1 ENVIRONMENTAL VARIABILITY AND MANAGEMENT: AN EXAMINATION OF LAND COVER CHANGE AND CONSERVATION IN SOUTHERN AFRICA By CERIAN GIBBES A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FUL FILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2011
2 2011 Cerian Gibbes
3 T o Phillip for your unending patience and support throughout this experience ; and to Sian for inculcating upon me the value of education
4 ACKNOWLEDGMENTS I wish to express my deepest gratitude to my advisor Dr. Jane Southworth Her advising style simultaneously supported and challenged me, ensuring that encouragement was offered when needed and that I extended my research an d professional development beyond my comfort zones. Her contagious enthusiasm is motivating and creates an exciting atmosphere which promotes the co ntinuous process of learning I am thankful to my advisory committee for helping me develop my professional and academic skills and reviewing my work My thanks to Dr. Eric Keys for daring me to challenge the accepted wisdom, for recognizing and encouraging my interest in the social components of the system, and for his truly open door policy. Dr. Timothy Fik h as played a central role in my undergraduate and graduate career. I have had the privilege of many brainstorming sessions with Dr. Fik, which produced creative ways to conceptualize and analyze data. More importantly though, through his encouragement my co nfidence in my research abilities has grown and I am grateful for his guidance throughout my entire time here at UF. Dr. southern Africa have been essential to my conducting research in this region. I thank him for h is eagerness to include graduate students in his many research endeavors in the region and for the many formal and informal discussions regarding conservation in the region which have helped shape my work and thinking. I appreciate Dr. Robert llingness to review my work and act as my external committee member. He has ensured that my work is relevant and can be effectively communicated to a wider audience. Although not officially part of my committee, I am also extremely appreciative of the guid ance (both research and career) provided to me by Dr. Peter Waylen. It is
5 due to his zealous teaching that I was first attracted to G eography at UF, and my overwhelming positive graduate experience is in part due to the pro graduate student atmosphere whic h flourished within the department while he was chair My thanks to the many individuals who collaborated with me both i n the field and in the lab, and with whom I have bounced ideas back and forth: Sanchayeeta Adhikari, Luke Rostant, Xia Cui, Pinki Mondal Dr. Youliang Qui, Narcisa Pricope, Erin Bunting, Bennety L ikukela Andrea Gaughan, Tim Fullman, Graham Child and the many others. I also want to thank Dr. Christian Russell for responding to my frantic cries for technical help, and for the encouraging w ords on frustrating days. I wish to thank Desiree Price a row, and for the supply of emergency candy and/or Advil This dissertation was funded by a University of Florida Alumni Fellowship, a NASA Research Assistantship ( NNX09AI25G ), the So cial Science Research Council and research grants from NSF/IGERT Working Forests in the Tropics program (NSF DGE 0221599), UF Center for African Studies UF Seed grant I thank my family for their uncondi tional support as I navigated my way through this process and my friends for their certainty that whatever I was doing during the last four years was worthwhile and impressive (still be determined!). Finally, I am grateful to my husband Phillip Morris who supported and continuously encouraged me. His reframing of what seems lik e a disaster into a manageable hiccup his assistance during summer field seasons, and his ability to make me laugh even on the most miserable of days are invaluable I look forward to our next adventure.
6 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 8 LIST OF FI GURES ................................ ................................ ................................ .......... 9 LIST OF ABBREVIATIONS ................................ ................................ ........................... 11 ABSTRACT ................................ ................................ ................................ ................... 12 CHAPTER 1 INTRODUCT ION ................................ ................................ ................................ .... 13 Land Change Science ................................ ................................ ............................. 13 Conservation ................................ ................................ ................................ ........... 14 Remote Sensing ................................ ................................ ................................ ..... 15 Dissertation structure ................................ ................................ .............................. 16 2 REMOTE SENSING WITHIN THE FIELD OF LAND CHANGE SCIENCE: PAST, PRESENT AND FUTURE DIRECTIONS ................................ ..................... 18 Past: Traditional Digital Remote Sensing Techniques in LCS Studies ................... 20 Present: Current Digital Remote Sensing Techniques Used in LCS Research ...... 23 Future: Remote Sensing Techniques Within LCS ................................ ................... 26 Chapter Summary ................................ ................................ ................................ ... 29 3 ILLUSIONS OF EQUITY: AN EXAMINATION OF COMMUNITY BASED NATURAL RESOURCE MANAGEMENT AND INEQUALITY IN AFRICA .............. 41 Community Based Natural Resource Management ................................ ................ 43 Environmental Capital ................................ ................................ ............................. 44 Perceived Benefits for the Community ................................ ................................ .... 46 CBNRM Shortcomin gs ................................ ................................ ............................ 48 The Community ................................ ................................ ................................ 48 Devolution of Power and Management ................................ ............................. 53 Cha pter Summary ................................ ................................ ................................ ... 55 4 QUANTIFYING ECOLOGICAL RESILIENCE IN A SEMI ARID SYSTEM .............. 62 Materials and Methods ................................ ................................ ............................ 66 Study Area ................................ ................................ ................................ ........ 66 Data collection and Analysis ................................ ................................ ............. 67 Precipitation data ................................ ................................ ....................... 67
7 Image collection and processing ................................ ................................ 68 Mean variance analysis ................................ ................................ ............. 70 Persistence analysis ................................ ................................ .................. 71 Results and Discussion ................................ ................................ ........................... 72 Changes in precipitation ................................ ................................ ................... 72 Mean varian ce analysis ................................ ................................ .................... 72 Persistence analysis ................................ ................................ ......................... 77 Chapter Summary ................................ ................................ ................................ ... 78 5 RESI LIENCE IN PRACTICE: RESPONSE OF THE SOUTHERN AFRICAN LANDSCAPE TO POTENTIALLY CATASTROPHIC CLIMATE SHIFTS ................ 91 Methods ................................ ................................ ................................ .................. 97 Hypergeo metric test. ................................ ................................ ........................ 97 Mean variance analysis. ................................ ................................ ................... 98 Persistence analysis. ................................ ................................ ........................ 98 6 AN APPLICATION OF OBJECT BASED CLASSIFICATION AND HIGH RESOLUTION SATELLITE IMAGERY FOR SAVANNA ECOSYSTEM ANALYSIS ................................ ................................ ................................ ............ 104 Methods ................................ ................................ ................................ ................ 111 Study Area ................................ ................................ ................................ ...... 111 Satellite Imagery and Field Data ................................ ................................ .... 113 Tree Crown Identification and Spatial Analysis ................................ .............. 115 Results and Discussion ................................ ................................ ......................... 119 Object Based Classification ................................ ................................ ............ 119 Spatial Distribution of Trees ................................ ................................ ........... 120 Scaling from Field to Landsat TM ................................ ................................ ... 122 Chapter Summary ................................ ................................ ................................ 124 7 CONCLUSIONS ................................ ................................ ................................ ... 134 Research Overview ................................ ................................ ............................... 134 Significance of this Research ................................ ................................ ................ 136 LIST OF REFERENCES ................................ ................................ ............................. 137 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 160
8 LIST OF TABLES Table page 2 1 Digital remote sensing as used in land change science research ..................... 35 2 2 Most common sources of remotely sensed data, as determined from the literature, with common characteristics listed. ................................ .................... 39 3 1 (CBOs). ................................ ................................ ................................ .............. 59 3 2 CBNRM growth in Botswana, 1993 2002. ................................ ......................... 60 3 3 Revenue generated from all CBNRM projects in Botswana, 1993 2002. ........... 60 3 4 Typology of CBNRM participation. ................................ ................................ ...... 61 4 1 Indicator measures. ................................ ................................ ............................ 88 4 2 Imagery attributes. ................................ ................................ .............................. 89 6 1 Density of large trees within each of the two land management types ............ 133
9 LIST OF FIGURES Figure page 2 1 Illustrated examples of land cover changes occurring on the landscape.. .......... 33 2 2 Developments of digital remote sensing within land change science: an expanding envelope of techniq ues. ................................ ................................ .... 34 3 1 Map of Southern Africa with dates of community based natural resource management (CBNRM) initiation. ................................ ................................ ....... 58 3 2 Devoluti on of rights and decision making in community based natural resource management (CBNRM). ................................ ................................ ...... 58 4 1 Study area situated within the proposed Kavango Zambezi Transboundary Conservation Area (KAZA). ................................ ................................ ................ 83 4 2 Merged IGBP classification of the study area, showing the majority of the landscape is classified as savanna. ................................ ................................ .... 83 4 3 Prec ipitation regime classes based on precipitation quintiles calculated using the mean annual rainfall from 1972 2000. ................................ .......................... 84 4 4 Hypothetical relationship between mean variance and vegetation status .......... 84 4 5 Results from the hypergeometric test of the tot al annual precipitation amounts. ................................ ................................ ................................ ............. 85 4 6 Results from the mean variance an alysis for the entire landscape, showing the ch anges in amounts of vegetation and heterogeneity ................................ 85 4 7 Results for mean variance analysis for each land cover class. ........................... 86 4 8 Results for mean variance analysis for each precipitation class ........................ 86 4 9 NDVI persistence figures highlight the spatial patterns in change in NDVI across the entire time period (1970 2009) and for each intermediate time step. ................................ ................................ ................................ .................... 87 5 1 Conceptual diagram of the relationship between mean NDVI / NDVI variance and savanna ecosystem state. ................................ ................................ ......... 100 5 2 Study area depicting the Kavango Zambezi tri basin region. ........................... 100 5 3 Results from the hypergeometric tests for each of the sta tion datas ets .......... 101 5 4 Results for mean variance analysis shown along with sample climate data. .... 101
10 5 5 Results from the analysi s of change in spatial heterogeneity of precipitation. .. 102 5 6 Results from the persistence analysis, where (a) shows directional change and (b) shows the cumulative NDVI change over the time period. ................... 102 5 7 Regions of consistently high, medium, and low mean NDVI values. ................ 103 6 1 Study area showing the larger regional c ontext and the boundaries of the two dominant land management types used in the region. ................................ ..... 128 6 2 Object based classification workflow. ................................ ............................... 128 6 3 Schematic to show the relationships between the actual landscape in terms of tree coverage or crowns ................................ ................................ .............. 129 6 4 Results of the object based classification for the region. ................................ .. 130 6 5 Distribution of trees withi size cohort. ................. 130 6 6 Spatial clustering of trees within the study region. ................................ ............ 131 6 7 Spatial distribut ion of Landsat pixels with greater than four trees per 30 x 30 m cell. ................................ ................................ ................................ ............... 131 6 8 Boxplots of the NDVI values for each tree coverage class (1 5) as determined by proportion tree cover per Landsat TM 30 x30 m pixel. .............. 132
11 LIST OF ABBREVIATION S CBNRM Community based natural resource management CBO Community based organization EVI Enhanced vegetation index LCS Land change science LULCC Land use and land cover change NDVI Normalized difference vegetation index OBC Object based classification TCA Tasseled cap analysis Ts Temperature VI Vegetation index
12 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fu lfillment of the Requirements for the Degree of Doctor of Philosophy ENVIRONMENTAL VARIABILITY AND MANAGEMENT: AN EXAMINATION OF LAND COVER CHANGE AND CONSERVATION IN SOUTHERN AFRICA By Cerian Gibbes May 2011 Chair: Jane Southworth Major: Geography Th e work presented here aims to understand land use land cover change in savanna landscapes. Specifically, the ways in which savanna landscapes are managed and monitored, and addresses factors such as 1) the spatial and temporal patterns of change, 2) the na ture of land cover change occurring, and 3) the underlying processes influencing these. I explore implications of landscape monitoring and management strategies employed in southern Africa, highlight challenges associated with such strategies, and explore the potential of new techniques in an effort to inform the management of savanna landscapes in southern Africa.
13 CHAPTER 1 INTRODUCTION Land Change Science The majority of the terrestrial landscape, as much as 50% of the ice free land surface, has been t ransformed and affected by processes such as land use and climate change (Haberl et al. 2007; Turner et al. 2007 ; Vitousek et al. 1997). Contributing to these global changes and of equal importance are regional and local changes. As such, local and regiona l landscape change and the implications of such change are fundamental to human environment research. Human environment research though dominated by geographers, in its most comprehensive form is relevant to and draws from a number of disciplines by combin ing human, environmental, and remote sensing sciences in an interdisciplinary framework increasingly referred to as Land Change Science (LCS) (Turner et al. 2007; Gutman et al. 2004) LCS s eeks to understand the iterative feedbacks between environmental an d social change by examining the dynamics of land use and land cover change (LULCC) within coupled human environment systems LCS research emphasize s the advancement of 1) observation, monitoring, and modeling of LULCC, 2) understanding of causes, impacts, and consequences of LULCC, 3) assessments of system vulnerability, resilience, or sustainability (Turner et al. 2007; Lambin and Geist 2005). change on surface albedo and hence o n energy exchange and climate was recognized (Lambin and Geist 2006) LULCC analyses of ten rely on long term historical reconstructions, fine resolution spatially explicit studies of landscape change, and garnering an understanding of human environment int eractions associated with LULCC
14 ( Ramankutty and Foley 1999; Kasperson et al. 1995; Dixon1 994 ) .This body of research examine s the relationships and feedbacks between LULCC and biodiversity, land degradation, resource management policies, and social and ecol ogical vulnerability ( Verberg et al. 2009; Castella et al. 2007; Metzger et al. 2006; Pimm and Raven 2000 ). Given the complexities of LULCC processes, remaining challenges to the LCS field include connecting local and regional LULCC narratives, and explor ing LULCC at multiple spatial scales and through multiple perspective lenses (Ramankutty et al. 2006). Conservation Understanding the interactions between LULCC and resource management policies such as protected area implementation or community based natu ral resource management (CBNRM) remains critical challenge to the LCS field (GLP 2005) LULCC is frequently used as a proxy for measuring the effect of resource management policies on social and ecological wellbeing. Much of the research emphasis has been placed on specifically on conservation strategies for example numerous studies iterate the effect of protected area designation on land cover patterns and local land use practices (Nagendra 2008; Hansen and DeFries 2007). Conservation, as a form of resou rce management, has traditionally relied on protected areas to enable landscape management practices aimed at ecological conservation, however, w ithin the last few decades has shift ed towards idea s of co occurrence of nature and human activity (Berkes 2004 ) This shift in the conceptualization of conservation led to current strategies focusing on a need to balanc e human wellbeing and ecosystem conservation as well as a need to intertwine multiple conservation strategies (De Fries et al. 2007 ; Daily and Ell ison 2002 ).
15 Protected area designation is now combined with p articipatory conservation in many parts of the world Participatory conservation (e.g. CBNRM) acknowledge s the cost of conservation to local people and attempts to include local stakeholders in the conservation process (at l east economically). There exists a dearth in the knowledge of the effect of PAs and participatory conservation strategies on social and ecological change. This is an example of one of the areas to which the LCS arena can contr ibute and possibly reduce the likelihood of ad hoc social and ecological outcomes of conservation strategies Remote Sensing LCS research foci ( identified above ) and conservation rely heavily on the use of passive remote sensing to provide a baseline und erstanding of the nature of environmental change occurring Assuming that the environmental parameters of interest can be measured using remotely sensed data, remote sensing offers a way in which to collect standardized repeat measures of environmental sta tus and change (Turner et al. 2003). Such measurements distinguish t he character of environmental change (both type of change and rate of change), the spatial location and spatio temporal variation of environmental change, and associations with parameters possibly driving this change (Southworth 2004 ) Much like LULCC analyses are pivotal to the LCS research agenda, remote sensing is one of the more frequently used tools (data sources and methodologies) for spatio temporal analysis of LULCC and as such the continued advancement of remote sensing analysis remains integral to the progression of LCS and management of natural resources. The diversity and wide availability of remote sensing data sources holds the potential to greatly improve the ways in which we study and understand land change
16 and it relationships to ecological and social change (Turner et al. 2007). This and the challenges of utilizing remote sensing for LCS and conservation are further expanded upon Chapters 2 6 of this dissertation. Dissertat ion structure The remainder of this dissertation centers on the three themes introduced above and is divid ed into five journal articles (C hapters 2 6) which are either published or currently under review (Southworth and Gibbes 2010, Gibbes and Keys 2010, C ui et al. under review, Gibbes et al. under review, Gibbes et al. 2010) followed by a conclusion section. Chapter 2 reviews passive digital remote sensing for the use of LCS and discusses the past, present and future directions for remote sensing applicat ions within the LCS field. This article provides a general overview of the remote sensing methodologies used commonly for land cover change analyses, discusses more novel approaches for monitoring and assessing land cover change and discusses the availabl e technology and main limitations to development in t his field Chapter 3 examines the dual goals of CBNRM as a way to protect the environment and enhance the socio economic equity of communities. This article is based largely on published literature, and addresses the constraints and opportunities for successful CBNRM in Africa, largely focusing on southern Africa which has been one of the early testing grounds for these environmental management strategies Chapter 4 presents an analysis of vegetation chan ge based on framework proposed by Westman and Leary (1986) and employed by Washington Allen et al. (2008) Vegetation state and change is examined through the use of vegetation indices and with reference to e z et a l. 2003; Nicholson et al. 2000) Considering this climate shift a disturbance to the system, the work
17 presented in this Chapter 4 explores the response of vegetation to the climate shift with specific regard to the state of the landscape pre disturbance. C hapter 5 expands upon the analysis presented in C hapter 4 by using a similar analysis framework, but the effect is examined through the use of annual NDVI time series data (1982 2009) The l ong term response of vegetation to this climate shift is explored while simultaneously controlling for the influence of shorter climate cycles. Chapter 6 moves beyond the use of vegetation indices to characterize the landscape and explores the utility of object based classification (OBC) and high resolution imagery for differentiating tree canopies from shrub or grasslands. Additionally this case study explores the possibility of using OBC and high resolution imagery as a scaling tool for linking field ob servations to more commonly used remote sensing measures and data sources.
18 CHAPTER 2 REMOTE SENSING WITHIN THE FIELD OF LAND CHANGE SCIENCE: PAST, PRESENT AND FUTURE DIRECTIONS 1 The impact of humans on the natural environment has increased dramaticall y over the last quarter of a century (Schweik et al. 2003). In response to this increase in human induced change, land change science (LCS) has become a leading research arena for addressing human environment interactions (Gutman et al. 2004). LCS joins th e human, environmental, and geographical information, often using remote sensing sciences, in an interdisciplinary effort to examine changes in land and their implications for global environmental change and sustainability (Turner et al. 2007). LCS frequen tly incorporates land cover change analyses, which assess shifts within a single land cover type (e.g. thinning of forests, Figure 2 1 a) or changes from one land cover type to another (e.g. Deforestation, Figure 2 1 b) (Masek et al. 2008; Mayaux et al. 2007) Land cover change analyses rely most heavily on the use of passive remote sensing systems. Changes in land cover are examined to gain a baseline understanding of how human decisions and actions are affecting the composition and configuration of the envir onment, which in turn may affect ecosystem functioning, biodiversity and climate (Southworth 2004). There are several factors that need to be addressed whi le monitoring land cover change (1) what kin ds of changes are taking place?, where are such changes o ccurring? (2) what are the rates of these changes? (3) what are the factors influencing each of the above? Digital remote sensing is a very attractive source 1 Reprinted with permission from Southworth, J. and Gibbes, C. (2010) Digital remote se nsing within the field of land changce science: Past, present and future directions. Geography Compass 4, pp. 1695 1712.
