 4.1) Where Ci means the change rate for land use type i, Ai(T1) denotes the area of land use type i in time T1. Areas with rapid green space loss may suggest high vulnerability to urban expansion. The change intensity analysis was calculated for two time periods (1992-2000 and 2000-2009) for the overall study area, the seven concentric belts, and the cells of the eight directional transects. 39 4.4 Thermal environment analysis Figure 4.6 demonstrated the work flow of the thermal environment analysis. Optical bands and thermal bands of the Landsat images were used to derive land cover information and surface temperature information separately. Image classification was performed based on the specified classification system for thermal analysis (i.e. grassland excluded). GIS spatial overlay and statistical modeling were used to quantify the relationships between LST change and landscape dynamics. Figure 4.6: Flow chart of the thermal environment analysis. Image preparation has been described in section 4.2.2. Spatial overlay (a) is the overlap between the land use layer and the LST layer while spatial overlay (b) is the overlap between land use change layer and the LST layer. 40 4.4.1 LST derivation The digital number (DN) values of both Landsat images were converted into spectral radiance based on the following formula: Radiance = gain × DN + offset (Equation 4.2) Where, gain and offset were derived from the head files of images and the Landsat current radiometric calibration coefficients (Chander et al., 2009). The retrieved radiance was converted to at-satellite brightness temperature (BT) based on the following equation, assuming that land cover had the same emissivity (Weng & Lu, 2008): Tb = K2 ln(1 + K 1 / Lλ ) (Equation 4.3) Where Tb is the at-satellite BT in Kelvin (K), Lλ is the spectral radiance in Wm-2 sr-1μm-1; K1 and K2 are the calibration constraint. K1=607.76 Wm-2 sr-1 μm-1, K2=1260.56K. To retrieve the LST, at-satellite BT was required to correct according to the real object properties. Therefore, the emissivity corrected surface temperature was computed following (Artis & Carnahan, 1982): Ts = Tb 1 + (λ × Tb / α ) ln ε (Equation 4.4) Where Ts is the land surface temperature in K, Tb is the at-satellite BT in K, λ is the wavelength of emitted radiance (λ=11.5μm) (Markham & Barker, 1985). α=hc/σ, (1.438×10-2mK), with σ as the Boltzmann constant (1.38× 10-23J/K), h as the Planck constant (6.626×10-34Js), and c as the velocity of light 41 (2.998×108m/s). ε is the surface emissivity, which derived from NDVI (Artis & Carnahan, 1982; Roerink et al., 2000). 4.4.2 Derivation of remote sensing indices Three remote sensing indices, NDVI, NDBI and modified normalized difference water index (MNDWI) (Xu, 2006), were computed to characterize different land use types. NDVI is an index that describes the greenness of areas, which is widely used in environmental studies (Chen & Brutsaert, 1998). NDBI is an indicator of urban areas, which can show an increment in built-up areas and barren land (Zha et al., 2003; Chen et al., 2006). MNDWI is used in this study because compared with NDWI, MNDWI can remove built-up land noise in the urban areas to some extent (Xu, 2008). When modeling the urban thermal environment, water areas revealed a remarkable difference in thermal characteristics. Therefore, in this study, MNDWI was selected to represent water areas. These three indices can largely represent most land use types in this study and are calculated as follows: NDVI = NDBI = RNIR − RRED RNIR + RRED (Equation 4.5) RMIR − RNIR RMIR + RNIR (Equation 4.6) MNDWI = RGREEN − RMIR RGRREN + RMIR (Equation 4.7) Where RNIR means the reflectance in the near infrared band; RRED and RGREEN respectively stand for reflectance in red and green bands; RMIR denotes the reflectance in the middle infrared band. 42 4.4.3 Impact of pattern change on LST The classified land use and the derived LST layers were incorporated into ArcGIS 9.3 to further analyze the relationship between the spatial pattern of LST and land use condition. Due to spatial autocorrelation, a large portion of the data was redundant for spatial analysis. Therefore, a spatial sampling was carried out to extract 5929 points which were evenly and densely distributed over the study area. The LST layers were then imposed onto the land use layers to explore the average temperature for each land use type (Spatial overlay (a) in Figure 4.6). The LST for different land use types in different years were compared to examine the thermal environment change. To examine to what extent land use change influenced the LST pattern, change detection was performed in ArcGIS to identify the areas where land use change took place from 1992 to 2006. Areas where land use changed from 1992 