19 point in time, in a consistent, spatially continuous and repeatable manner. Such remotely sensed data are also available at a range of spatial and temporal scales. The examination of land and environmental changes relies heavily on theoretical and methodologica l approaches from a variety of disciplines, however, in the past two decades there has been an increasing reliance on the use of remotely sensed data and analyses, regardless of discipline. The assumption underlying the use of remote sensing within this fi eld is that key environmental parameters that relate to the human environment interactions of interest can be remotely detected (Turner et al. 2003). Successful examination of environmental changes depends on repetitive collection of data at varying tempor al and spatial scales. This necessity has resulted in remote sensing becoming a useful and practical tool for examining land cover change and its ecological and socioeconomic impacts. Additionally, remote sensing enables ctral range visible to humans and assess energy exchange in other portions of the spectrum (e.g. near infrared and thermal wavelengths). Historically, the most frequently used types of sensors are those of passive remote sensing, which relies on measuring the reflectance and emission of the opposed to active systems, which send out a beam of energy and measure the return signal strength and return rate in order to determine surface properties. The increased number and availability of air and space borne sensors has improved our ability to observe, monitor and characterize the landscape (Turner et al. 2007). Satellite imagery offers repeat data of large areas of terrestrial ea rth, and as such has been recognized as an ideal tool for exploring land change patterns over
20 space and time (Iverson et al. 1989). Remotely sensed data enable assessments of change in the quantity and distribution of land cover types, which can then be li nked to changes in carbon stocks, habitat and resource availability, and biodiversity. The current availability of such data at varying spatial and temporal scales also facilitates cross scale studies that examine the multiscalar temporal and spatial dynam ics of land change. The use of remote sensing within LCS has helped reveal the extensive impact of humans on the landscape (Kepner et al. 2000, Mertens and Lambin 2000; Ochoa Gaona and Gonzalez Espinosa 2000). Analysis of land cover change via the use of s atellite imagery has shown that, between 1990 and 1997, Southeast Asia had the highest deforestation rates (as compared to Africa and Latin America) (Achard et al. 2001); that land cover changes in Africa include erratic variations in land cover conditions linked to variation in climate (Lambin and Ehrlich 1997); and that urban expansion is the greatest current threat to the loss of croplands globally (Doos 2002). Past: Traditional Digital Remote Sensing Techniques in LCS Studies Assessing and monitoring th e dynamics of land change has been heavily dependent on the use of such passive satellite imagery to generate land cover classifications. This is evident by the extensive number of land change publications that utilize land cover classification for analyse s of landscape change (e.g. Peddle et al. 2004; Ruiz Luna and Berlanga Robles 2003; Topp and Mitchell 2003; Jansen and Di Gregorio 2002; Geoghegan et al. 1998) and by the development of a number of global land cover classification products such as GeoCover and ESA land cover products that offer readily available global land classifications. Land cover classifications have offered a means to simplify the landscape and observe drastic changes among land cover classes ( Table 2 1 a). Traditional classifications generally rely on spectral data
21 (Southworth et al. 2004) from satellite imagery and employ a variety of algorithms, from unsupervised statistical techniques through more advanced and user defined supervised maximum likelihood classification (MLC), all of w hich rely on the basic principal of clustering the data into classes representative of the land covers present in the landscape (Xie et al. 2008). Such discrete land cover classifications have been used to generate detailed local and regional land cover tr ajectories (Ramsey et al. 2006; Rignot et al. 1997; Thompson 1996; Torres Vera et al. 2009; Wessels et al. 2004). The use of discrete remote sensing methodologies highlights land cover conversions, and provides neatly categorized landscapes, which may be u seful for management purposes. The need to link land change research to the applied management of landscapes has supported the use of classifications ( Table 2 1 a, Figure 2 2). Furthermore, land cover classifications provide data sources that can be readily used for statistical and mechanistic modeling. For example, Sarkar et al. (2009) used a classification of Landsat TM data as an input data source for dynamic variable modeling of land cover change in the Andes and the Amazon and determined that predicted changes for these regions greatly influenced habitat fragmentation. Likewise, Weng (2002) utilized a land cover classification to model urban expansion in the Zhujing Delta region. To some extent the increase in modeling of land change has resulted in LCS remaining heavily dependent on discrete classifications of land cover, which can be readily imported into models. Although the use of land cover classifications has enabled a basic understanding of the discrete land cover changes, limitations associated wi th this approach include the subjectivity of the use of classifications, the pure pixel assumption, the lack of inclusion
22 of spatial data, and inability to assess within class variability (Southworth et al. 2004). It is important to consider that classific ations are abstractions as they depict a simplified representation of reality (Di Gregorio and Jansen 2000) which is frequently influenced uniqueness of classifications has challenged the development of a global land cover taxonomy and resulted in meta analyses of land cover change that rely on multiple independently defined land cover classifications (Geist and Lambin 2001; Rudel 2008). The other limitations of classifi cations mentioned above are to some degree being addressed through the incorporation of vegetation indices and more advanced classification methodologies ( Table 2 1 a, Fig ure 2 2) which are becoming more common and are further discussed below and within the context of current LCS needs. The use of continuous data products, most commonly vegetation or spectral indices, emerged in response to the inability of land cover classification analyses to explore within class changes, and as a means to objectively link data regarding emitted and reflective energy to the biophysical processes occurring. The relationship between vegetation indices and biomass has been investigated (Carlson and Ripley 1997; Jiang et al. 2006; Wang et al. 2005) and based on an understanding of this relationship, vegetation indices are commonly used to examine the inter annual and intra annual changes in quantities and distributions of green biomass. The most widely used of these indices has to date been the Normalized Difference Vegetation I ndex or NDVI (Gutman et al. 2004), which is based on the principal that vegetation is highly reflective in the near infrared and highly absorptive in the visible red (Xie et al. 2008). Martiny et al. (2006) demonstrate the use of NDVI to explore the relati onship between land cover and
23 climate in semiarid Africa, a relationship that is a current foci of LCS, whereas Daniels et al. (2008) and Southworth et al. (2004) show the use of NDVI for within class change analyses. Although NDVI is a popular vegetation index (VI), there is concern regarding its suitability for all landscapes, particularly those with high biomass where this index is shown to saturate (Huete et al. 2002; Sellers 1985). The Enhanced Vegetation Index (EVI; Huete et al. 1994), temperature (Ts ratios (Goward et al. 1985) and Tasseled Cap Analysis (TCA; Kauth and Thomas 1976), among others, are also used to examine to examine terrestrial biomass changes. EVI has proven useful in regions where NDVI saturates (Huete et al. ios rely on the use of biophysical parameters to asses land change and, as such, enhance basic vegetation indices by incorporating temperature related characteristics of the landscape that play a fundamental role in many biological processes (Petropoulos e t al. 2009; Southworth 2004). TCA is thought to be a useful approach for land change in dryland areas where moisture availability and vegetation are closely linked (Jolly and Running 2004), this is of particular relevance to LCS as landscape changes in dry lands are considered extensive, and relevant to the wellbeing of 250 million people in the developing world (Reynolds et al. 2007). Present: Current Digital Remote Sensing Techniques Used in LCS Research Although conventional methods of supervised and unsu pervised classification of remote sensing imagery have been popular for the past 30 years (Asner et al. 2003; Hay et al. 2005; Table 2 1 a), there has been a shift toward more advanced classification techniques (Blaschke et al. 2000, 2004; Castilla 2003; Ha y et al. 2001, 2005; Hay et al. 2003) in recent years ( Table 2 1 b, Figure 2 2). This shift has been the result of newer technological developments in remote sensing studies and various criticisms of the
24 inadequacies of conventional methods of remote sensin g analyses. The newer technological developments are the increase in commercially available higher resolution image products (Hay et al. 2005; Wulder 1998; Wulder et al. 2004) like Quickbird and IKONOS with spatial resolutions of less than 4.0 meters (Jens en 2005) and finer spectral resolution images (hyperspectral images) like Hyperion and AVIRIS (Jensen 2005). Criticisms of the traditional techniques arose because of the inadequacy of the conventional pixel based classification method to address the spat ial characteristics of a pixel (Wang et al. 2004) and the heavy focus on the spectral information of a single pixel (Shackelford and Davis 2003) as opposed to true geographical objects (Hay et al. 2005) that has been defined by Wang et al. real entity the user intends to single pixel as opposed to image objects results in ineffective land cover extraction when there are similar spectral characteristi cs of different land cover types or the same land cover class has different spectral responses (Wang et al. 2004). The increased availability of satellite imagery holds the potential for improved land surface monitoring and analysis. In the last decade sat ellite imagery has become more widely and freely available ( Table 2 2 ), increasing the options for data used for remote sensing analyses of land cover change. The increased quantity of data is accompanied by an increased range of data types, as each of the image sources have specific though occasionally overlapping spectral, spatial and temporal characteristics (Xie et al. 2008). The free, public release of Landsat data and the increased availability of low cost data sources such as ASTER contribute to the current wealth of remote sensing
25 data. Though Landsat has traditionally been the most commonly used satellite data source, the malfunction of Landsat 7 has resulted in the need for researchers to consider alternative sources of imagery. Many researcher s concur that MODIS, with its freely available data of moderate spatial and high temporal resolution across a large spectral range, and ASTER, which most closely resembles the Landsat products spatially, temporally and spectrally, holds the promise of bein g the most important datasets for LCS research in the coming decades specifically for the study of both regional and global scale land cover characterization, monitoring and prediction (Hayes et al. 2008; Justice and Townshend 2002). Despite the increasi ng choices of data sources, the cost limitation of using satellite imagery still remains a restraint for LCS research ( Table 2 2 ). Many of the high resolution data sources, for example SPOT and IKONOS imagery are not within the cost realm for many LCS rese archers. Recent shifts in remote sensing analyses have included the incorporation of spectral mixture analysis (SMA; Kuemmerle et al. 2008; Myint and Okin 2009), object oriented classifications (Laliberte et al. 2004; Platt and Rapoza 2008), fuzzy classifi cations (Leyk and Zimmermann 2007; Tang et al. 2005) and texture analysis (Hudak and Wessman 2001; Scarpa et al. 2009; Dobrowski et al. 2008; Table 2 1 b, Figure 2 2). Although many of these approaches were developed in the mid 1990s (Lu et al. 2004) the ov erreliance on traditional classifications has dominated the field until the recent need to explore alternative approaches for characterizing the complexity of landscape composition and change. The incorporation of these approaches into the suite of remote sensing methodologies used in LCS research improves our ability to
26 simplicity of land cover characterization. For example, Hudak and Wessman (1998, 2001) recognize the correla tion between image texture and woody stem count and canopy cover and hence the utility of this approach for quantifying bush encroachment in a savanna system. Similarly Sha et al. (2008) use fuzzy classification to differentiate grassland groupings in Inne r Mongolia, a region that is based on a traditional Likewise, Bhaskaran et al. (in press) demonstrate the improved capability of using both spectral and spatial data in an object oriented classification for differentiating among urban features. In addition to relying on a wider suite of remote sensing analyses, land change scientists are also turning to applications that rely on hyperspectral imagery ( Table 2 1 b, Figure 2 2 ). Hyperspectral data include hundreds of spectral bands that are better able to capture within class differences in reflectance and absorption (Watanachaturaporn et al. 2004). Galvao et al. (2005) successfully discriminate among sugarcane varieties in Bra zil using hyperspectral data. These data hold the potential to enable more detailed and complex representations of land change (e.g. the differentiation of types of agricultural fields as opposed to the lumping of all fields into the same category); howeve r, the expansion of land cover change research utilizing these data is predominantly limited by the cost (Xie et al. 2008) of both imagery and associated computing needs for the interpretation of these data products. Future: Remote Sensing Techniques Withi n LCS Much attention has been paid to the issues of human induced land cover change within the last few decades as evidenced by such multidisciplinary, multinational
27 ( http://lcluc.umd.edu/index.php ) or the International Geosphere Biosphere Programmes ( http://www.igbp.kva.se/ ). However, the dynamic nature of the information required for research, management, a nd planning initiatives has not truly been obtained by many of the currently used remote sensing approaches. The emphasis on use of land cover classifications of remotely sensed data for land cover change analyses ( Table 2 1 a) has some serious limitations that can affect interpretation of studies of land cover change (Gutman et al. 2004). By reducing the landscape to broad (and often subjective) classes we trade detail and variation for generality of patterns (Verburg et al. 2002). The subjectivity of land cover classification means that classifications often cannot be standardized and thus study comparisons may not be possible. Furthermore, the reliance on land cover classifications has led to more emphasis on studies that examine land cover conversions rat her than modifications (Lambin and Geist 2006). However, land cover modifications or changes within a land cover class can be as influential on the environment as land cover change (Foody 2001; Lambin and Ehrlich 1997). The future directions of LCS ( Table 2 1 c, Figure 2 2) will likely rely on the incorporation of remote sensing methodologies being used in other fields (much like the development of LCS has relied on the incorporation of ideas and theories from other fields). For example, LIDAR and RADAR (use d in ecology and forestry to characterize vegetation structure) and thermal analyses (used in urban and climate studies) hold potential as remote sensing methodologies, which could further enhance our understanding of land characteristics, and land climate relationships (Gluch et al. 2006; Lecke 1990; Lefsky et al. 1999; Patenaude et al. sensing methodologies such as thermal based and multisensor analyses holds the
28 potential to improve the direct linkage between satelli te and ground data. Enhancing the relationship between what is observed on the ground and how it is characterized from space will potentially improve the inputs used to model land cover change, and thereby advance LCS research ( Table 2 1 c). The use of mult iple spectral datasets is becoming more common, due in large part to the need to examine land cover change at multiple and interacting scales. The merging of multiple datasets is likely to be highly advantageous for LCS analyses and holds the potential to generate analyses that more rigorously address the social actions and decisions impacting the land cover changes at multiple scales. The merging of data from multiple temporal, spatial and spectral scales could lead to improved characterizations of the lan dscape and the identification of the scale at which various drivers are influential. Both land cover change itself and the processes driving change can occur at multiple scales, and thus the incorporation of data with varying temporal, spatial and spectral resolutions ensures that the effect of scale on land change can be accounted for. Stickler and Southworth (2008) intertwine the use of Quickbird and Landsat data to identify the scale at which habitat selection for redtail monkeys is best captured, thereb y relying on multiple data sources with varied spatial and temporal resolution to determine the most appropriate scale for examining the process of habitat selection. Thenkabail et al. (2004) and Soudani et al. (2006) explore inter sensor relationships bet ween IKONOS, Landsat ETM+ and SPOT in an attempt to develop an understanding and interpretation for the relationships across sensors and between ecological variables and spectral indices used to describe the landscape. Such work recognizes not only that la nd changes and processes are scale dependent but also that
29 the multitude of data sources can potentially be combined to increase the continuity of remotely sensed data ( Figure 2 2). Examining temporal change requires the continuation of land observations a nd with the development of satellite technology the relationships across sensors must be explored to determine compatibility of different data sources. For example, Tucker et al. (2005) process AVHRR NDVI within the SPOT and MODIS NDVI dynamic range, enabl ing the maintenance of the advantages of MODIS and SPOT data while retaining historical information. In addition to testing the compatability of various data sources, data fusion techniques are becoming more commonly used to combine multiple data sources f or a single analysis. Data fusion techniques aim to fully exploit the spectral, spatial and temporal information provided by the range of data sources currently available (Pohl and Van Genderen 1998; Table 2 1 c, Figure 2 2). As methodological approaches ar e adapted from other disciplines to LCS, and the use of techniques such as object oriented classification and texture analysis expand, and data sources are combined to maximize spatial, spectral and temporal information ( Table 2 1 c) the need for improved a ccuracy assessments of remote sensing analyses of land cover will continue to be central to the appropriate use of remote sensing. New techniques may demand more creative ways for assessing accuracy and the use of multiple data sources could contribute new forms of testing quantitative and locational accuracy. Accurate accounting of data quality and accuracy for the derived remotely sensed data are a key current and future challenge to this field. Chapter Summary The use of remotely sensed data for land cov er change analyses has a rich history ( Table 2 1 Figure 2 2) but can be seen to have serious limitations if not used in
30 combination with social and ecological data (Gutman et al. 2004). Such integrations strengthen the research and are a necessity to furt her real understanding of this field. As such, remote sensing represents an important data source for land change scientists (Turner 2003). Due to the many limitations of the traditional land cover classification approaches, techniques such as SMA (Schweik and Green 1999), generalized linear models (Morisette et al. 1999), genetic classifiers (Pal et al. 2001) and fuzzy classifiers (Foody and Boyd 1999) have been developed. However, it is also necessary to investigate different techniques of land cover repr esentation beyond variations on classification for land cover modeling purposes. Continuous analyses of remotely sensed data have many advantages. They rely more closely on the original data collected by the satellite and can be linked directly to land sur face processes (Southworth et al. 2004), so producing more realistic depictions of the surface and the surface processes (Gutman et al. 2004). They are more flexible (given that a pixel no longer must belong to an individual cover class but can represent t he true gradient or mosaic) and they are more useful inputs to many earth system models, where they can be used to represent the initial boundary conditions (Gutman et al. 2004). Additionally, the subjectivity associated with the generation of land cover c lasses is not an issue in continuous analyses enabling cross study comparisons. However, accurate characterizations of data product accuracy can become even more challenging, and accuracy of data quality and accuracy assessments will become paramount. With the plethora of new satellite platforms and data availability (especially at reduced cost or free, Table 2 2 ) there is now a wealth of available data sources, with varying spatial and spectral resolutions, which are available to address land change
31 resear ch, as compared to just two decades ago where most research relied solely on a single platform, e.g. the Landsat platform or AVHRR. Finally the integration of data sources across multiple platforms and sensors is being fully investigated, as well as the us e of both passive and active sensors, and combinations of analysis tools. It is through these future research foci that many new developments will occur in terms of remote sensing techniques, specifically within the field of LCS ( Figure 2 2). This combined use of multiple remote sensing techniques (categorical, continuous, hybrid) should be utilized in order to maximize these tools and improve our ability to understand both broad patterns of change and finer variation in land cover. This will enhance the fi eld of LCS, as we develop better techniques that can link more directly to the actual more from the realm of the physical sciences, integrating more of the continuous ou tput from the satellite analyses, such as surface temperatures, evapotranspiration, etc., and to do so across multiple spatial and temporal scales, e.g. with MODIS monthly products available across the last decade. Currently, the limitations of remote sen sing within the field of LCS are the overreliance on land cover classifications, which are often the necessary input to land cover models or fragmentation analyses (Southworth et al. 2004), as well as potentially prohibitive costs associated with many of t he newer products such as AVIRIS, IKONOS and QUICKBIRD ( Table 2 2 ). Likewise the computational requirements to run continuous model formats rather than discrete land cover classes as inputs are significant and can again limit applicability to the researc her. For future developments the field of LCS needs to borrow more remote sensing techniques and tools from other
32 fields (e.g. ecology) and to continue to use the multitude of data products available rather than repeat the overreliance of a single product, as we had in the past. Finally the use of not only high spatial resolution, but also high temporal resolution will help us to better understand the drivers of change, one of the main foci of the LCS field, rather than the decadal slices of land cover that we had relied on in the past (Gutman et al. 2004; Table 2 1 Figure 2 2 ). The future is bright within this field as we develop high resolution, multitemporal and spatial analyses, based on a multitude of image products and requiring a suite of remote sens ing techniques ( Figure 2 2). Such understanding is essential as we build more integrated models of land change as necessitated by the field (Lambin and Geist 2006), better integrating social and biophysical drivers, based in part on multisource data integr field and as such LCS needs to recognize and encourage experimentation with multiple methodologies. Such an approach will permit the use of the plethora of data products and techniques available to researchers, and support the selection of the best tools for each situation on a case by case basis.
33 Figure 2 1. Illustrated examples of land cover changes occurring on the landscape V iewed from photographs, and by satellite s ensors, in terms of both continuous and discrete data types for a savanna landscape in southern Africa, illustrating a thinning of trees or landscape degradation, and a forested landscape in Uganda that has undergone land clearing for agriculture, i.e. def orestation.
34 Figure 2 2. Developments of digital remote sensing within land change science: an expanding envelope of techniques (# relates to examplar articles from Table 2 1).
35 Table 2 1. Digital remote sensing as used in land change science research for (a) traditional techniques, (b) current techniques and (c) new and future techniques. (a) # Technique Paper Author Year 1 Supervised classification Land cover change trajectories in southern Camerooon Mertens and Lambin 2000 2 Unsupervised classificatio n Development of a Land Cover Characteristics Database for the Conterminous United States. Loveland et al. 1991 3 Supervised classification Classifying successional forests using Landsat spectral properties and ecological characteristics in eastern Amazo nia. Vieira et al. 2003 4 Classification A remote sensing GIS evaluation of urban expansion and its impact on surface temperature in the Zhujiang Delta, China Weng 2001 5 Spectral indices: NDVI Global land cover classifications at 8 km spatial resolutio n: the use of training data derived from Landsat imagery in decision tree classifiers. DeFries et al. 1998 6 Classification Parametric land cover and land use classifications tools for environmental change detection Jansen and DiGregorio 2002 7 Classif ication A standard land cover classification scheme for remote sensing applications in South Africa Thompson 1996 (b) # Technique Paper Author Year 8 Sub pixel classification Using genetic algorithms in sub pixel mapping Mertens et al. 2003 9 Object ori ented classification A performance evaluation of a burned area object based classification model when applied to topographically and non topographically corrected TM imagery Mitri and Gitas 2004 10 Texture analysis Image Texture Processing and Data Integ ration for Surface Pattern Discrimination Peddle and Franklin 1991
36 Table 2 1. Continued # Technique Paper Author Year 11 Sub pixel classification On the estimation of spatial spectral mixing with classifier likelihood functions. Schowengerdt 1996 12 De cision tree classification relying on texture data The use of decision tree and multiscale texture for classification of JERS 1 SAR data over tropical forest. Simard et al. 2000 13 Object oriented and multi sensor An integrated approach to land cover cla ssification: an example in the Island of Jersey. Smith and Fuller 2001 14 Decision based classification Monitoring urban land cover change: An expert system approach to land cover classification of semiarid to arid urban centers. Stefanov et al. 2001 15 Image fusion used to enhance resolution Fusion of satellite images of different spatial resolutions: Assessing the quality of resulting images. Wald et al. 1997 16 Object oriented Object based classification of remote sensing data for change detection. W alter 2004 17 Sub pixel Urban land cover change detection through sub pixel imperviousness mapping using remotely sensed data. Yang 2003 18 Object oriented Object oriented image analysis for mapping shrub encroachment from 1937 to 2003 in southern New Me xico. Laliberte et al. 2004 19 Object oriented and SPOT Segment Based Land Use Classification from Spot Satellite Data. Johnsson 1994 20 NDVI and thermal The surface temperature vegetation index space for land cover and land cover change analysis. Lambi n and Ehrlich 1996 21 Non Linear mixing and Neural Networks Non linear mixture modelling without end members using an artificial neural network. Foody et al. 1997 22 Sub pixel classification and fuzzy membership Sub Pixel Land Cover Composition Estimati on Using a Linear Mixture Model and Fuzzy Membership Functions. Foody and Cox 1994 23 Sub pixel classification Ordinal Level Classification of Sub Pixel Tropical Forest Cover. Foody 1994
37 Table 2 1. Continued # Technique Paper Author Year 24 High spatia l resolution Application of multi scale spatial and spectral analysis for predicting primate occurrence and habitat associations in Kibale National Park, Uganda Stickler and Southworth 2009 25 High spatial resolution High spatial resolution remotely sense d data for ecosystem characterization Wulder et al. 2004 26 Spectral Mixture Analysis Modelling land cover types sing multiple endmember spectral mixture analysis in a desert city Myint and Okin 2009 27 Fuzzy classification Improving land change detectio n based on uncertain survey maps using fuzzy sets Leyk and Zimmermann 2007 28 Fuzzy classification Reasonings about changes of land covers with fuzzy settings Tang et al. 2005 29 Texture and High spatial resolution Textural analysis of high resolution im agery to quantify bush encroachment in Madikwe Game Reserve, South Africa, 1955 1996 Hudak and Wessman 2001 (c) # Technique Paper Author Year 30 Thermal data analysis A mono window algorithm for retrieving land surface temperature from Landsat TM data an d its application to the Israel Egypt border region Qin et al. 2001 31 Thermal data analysis Measurement and Analysis of Thermal Energy Responses from Discrete Urban Surfaces Using Remote Sensing Data. Quattrochi and Ridd 1994 32 Thermal data analysis A nalysis of vegetation within a semi arid urban environment using high spatial resolution airborne thermal infrared remote sensing data. Quattrochi and Ridd 1998 33 LiDAR and multi sensor Mapping deforestation and secondary growth in Rondonia, Brazil, usin g imaging radar and thematic mapper data. Rignot et al. 1997 34 Laser Altimeter Remote sensing applications to hydrology: Airborne laser altimeters. Ritchie 1996 35 Multi sensor Multi temporal MODIS Landsat data fusion for relative radiometric normaliza tion, gap filling, and prediction of Landsat data. Roy et al. 2008
38 Table 2 1. Continued # Technique Paper Author Year 36 Object oriented and multi sensor An integrated approach to land cover classification: an example in the Island of Jersey. Smith and Fuller 2001 37 Hyperspectral and radar data fusion Fusion of hyperspectral and radar data using the IHS transformation to enhance urban surface features. Chen et al. 2003 38 Thermal data analysis Land surface temperature retrieval from LANDSAT TM 5. So brino et al. 2004 39 Thermal data analysis Remote sensing of the urban heat island and its changes in Xiamen City of SE China. Xu and Chen 2004 40 LiDAR Estimation of tropical forest structural characteristics using large footprint lidar. Drake et al. 2 002 41 Thermal data analysis Change analysis of land surface temperature based on robust statistics in the estuarine area of Pearl River (China) from 1990 to 2000 by Landsat TM/ETM+ data Zhang 2007 42 NDVI and thermal The surface temperature vegetation i ndex space for land cover and land cover change analysis." Lambin and Ehrlich 1996 43 Thermal and addressing scales Scaling Effect on the Relationship between Landscape Pattern and Land Surface Temperature: A Case Study of Indianapolis, United States. Liu and Weng 2009 44 LiDAR and multi source Multi source land cover classification for forest fire management based on imaging spectrometry and LiDAR data. Koetz et al. 2008 45 LiDAr and Object oriented Object based land cover classification using high post ing density LiDAR data. Im at al. 2008 46 SAR, LiDAR and multi source Joint analysis of SAR, LIDAR and aerial imagery for simultaneous extraction of land cover, DTM and 3D shape of buildings. Gamba and Houshmand 2002 47 Hyperspectral analysis Discriminat ion of sugarcane varieties in southeastern Brazil with EO 1 hyperion data Galvao et al. 2005 48 Continuous temporal analyses Effects of precipitation and soil water potential on drought deciduous phenology in the Kalahari Jolly and Running 2004 49 Contin uous temporal analyses Compared regimes of NDVI and rainfall in semi arid regions of Africa Martiny et al. 2006
39 Table 2 2. Most common sources of remotely sensed data, as determined from the literature, with common characteristics listed. Sensor satelli te Spatial resolution Temporal availability Cost Advantages Limitations Source AVHRR 1.1 km Beginning November 1978 Freely available Cost, spatial extent (global availability) Spatial resolution ht tp://noaasis.noaa. gov/NOAASIS/ml/ avhrr.html MODIS 250 m 1 km Beginning February 2000 Freely available Cost, spatial extent (global availability) Spatial and temporal resolution http://modis land. g sfc.nasa.gov/ Envisat 300 m at NADIR/1200 m global Beginning May 2002 Starts at $32.00 per scene Global availability, multitude of instruments and products available Temporal resolution (recent) http://earth.esa.int/ pub/ESA_DOC/ ENVISAT/ENVI87a.pdf Landsat MSS 68x 83 m commonly resampled to 57 m July 1972 October 1992 Freely available Cost, ease of access, commonly used o longer available http://eros.usgs.gov/#/ Guides/l andsat_mss Landsat TM 30 m MS 120 m TIR Beginning July 1982 Freely available Cost, ease of access, commonly used Data quality degraded http://www. landcover.org/data/ landsat/ Landsat ETM 15 m Pan 30 m MS 60 m TIR Beginning April 1999 Freely availab le Cost, ease of access, commonly used Data problems in filling http://www. landcover.org/data/ landsat/
40 Table 2 2. Continued Sensor satellite Spatial resolution Temporal availability Cost Advantages Limitations Source IK ONOS 2 1 m Pan 4 m MS Beginning January 2000 Starts at $7.00 per km sq Resolution, tasking possible Cost, spatial extent http://www. geoeye.com/ CorpSite/ Quickbird 0.61 m Pan 2.4 m MS Beginning October 2001 Starts at $22.00 per km sq Resolution Cost, spatial extent http://www. geoeye.com/ CorpSite/ ASTER 15 m VNIR 30 m SWIR 90 m TIR Beginning December 1999 Starts at $128.00 per scene Frequency, resolution, thermal ba nds, similar to Landsat continuity Cost, spatial extent http://terra.nasa.gov/ Brochure/Sect_4 2. html SPOT 10 m Pan 20 m MS Beginning May 1986 Starts at $0.30 per km sq Resolution, tasking possible Cost, spatia l extent http://www.spot.com/ RADARSAT 8 100 m Beginning November 1995 Freely available Cost, global availability Interpretation can be complex http://www. radarsat2.info/
41 CHAPTER 3 ILLUSIONS OF EQUITY: AN EXAMINATION OF COMMUNITY BASED NATURAL RESOURCE MANAGEMENT AND INEQUALITY IN AFRICA 2 Human environment geography strives to explain patterns of resource use, land cover conversions, and environmental change while at the sa me time proposing tractable solutions to contemporary problems faced outside of academia (Lambin et al. 2001). This tradition stretches back well beyond the current debates (Sauer 1925; Brookfield 1964; Blaikie and Brookfield 1987; Bassett 1988; Kates 2002 ; Robbins 2004; Tuner and Robbins 2008) yet remains at the core of our discipline and deserves focused attention (Turner 2003). With the rise in sustainability science many academic geographers study how conservation and development programs play out on th e land and in society (Kates 2002). Of these efforts, considerable attention has been paid to effort, community based natural resource management (CBNRM). Community base d natural resource management is an approach to natural resource management that seeks to incorporate the participation of community members and resource users in decision making (Gruber 2010; Soeftestad 2006; Zanetell and Knuth 2004). CBNRM evolved in res ponse to the limitations of top down natural resource strategies and has been introduced and implemented in a number of developing countries (Gruber 2010). CBNRM is used as a mechanism to advance both environmental protection and the socio economic status of local communities (Armitage 2005). In theory CBNRM can devolve power, increase equity and improve resource management (Larson and Ribot 2004; Tacconi 2007). This article examines the theory 2 Reprinted with permission from Gibbes, C., and Keys, E. (2010). The illusion of equity: Community based natural resource management i n southern Africa. Geography Compass 6, pp. 1 15
42 and practice of CBNRM in southern Africa, which has been widely implemented throughout the region (Child 2009). We explore the role of CBNRM strategies in improving social equity, focusing on the limitations associated with this natural resource management strategy and the aspects of its implementation which, if impro ved, could enhance the likelihood of CBNRM programs increasing equity within and among communities. We take a critical approach to the theory and practice of CBNRM in southern Africa. By critical we mean that while the ideals of CBNRM may be laudable, ther e are both pitfalls to its exercise and opportunities to improve its implementation. As defined, CBNRM should integrate ecological sustainability, economic efficiency and social equity (Pagdee et al. 2006). Although occasionally successful at the first two objectives, the enhancement of social equity often fails due to assumptions regarding communities and devolution. The Bruntland Report and other writings led the charge to develop sustainable practices (WCED 1987). Worldwide, individuals, institutions, an d communities strive to protect environments from human disturbance while enhancing welfare (Hellig 1994; Southworth et al. 2006; Zimmerer 2000; Zimmerer and Young 1998). Within academic geography this drive is exemplified by literature that explicitly tri es to link conservation with development, recognizing the coupled nature of the human environment relationship. In various formulations man and the biosphere, sustainable development, and others conservation with development schemes are eyed hopefully at fostering sustainable and equitable nature society interactions (Brower 1990; Coomes and Barham 1997;
43 The article utilizes critical document analysis, literature analysis, and informal key informant intervie ws to ask: (i) what are the perceived benefits and associated assumptions that limit the success of incorporating improvements in social equity with CBNRM? (ii) How does CBNRM implementation and success vary across southern African rural settings? (Figure 3 1). The article first describes in more detail the evolution of CBNRM programs and links to environmental capital, we then discuss the benefits of CBNRM programs, followed by an examination of the shortcomings, and conclude with a discussion of the consi derations needed to improve CBNRM program implementation. Community Based Natural Resource Management Approaches to natural resource management have changed within the last two decades and currently there is a move toward natural resource management that a ddresses the socioeconomic needs of communities and incorporates communities in the management process (Child 2009; Fabricius et al. 2004; Hulme and Murphree 2001; Nelson and Agrawal 2008). Current conservation practices attempt to bridge conservation and management through the use of CBNRM (Dressler and Buscher 2008). CBNRM programs are implemented worldwide and attempt to spur rural economic development while improving natural resource management (Boggs 2000; World Wildlife Fund for Nature 2006). The real ization that communities are an integral part of the ecosystem and that conservation benefits are gained by including communities in resource management processes and decisions leads to the intertwining of concerns about biodiversity and development (Etkin 2002). In an attempt to avoid conflicts of interest between environmental conservation and socio economic development, and in response to the failures of centralized conservation, CBNRM aims
44 to decentralize and enable communities to manage their resources and gain economic benefit (Bradshaw and Bekoff 2001; Burke 2004; Goldman 2003; Indrawan 2010). CBNRM consists of a variety of components, including: participation, ownership and incentives, benefitting transparency and state cooperation. Ideally CBNRM pro jects and policies embody both the process of gaining empowerment and environmental awareness and the result of increased economic and environmental sustainability, thereby representing a holistic approach to conservation with development. These components and other principles of CBNRM are elaborated upon by Murphree (1997) and will be readdressed in later portions of this article. Environmental Capital Capital exists in a variety of forms including physical, human and environmental. It tock that yields a flow of valuable goods and services into the (Costanza and Daly 1992, 38). CBNRM is an approach used to foster sustainable use and development of environmental capital. The idea of environmental capital [sometimes referred to in Harte 1995)] implies that the environment or components of the environment can be welfare. It is important to realize that, as with other forms of capital, environmental capital has the potential to improve economic welfare of whole societies, but does not necessarily alter the level of inequality in society. Environmental capital may be divided into renewable and non renewable capital. Most often CBNRM addresses the management o f renewable environmental capital, which includes elements such as vegetation, wildlife, and ecosystem services. In regions where ecological limitations (e.g. precipitation, soil quality) restrict the gains for traditional economic activities, such as live stock and field agriculture, the wise
45 use of environmental capital is of particular importance. Many regions within Africa possess such ecological limitations as highly variable climates constrain agricultural incomes. Within this agriculturally limited co ntext are a wide variety of charismatic, endemic species. Many of these species are megafauna prized by foreigners for hunting and viewing, thus it seems appropriate that in an attempt to maximize economic benefits from the landscape, African governments s hould actively promote and manage their wildlife. This idea is supported by Child (1989) who discusses the increased economic benefits from utilization of land for wildlife rather than for ranching. Upon recognition of the value of their renewable natural resources many African countries have incorporated CBNRM to manage their wildlife (Lindsey et al. 2009). One of the earliest and touted as the most successful examples of CBNRM is the Communal Areas Management Programme for Indigenous Resources (CAMPFIRE). CAMPFIRE resulted from the recognition that wildlife utilization is a highly competitive form of land use in the semi arid regions of southern Africa (Jansen et al. 1992; PwC 2001) and the idea that rights and benefits ought to be devolved to local commun ities (Taylor 2009). Redirecting wildlife to the realm of economics rather than solely for conservation provided an opportunity for communities and households to attain direct economic benefit (see Table 3 1 ) from the presence and management of wildlife (J ones and Murphree 2001; Taylor 2009). CBNRM is multidimensional and in addition to increasing the sustainability of ecosystems, includes equity of power, rights, and benefit sharing from resources (Agrawal and Ribot 1999; Dressler et al. 2006). Despite be ing one of the central goals of CBNRM, and being deemed as instrumental for ensuring the sustainability of CBNRM programs, the importance of the
46 equity component of CBNRM is drastically overshadowed by the overwhelming concern with ecological sustainabilit y. The issue of equity has been relatively neglected (Cochran et al. 2009). For the purpose of this article, equity will be discussed at and below the community level of organization incorporating households and individuals where appropriate. We focus on e quity because the nature of equity influences growth and development (Bebbington et al. 2008), and as such is not only one of the central goals of CBNRM, but also affects development another goal of CBNRM. Equity is considered from a prioritarianism poin t of view, which is based on the weak equity axiom (Sen 1973). The weak equity axiom is concerned with distribution at the level of the individual as opposed to aggregate welfare. This stance emphasizes the relative nature of equity, where the worst off bo dy (be it a group or individual within a community) is determined based on a comparison of that body to the state of others. Priority in terms of increasing rights, access or benefits is then given to the body which is deemed worst off. Therefore, if CBNRM increases equity it should increase the power, rights and benefits to the weakest institution (community, private sector, government, NGOs) within natural resource management and to the weakest group or individual within the community without reducing the wellbeing of any other institution or individual. Alternatively, to increase equity within a group (of communities or individuals) the power of the strongest could be reduced to be comparable to that of the rest of the group. Perceived Benefits for the Co mmunity base (Getz et al. 2002; Turner 2004). CBNRM in Africa most frequently links wildlife a nd
47 economic gain. Generally, communities are given wildlife quotas which they then decide how to utilize. Our conversations with tour operators and government agents indicated that animals may be hunted locally by community members, killed if they are prob lem animals (not part of the quota) or sold to tour operators for hunting purposes. Additionally, some communities sell photography rights to tour operators, allowing the operator to bring tourists onto communal land and take photographs, but forbid the us e of animals for game hunting. Case studies suggest that in some situations communities have managed to make money from the wise management and sale of wildlife (Child 1997; Emerton 1999; Boggs 2000; Fabricius et al. 2004; Madzwamuse and Fabricius 2004). I n situations where money is made from the selling of wildlife for hunting or photographic purposes, benefits to the communities are generally realized through the creation and implementation of community projects. These projects take many forms but popular choices include school or health center construction or the boring of wells. In principle, profits, entitlements, and responsibilities should be equitably distributed and guaranteed among community members (Pagdee et al. 2006). Other than economic gains, CBNRM attempts to devolve power to communities (Murphree 2005). This management approach should return ownership of resources and decisions regarding resources to those immediately and most directly affected by the use of these resources. Decentralization of responsibilities and power aims for more equitable and efficient forms of management of resources. This approach considers resource management which integrates multiple stakeholders equally more efficient than centralized management which is thought to be inefficient and infringe upon
48 returns rights of proprietorship to communities; while not privatizing resources, since the rights are given to the community and not to indiv iduals. Implementation of CBNRM has taken different shapes in Southern Africa with spatial and organizational structures limiting cross national comparisons. While summary statistics are unavailable, CBNRM experiments have been met with some success. Table s 3 2 and 3 3 outline the experiences of CBNRM in Botswana from 1993 until 2002. Table 3 2 shows that the cumulative number of villages under CBNRM rose from five in 1993 to 120 in 2002. By o only three districts 9 years before. Income from CBNRM in Botswana (Table 3 3 ) also steadily rose between 1993 and 2002 from US $3612 to US $1,271,725 in 2002. Similar data were not available for other countries in the region. While these achievements po int to the possibilities of increased participation and profit from entering into CBNRM there are also difficulties associated with CBNRM. CBNRM Shortcomings The Community Although the possible benefits for the community from CBNRM are manifold, it is unli kely that these benefits are received and equally distributed. Unequal distribution stems from assumptions often associated with CBNRM implementation. One of the first assumptions is that the community is a cohesive well defined unit. Communities are not e asily nor consistently defined, and although recognized as one of the most enduring concepts in social science is also one of the most vaguely defined (Agrawal 1995; Murphree 1999). Communities are highly complex and pluralistic entities (Nightingale 2003) in which power and authority are constantly changing and are influenced by group affiliations, socio political identities and external factors (Leach et al. 1999;
49 Poteete and Ribot 2009). Fabricius et al. (2001) argue that communities in Africa are nebul ous and fluid and often a figment of the imagination of project managers and donors seeking targets for project implementation. Lack of a defined community, resulting from the complexity and fluidity of communities, can lead to conflict over resources and decreased effectiveness of CBNRM. Such conflict resulted within the Makuleke peoples in South Africa. Divisions within the leadership led to conflict over the distribution of monetary benefits derived from the resources in their management area (Fabricius et al. 2001). Without appropriately identifying the community and creating strategies to manage change within the definition of the community CBNRM is unlikely to enable conservation and development. The supposition that economic and social benefits for th e community reach all of its members rests on the second assumption regarding communities. This is that communities, once defined, are homogeneous. However, it has been shown that communities are inherently heterogeneous (Leach et al. 1999; Katz 2004). Eve n within a cohesive community, heterogeneity influences power structures, participation, decision making and benefit distribution. The vision of small integrated communities using locally evolved norms and rules to manage resources fails to attend to diffe rences within communities (Agrawal and Gibson 1999). Differences within the community may be based on wealth, age, origin, gender and other aspects of social identity (Leach 1999). Feminist literature has demonstrated how power and authority can be distrib uted based on gender, often leading to inequitable distribution of resources (Goetz and Gupta 1996; Rocheleau and Ross 1995; Rocheleau and Thomas Slayter 1995;
50 Rocheleau et al. 1997). Unequal power and authority has also been recognized when examining soci al interactions of groups with varying ethnic backgrounds (Grinker 1994). Such differences can influence which members of the community are actively involved in the management of natural resources. Power structures that exist within the community prior to the arrival of CBNRM may lead to chiefs or spiritual leaders being overly influential in the management of resources. Or conversely, and more commonly, traditional power structures may be challenged by the new approach to resource management that often pro motes the democratic election of committee members responsible for implementing management decisions. This redistribution of power can cause conflict and these social tensions limit the success of CBNRM programs (Fabricius et al. 2004; Alesina and Ferrara 2002). Pre existing heterogeneity may also lead to inequalities wherein participation may be limited to a specific sub group within the community. For instance, in our visits we noticed that in Botswana women do not generally speak at the traditional commu nity meetings (locally known as kgotla). The evidence of socially stratified distribution of resources challenges the notion that women or other disempowered groups are equitably represented in community meetings. Participation comes in a variety of forms (Table 3 4), ranging from that which is externally driven to that which is internally initiated. Hypothetically CBNRM enables functional participation for community members. That is, community member participation should incorporate multiple perspectives a nd individuals should have a stake in management of resources. However, heterogeneity within the community may limit participation possibilities for certain groups. Not only are