to 2006 were overlaid with the LST layers to calculate the LST differences before and after the occurrence of land use transformation (Spatial overlay (b) in Figure 4.6). LST variation is subject to three main factors: seasonal difference, time period of a day, and land use condition. To investigate the LST variation associated with only land use change, the other two factors need to be excluded. Since the Landsat satellite follows a sun-synchronous orbit, it ensures that objects in the same place have a similar local time. The two images used in this study were both captured at around 10:30AM local time, thereby excluding the influence of the day time factor. As for the factor of seasonal difference, temperature difference for the same land use type between 43 two years was assumed to be caused by seasonal factors. Therefore, to exclude the seasonal factors, LST difference for the same land use type was normalized against all other LST difference: = dTij Tj (2006) − Ti (1992) (Equation 4.8) ∆= Ti Ti (2006) − Ti (1992) (Equation 4.9) dT = n dTij − ∆Ti (Equation 4.10) Where dTij is the temperature difference between land use type j in 2006 and land use type i in 1992; △Ti is the temperature difference for the same land use type i in 2006 and 1992; dTn is the normalized temperature by subtracting △Ti from dTij. 4.4.4 Modeling LST based on land use indicators Land use was represented by remote sensing indices. Areas with more greenness exhibited higher NDVI value. Impervious land was quantified by NDBI. High NDBI value generally signifies the areas with intensive developments. MNDWI is sensitive to water body, so it exhibits a high value for areas covered by water. The multivariate regression was conducted to model the LST based on the combination of explanatory variables. The multivariate regressions based on ordinary least square (OLS) and geographically weighted regression (GWR) were performed. As opposed to OLS which might mask out significant local variation, GWR is a localized regression to examine the spatially non-stationary phenomenon (Fotheringham et al., 2002; Lloyd & Shuttleworth, 2005) and is expressed as: 44 y= β 0(ui, vi ) + ∑ β k (ui, vi ) Xki + ε i (Equation 4.11) k Where β0 is the constant that depends on the specific location i; βk is the coefficient of independent variable Xk at the location i; εi is the residual term at the location i; (ui, vi) indicates the coordinates of the point i. GWR allows local parameters to be estimated based on adjacent points. Coefficient βk is controlled by the weight which is assigned according to the spatial proximity of point i to its adjacent points. Weight can be obtained through two types of functions, fixed or adaptive (Páez et al., 2002a, 2002b). In this study, adaptive function is used which takes on the form as below (Fotheringham et al., 2002): dij wij = [1 − ( ) 2 ]2 if j ∈ {S } b wij = 0 otherwise (Equation 4.12) Where dij is the distance from point i to j; S is the set that indicates the specified N nearest neighbor points; b is referred as bandwidth, the distance from the Nth nearest neighbor point to i. 4.5 Green biomass analysis VD was used as the proxy to measure the urban green biomass. VD differs from vegetation cover because vegetation cover only measures a two dimensional land use phenomenon. VD, on the other hand, reflected vegetation amount in all vertical layers above ground. When comparing the relative green biomass in the urban environment, VD density was therefore an ideal indicator (Nichol & Wong, 2005). Besides, the mapped VD distribution and the analyzed temporal VD change can be used to guide the afforestation 45 activates and vegetation conservation. Figure 4.7 demonstrated the work flow of the green biomass analysis. Figure 4.7: Flow chart of the green biomass analysis. RS indices include Green band value, Chlorophyll index, NDVI and Simple ratio vegetation index. 4.5.1 Calculation of VD VD was computed based on the formula from Nichol and Lee (2005): VD= % 100 × ∑ WXAX / ATWT (Equation 4.13) Where WX is the weight for vegetation type X, which indicates the relative density for different forms of vegetation; AX is the areas covered by vegetation X in each plot; AT is the total area in each plot; WT is the sum of all weights. Based on the weighting methods used in Nichol and Lee (2005), weights were assigned according to the mean value of leave area index (LAI) for each type of land cover recorded in Scurlock et al. (2001) and according to the interpretation of the reference Quickbird image and planning maps in the 46 study area (Table 4.2). Within the same land use class, VD can be determined by the color, shape and texture reflected in the satellite images. Table 4.2: Weights of VD for different vegetation cover. Class Weight Description Sparse agriculture land 0.2 ≤ w[...]