51 certain individuals limited in participation at the community meeting scale, b ut also at the council and committee scale. As discussed above, committees of councils are elected (ideally democratically) to ensure that resource management decisions are implemented and to distribute benefits to the community. During our research we not ed that with CBNRM programs in Southern Africa, participation in committees and councils is frequently dominated by older males, suggesting that participation at this scale is skewed toward certain groups within the community. Heterogeneity reflected in po wer structures and participation in CBNRM programs also affects distribution of benefits. Inequity in power and participation often lead to inequity in distribution of benefits (Adhikari 2005). Elite control of resource management practices has occurred gl obally and is often accompanied by corruption which enables the elite group to maintain power and benefits (Iversen et al. 2006; Kummer and Turner 1994). In Southern African CBNRM programs, money gained from natural resource management is often captured by influential people within the community restraining the equitable return of money to community members (Child 2004). The idea of equitable distribution of benefits neglects the existence of power structures and participation which are likely to encourage corruption. A component of CBNRM which is rarely sufficiently implemented the intra community sharing of management and decisions over resources is illusory and as a result benefit sharing is likely to be biased without policies to address community hetero geneity and monitor the distribution of benefits (Ribot 1999). The monitoring of benefit distribution itself can become problematic if done internally, since this would give power to those community members who are responsible for monitoring the distributi on of benefits. The power associated with
52 monitoring the distribution of benefits can result in conflict between traditional authorities and those responsible for benefit distribution monitoring and such conflicts occur even in relatively cohesive communit ies with formal legal rules for determining benefit distribution (Koch 2004). Another issue that contributes to the possibility of CBNRM increasing inequality of proprietorship and benefits of resource management is the lack of consideration regarding live lihoods of individuals within the community. As discussed by Bebbington (1999), Ellis and Mdoe (2003), and Katz (2004) livelihood strategies vary among individuals. Resource management is time consuming and can be high in real or opportunity costs, in part icular for poorer members of communities. Frequently the poorer members of society are less able to make decisions based on long term wellbeing and must rather consider immediate costs of changing from one form of revenue to another (Downing 1991; Wood 200 3). CBNRM affects different socio economic strata in communities differentially. In many communities, there are poorer people who lightly participate in CBNRM and receive few benefits but are subject to the problems associated with greater wildlife numbers such as house relocation and crop damage. The need to consider potential revenue loss from shifting land use practices and changes in resource access, (such shifts may include converting agricultural fields to wildlife zones, a common change associated wi th CBNRM in southern Africa) affects individuals within the community differently. Generally these changes in resource management practices have greater effects on community members who have lower levels of physical, human, and social capital, and are thus less likely to be able to afford to shift from current land and resource practices to those which are deemed acceptable
53 for CBNRM (Shackleton et al. 2008). In fact, if CBNRM programs were to really focus on increasing equity, the overemphasis on conservat ion would not exist since conservation is a rural livelihood strategy, and often marginalized populations Africa rely on non rural livelihood diversification strategies, to minimize risk (Ellis 1998). Additionally, changes in land use and resource access a re often imposed upon all community members without regard for the varying abilities to economically accommodate such changes. This portends that the impacts of the implementation of CBNRM on livelihoods within the communities vary by individual, a fact wh ich is often overlooked and can result in increased inequality as certain individuals benefit from the revenue generating opportunities that CBNRM offers, while others do not. Devolution of Power and Management A founding principle of CBNRM is the devoluti on of power and the institution of rights regarding natural resources. Devolution of power related to environmental management is occurring throughout Africa, however, it is haphazardly and incompletely implemented (World Bank 2000). CBNRM is thought to re turn ownership of 1973), promotes linking costs and benefits at the same scale. Thus, if communities manage the wildlife and are impacted by the costs of these actions, t hey should receive benefits. Ideally, benefit sharing and cost enduring should be proportional (Murphree 2005). If a community bears the majority of the responsibility for resource management then it should receive the greatest portion of benefits from the conservation of the wildlife. Current forms of CBNRM in Africa, however, consist mostly of governments setting quotas (number of animals assigned for the community to use), communities selling quota animals to tour operators and tour operators selling the photography rights
54 or hunting rights to international tourists. Although communities can benefit some from this process the devolution of power and rights is incomplete in most cases. The community is left in a disadvantaged position as a receiver of quot devolved to communities, the ultimate goal of conserving wildlife and utilizing it for hunting and tourism purposes is predetermined by governments an involved. Effectively, this results in local communities managing wildlife for the income derived from tourism. The community is often the body responsible for recording animal population counts, bearing the burden of destroyed agricultural fi elds from problem animals and having access to resources limited. Few cases (e.g. CAMPFIRE, Botswana) of CBNRM have managed to integrate training for communities to set quotas internally and be acknowledged and considered by the government bodies responsib le for wildlife. Governments are hesitant to relinquish the control of decision making power to communities (Batterbury and Fernando 2006) because often they do not consider communities to be sufficiently skilled to determine quotas. Failure to sufficientl y transfer decision making powers results in decentralization being a pretense (Ribot 2004). As a result, communities often do not reach the end of the resource management spectrum where there is a complete transfer of authority and responsibility (Figure 3 2). Southern African communities benefit from the sales of animals to tour operators, but to a much smaller degree than the benefits received by the tour operator from the tourist. This is where the disconnect between costs and benefits leaves the commun ity in a disadvantaged position. Again, CAMPFIRE is one of the few programs that have
55 operators to provide the best possible package including both monetary and nonmonet ary benefits for the community. This is rare in CBNRM due to the historically imposed power structures and heterogeneity in power distribution related to race relationship in much of Africa although may represent a road map to more successful CBNRM elsewhe re (Child 2006). Jones (1999) discusses such challenges associated with operator community links as they pertain to the implementation of CBNRM in Namibia. Tour operators tend to be white Africans, this can lead to tension, insufficient communication of ex pectations, and a lack of trust between operators and communities (Child 2006), therefore limiting successful negotiations. Additionally, there is often a lack of information within the community about the value of their wildlife. Without sufficient knowle dge regarding the market for their commodity, communities are inevitably lesser stakeholders in the natural resource management process and unable to benefit sufficiently from resource management. Chapter Summary The need for sustainable management of spec ies and ecosystem services is undeniable. This article does not argue against adaptive management practices that incorporate local knowledge and skills, but instead questions the touted benefits of CBNRM as currently implemented throughout much of Southern Africa. In particular, the ideal of increased equity of decision power, rights and benefits both for communities and within communities is challenged. The concern is that the increased equity of communities as a social unit within resource management stra tegies that include CBNRM is a misconception. In fact, communities are too frequently quota and price takers, with limited negotiating power. Furthermore, the claim that CBNRM promotes equity (from the prioritarian perspective) does not adequately address equity within the
56 community and disregards preexisting community inequalities related to power, social norms, capital, and vulnerability. Considering increased equity as the increase of power, rights, and benefits to the worst off institution (community, p rivate sector, government, NGOs) within natural resource management and to the weakest individual in the community, it does not appear that CBNRM is successfully or consistently increasing equity at either scale (Berry 1997; Cleaver and Toner 2006). This i s not to say that CBNRM is not useful for development and growth of societies, but that implementation of CBNRM must give greater consideration to social structures at both the community and intra community scales to ensure that growth is accompanied by de creased inequality. Policy related to CBNRM programs needs to require high levels of transparency and accountability for all parties involved in the maximization of resources (communities, governments, private sector and NGOs), to limit the continuance and enhancement of inequality at the community and intra community scale. This seems like an intricate task considering the variability across and within societies; however, as with ecological elements of CBNRM such detailed case by case examination is requir ed if CBNRM is to be successful and sustainable. As geographers continue the search for sustainable solutions to human environment relationships they engage with conservation with development efforts like uman environment relationships (Turner 1997, 2003) positions it well to assess the implications of integrated conservation development schemes due in part to tradition but more importantly to the science. To maintain relevance in the application of conservation with development programs academic
57 Geography should continue to encourage studies that assess how the goals of CBNRM like projects are met and provide analysis of what causes the relative s uccess of failure of these efforts.
58 Figure 3 1. Map of Southern Africa with dates of community based natural resource management (CBNRM) initiation. Figure 3 2. Devolution of rights and decision making in community based natural resource management (C BNRM).
59 Table 3 1. Community Based Organizations ( CBOs ) [ Adapted from Wirbel aeur et al. (2007) A preliminary assessment of the natural resources management capacity of community based organizations in Southern Af rica. Cases from Botswana, Mozambique, Namibia, Zambia, and Zimbabwe. WWF SARPO Occasional Paper 17. Harare, Zimbabwe: WWF SARPO. ] CBO Year CBNRM Granted Surface Area (sq km) Population Households 2003 Earnings ($US) Earnings per household Earning per sq km Gonono Ward 4 1988 798 4823 1020 $8,764.00 $8.59 $10.98 Malipati Ward 14 1991 nd nd nd $11,269.00 nd nd Nebiri Ward 7 1988 302 1575 329 $29,805.00 $90.59 $98.69 Tsholosho Ward 7 1991 1319 5063 912 $28,400.00 $31.14 $21.53
60 Table 3 2. CBNR M growth in Botswana, 1993 2002. [ Adapted from Wirbel aeur et al.(2007) A preliminary assessment of the natural resources management capacity of community based organizations in Southern Africa. Cases from Botswana, Mozambique, Namibia, Zambia, and Zimbab we. WWF SARPO Occasional Paper 17. Harare, Zimbabwe: WWF SARPO.] Hierarchical organizations 1993 1995 1997 1999 2001 2002 Districts involved in CBNRM 2 3 6 8 8 9 Community Based Organizations involved in CBNRM 2 6 19 45 61 83 Community Based Organizatio ns registered 1 4 10 26 46 67 Villages in CBNRM (Cumulative 5 12 30 91 99 120 Table 3 3. Revenue generated from all CBNRM projects in Botswana, 1993 2002 [ Adapted from Wirbel aeur et al.(2007) A preliminary assessment of the natural resources manageme nt capacity of community based organizations in Southern Africa. Cases from Botswana, Mozambique, Namibia, Zambia, and Zimbabwe. WWF SARPO Occasional Paper 17. Harare, Zimbabwe: WWF SARPO.] Year Revenue (in Botswana Pula) In 2009 $US 1993 BWP 24,000.00 $ 3,612 1997 BWP 1,410,000.00 $ 212,205 1999 BWP 2,270,000.00 $ 341,635 2001 BWP 6,420,000.00 $966,210 2002 BWP 8,450,000.00 $ 1,271,725
61 Table 3 4. Typology of CBNRM participation [ Adapted from Pretty, J., Guijt, I., Scoones, I. and Thompson, J. (1 994). A Environm ent and Development, London, UK .] Type of Participation Passive participation externally initiated Participation in information giv ing Participation by consultation Participation for material incentives Functional participation Interactive participation Self mobalization internally initiated
62 CHAPTER 4 QUANTIFYING ECOLOGICAL RESILIENCE IN A SEMI ARID SYSTEM Semi arid sa vanna ecosystems are experiencing changes in structure, function, and composition driven by a combination of natural and anthropogenic factors (Galvin and Reid 2010). With global climate change predictions indicating that these regions will experience incr eased frequency and intensity of extreme events (in particular droughts), there is a high likelihood of drastic vegetation change likely accompanied by decreases in biodiversity. Such changes would lead to modifications in resource availability for populat ions living within semi arid savannas. Therefore concerns about land degradation are inherently intertwined with concerns about human well being and poverty. The socio ecological implications of possible degradation in semi arid regions necessitate an impr oved understanding of the ecological resilience of these regions. Savanna landscapes occupy the continuum between forest and grasslands, consisting of a continuous layer of grasses interrupted by trees and shrubs (Hill et al. 2010). They include landscapes with varying proportions of grasses, trees and shrubs, and occupy regions of highly variable rainfall, limiting the development of a closed canopy (Scholes and Archer 1997).The inherent heterogeneity of savannas is partially maintained by complex interact ions amongst biophysical and human caused drivers of change. While water availability and soil nutrients are the primary determinants of savanna composition (Scholes and Archer 1997), they are coupled with the local factors of fire, herbivory, and anthropo genic land use to create and maintain spatial heterogeneity within savanna landscapes (Peters and Havstad 2006, Scholes and Walker 1993). Changes in savanna composition are occurring globally as a result of
63 variations in the interactions of these local dri vers of landscape change and changes in climate (Sankaran et al. 2005). Climate change is expected to exert significant influence on savanna vegetation, as spatial and temporal patterns of precipitation are strongly correlated with vegetation composition ( Chamaille Jammes et al. 2006, Vanacker et al. 2005, Richard and Poccard 1998, Fuller and Prince 1996, Townshend and Justice 1986). Changes in global ocean atmosphere interactions and resultant precipitation patterns have been observed in the mid to late 19 et al. 2003, Nicholson 2000), potentially initiated by differences in circulation in the North Pacific (Hare and Mantua 2000). Over the savanna regions of southern Africa this shift has been associated with increasing air temperatures and prol onged dry periods (Batisani and Yarnal 2010, Nicholson 2000, Nicholson et al. 2001) often in association with warm phases of ENSO. This increased connectivity with the major cause of global climate variability is postulated to be the result of increased in fluence of the ENSO phenomenon on the Indian Ocean (Reason and Mulenga 1999, Shi et al. 2007), a principal source of moisture to the study area (Mason and Jury 1997). Partially driven by climate, southern African savannas are thought to be currently experi encing dryland degradation, frequently associated with changes in floral and faunal biodiversity (Chown 2010, Magadza 1994). Such degradation can be quantified in terms of vegetation attributes, including (though not limited to) plant cover, heterogeneity, and productivity (Washington Allen et al. 2004, 2008). Degradation in the southern African context is defined by shifts in vegetation composition, specifically, shifts towards shrub dominated landscapes with increases in bare soil, and decreases
64 in tree c over and perennial grasses (Goheen et al. 2007, Ringrose et al. 1990). Despite significant research, the causes of degradation in southern Africa are still uncertain, though specific focus has been placed on increased grazing pressure (Skarpe 1990), and ch anges in climate (Sankaran et al. 2005). Long term monitoring of savannas is needed to enhance the understanding of broad scale changes in vegetation and the possible relationship between such changes and ecosystem resilience. Ecosystem resilience, defin ed as the capacity of a system to absorb disturbances and maintain essential structures, processes, and feedbacks (Adger et al. 2005), can be difficult to quantify because it is a condition relative to an earlier state (Carpenter et al. 2001). However, exa mining ecosystem characteristics which serve as indicators of degradation across large spatial and temporal scales is one way by which the effect of disturbances on landscapes may be considered. Ecosystem indicators such as vegetation status can be analyz ed using spatially aggregate and/or spatially explicit methods to measure resilience (Washington Allen et al. 2008). with regard to ecological resilience: elasticity, ampl itude, malleability, and damping (Table 4 1). Each of these measures the state of the indicator (i.e. vegetation) in comparison to a reference state, and considers the capacity of the system to absorb disturbances, while retaining structure and function. The expression of each measure is dependent upon the ecosystem characteristic being observed and thus will differs by characteristic, but allows for comparison of any one of them across space and time rely largely on field data and a simulation model, however, similar approaches have been implemented utilizing
65 remotely sensed vegetation data (Simoniello et al. 2008, Washington Allen et al. 2008). The quantity and heterogeneity of vegetation within Afri can savannas are ecosystem characteristics which reflect the state of the system and can be quantified over large spatial and temporal extents using remote sensing. Vegetation indices offer ideal measures of amounts of vegetation and variations in vegetati on distribution both over space and in terms of vegetation type, and the relationship between vegetation indices and biomass is well established (Carlson and Ripley 1997). The Normalized Difference Vegetation Index (NDVI) is particularly useful for measuri ng amounts of photosynthesizing biomass in semi arid and savanna regions (Martiny et al. 2006), where is does not saturate (Richard & Poccard 1998). While there are potential limitations to the use of NDVI in this region due to soil type and exposure (Nic holson and Farrar 1994), the time frame for this study, 1975 onwards, and the multiple platforms used (Landsat MSS and Landsat TM) from which NDVI may be obtained, make the index an ideal beginning point which links well with previous regional research (Ri chard and Poccard 1998, Scanlon et al. 2005, Stige et al. 2006). This study utilizes remotely sensed data, and a framework proposed by Westman Allen et al. (2008), to examine the ecological resilience of south ern African savannas. The state of the vegetation as measured through a vegetation index is determined in order to explore temporal and spatial changes in the distribution of vegetation across the landscape following disturbances to the system resulting fr (Chavez et al. 2003, Nicholson et al. 2001), and the response of vegetation is examined with reference to the state of the pre disturbance landscape. First mean annual
66 precipitation from 1950 2008 are analyzed to identify the effect of the global climate shift on local precipitation patterns. The following questions are then asked; 1) What are the long term vegetation trends post disturbance? 2) Does the system return to a similar state with regard to vegetatio n amount and variance? and 3) How does the vegetation trend vary spatially? Materials and M ethods Study Area The study area (~ 80728 km2) is located in southern Africa including portions of Angola, Zambia, Botswana, and Namibia (Figure 4 1). The study ar ea encompasses much of the proposed Kavango Zambezi Transboundary Conservation Area (KAZA), which aims to link wildlife conservation and development efforts across the four countries and Zimbabwe, and as such, is of significant ecological and social import ance to southern Africa. Community and national economies are heavily dependent upon wildlife tourism and therefore changes in vegetation which may directly impact wildlife species distributions are not only of ecological concern but also of economic impor t. This region straddles the downstream portions of 3 major regional water courses, the ins have remained unconnected, with the waters of the Okavango and Kwando flowing into closed basins. The climate is tropical savanna with three seasons: a hot dry season (August October), a hot wet season (November April) and a cool dry season (May July). Annual precipitation ranges from 400 2200 mm/yr and is characterized by high temporal interannual variability of precipitation, with totals increasing and the coefficient of variability decreasing towards the northern portion of the study area (Gaughan an d
67 Waylen under review). Soils are primarily Kalahari sand, which is nutrient poor. The vegetation is dominated by a variety of annual and perennial grasses, dominant woodlands (ex. Colophospermum mopane woodlands), mixed woodlands, and mixed shrublands. Da ta collection and Analysis Precipitation data Station data from Maun and Shakawe Botswana for 1950 200 7/2008 are examined in terms of their precipitation patterns pre and post the documented global climate shift of the1970s. The se station s possess some of the few long term, reliable, precipitation records within the study area. The data are sub divided into two time periods, 1950 197 2 and 19 79 2008, leaving a nine year buffer surrounding the putative climate shift, thereby avoiding the possibility of an inaccurate identification of the shift, or the inclusion of data from any transitional period between the two periods and regimes. The buffer period was identified using the guidance of the literature ( Chavez et al. 2003, Nicholson 2000 ) and by conducting a cumulative sum (CUSUM) analysis on the precipitation time series thereby identifying where the shift occurred within each of the datasets. Given the considerable interannual variability of precipitation and its highly skewed nature in semi arid regions, standard parametric tests of changes in means and variances are inappropriate. Instead, the entire historic record, 1950 2008, is divided into terciles. If there are no systematic changes in rainfall properties, any sub sample of that population (such as 1950 70 or 1980 2008) should contain roughly equal proportions of observations from each tercile class. Under such an hypothesized random sampling scheme the probability of observing any number of observations from one particular tercile is determined by the hypergeometric probability distribution in
68 which N represents the total population (1950 2008 = 59), the size of the sub s ample (i.e. 1950 70, n = 21, or 1980 2008, n=29), and the number of the total population possessing the characteristic of interes t (e.g. number of observations in the upper tercile or those in the lower, k = approximately 19). The actual number of observations in a particular tercile during one of these two periods can then be compared to the probability of such an outcome under th e null hypothesis of randomness, and the null rejected, or not, accordingly (Martineu et al. 1999). Thus, in the pre 1970 period one would expect to see approximately one third (~7) of the observations characterized as upper tercile and one t lower tercile, and between 1980 2008 there should be nine of each, both with known probabilities of other numbers of observations distributed about them. Image collection and processing A multi temporal and multi spectral dataset was used for this study (Table 4 2). Fifteen Thematic Mapper (TM) images and seven Landsat Multispectral Scanner (MSS) images that covered the study area were acquired, all with less than 10% cloud cover, yielding a dataset that spans the time period 1973 2009. Dry season imagery was acquired to ensure limited cloud cover, for seasonal consistency, and to monitor the more persistent vegetation cover of perennials and to avoid shorter term variation from annuals and other sources of ephemera l vegetation cover (e.g., greening up following a rain event, rather than longer term variation as a result of longer temporal changes as noted by Archibald and Scholes in 2007 and Townshend and Justice in 1986). Landsat TM and MSS compatibility was ensure d using direct matching of red and near infrared bands across the two platforms (Jensen 2005). All image analysis was conducted using ERDAS Imagine 9.2.
69 The images were georectified using a nearest neighbor resampling algorithm (RMSE of less than 0.5 for a ll images) and projected to UTM WGS 84 Zone 34 South. Image calibration was completed to correct for sensor and atmospheric variation using CIPEC methodology (Green et al. 2005) and Landsat MSS and TM post launch calibration gains and biases from tables an d formula provided by Markham and Barker (1986). The image footprints were mosaiced to produce a continuous dataset for each date (7 scenes for Landsat MSS and 5 scenes each for Landsat TM). The Normalized Difference Vegetation Index (NDVI) was then derive d for each mosaic date. This index is used widely, allowing the findings to be interpreted easily and compared to other land cover studies across this region and to other semi arid environments globally (Guerschman et al. 2009, Scanlon et al. 2005, Schmidt and Karnieli 2000). In addition to the Landsat data, three tiles of the MCD12Q1 land cover (MODIS Terra + Aqua land cover type yearly L3 global 500m sin grid) product of 2008 were downloaded. The land cover data was reprojected to a Universal Transverse Mercator (UTM) WGS84 projection (Zone 34 South) and clipped to study area. The primary land cover scheme defined by the International Geosphere Biosphere Programme (IGBP) was used to examine vegetation trends within pre specified land covers. The IGBP clas sification product is a broad classification which is based on the canopy component philosophy presented by Running et al. (1994). The eleven land cover types in IGBP, were merged to seven classes (barren/barely vegetated, croplands, wetlands, grasslands, savannas, shrublands, and forest) according to field survey data (Figure 4 2).
70 NOAA Global Precipitation Climatology Project (GPCP) version 2 monthly global precipitation datum was acquired to cover the study area. This datum is a monthly analysis of surfa ce precipitation estimated from low orbit satellite microwave data, geosynchronous orbit satellite infrared data, and surface rain gauge observations, and with a resolution of 2.5o latitude x 2.5o longitude (Adler et al. 2003). The gridded datum contained monthly rainfall values, which were summed to provide rainfall totals for the water year. This datum was subset into quintiles to partition the study area for separate analyses of NDVI over time, based on precipitation input to the system (Figure 4 3). Me an variance analysis Remote sensing offers the possibility to analyze changes in ecological resilience at the landscape scale using vegetation amount and heterogeneity as indicator measures. NDVI was characterized using a mean variance analysis which can b e computed from image statistics (Pickup and Foran 1987) and which characterizes the temporal behavior of NDVI (Washington Allen et al. 2008). The approach is employed to describe the trajectory of vegetation states across the 37 year time period in terms of the mean NDVI, characterizing the overall amount of vegetation within the landscape, and the simultaneous variance of NDVI describing the heterogeneity in vegetation. Figure 4 4 shows the hypothetical relationship between mean variance and vegetation s tatus. Each quadrant in Figure 4 4 offers a relative measure of heterogeneity (variance) and vegetation presence (mean). Variance is representative of the degree of landscape heterogeneity, while the mean gives an indication of vegetation quantity. Quadran t 1 displays low mean and low variance values, and can be considered to be representative of degraded landscapes. Quadrant 2 exhibits low mean values but high variance, and indicates landscapes that possibly possesses a lot of bare
71 ground and which are sus ceptible to disturbance. Quadrant 3 contains landscapes with high mean and low variance values suggesting that much of the landscape has vegetation cover. The high mean and high variance of quadrant 4 can be considered the most ideal for savanna as the lan dscape has high vegetation cover and is also heterogeneous, which according to the literature is a more natural or ideal state for savanna ecosystems (Hill et al. 2010, Peters and Havstad 2006, Scholes and Walker 1993). As such, if a landscape moves away f rom Quadrat 4 we would say it is being degraded or changed in some negative way. If it moves towards Quadrat 4 we could argue it has recovered from a perturbation (in this instance a decrease in precipitation) and the landscape is thus resilient at this ti me scale. The mean variance plots for the entire landscape, individual land covers as determined by the IGBP land cover classification and the quintile classes were generated for this study. Persistence analysis To assess NDVI change in a spatially explic it manner an NDVI persistence layer was calculated. The layer characterizes the direction of change in NDVI relative to the 1990 2009) the direction of change (i.e. increase or decrease in NDVI) determined using the following nomenclature: t < t i = +1 t > t i = 1 t = t i = 0 i is NDVI at each of the following time steps. By assigning a the appropriate value to pixels depended upon the direction of change, a cumulative persistence layer may be calculated by summing each of the individual maps.
72 Results and Discussion Changes in precipitation Results from the hypergeometric test (Figure 4 5) indicate a significant increase in the number of annual precipitation totals falling in the driest tercile of all totals post Significance levels are approxima tely 0.05, but vary a little because of the discrete nature of the hypergeometric distribution. The later portions of the rainy season seem to be primarily responsible for this reduction in annual totals; the procedure is repeated on monthly totals during the rainy season. February and April record significantly fewer years in their respective monthly lowest terciles, 1950 197 2 while months of January through April return significantly greater membership, 19 79 200 7/08 implying that the totals in the lat ter half of the rainy season have indeed diminished since the mid 1970s This conclusion is reinforced by analysis of the memberships of the wettest tercile, which were significantly greater in the same months prior to the proposed shift in climate, and l ess afterwards. Mean variance analysis NDVI mean variance portraits for the entire landscape (Figure 4 6) indicate that prior to the proposed climate shift, the landscape as a whole had relatively well vegetated (mean NDVI) with fairly low spatial heteroge neity (as indicated by the NDVI variance). NDVI in 1984 represents the landscape immediately following the climate shift (drying) and reveals a landscape with decreased amounts of vegetation but increased heterogeneity, and reflects the impact of the prec ipitation reduction across this larger landscape, in the form of less vegetation and greater patchiness of covers. Vegetation appears to recover through 1990 in terms of amount, with low spatial
73 variance but by 2009 has maintained the increased mean vegeta tion amount but also has higher spatial variance. As such it appears that both the amount of vegetation has recovered to a pre disturbance regime and also the vegetation is more heterogeneous in 2009, potentially indicating a return to a more natural savan na state, i.e. the inherent heterogeneity of savannas (Scholes and Archer 1997). The temporal trajectory of mean variance at the landscape scale is generally mirrored across the IGBP land cover classes, with the exception of the barren/barely vegetated lan d cover class (Figure 4 7). NDVI measures move across the plots from quadrant 3 (high mean and low variance) prior to the shift, to the top left quadrant 2 in 1984, returning eventually to similar mean NDVI values but higher variance, in 2009. The majority (> 50%) of the study area is classified as savanna, this is not unexpected. The land classes differ in the extent to which mean NDVI deceases and NDVI variance changes. The highest mean NDVI are observed within Shrub and Savanna (and some of the barely v egetated lands which, based on field observation for a portion of the study area, are suspected to include many areas in which shrubs are present), and the lowest values are found in the Grass and Forest classes (and then the Wetland and Crops). Therefore the increase in mean NDVI may represent an increase in Shrub or Savanna land cover and a decrease in Forest and Grasses. In terms of variance, the land cover classes with the highest NDVI variance values are Forest and Wetland with Savanna and Bare represe nting lower NDVI variance values. The increased variance for Forest and Wetland may reflect an increased patchiness in these covers; whereas Savanna and Bare have a decreased variance which would be reflective of a more homogeneous
74 vegetation cover. As suc more homogenous covers whereas the Forest and Wetland classes are becoming patchier and more fragmented across the region. These mean variance patterns for individual covers, once viewed as an entire landscape result, could reflect an increased savanna/shrub cover, with high mean NDVI values and also an increase in landscape heterogeneity due to the different patterns of change across the different cover types resulting in an overall increase in patchiness or heterogeneity at this landscape level. The 1970 and 1990 results (Figure 4 7) show similar ranges and patterns of variance across land covers, although the mean values are higher for all classes in 1990. 1984 and 2009 present a greater range of variances which may relate to the more extreme precipitation events 1984 drought and 2009 large scale flooding. However, while the range and pattern across land covers is similar, the mean values for 1984 are the lowest of all years and all land cove rs as a result of the persistent drying of the climate after the mid 1970s. On the other hand, the very wet year of 2009 records mean the Barely Vegetated, return slightly lower mean, and considerably higher variances. shows a dramatic and consistent spatial trend of decreased NDVI across the entire landscape in 1984. By 1990 rec overy appears to have set in across all land cover classes resulting in high mean NDVI and low NDVI variance, with values being higher in terms of vegetation amount but similar for variance across all classes. Finally, 2009 may represent a return of vege tation (mean and variance) similar to that of pre disturbance, but NDVI varianc es are much larger in the 2009, probably due in part to
75 the higher precipitation and some flooding in that particular year, as well as potentially reflecting a return to a more ecosystems, where high heterogeneity reflects the multiple human and natural, complex drivers of savanna systems (Hill et al. 2010, Scholes and Walker 1993). If we look across land covers, and less by temporal groupings alone (in essence linking the dots for each land cover class sequentially by time in Figure 4 7), we can see that Savanna and Barely Vegetated classes represent relatively flat patterns of change over time in terms of their trends and show quite limited dissimilarity in NDVI variance a decreased mean NDVI (as for Savanna and Barely Vegetated) but a significant increase in NDVI variance. Likewise the trend of inc reasing mean NDVI from 1984 to 1990 holds for all covers. However, for Crops, Forest, Grasses, Shrub and Wetland, this trend is also accompanied by a significant decrease in NDVI variance, followed in 2009 by a decreasing NDVI mean (as for all land covers) and another significant increase in NDVI variance. When we view the land cover map (Figure 4 3) we see that the dominant land covers are Savanna, followed by Wetland and Shrubs, and so these are the major patterns of interest for this landscape. The mean variance patterns considered across the precipitation regimes (quintiles) are shown in Figure 4 8. Again the trajectory of NDVI mean and variance is similar to that seen in the overall landscape, with decreasing mean NDVI values from 1984, followed by a recovery period. Interestingly the mean NDVI values by date are similar across the different precipitation zones, the real difference across zones lies instead in the variance. The first quintile represents the lowest mean annual rainfall
76 region, incr easing to the fifth quintile which has the highest mean annual rainfall (Figure 4 3). Overlap exists between zone 1 and the wetland land cover class (Figure 4 2), with this driest annual precipitation region, actually being composed of much of the Okavango Delta region, and as such much of the water input to this part of landscape is related to runoff and water flow, separate from just the precipitation input experienced by the rest of the landscape. In general this first quintile represents the highest ND VI variances, which occurs despite this region receiving low precipitation input and due to the presence of the wetland being fed by river flow and hydrological inputs. However, the pattern for NDVI variance differs, with the middle quintiles (second, thir ds and fourth) showing very little initial change in NDVI variance. The driest part if the study area (first quintile) has the highest NDVI variance across each of the first three time steps, however in 2009 the NDVI variance of the wettest part of the stu dy area exceeds that of the driest. This pattern may well link to the extreme year of 2009, where flooding covered much of the region, and may well be reflected in this pattern of higher NDVI variance. Considering the overall landscape mean variance tren ds in terms of NDVI mean resulting from the shift in climate being felt across the entire landscape (Figure 4 6), across all land cover types (Figure 4 7) and across all pr ecipitation zones (Figure 4 8) thus truly showing the significant impact this change of regime had across the region. This was followed by a decade of recovery and adaptation to the new precipitation regime savannas are adaptable environments and indeed are known by their heterogeneity over space and time. Thus by 1990 we see the highest mean NDVI
77 values and still relatively low variance. By 2009 the landscape appears to have stabilized once more, with mean NDVI values being similar to the initial state o f the values, which may well reflect a slightly different vegetated state than was previously present in this region. The occurrence of extreme events, such as in 2009 whe n flooding occurred throughout much of the southern portion of the study area (Okavango Delta area and Caprivi Strip, Namibia), is also reflected in the variance of NDVI values for 2009 (Figures 4 6, 4 7 and 4 8). Overall though we can see that a higher me an NDVI and a higher variance of NDVI potentially represents a resilient landscape returning to its inherent heterogeneity of cover, as expected for savanna ecosystems (Hill et al. 2010, Scholes and Archer 1997, Scholes and Walker 1993). Persistence ana lysis The trends and imagery discussed so far represents a predominantly aspatial analysis. In order to address the spatial patterns across time we undertook a persistence trend analysis conceptually similar to that of Lanfredi et al. (2004), which reveal s the degree of consistency of temporal trends over time. Given that the driver of change across the landscape has been a significant shift in climate precipitation and subsequent recovery period, we would not necessarily expect to see consistent trends ov er time yet it is interesting to consider each of the time step intervals of analysis and examine whether the spatial patterns generally match those observed by time, land cover class, and precipitation zone. Such a comparison also permits the detection of portions of the landscape which consistently differ from the overall temporal trends. Patterns of spatial persistence of NDVI values for each date are compared to the initial
78 conditions of 1970 (Figure 4 de crease in NDVI over the entire study region (Figure 4 9a). Likewise, 1970 to 1990 illustrates the rebound and recovery in full effect with increased NDVI values, as we saw in Figures 4 5, 4 6 and 4 7. However, not all regions follow this dominant trend, sp ecifically the wetland regions and the highest and lowest precipitation zones which evince more negative trends in NDVI change. By 2009 the landscape seems to be returning to its initial state (according to Figures 4 6, 4 7, and 4 8) as reflected by almost equal areas of increase and decrease in NDVI (Figure 4 9c), but overall declining slightly. Given that the persistence threshold here is set at 0.01 in order to filter out small fluctuations and only highlight major changes, then the trends shown in Figur e 4 9 can be seen to support the existing contentions, while furnishing interesting spatial information about which areas or cover types seem to behave contrary to the dominant pattern. Figure 4 9d summarizes these persistence trends by summing the individ ual coverages at a pixel level to reveal the overall trend of a slight decrease in NDVI. Areas of consistently declining NDVI are the Wetland regions and precipitation zones 1 and 5 (highest and lowest quintiles). Smaller areas of consistent trends are fou nd in the Savanna land cover class. Thus the spatial trends add an interesting component to the analysis, particularly when linked to the land cover and precipitation information. Chapter Summary The concepts of resilience and the indicators and metrics m entioned in Table 4 1, can be related to the landscape under study. Inevitably the dates selected for the study control to some degree the findings, yet selection of the largest range of available dates and data, compilation of the imagery and the number of time periods studied
79 represents a significant data source and provides a novel in depth study of this landscape. Linkages between precipitation and vegetation addressed some in previous studies, are here viewed from a longer term perspective within a resilience framework and consider the climatic shift of the late 1970s as a significant environmental perturbation, with subsequent periods representing the response and potential recovery mine the landscape more holistically and attempt to address the longer term survival, or likelihood of adaptation, of this system as future climatic shocks are predicted to occur. Referring to the terms in Table 4 1, our landscape appears to display an ela sticity of around 35 year. The amplitude of the change in vegetation resulting from the perturbation of the climate system was extreme throughout the entire landscape (in terms of space), present across all land covers, and numerically expressed as a drop in the mean NDVI of between 0.3 and 0.4 a massive decline in healthy green biomass although the slightly later image date for the 1984 image may contribute to this a little, the value is still very high even after accounting for this. This constitutes a significant and widespread response, as is clearly seen in Figure 4 9. The malleability of the landscape examination of the range of values of mean and variance of the variable across various land covers, suggests a potential shift towards more Savanna and Shrub. Increases in extensive flooding during the last years of the study. A shift in cover types therefore cannot be completely ascertained, although one seems likely given the dramatic changes noted
80 across the region. System dampening can be inferred in Figures 4 6, 4 7, a nd 4 8 although a finer temporal analysis would be necessary to provide a more substantive ent paths of degradation and post climate shift collapse, compared to that of restoration and recovery as evidenced by the shape of the pathways over time, across land cover type and across precipitation zones. The results are substantive as when addressed across various types of gradients or classifications, the results still hold true. The addition of the spatial persistence trends adds weight to the discussion, depicting some clear spatial patterns of NDVI decline and increase, but overall a landscape in which over 70% of the region is close to its initial position from 35 years earlier (NDVI values in persistence coverages on Figure 4 9d of 1, 0 or +1), seemingly showing a resilient and adaptable landscape, even in the face of a significant and long ter m climatic shock. It is difficult to observe a clear unambiguous climatic shock to any system, and then to have a period of stability to observe the recovery of the system. However, the second half of the 1970s provides a fairly universally accepted shift in the operation of global circulation (Chavez et al. 2003, Hare and Nantua 2000), and its local impacts are quite clear and persistent throughout the time period under consideration here (Batisani and Yarnal 2010, Mason 2001), although there is some ques tion as to whether the data from 2009 may actually represent a shift back to wetter conditions (Swanson and Tsonis 2009). Nonetheless, the last two decades of the 20th century offer an ideal period to investigate the impacts of both a decline in annual pr ecipitation, and a greater variability
81 associated with ENSO, upon vegetation in an ecosystem that is presumed to be extremely sensitive to such fluctuations. In terms of landscape resilience, we argue that the system has responded to the adapted and responded in such a way that the mean vegetation has indeed returned to previously evi dent. This implies that the region is ecologically resilient to the climate variability it has experienced, though this does not necessarily mean that it will stay this way under future predicted climate changes. Other arguments could be applied to explai n the observed patterns, such as the natural growth cycle of savanna vegetation, which is postulated to occur on a predominantly 30 year cycle (Meyer et al. 2007). This with a natural death and regeneration of the landscape, although it is unlikely that such a phenomenon would occur simultaneously over such a large landscape and so consistently in the many land cover types (Figure 4 7). Future research will decompose this tempo ral trend further, utilizing monthly data throughout the entire period of study. This will require an increase in spatial scale (AVHRR and MODIS), and so the fine grained spatial resolution and detail from this study will be sacrificed for improved tempor al resolution. Additionally, this research can be used to ascertain types of land cover shifts as these images provide a basis for more in depth examination of land cover change and sub pixel linear unmixing to determine components of vegetation at this f iner spatial scale will also be undertaken. Clearly there are some major shifts even within this resilient system and the land cover data used here (Figure 4 2) is static and
82 so changes in cover type were not evaluated, only changes in NDVI. However, this research has clearly established this dryland region to be one of resilient and adaptable ecosystems, even after such a significant climatic shift, and future research will link these changes much more explicitly to climate at an annual time step and to fi ner land cover classifications which will vary across time.
83 Figure 4 1. Study area situated within the proposed Kavango Zambezi Transboundary Conservation Area (KAZA) Figure 4 2. Merged IGBP classification of the study area, showing the majority of th e landscape is classified as savanna.
84 Figure 4 3. Precipitation regime classes based on precipitation quintiles calculated using the mean annual rainfall from 1972 2000. Figure 4 4. Hypothetical relationship between mean variance and vegetation status [Adapted from Washington Allen, R. A. Ramsey, R., West, N. E. and B. E. Norton, B.E. (2008) Quantification of the ecological resilience of drylands using digital remote sensing. Ecology and Society 13, pp. 33. ]
85 Figure 4 5. Results from the hypergeomet ric test of the total annual precipitation amounts, based on their water (May April) year from periods 1950 197 2 and 19 79 2007/2008 Figure 4 6. Results from the mean variance analysis for the entire landscape, showing the changes in amounts of vegetati on and in heterogeneity within the landscape across the twenty nine year time period.
86 Figure 4 7. Results for mean variance analysis for each land cover class (barren/barely vegetated, croplands, wetlands, grasslands, savannas, shrublands, forest). Fi gure 4 8. Results for mean variance analysis for each precipitation class (quintiles).
87 Figure 4 9. NDVI persistence figures highlight the spatial patterns in change in NDVI across the entire time period (1970 2009) and for each intermediate time step.