... examine the variation of the green space change rate in the city The second and the third analyses focused on the environmental implications of the changing landscape patterns The second part concerned with urban thermal environment change Integrated methods of Geographical Information Systems (GIS) and remote sensing were used to investigate the impact of land use change on the dynamics of land surface temperature... 2009) 2.1.2 Landscape metrics in sustainable development Investigating urban landscape space often requires researchers to first quantify the urban landscape composition and configuration so that a sound understanding of the distribution of urban landscape pattern can be attained (Forman & Godron, 1986) One of the milestones of landscape ecology was the extensive use of landscape metrics in spatial... address the following research questions: (1) How does the landscape pattern change in response to the rapid urbanization process and the greening policies? (2) To what extent do landscape pattern changes influence the local thermal environment? (3) How does the green biomass vary in the changing landscape pattern? 6 1 6 Outline of the thesis Following this brief introduction on the general rationale of the. .. dynamics 1.5 Scope and aims of this study This study investigated the dynamics of landscape patterns and the associated environmental implications in a city with rapid urbanization, Kunming, China In Kunming City, rapid urban growth encroached much green space on the urban fringe and suburbs, while the intensive urban development took place in many semi-natural places within the urban areas in the past several... with land use change, and the measurement of urban green biomass Chapter Five presents the results and discusses the dynamics of the urban green space pattern in response to the urbanization and greening policies Chapter Six displays and discusses findings of the thermal environment change due to landscape alterations Chapter Seven reveals the results of the amount of green biomass derived from VD and. .. ameliorate the urban environment and to enhance citizens’ quality of life, the municipal government has implemented several urban greening policies since 2000 To examine the effects of driving forces on landscape pattern change and the subsequent environmental implications, this study consisted of three analyses The first part focused on the landscape dynamics in response to the rapid urbanization and greening... studies on the urban thermal environmental dynamics caused by the change of land use 2.2.1 Impact of land use change on urban thermal environment Since each land use type has its unique thermal, moisture, and optical spectral properties, the change of land use affects local thermal environment (Oke, 1982) The expansion of impervious surface alters the heat capacity and radiative properties of land surface... implications were thus investigated under the rapid urbanization and greening policies Figure 1.1 illustrates the framework of this study 4 Figure 1.1: Process of landscape pattern change Urbanization and greening policies serve as driving forces to modify landscape patterns The change of landscape patterns results in a variety of implications, such as thermal environment and green biomass dynamics. .. Nakagoshi, 2006) 11 The spatial patterns and variations obtained from the gradient analysis in conjunction with landscape metrics, however, often depend on the positions and the orientations of transects (Hahs & McDonnell, 2006; Yeh & Huang, 2009) Linear gradient analysis examines patterns along a predefined direction which may be limited in capturing the spatial variation of land use patterns Urban morphology... metropolitan area (Li et al., 2005) Other examples included a greening plan based on landscape ecology in Nanjing (Jim & Chen, 2003) and a greenway augment plan in Xiamen Island (Zhang & Wang, 2006) 1.4 Process of landscape pattern change Understanding the process of landscape pattern change requires the examinations of the driving forces and the subsequent implications There are 3 complicated interactions among ... urbanization and greening policies in the Kunming metropolitan areas, China Urban thermal environment and green biomass were investigated in the context of landscape pattern change The concentric and. .. understanding of the distribution of urban landscape pattern can be attained (Forman & Godron, 1986) One of the milestones of landscape ecology was the extensive use of landscape metrics in spatial... areas and corridors in the urban areas, aiming to meet the standard of the National Garden City 3.7 Area of the research site Urban regions comprise cities and their interdependent suburban areas