88 Table 4 1. Indicator measure s. [Adapted from Washington Allen, R. A. Ramsey, R., West, N. E. and B. E. Norton, B.E. (2008) Quantification of the ecological resilience of drylands using digital remote sensing. Ecology and Society 13, pp. 33. ] Indicator mea sure Definition Elasticity The period of restoration to a reference condition following a disturbance Amplitude Magnitude of change resulting from a disturbance Malleability Degree to which the state established after a disturbance differs from the orig inal state Damping Pattern of oscillation in a system following disturbance
89 Table 4 2. Imagery attributes. Image footprint (path/row) Image date (dd mm yyyy) Scanner Root mean square error 187/73 08 06 1979 Landsat MSS 0.327 187/74 18 05 1976 Landsa t MSS 0.497 188/72 30 06 1975 Landsat MSS 0.498 188/73 18 06 1979 Landsat MSS 0.499 188/74 12 06 1975 Landsat MSS 0.476 189/72 27 05 1973 Landsat MSS 0.496 189/73 27 05 1973 Landsat MSS 0.449 175/72 09 06 1984 Landsat TM 0.426 175/72 26 03 2009 Land sat TM 0.430 175/73 09 06 1984 Landsat TM 0.389 175/73 26 03 2009 Landsat TM 0.470 175/74 09 06 1984 Landsat TM 0.386 175/74 26 03 2009 Landsat TM 0.380 176/72 02 07 1984 Landsat TM 0.362
90 Table 4 2. Continued Image footprint (path/row) Image date ( dd mm yyyy) Scanner Root mean square error 176/72 20 05 2009 Landsat TM 0.452 176/73 02 07 1984 Landsat TM 0.447 176/73 20 05 2009 Landsat TM 0.441
91 CHAPTER 5 RESILIENCE IN PRACTI CE: RESPONSE OF THE SOUTHERN AFRICAN LANDSCAPE TO POTENTI ALLY CATASTROPH IC CLIMATE SHIFTS The function and spatial distribution of terrestrial ecosystems are heavily influenced by past and current climate change and variability (Masek 2001). These ecosystems and their associated vegetation patterns serve as a resource base for socio economic development and maintain numerous intangible ecosystem services. Climate changes will lead to shifts in the amount as well as the variability of prevalent precipitation patterns which will in turn alter vegetation patterns (Mason and Jouber t, 1997). Thus, quantitative assessments of the influence of climate change on vegetation patterns are central to understanding rates of change in response to climate shifts, and the spatial distribution of the impact. An extensive body of literature addr esses widespread changes in Pacific basin climate which took place in the mid 1970s and influenced the characteristics of El Nio Southern Oscillation (ENSO), and climate patterns associated with ENSO (Namias 1978, Trenberth 1990, Ebbesmeyer et al. 1991, G raham 1994, Trenberth and Hurrell 1994). The effect of this climate shift has been examined via a multitude of climate and biological time series (Mantau et al. 1997, Shi et al. 2007). This mid 1970s climate shift has been associated with a shift in precip itation patterns across much of southern Africa, resulting in post 1970s drought conditions being more closely associated with ENSO events (Fauchereau et al. 2003). The African continent in general is considered one of the most vulnerable continents to cl imate change and global climate change predictions indicate that much of Africa will experience increased frequency and intensity of dry periods (IPCC 2007). The influence of climate shifts, such as that which occurred during the 1970s, on
92 vegetation patte rns is particularly acute in the semi arid savannas of southern Africa where climate, specifically mean annual precipitation, is the dominant regional control over vegetation distribution (Townshend and Justice 1986, Fuller and Prince 1996, Richard and Poc card 1998). As such, shifts in precipitation patterns or prolonged dry periods can potentially result in dramatic shifts in vegetation and irreversibly alter the state of these savanna landscapes. African savannas by nature are highly heterogeneous mixed woody herbaceous systems (Scholes & Walker 1993, Hanan & Lehmann 2011). These ecosystems exist in multiple states and movement from one state to another is dependent upon precipit ation vegetation relationships (Zen and Neelin 2000) and at a local spatial scale on a combination of herbivory and fire (Dublin et al. 1990). The state of savanna ecosystems, and their ability to absorb shocks can be quantified in terms of vegetation attributes, including quantity of vegetation cover, heterogeneity, and productiv ity (Washington Allen et al. 2004 a,b, 2006). S avannas ideally exist in the upperright quadrant of F igure 5 1, however, the savannas are highly adapted to disturbances, and ultimately their resilience lies in the ability of the landscape to shift from one quadra n t to another as it responds a nd recovers from disturbances. D ue to the nature of savannas when considering shifts from one state to another, both vegetation amount and heterogeneity ought to be considered a s decreases in eithe r of these characterist ics can lead to an altered state for the savanna system ( Scholes & Walker 1993 ) Examining ecosystem characteristics such as the afore mentioned vegetation attributes allows for an assessment of the effec t of disturbances on landscape. Furthermore these ve getation attributes are quantifiable measures which can be associated with landscape
93 elasticity, amplitude, malleability, and damping and hence offers a look at preliminary conside ration of vegetation attributes is most frequently conducted with regard to a reference state (Washington Allen et al. 2008), examinations of post shock landscape dynamics offer valuable insight related to shifts in state of landscape and system resilienc e. Repeat spatial and temporal information on the distribution of vegetation is a requisite for understanding the response of vegetation to climate change. The current availability of remotely sensed measures of vegetation indices, offers ideal measures o f vegetation distribution. Additionally, the temporal span of such measures is becoming sufficiently long enough to utilize for time series analyses and to link to past global climate changes. Due to the long temporal record and its tested suitability for measuring vegetation change, the Normalized Difference Vegetation Index (NDVI) is the most widely use vegetation index. Although, soil exposure and type potentially limits the use of NDVI in semi arid regions (Huete et al. 1985 ), NDVI has shown to be a go od indicator of vegetation characteristics, including biomass, percent green cover, biomass production (Tucker, 1979; Asrar et al. 1984; Sellers, 1985). We utilize a framework proposed by Westman and Leary (1986) and Lanfredi et al. (2004) to examine th e large scale response of savanna vegetation to potentially catastrophic climate change. Specifically we quantify spatial and temporal long term (Chavz et al. 2003; Nic holson et al. 2000). NDVI time series data is used to examine vegetation dynamics in response to climate changes across a semi arid region within
94 southern Africa, delineated by the Okavango, Kwando, and upper Zambezi catchments. The total study area covers 683,000 km 2 encompassing land areas within Zambia, Angola, Namibia, and Botswana (Figure 5 2). Annual precipita tion ranges from 400 mm/yr to 600 mm/yr. The region is dominated by savanna, and thus the natural state of the landscape is characterized by hig h heterogeneity and consists of mixtures of grasslands and woodlands. Hypergoemetric tests were conducted on data from six st ations within the study region shift on precipitat ion patterns in the region. The precipitation data from each station was thought to have occurred sometime between 1974 and 1978, however since there exists uncertainty a bout the speed with which the change occurred, we utilize a six year period (1973 197 8 ) to separate the pre climate shift precipitation data from the post shift precipitation data. In the presence of no syste matic shift in rainfall totals, sampl e s of the population (such as 1950 1972 or 1979 2008) have a distribution of observations across each tercile roughly proportionate to the ratio of the sample to population size. The results from the hypergeometric test indicate that for all six station s s ignificantly greater than expected dry years and fewer than expected wet year were (Figure 5 3) Conversely there exists no significant difference between the number of expect and observed wet/dry years during th e pre 1970 the global shift manifested itself in the form of a drier climate and environment.
95 The long term temporal dynamics of NDVI were characterized using a mean variance anal ysis (Pickup and Foran 1987). This graphical aspatial analysis characterizes the temporal behavior of NDVI with regard to changes in vegetation amount (mean) and landscape heterogeneity (variance). To control for the effect of shorter climates cycles a sev en year running average was completed on the raw data. Figure 5 4 shows the resultant mean variance plot for the entire landscape from 1982 2009. The trajectory initially shows a decrease in landscape heterogeneity (NDVI v t his is likely a lagged change in vegetation due to the and adapt to the new climate regime, as is apparent by the increases in the amount of vegetation present and a re turn to system with heterogeneous cover. It would appear as it shifts from a state of low vegetation cover and heterogeneity immediately after the climate shock to one o f high vegetation cover and heterogeneity. Although the time series analysis used here does not enable a comparison of vegetation amount and heterogeneity to that of a pre climate shift landscape, evidence indicates that this savanna landscape during the 1 heterogeneous (Cui et al. under review). The overall increase in vegetation amount across the landscape may be due to shifts in vegetation composition. In much southern Africa vegetation changes over the past three decades have been largely typified by decreases in tree and herbaceous cover, and increases in dense shrub cover. Changes in NDVI variance could be attributed to changes in the spatial homogeneity of rainfall across the region. Figure 5 5
96 shows the periods when the rainfalls are more heterogeneous or more homogeneous than would expected purely based on monthly mean precipitation alo ne (Waylen et al. in prep) If NDVI is a simple function of monthly precipitation then these are periods when we wo uld be expecting the mean variance plots to wander off of any sort of linear function because for the same mean rainfall (NDVI) in two different years we might have higher/lower variance, simply because the spatial variability of rainfall over that area wa s higher/lower than expected too which could then be translated into the spatial variance of NDVI shock was conducted using a persistence analysis. This analysis considers both the directionality of change in NDV I values, and the cumulative amount of change in NDVI per pixel across the study are a As seen in Figure 5 6 (a) the direction of change in NDVI for the majority of the landscape is towards increased NDVI values, t his is as expected considering that the mean variance analysis shows an overall increase in the amount of vegetation within the landscape. Figure 5 6 (b) identifies the cumulative change (i.e. total change in terms of NDVI value) in NDVI per pixel over the 27 year time period. Figure 5 6 (a) and (b) highlight areas within the landscape where the showing where NDVI has decreased for at least for up to 12 of the 27 year time per iod, and by as much as .2517. The spatial clustering of the areas which have decreasing NDVI values are associated with parts study region which have the highest overall NDVI values (Figure 5 7 ), suggesting that climate shift differentially affects vegetat ion within the landscape and that the parts of the landscape with large quantities of vegetation (i.e.
97 high NDVI values and likely more closed canopy woodlands ) were more greatly affected by the drying of the environment. The effect of the climate shift i the vegetation to this new climatic state could be considered analogous to the effect of future climate change on this landscape. The IPCC (2007) p redictions of further decreases in precipitation for this regi on will influence the vegetation characteristics of the landscape and could again force the landscape to shift states. Savannas are adapted to highly variably precipitation patterns; however, if future climate change is manifested in prolonged existence of dry periods, the ability of the landscape to return climate shift may be challenged. Methods Hypergeometric test. The hypergeometric test compares the clustering patte rn of the observed rainfall to a random drawing of n elements among the N members of the initial population (Martineu et al. 1999). If rainfall is stationary from 1950 2008, we expect to find similar proportions of wet and dry years within each of the two 1950 2008) relative to the sample size of each group. Across the 20 year time period from 1950 1970 we would expect to see approximately one d one third (~7 ) Similarly across the 28 year time period from 1980 2008 we would expect to see approximately one third (~9) of the observations characterized
98 Mea n v ariance analysis. Figure 5 1 shows the hypothetical relationship between mean variance and veget ation status. Each quadrant in F igure 5 1 offers a relative measure of heterogeneity and vegetation presence. Variance is representative of the degree of la ndscape heterogeneity, while the mean gives an indication of vegetation quantity. Quadrant 1 has low mean and low variance values, and can be considered representative of degraded landscapes, those with low vegetation and little variation across the landsc ape in the amount of vegetation. Quadrant 2 has low mean values but high variance, and indicates landscapes that possibly have a lot of bare ground and are susceptible to disturbance. Quadrant 3 contains landscapes with high mean and low variance values in dicating that much of the landscape has vegetation cover, however the skewness of the data may indicate that part of landscape is susceptible to disturbance. Quadrant 4 has high mean and high variance values. Landscapes in this quadrant can be considered t he most ideal as the landscape has vegetation cover and is not susceptible to disturbance. Persistence analysis. 2009) the direction of change (i.e. increase ) is determined using the following nomenclature: t < ti = +1 t > ti = 1 t = ti = 0 i is NDVI at each of the following time steps. By assigning a value of +1 to pixels which have had an
99 increase in NDVI, 1 to pixels which show a decrease in NDVI, and 0 indicating no change, the cumulative persistence layer is then calculated by summing each of the persistence maps. Additionally, to examine the overall cumulative change in NDVI in a sp atially explicit manner, we calculated a persistence layer which compares the NDVI state of the year in question to the NDVI value of the prior observation This was done by calculating t he total amount of change from year to year and then summing each of these calculations to ascertain a cumulative change persistence output
100 Figure 5 1. Conceptual diagram of the relationship between mean NDVI / NDVI variance and savanna ecosystem state. Figure 5 2. Study area depicting the Kavango Zambezi tri basin reg ion.
101 Figure 5 3. Results from the hypergeometric tests for each of the station datasets, and below the 33 rd and 66 th percentile respectively Figure 5 4 Results for mean variance analysis shown along with sample climate data.
102 Figure 5 5 Results from the analysis of change in spatial heterogeneity of precipitation showing (top) the actual monthly residuals (difference between observed and expected values of spatial variability) and (bottom) a twelve month running mean to filter out seasonality and highlight major annual fluctuations Figure 5 6 Results from the persistence analysis where (a) shows directional change and (b) shows the cumulative NDVI change over the time peri od.
103 Figure 5 7 Regions of consistently high, medium, and low mean NDVI values.
104 CHAPTER 6 AN APPLICATION OF OBJECT BASED CLASSIFICATION AND HIGH RESOLUTION SATELLITE IMAGERY FOR SAV A NNA ECOSYSTEM ANALYSIS 3 Savannas are geographically extensive and socio economically important, covering approximately 25% of the terrestrial landscape and supporting a growing proportion of t et al. 2002; UNEP 2002; Scholes and Archer 1997) Savannas contribute greatly to global net primary productivity (NPP) and play a significa nt role in the carbon cycle (Williams et al. 2007; Still et al. 2003) Savanna ecosystems are undergoing rapid changes in composition and structure driven by natu ral and anthropogenic causes (Holdo et al. 2009) These changes hold the potential to greatly influence socio ecological function ing within savanna systems (Holdo et al. 2009; Scanlon et al. 2005; Ringrose et al. 1998) African savannas in particular are projected to be under risk of extensive change largely d ue t o changes in climate patterns (Archer et al. 2001) which may exacerbate the challenges presently facing humans living in this region. The discussion of ecological change in African savannas focuses on shifts in tree and shrub cover, specifically the d ecline in tree cover and change in spatial arrangement of trees, which impacts the productivity of the system, modifies availability of resources for both wildlife and humans, and could have large impacts on earth atmosphere inte ractions (Ludwig et al. 200 8; Beerling and Osborne, 2006; Ringrose et al. 2002) It is thus imperative that the quantification and characterization of tree canopies in savannas is improved to inform management policies aimed at ensuring sustainable use of resources in savanna region s and to 3 Reprinted with permission from Gibbes, C., Adhikari, S., Rostant, L., and Southworth J. (2010). Utility of object oriented classification and high resolution satellite imagery for semi arid savanna ecosys tem analysis. Remote Sensing 2, pp. 2748 2772.
105 better understand the impact of changes in s avannas at multiple scales (Holdo et al. 2009; Ludwig et al. 2000) Unlike forests, savannas have a discontinuous tree canopy, and are defined by the complex interactio ns between trees and grasses (Belsk y 1994) Tree cover is fundamental to savanna functioning, moderating the floristic and faunal composition, struct ure and function of savannas (Scholes and Archer 1997) Trees increase structural complexity and alter resource availability, creating microha bitats by altering soil temperature, fertility, and biomass allocation, increasing diversity within the system (Cramer et al. 2010; Callaway 1998; Belsky and Canham 1994; Scholes and Archer 1997; Belsky 1989) Declines in tree cover may thus reduce availab ility of herbaceous resources, which can result in degradation of habitat for wildlife, and loss of resources for local human populations. In light of this, the critical challenge for understanding the functioning in savanna systems lies in understanding t he spatiotemporal dynamics of trees. Spatiotemporal variation in tree cover, and the relationship between trees and savanna composition and function is dependent upon tree size, age and length of presence (Scholes and Archer 1997). Field experiments have b een used to explore the relationship between trees (of varying size and age cohorts) and other savanna components (grass, herbivores etc.). For example, Cramer et al. (2010) explore the relationship between leguminous trees and C 4 grasses, determining tha t the ability to fix N 2 enables trees and C 4 grasses to coexist despite limited availability of N 2 Goheen et al. (2007) show that large mammalian herbivores suppress the reproduction of Acacia, and can influence population dynamics within savannas. These approaches
106 successfully establish baseline understandings of the influence of trees within specific contexts; however they are limited in spatial and temporal scale and as such the role of trees in savanna functioning is still insufficiently quantified. Th e lack of time series data which captures tree demographic characteristics (ex. crown size) and is specifically related to tree cover change across the landscape contributes to the limited understanding of the initial states of tree biomass and environment al thresholds present within savanna systems. Furthermore, the dearth of data quantifying tree demographics and spatiotemporal changes restricts our understanding of the regional and global impacts of ecological changes in savannas and the vulnerability of savannas to global change. Appropriate management of extensive savanna landscapes requires the need for tools which enable monitoring across larger spatial scales than is feasible with field based studies alone. Savanna landscapes in southern Africa pos sess ecological limitations such as highly variable climates which constrain agricultural profits; however there exists a wealth of charismatic and endemic wildlife species of high economic value within the region (Rozemeijer et al. 2000; Western 1989) Co untries in this region have therefore relied on a wildlife centric approach to managing their savanna landscapes. This approach has emphasized local management scales, with great variation existing not only in the actual management of vegetation but in the management of the drivers of vegetation structure herbivory, fire, human activity, and biotic factors such as tree grass interactions (Scholes and Walker 1993) Concern about the widespread decrease in the number of large trees, changes in the clustering patterns of trees, and corresponding impacts of these habitat changes on wildlife have focused management
107 approaches on the larger landscape matrix (van Aarde and Jackson 2007) again necessitating vegetation monitoring tools which can accommodate larger spatial and temporal scales than field measurements. The increased isolation of tree dominated savanna habitat in a largely shrub dominated landscape matrix, and the recognition of the need for scale appropriate management of megafauna has led to a shift t oward landscape scale management practices, which cross a dministrative boundaries (van Aarde and Jackson 2007; Blanc et al. 2003; Whyte et al. 2003; Katerere et al. 2001) The collaborative management of these larger land areas demands a standardized appro ach to monitoring and surveying habitat status and changes in tree cover. The trajectory of a savanna and corresponding wildlife habitat is dependent upon initial tree cover, the distribution of trees across size cohorts and the spatial distribution of tre e clusters. Large scale plot studies require extensive manual sampling which is a practical limitation in much of southern Africa, where human populations are low per/km2 and poor accessibility limits the applicability of large scale standard ized field base d habitat monitoring efforts (Scanlon et al. 2007; Mittermeier et al. 2003) Currently in southern Africa much of the savanna which is being incorporated into various collaborative or transboundary management schemes is not system atically su rveyed or monitored (Blanc et al. 2003) Remotely sensed data provides an alternative to field based vegetation plots and captures landscape scale vegetation data (Scanlon et al. 2007) Due to its repeat nature remotely sensed data is useful for tracking c hanges in tree cover over longer periods of time and at more varied temporal scales than what is typically done with field experiments. Remote sensing offers access to longer term, continuous data for larger
108 extents than traditionally used for ecological m onitoring (Gillanders et al. 2009; Roughgarden et al. 1991) Analysis of medium resolution data (such as Landsat TM) has successfully partitioned the phenological pa tterns of trees versus grasses (Sekhwela and Yates et al. 2007; Do et al. 2005; Shackleton 1999) However, in highly heterogeneous landscapes, such as savannas, it is limiting to solely rely on medium resolution data to capture tree demographic information (ex. tree crown shape, or membership in a given age or size cohort). The incorporation of high resolution satellite imagery potentially offers the ability to bridge the scale gap between plot field studies and larger spatiotemporal studies which rely on medium to course resolution imagery. Incorporating high resolution data (for the purpose of this research we focus on satellite imagery although aerial photography could also be used as a source of high resolution data) addresses the need for repeat quantification of tree cover, clustering patterns and demographics. Furthermore integration of hig h resolution data into analyses of vegetation patterns holds the potential to establish the relationship between tree cover and patterns on the ground, high spatial resolution pixel signatures and potentially even up to the spectral signatures of a medium resolution pixels (ex. Landsat 30 m 30 m). Thenkabail (2004) demonstrates the use of high resolution data to better characterize the relationship between ecological variables and medium resolution spectral indices. Although medium resolution has proven u seful for research questions which address phenology without emphasis on vegetation structure and demographics, insufficient research has focused on the incorporation of high resolution imagery in studies of landscape chan ge within savanna systems (Nagendr a et al. 2010; Stickler and Southworth 2008) ]. Thus further work incorporating multi resolution remotely sensed
109 data in combination with field data is needed to assess the suitability of the plethora of satellite imagery curren tly available to researchers (Nagendra et al. 2010) The increased number and availability of air and space borne sensors has improved our access to remotely sensed data and enhances our ability to monitor and characterize components of the landscape (Turner et al. 2007) The combined use of high resolution remotely sensed data and socioeconomic data has proven useful for understanding the causes and consequences of ecological change and thus is usef ul for ecosystem management (Walsh et al. 2008) This study contributes to the savanna ecology and remote sensing literature by exploring the utility of IKONOS high resolution data and advanced remote sensing methodologies for characterizing tree cover, clustering and demographics in southern African savannas. We investigate the use of objec t based classification and IKONOS imagery as a possible tool for scaling from field observations to medium resolution data. In response to criticisms of the inadequacies of conventional methods of remote sensing analysis which relies heavily on spectral ch aracteristics of a single pixel, along with increase in commercially available higher resolution images, object based classifications have gai ned popularity in recent years (Walsh et al. 2008; Turner et al. 2007; Hay et al. 2005; Laliberte et al. 2004; Wul der et al. 2004; Castilla 2003; Hay 2003; Blaschke and Hay 2001; Hay et al. 2001) Though not a new technique (Platt and Rapoza 2008; Weisberg et al. 2007; Laliberte et al. 2004; McGlynn and okin 2006) the overreliance on traditional pixel based maximum l ikelihood classifications has dominated studies of land cover change. Object based classification relies on both spectral and spatial (size, shape, texture, association with neighboring objects) data to characterize the landscape, and thus proves useful in
110 landscapes where there are similar spectral characteristics of different vegetation types, as is the case in southern African savannas, where trees and shrub have similar spectral signatures, but d iffering ecological roles (Huttich et al. 2009; Wang 1990) As such we expect that object based classification of IKONOS imagery will prove useful for: identifying trees and quantifying spatial patterns of trees in savanna systems; for exploring variation in these characteristics across different land management units; and for scaling between field observations and medium resolution data. We employ a descriptive case study approach, similar to that use d by Hartter and Southworth (2009), Nagendra et al. (2007) and Munroe et al. (2007) to identify spatial patterns of trees and explore possible linkages between land management strategies and variation in the spatial distribution of trees. Due to the lack of repetition of each land management type the generalizability of the study is limited; however the study design offers a useful descriptive analysis of tree distribution across the study region, and a useful example of an application of object based analysis conducted on high resolution satellite imagery. Specifically, we ask (1) Can object based classification of h igh resolution imagery be used to identify and monitor tree demographics and distribution in a savanna system? (2) How does the distribution of trees vary spatially across two study areas representing different management strategies of government managed p rotected areas versus community conservation area? (3) What are the current clustering patterns for large trees within the landscape? and (4) Can we then scale OBC data from IKONOS to Landsat to enable a regional scaling of the findings?
111 Methods Study Area The study area covers approximately 278.6 km 2 and is located in the Caprivi, Namibia (Figure 6 1). A comprehensive treatise of the Caprivi region may be foun d in Mendelsohn and Roberts (1997) and we draw mainly from this reference in the subsequent site description, accompanied by more contemporary sources, and environmental history interviews conducted in 2007 and 2008. Most of the Caprivi region is a part of the larger Kalahari woodlands landscape, which stretches into all of the surrounding countries ( Botswana, Angola, Zambia, and Zimbabwe). The study area is divided between the Bwabwata National Park Kwando Core Area (KCA) on the western side of the Kwando River (132.1 km 2 ), and the communal conservancy areas on the eastern side (146.48 km 2 ). The KCA h as historically had local people within it, spread of sleeping sickness. Subsequent to this, local peoples were officially restricted from entering this area when the Capriv i Ga me Park was declared in 1968 (Mayes 2008) The South African Defense Force (SADF) occupied the northern part (north of continued to occupy this area until Namibian independe nce in 1990. Since the mid resource use mainly to photographic tourism. In the conservancy region, improvements in the treatment of cattle diseases allowed these populations t o reb ound beginning in This in turn created increased grazing pressure in communal regions, as well as an increase in draught power allowing people to clear land for more cultivation. The human population in Capriv i as a whole has
112 increased, leading to an increase in the pressure placed on vegetation resources. The land uses on either side of the river therefore lie in contrast to one another, with the KCA having considerably less anthropogenic conversion and less d irect human manipulation (e.g., the use of fire to manage grass growth was until 2009 suppressed, when in the KCA the management of savanna vegetation started utilizing managed or prescribed burns with anthropogenic fires to manipulate vegetation growth), while the communal conservancies are multiple use lands with areas designated for settlement, wildlife corridors, agriculture, and cattle. The communal lands and vegetation are directly manipulated to accommodate the variety of land uses occurring within t his area, for example, fire is actively used to clear land for agriculture or to encourage grass growth for cattle. In contrast, interviews with government officials (park managers) indicate that in the KCA fire was not used as a management tool to enhance grass growth, and in certain cases naturally occurring fires were controlled from spreading through the use of fire breaks. In the KCA management practices are determined by local and regional government officials, while in the community conservation area s savanna management practices are decided by committees consisting of local community members. However, recent (2007) implementation of a collaborative management scheme for the study area has resulted in collaborative management practices across the land units and generated the demand for a standardized approach to monitoring wildlife quantities and tree cover. The study area is typical of much of savanna present in the surrounding countries, consisting of woodlands interspersed with open grassland areas. Common tree species found in the study area include: camel thorn ( Acacia erioloba ), karamoja ( Acacia tortilis ),
113 rhodesian teak ( Baikea plurijuga ), wild seringa ( Burkea Africana ), silver terminalia ( Terminalia sericea ), and russet bushwillow ( Combretum her eorense ). Rainfall is seasonal (influenced by the movement of the inter tropical convergence zone (ITCZ)) and ranges from 400 to 700 mm with sticallan annual average of 600 mm accompanied by an average a nnual temperature of 21.8C (Rice 1997) The area pro vides habitat for diverse wildlife, including 430 species of birds, and various game such as the sitatunga ( Tragelaphus spekii ), red lechwe ( Kobus leche ), buffalo ( Syncerus caffer ), and elephant ( Loxodonta africana ). Satellite Imagery and Field Data High resolution IKONOS imagery (4 m 4 m visual infrared bands) and medium resolution Landsat TM imagery (30 m 30 m) was acquired for the study area, from the Geoeye Foundation and the Council for Scientific and Industrial Research (CSIR) respectively. The i mages selected are from the dry season (IKONOS: May 21st 2006, Landsat: May 1st 2007) to minimize cloud cover and phenological variation between the two data sources. The one year time lag is unavoidable, as is the case with many studies which rely on prev iously collected satellite imagery, however both years experienced similar precipitation patterns, which is the dominant driver of interannual variation in vegetation in the region. The Landsat image was subset to match the spatial extent of the IKONOS ima ge and geometric registration was conducted using image to image registration and a nearest neighbor resampling algorithm, with root mean square error (RMSE) of <0.3 pixels. The geometric accuracy for the IKONOS image was verified using 29 ground control p oints collected in the field (RMSE < 6). The digital numbers were converted to reflectance values using radiometric calibration and incorporating post launch calibration gains and biases (Green et al. 2005) Water bodies
114 and clouds were masked out of the t wo images using a binary mask. An unsupervised classification enabled spectral identification of cloud cover and water bodies. The unsupervised classification was combined with spatial data (shapefiles of water bodies) acquired from the Ministry of Environ ment and Tourism to determine the extent of the riparian zone and ensure that tributaries to the Kwando river were also included in the mask. Removal of clouds and water bodies was done to reduce the likelihood of misclassifications during the object based classification. Field data was collected in the form of training samples (n = 77) and vegetation transects (n = 32). The sample design was limited by accessibility t o c ertain areas within the study site and as the design used was stratified random. Both a pproaches identified spatial location of individual trees (n = 118) and associated tree demographic data (species, canopy size, dbh, height), dominant vegetation type (woodland, grassland, and shrub, and bare), and general land cover measurements (% canopy cover, dominant understory), and land use history. The training sample protocol was adapted from the CIPEC protocol (Green et al. 2005), while the Walker (1976) vegetation transect methodology was used as this has proved suitable for characterizing vegeta tion in southern African savannas. The combined use of training samples and vegetation transects enabled the collection of spatial location of trees and of tree demographic data, while simultaneously maintaining the spatial distribution of field data colle ction across the study area During the field season we also conducted 27 key informant interviews aimed at understanding who determined management practices for both KCA and the community conservation management areas, and what factors contributed to (spe cifically trees) development, and how vegetation disturbances (ex.
115 fire, herbivory) are managed and used. The key informant interviews helped develop an understanding about the possible drivers of tree distribution differences in KCA versus community conse rvation management areas, this is further discussed in the results and discussion section. Tree Crown Identification and Spatial Analysis Image processing was conducted in ERDAS Imagine 9.3, using the Imagine Objective tool which enables automated feature extraction. An object based classification was used to identify trees in the IKONOS image. Object based classification is defined as assigning classes to image objects (Hay et al. 2003; Schneider and Steinwender 1999) which are the result of segmentation o f an image into discrete non overlapping un its based on specific criteria (Mitra et al. 2004; Hall and Hay 2003; Hay et al. 2003; Hay et al. 2002) The object based classification consisted of five processes: probability matrix computation, image segmentati on, raster to vector conversion, vector processing, vector cleanup (Figure 6 2 ). The probability matrix computation is a supervised procedure which was trained using the location of known tree crowns. Although we tested a variety of input variables derived data from the original IKONOS image (e.g. NDVI, texture) the best results were acquired using the original four band visual infrared IKONOS image. The probability matrix computation assigns a probability of being a tree or not to each pixel. The probabili ty of being a tree or not is determined based on the similarity of spectral values of the given pixel to the spectral values of the pixels of known trees identified during the training process. Segmentation of an image is the process of partitioning a digi tal image into multiple regions to simplify and/or change the representation of an image into something that is more meaningful and easier to analyze (Goward and Prince 1995)
116 The segmentation based on spectral values and spatial characteristics created ob jects that are useful in classifying land cover in regions such as southern Africa where trees and shrubs are spectrally similar but different spatially and ecologically. The segmentation process was performed on the original four band visual infrared IKON OS image; however, the pixel probability layer was also used to calculate the probability zonal mean for each generated segment. The segmentation approach used combined splitting and merging of the input image with the use of edge detection to identify seg ments with similar characteristics. The parameters used for image segmentation included: Euclidean distance (the compute settings function was used to determine the minimum value difference and the variation factor, 36 and 3.5 respectively), edge detection with an automatically generated threshold of 53, and a minimal edge length of three pixels. The segmented image was then converted into a vector file, which was used for the vector object processing and cleanup procedures. Vector object processing consist s of another supervised classification which utilizes geometry to refine the identification of trees. A variety of geometric cues are available for use in vector object processing, however after experimenting with the cue options we selected the following cues for use: area, a perimeter area ratio, and shadow. The thresholds used for the cues were determined by training the classification with the area and perimeter area characteristics of the trees observed in the field. The area characteristic had a minim um value of 48.6 square meters and a maximum of 405.55 square meters, and the perimeter area ratio had a minimum of 10.7 and maximum of 20.56. The shadow cue measures the association (determined based on adjacency) of the vector objects generated from the
117 first three steps of the object based procedure to a shadow polygon created using a separate unsupervised classification. The vector cleanup process consisted of applying a probability filter to the vector file. The probability filter removed all object wi th less than 0.9 probability of being a tree crown, thereby ensuring that the final vector layer only included polygons that had a high certainty of being tree crowns. An assessment of the accuracy of the object based classification was conducted comparing the location of tree polygons as identified in the final vector layer from the object based classification with the actual spatial location of individual trees as determined in the field. The resulting object based classification was then used to assess t he current differences in initial tree cover across the two dominant management types. The spatial pattern analysis. Since much of the concern within the study area is regar ding the decrease in large trees, this analysis is limited to trees with crown diameters of 12 tree canopy collected during training sample and vegetation transect collect ion. This threshold aimed to limit the analysis to the larger trees and target tree species which are thought to be keystone species for African sav annas, such as Acacia erioloba ( Moustakas et al. 2006 ) and reduces the size of the dataset for processing p urposes. Point pattern analysis was used to identify tree clusters and the possible corresponding vegetation development processes. The Getis Ord Gi* statistic was used to analyze the spatial clustering of individual trees based on crown size. This statist ic measures the local association amongst features and exp lores the spatial clustering ( Ord and Getis 1995) and as such is a useful tool to detect clusters of trees with similar crown sizes,
118 which might result from environmental conditions. The Getis Ord Gi* statistic calculates a standardized z score for each tree, which identifies the magnitude of deviation from the expected tree crown size as determined by surrounding tree crowns, thus allowing you to identify significant clustering of large and small t ree crowns. Since the Getis Ord clustering. The tree cover extracted from IKONOS data was also linked to medium resolution Landsat imagery (Figure 6 3 ) in two ways. First, the spatial distribution of trees within a 30 m 30 m cells was assessed. The 30 30 m cells correspond in both size and alignment to the Landsat TM data. This approach allowed us to calculate the number of trees per cell of the medium resolution Landsat imagery (i.e., the number of trees per 30 m 30 m grid cell). The possibility that a cell containing multiple small trees had similar spectral values to one containing a single large tree was addressed by looking not only at tree counts per cell but also at the proportion of tree cover per Landsat pixel. Proportion of each Landsat pixel covered by tree polygons was calculated and compared to the corresponding normalized difference vegetation index (NDVI) value of the Landsat image (Figure 6 3 ). NDVI is the most widely used vegetation index, and is an indicator of vegetation growth ideally suited for semi arid regions where the index does not sa turate at high foliage bioma ss ( Goward and Prince 1995; Nicholson and Farrar 1994 ]. Additionally, this index is widely used, allowing our research findings to be easily interpreted and compared to other land cover studies. Tree proportion classes are developed by grouping pixels base d on amounts of polygon coverage. For example,
119 tree class 1 consists of all pixels with 0 20% tree coverage, class 2 consists of all pixels with 21 40% coverage, and so forth, ultimately resulting in 5 tree cover classes. The variation in NDVI as a functio n of proportion of trees is quantified by examining the difference amongst NDVI values for each tree cover class and is using graphical outputs and the Kruskal Wallis test. The Kruskal Wallis test is a non parametric approach, with the test statistic H, us ed to determine wheth er two or more groups differ (Field 2009) Results and Discussion Object Based Classification The object based classification results in a vector file containing polygons which are associated with one of two classes: class 1 trees, cla ss 2 non tree. For the purpose of this study we are interested in utilizing object based classification to discriminate trees from the background mosaic landscape. Figure 6 4 shows the classification for a subset of the study area as well as both the IKONO S and Landsat data for the subset and larger study area. The object based classification improves upon a pixel based classification by incorporating shapes and spatial associations (e.g., association with shadow) of tree crowns; additionally, the resulting vector layer includes shape and size in the characterization of tree crowns. The results from the object based classification were considered in terms of number of correctly identified tree crowns, as determined by comparing the location of trees identifi ed in the object based classification to the actual spatial locations of individual trees identified in the field. The overall accuracy was 84%. The ability to monitor tree presence, growth via measurements of size and spatial pattern, and change in savann as through the examination of realistic objects (trees) as opposed to pixel based classifications of tree
120 cover is particularly useful in an ecosystem where tree presence and demography greatly influences ecological processes and t he conditions of microhab itats (Callaway 1998; Scholes and Archer 1997; Belsky and Canham 1994; Weltzin and Coughenour 1990; Belsky et al. 1989) As Figure 6 4 shows, the object based classification results in a vector layer, which contains spatial characteristics such as size and shape of crown that can potentially be linked to tree species and ages. Classifications such as the one demonstrated here provide standardized baselines for initial tree distribution and demographics across the multiple management units (protected area an d community conservations areas) present in this landscape. In the study area the concern that selective removal of large trees is occurring (as a result of herbivory) could be quantified and monitored using an object based classification of the ecosystem. Spatial Distribution of Trees An examination of differences in tree distribution across the two management areas indicate that there is a greater density of trees present in the protected area than in the community conservation areas (Table 6 1). Thus the re is more land in the community conservation areas which consist of grass, shrub, and bare land covers. This pattern is most pronounced when looking at the trees within the diam crown size cohort (Figure 6 5 ). These findings are consistent with the field observations. Key informant interviews indicated that this difference is likely the result of the different management practices utilized in the protected area and co mmunity conservation areas. In the community conservation areas clearing land for agriculture and/or homesteads, and collection of timber resources reduces the presence of trees. Fire has historically been actively used in the community conservation areas and until 2009 was suppressed in the protected area. This difference likely contributes to historical
121 differences in tree development and thus in differences in current observed tree demographics. Within the community conservation areas the large trees are predominantly located in the western portions of the management units, along the river. The community management strategies include designating the riverfront lands as wildlife areas and placed land use restrictions on land closer to the river, likely res ulting in a different resource use pattern in the western versus eastern portions of the community managed lands. Agricultural lands are maintained further away from the river in areas where the soil is considered more fertile and agricultural plots are cl oser to the main road and homesteads. Agriculture in this area, as in much of sub Saharan Africa, is largely rainfed (Barron et al. 2003) and proximity to the river is less of a concern than soil quality and market accessibility. The land along the riverfr ont is managed in a manner more similar to that of the protected area than the remainder of the community conservation areas. The Getis Ord Gi* statistic was mapped (Figure 6 6 ) for the selected size cohort characteristics or a combination of both. Results indicate significant clustering of trees at the upper end of this size cohort (ie. tree s with extremely large crown size relative to the total tree population in the study area) in southern portions of the study area, while trees with smaller crowns within this size cohort tended to have more evenly distributed clusters across the landscape. The tree spatial location of tree clusters is probably the result of both historical stand development processes and the influence of variation in other biotic parameters (such as soil moisture, herbivory) which influence tree growth. Significance was det ermined using the Bonferroni adjusted critical value of 1.68. As
122 spatial processes gain increasing attention in ec ology and ecosystem management (Groen et al. 2008; Levin 1992; Turner 1989) the identification of tree clusters within the landscape aids wit h developing an understanding of the interactions between spatial processes and heteroge neity in savanna vegetation (Wiens 1989) Standardized characterizations and quantifications of tree clustering offer useful ecological information for monitoring and m anaging the spatially heterogeneous processes (e.g., herbivory, fire, anthropogenic land use) which influence the maintenance of savanna vegetation. The origin and continuation of clustering of larger trees in the southern portion of the protected area is likely attributed to development patterns and spatial heterogeneities in the ecological mechanisms which influence savanna composition. Additionally, key informant interviews confirmed that although the South African Defense Force (SADF) historically occup ied parts of the protected area, land and resource use was limited to the southern portion of the protected area, as such this historical variation in land use patterns within the protected area also likely contribute to the current vegetation patterns. Sc aling from Field to Landsat TM The frequency count of trees per 30 30 m grid cell (cells correspond to the actual 30 x 30 m Landsat pixels), facilitates the assessment of the relationship between NDVI and number of trees per cell. The number of trees per cell ranged from 0 to 5, this range was similar to that observed during vegetation transect collection. Cells with 3 or fewer trees are evenly distributed over the landscape, while those with 4 or 5 trees appear more clustered (Figure 6 7 ). The mean NDVI values for cells containing 1 5 trees ranged from 0.170 to 0.2 suggesting little variation in the NDVI values based on the number of trees per cell. This would suggest that NDVI is not particularly useful for
123 discriminating tree cover, however, it could al so be a factor of the size of the tree crown. A cell containing a single large tree may in fact have similar proportions of canopy cover to a cell containing numerous small trees, thus resulting in similar NDVI values for those two cells. To address this w e examine the proportion of tree cover per Landsat NDVI pixel. Graphical analysis of NDVI values for the five tree coverage classes (determined by grouping proportion coverage %) show similar distribution patterns and mean values for all five tree coverag e classes (Figure 6 8 ). The results from the Kruskal Wallis test (H = 4.39, significance level 0.05) support the graphical analysis, indicating that the distribution of NDVI values for each coverage class is not statistically different. The mean NDVI for al l pixels with tree coverage is 0.289, while that for cells with no portion of the cell covered by trees is 0.211. The similarity in mean and distribution of NDVI values across the tree coverage classes and the small difference in mean NDVI value for cells with tree coverage versus those without suggests that the land covers in the background matrix (i.e., class 2: shrub, grass, or bare lands) contribute greatly to NDVI values observed for the study region. Similar to Moleele et al. (2001) and Ringrose (1989 ) we find that although NDVI offers a good approximation for overall vegetation cover, more potentially rigorous quantification of vegetation structure require the incorporation of advanced remote sensing techniques such as object based classification. The high accuracy of the object based classification combined with the ability to quantify tree demographics per Landsat pixel indicates that object based classification of high resolution imagery is a potentially successful scaling tool for linking field and coarse resolution vegetation studies.
124 Chapter Summary To address the limitations of traditional pixel based methods of remote sensing analysis and maximize the use of higher resolution imagery within this landscape we employ an object based classification to characterize vegetation structure within a savanna landscape. High resolution satellite imagery has become increasingly available yet the scientific applicability of the se datasets remains limited (Nagendra et al. 2010) Here we examine the utility of such a dataset for classifying a savanna landscape through an object based approach and then link it to Landsat data, to scale up to a more regional landscape level. We use field data to verify and test the suitability of the object based classification an d find that the classification of the IKONOS imagery successfully identifies tree locations and depicts demographic characteristics (ex. crown size). Tree distribution and clustering corresponds to field data collected for this area and the spatial pattern ing across the land management units is as observed and expected, with larger quantities of trees and larger sized trees found within the protected area and greater quantities of other land covers found within the community conservation areas where local r esidents live and farm. In addition to characterizing tree location and demographics, the results from the object based classification prove useful for point pattern analyses offering an assessment of the spatial location and characteristics of the trees r elative to surrounding vegetation. Point pattern analyses contribute to the determination of presence or absence of spatial interactions, and thus enhance inferences about processes based on pattern. Although many studies use vegetation indices for assessm ents of savanna vegetation, our findings suggest that the addition of spatial data is critical for the accurate characterization of vegetation components within this ecosystem. This is
125 congruent with stud ies of savanna vegetation (Moleele et al. 2001; Ring rose and Matheson 1987) where commonly used vegetation indices such as NDVI are limited in their ability to characterize the complexity of vegetation composition. The limited differences found here in NDVI values across the tree count and tree cover class es reemphasizes the need to combine spectral and spatial data to characterize savanna landscapes since relying solely on spectral datasets to characterize trees in savanna landscapes could lead to misclassification of vegetation. Similarity in spectral ref lectance of shrubs and trees in this region makes discriminating structure solely using spectral information challenging, especially at the courser Landsat scale. We show that in ecosystems such as savannas where vegetation structure is distinguishable by object shape and spatial characteristics the incorporation of this additional information is useful for differentiating structural types. Due to the difficulty in scaling from field observations to satellite data integrating these data sources and analyses is one of the challenges in studies of plant phenology and ecosystem change (Fisher and Mustard 2005) This study utilizes high resolution IKONOS imagery to bridge scale and link field observations and medium resolution Landsat TM imagery in an explicit m anner, thereby providing a method to integrate field data (plot data) with continuous measures of biomass derived from satellite data. Furthermore, the results from this study suggest that the scale of environmental change in savanna systems is critical. W hile quantification of overall changes in biomass is possible with course resolution NDVI, characterizing and partitioning tree biomass and monitoring tree cover trajectories re quires finer scale analysis (Weisberg et al. 2007)
126 The use of a hierarchical m ulti scale approach for characterizing southern African savanna landscapes offers detailed and ecologically relevant information. Currently the limitations of using remote sensing for ecological studies include the over reliance on traditional maximum like lihood classifications (Southworth 2004) Although pixel based approaches certainly can be useful for characterizing the land cover, and are particularly useful when landscape components of interest have very different spectral signatures, this case study demonstrates the use of an alternative approach to pixel based classifications which incorporates both spectral and spatial information. The addition of spatial information to the classification process is particularly useful when trying to identify landsc ape objects which may have similar spectral characteristics. The object based approach applied here captures tree demographics, which when assessed spatially and temporally can be used to infer process from pattern. One challenge of the object based approa ch is the ability to discriminate between polygons of individual trees as opposed to those representing patches of trees. We attempted to address this challenge by calibrating the area cue within the vector object processing using individual tree crown loc ations as determined in the field. This ensured that rather than generating large polygons of similar spectral values, the output consisted of multiple smaller polygons. Although this does not completely remove the presence of polygons which represent mult iple trees as opposed to individual trees, it certainly reduces the occurrence. The obvious logistical limitation of the analysis shown here is the cost of the high r esolution imagery. Rocchini (2007) shows that the cost of hyperspatial imagery limits acce ss to these datasets, however as multi institutional collaborations form and commercial high resolution imagery becomes
127 available through data grants access to imagery sources such as IKONOS is increasing. Additionally, the trade off between high spatial a nd spectral resolution means that imagery with high spatial resolution is not necessarily appropriate for all research questions. While possibly useful for characterizing landscape components that are require higher spatial resolution than commonly relied upon Landsat imagery (30 m 30 m), the loss of spectral information may limit the ability to differentiate between certain characteristics. Possibly more limiting is challenges associated with data management of the large amounts of data that accompany pr ocesses such as image segmentation of high resolutio n imagery. Laliberte et al. (2007) determined that combining object based classification and decision tree modeling is a useful approach for managing the large amounts of spatial data inherent in high res olution imagery. Object based classification of high resolution imagery provides the tools to differentiate vegetative classes shrub and trees with little spectral separability in savanna ecosystems. This method is useful for heterogeneous landscapes where spatial characteristics define vegetation groups, spectrally similar vegetation types have differing ecosystem functions and the integration of multi scale analyses are needed to effectively quantify ecological change. Having tested the applicability of t his methodology and data source for characterizing tree cover in southern African savannas, future research will examine direct linkages between proportional tree coverage as derived from the object based classification and proportions derived from a sub p ixel classification of Landsat TM imagery.
128 Figure 6 1. Study area showing the larger regional context and the boundaries of the two dominant land management types used in the region. Figure 6 2. Object based classification workflow.
129 Figure 6 3. Sc hematic to show the relationships between the actual landscape in terms of tree coverage or crowns, the idealized object based classification results in polygon form, and then related to the same area but at a Landsat scale, illustrated with an NDVI image. The results shown are for the actual study region, photos though were representative from that region for (a) area with <25% canopy closure, (b) an area of approximately 50% canopy closure, and (c) and area of greater than 75% canopy closure
130 Figure 6 4 Results of the object based classification for the region, illustrated with the larger study are and then a blow up of the focus region to enable the reader (left ) a focus region i n the study area (color composite RGB = near infrared, red, green where red represents vegetation and cyan represents bare earth), (middle ) the resultant OBC results of individual tree polygons calcula ted from the IKONOS data, and (right ) the Landsat NDVI image of the focus region, clearly illustrating the differences in scale. Figure 6 5. size cohort.
131 Figure 6 6. Spatial clustering of trees within the study region for (a) Spatial results of the Getis Ord Gi* statistic with Bonferroni correction applied to determine statistically sign ificant tree clusters (b) Landsat TM NDVI shown for the same study region Figure 6 7. Spatial distribution of Landsat pixels with greater than four trees per 30 x 30 m cell.
132 Figure 6 8. Boxplots of the NDVI values for each tree coverage class (1 5) as determined by proportion tree cover per Landsat TM 30 x30 m pixel.
133 Table 6 1. Density of large trees within each of the two land management types KCA, government managed (~132.1 km2) versus community managed lands (~146.48 km2). Large trees are define d as those with crown diameter KCA (government managed protected area) Community Conservancy Areas (Community managed) Total no. of trees/km 2 580.2 352.7 No. of large trees/km 2 11.5 3.4
134 CHAPTER 7 CONCLUSIONS Research Overview The overarc hing goal of this dissertation is to examine landscape change and management in semi arid savannas. To this end, I examine the use of remote sensing for measuring landscape change, dissect the assumed benefits of current land management / conservation stra tegies, and quantify the spatial and temporal patterns of land cover change at multiple scales. Each of these research foci are presented here in separate research papers encompassing this dissertation. Chapter 2 establishes that remote sensing as a critic al tool for LCS research holds further potential for advancing this field While the data sources and methodologies used within the LCS arena, have undoubtedly contributed to our current understanding of landscape patterns and the driving pro cesses, further advancement within this fi eld will necessitate the simultaneous use of multiple remotely sensed data source and the intertwining of both categorical and continuous methodologies that move beyond maximum likelihood classifications. Chapter t hree considers the effect of CBNRM strategies on social equity The assumption that CBNRM is an economically or social advantageous institution for all individuals is questioned. The discussion posed challenges conservation research and implementation to m ove beyond the broad sweeping evaluations of CBNRM and more consistently consider the nuisances associated with implementation of this (or any) conservation strategy. The findings from C hapters 4 and 5 indicate that resulte d in a drier environment within southern Africa which resulted in shifts in ecosystem state as measured by specific vegetation attributes vegetation amount and
135 heterogeneity. Yet, based on the results from C hapter 4 after approximately thirty years, the system appears to have returned to a somewhat similar state The results from the annual time series analysis conducted for C hapter 4 offer a richer inspection of the ana ly sis, in C hapters 3 and 4 offer two distinct and yet complimentary temporal scales e ach providing a valuable perspective of vegetation change as affected by climate shifts The first th e state of landscape prior to the climate shift, while the latter provides a finer temporal resolution of analysis. Finally C hapter 6 attempts to incorporate the idea of moving remote sensing analyses of lands cover beyond vegetation indices, and specifi cally begins to address the limitations of the quantifying vegetation using vegetation indices. These limitations are considered specifically for savanna ecosystems where, while the overall amount and heterogeneity of vegetation are valuable characteristic s for monitoring landscape change at the regional scale, the discrimination of vegetation components (trees versus shrubs versus grasses) is necessary for management purposes. The ability to measure land cover changes in terms of shifts in vegetation compo sition is more pertinent to questions regarding local LULCC and land/resource managemen t. Thus, the work presented in C hapter 6 draws from the position presented in C hapter 2 that remote sensing data and methodologies need to be further maximized in order to better address questions of LULCC, while attempting to characterize the landscape in a manner relevant to local land use and management.
136 Significance of this Research This dissertation contributes to each of the research foci of LCS Firstly and most e xtensively this dissertation explores the approaches used for observing and monitoring LULCC and investigates multi scalar (temporal and spatial) quantifications of LULCC, while specifically furthering these measures for semi arid systems. Exploring the ca uses, impacts, and consequences of land use remains a fundamental ambition of LCS to which reviews of the social impacts of conservation strategies contribute. The employment of vegetation attributes as proxy measures for ecosystem state, while not new, ha s not been extensively tested particularly for semi arid regions. The regional assessment of land cover change conducted in C hapters 4 and 5 ; followed by a local evaluation of the status of land cover in C hapter 6 offers an innovative way for LULCC studies to begin to place local stud ies within a regional context. T his approach contributes to LCS assessments of v ulnerability and resilience
137 LIST OF REFERENCES Achard, F., Eva, H. and Mayaux, P. (2001). Tropical forest mapping from coarse spatial resoluti on satellite data: production and accuracy assessment issues. International Journal of Remote Sensing 22, pp. 2741 2762. Adler, R., Huffman, G., Chang A., Ferraro, R., et al. (2003). The Version 2 Global Precipitation Climatology Project (GPCP) Monthly Pr ecipitation Analysis (1979 present). American Meteorological Society 4, 1147 1167. Archer, S Boutton, T. and Hibbard, K. A. (2001). Trees in grasslands: biogeochemical consequences of woody plant expansion. In: Schulze, D., Heimann, M., Harrison, S., Holl and, E., Lloyd, J., Prentice, I., Schimel, D. (eds) Global Biogeochemical Cycles in the Climate System. Academic Press: San Diego, CA, USA. Archibald, S. and Scholes, R. J. (2007). Leaf green up in a semi arid African savanna separating tree and grass re sponses to environmental cues. Journal of Vegetation Science 181, pp. 583 594. Asner, G. P., Hicke, J. A. and Lobell, D. B. (2003). Per pixel analysis of forest structure: vegetation indices, spectral mixture analysis and canopy reflectance modeling. In: W ulder, M. A. and Franklin, S. E. (eds) Remote sensing of forest environments. Concepts and case studies. Boston, Dordrecht, London: Kluwer Academic Publishers, pp. 209 254. Asrar, G., Fuchs, M., Kanemasu, E. and Hatfield, J. (1984). Estimating absorbed pho tosynthetic radiation and leaf area index from spectral reflectance in wheat, Agron. J. 76, pp. 300 306. Barron, J., Rockstrom, J., Gichuki, F. and Hatibu, N. (2003). Dry spell analysis and maize yields for two semi arid locations in east Africa. Agr. Fore st Meteorol 117, pp. 23 37. Batisani, N. and Yarnal, B. (2010). Rainfall variability and trends in semi arid Botswana, Implications for climate change adaptation policy. Applied Geography 30, 483 486. Beerling, D. and Osborne, C. (2006). The origin of the savanna biome. Global Change Biol. 12, pp. 2023 2031. Belsky, A. (1994). Influences of trees on savanna productivity: Tests of shade, nutrients and tree grass competition. Ecology 75, pp. 922 932. Belsky, A., Amundson, R., Duxbury, J., Riha, S., Ali, A. a nd Mwonga, S. (1989). The effects of trees on their physical, chemical, and biological environment in a semi arid savanna in Kenya. J. Appl. Ecology 26, pp. 1005 1024.
138 Belsky, A. and Canham, C. (1994). Forest gaps and isolated savanna trees. BioScience 44, pp. 77 84. Berkes F. (2004). Rethinking community based conservation. Conservation Biology 18, pp. 621 630. Bhaskaran, S., Paramananda, S. and Ramnarayan, M. (in press). Per pixel and object oriented classification methods for mapping urban features usin g Ikonos satellite data. Applied Geography Blanc, J., Thouless, C., Hart, J., Dublin, H., Douglas Hamilton, I., Craig, C. and Barnes, R. (2003). African Elephant Status Report 2002: An Update from the African Elephant Database, International Union for the Conservation of Nature : Gland, Switzerland and Cambridge, UK. Blaschke, T., Lang, S., Lorup, E., Strobl, J. and Zeil, P. (2000). Object oriented image pr remote sensing environment and perspectives for environmental applications. In: Cremers, A. and Greve, K. (eds) Environmental information for planning, politics and the public vol. 2. Metropolis Verlag: Marburg, pp. 555 570 Blaschke, T. and Hay, G. (2001). Object based image analysis and scalespace: Theory and methods for modeling and evaluating multiscale landscape structure. Int. Arch. Photogramm. Remote Sens 34, pp. 22 29. Blaschke, T., Burnett, C. and Pekkarinen, A. (2 004). Image segmentation methods for object based analysis and classification. In: de Jong, S. and van der Meer, F. (eds) Remote sensing and digital image analysis Dordrecht: Kluwer Academic Publishers, pp. 211 236. Callaway, R. (1998). Are positive inter actions species specific. Oikos 82, pp. 202 207. Carlson, T. and Ripley, D. (1997). On the relation between NDVI, fractional vegetation cover, and leaf area index. Remote Sensing of Environment 62, pp. 241 252. Castilla, G. (2003). Object oriented analysis of remote sensing images for land cover mapping: conceptual foundations and a segmentation method to derive a baseline partition for classification. PhD thesis, Polytechnic University of Madrid. [Online]. R etrieved on 14 September 2010 from: http://oa.upm.es/133/1/07200302_castilla_castellano.pdf Castella, J., Pheng Kam, S., Dinh Quang, D., Verburg, P. and Thai Hoanh, C. (2007). Combining top down and bottom up modelling approaches o f land use/cover change to support public policies: Application to sustainable management of natural resources in northern Vietnam. Land Use Policy 24, pp. 531 545. Chapin, F.S., Matson, P.A. and Mooney, H.A. (2002). Principles of Terrestial Ecosystem Ecol ogy. Springer Verlag: New York, NY, USA.
139 Chamaille Jammes, S., Fritza H., Murindagomo, F. (2006). Spatial patterns of the NDVI rainfall relationship at the seasonal and interannual time scales in an African savanna. International Journal of Remote Sensing 27, pp. 5185 5200. Chavez, F., Ryan, J., Lluch Cota, S. and Niquen, M. (2003).From Anchovies to Sardines and Back, Multidecadal change in the Pacific ocean. Science 299, pp. 217 221. Chen, C. M., Hepner, G. F. and Forster, R. R. (2003). Fusion of hypersp ectral and radar data using the IHS transformation to enhance urban surface features. Journal of Photogrammetry and Remote Sensing 58, pp. 19 30. Chown, S. (2010). Temporal biodiversity change in transformed landscapes: a southern African perspective. Phil Trans. R. Soc 365, 3729 3742. Cramer, M., van Cauter, A. and Bond, W. (2010). Growth of N2 fixing African savanna Acacia species is constrained by below ground competition with grass Acacia grass competition in savanna. J. Ecol 98, pp. 156 167. Cui, X. Gibbes, G., Southworth, J. and Waylen, P. (u nder review ) Quantifying ecological resilience in a semi arid system. Journal of Arid Environments Daily, G. and Ellison, K. (2002). The new economy of nature: the quest to make Islan d Press, Washington, D.C., USA Daniels, A. E., Painter, K. and Southworth, J. (2008). Milpa imprint on the tropical dry forest landscape in Yucatan, Mexico: remote sensing and field measurement of edge vegetation Agriculture, Ecosystems and Environment 12 3, pp. 293 304. DeFrie s, R., Hansen, A., Turner, B., Reid, R. and Liu, J. (2007). Land use change around protected areas: management to balance human needs and ecological function. Ecological Applications 17, pp.1031 1038. DeFries, R. S., Hansen, M., Towns hend, J. R. G. and Sohlberg, R. (1998). Global land cover classification at 8 km spatial resolution: use of training data derived from Landsat imagery in decision tree classifiers. International Journal of Remote Sensing 19, pp. 3141 3168. Di Gregorio, A. a nd Jansen, L. J. M. (2000). Land cover classification system, concepts Organization of the United Nations Publishing Service. Dixon, R., Brown S., Houghton, R., Solomon, A., Trexl er, M. and Wisniewski, J. (1994). Carbon pools and flux of global forest ecosystems. Science 263, pp. 185 190. Do, F., Goudiaby, V., Gimenez, O., Diagne, A., Mayecor, D., Rocheteau, A. and Akpo, L. (2005). Environmental influence on canopy phenology in the dry tropics. Forest Ecol. Manag 215, pp. 319 328.
140 Dobrowski, S. Z., Safford, H. D., Cheng, Y. B. and Ustin, S. L. (2008). Mapping mountain vegetation using species distribution modeling, image based texture analysis, and object based classification. Appl ied Vegetation Science 11, pp. 499 508. Doos, B. (2002). Population growth and loss of arable land. Global Environmental Change 12, pp. 303 311. Drake, J. B. Dubayah R.O., Clark, D., Know, R., et al. (2002). Estimation of tropical forest structural charac teristics using large footprint lidar. Remote Sensing of Environment 79, pp. 305 319. Dublin, H. T., Sinclair, A. and McGlade, J (1990). Elephants and fire as causes of multiple stable states in the Serengeti Mara woodlands. Journal of Animal Ecology 59, 1147 1164. du Plessis, W.P. (1999). Linear regression relationships between NDVI, vegetation and rainfall in Ethosha National Park, Namibia. Journal of Arid Environments 42, pp. 235 260. Ebbesmeyer, C., Coomes, C., Cannon, G. and Bretschneider, D. (1989). Linkage of ocean and fjord dynamics at decadal period. In: D.H. Peterson [ed.] Climate Variability on the eastern Pacific and western North America, Geophys. Monogr. Am. Geophys. Union 55, pp. 399 417. Fauchereau, N., Trzaska, S., Rouault, M. and Richard, Y. (2003). Rainfall Variability and Changes in Southern Africa during the 20th Century in the Global Warming Context. Climatic Change 29, pp. 139 154. Field, A. (2009). Discovering Statistics Using SPS S, SAGE Publications: London, UK. Fisher, J. and Mustar d, J. (2005). Cross scalar satellite phenology from ground, Landsat, and MODIS data. Remote Sensing of Environment 109, pp. 261 273. Foody, G. M. (1994). Ordinal level classification of sub pixel tropical forest cover. Photogrammetric Engineering and Remot e Sensing 60, pp. 61 65. Foody, G. M. (2001). Monitoring the magnitude of land cover change around the southern limits of the Sahara. Photogrammetric Engineering and Remote Sensing 67, pp. 841 847. Foody, G. M. and Boyd, D. S. (1999). Detection of partial land cover change with ecotone movement. International Journal of Remote Sensing 20, pp. 2723 2740. Foody, G. M. and Cox, D. (1994). Sub pixel land cover composition estimation using a linear mixture model and fuzzy membership functions. International Jour nal of Remote Sensing 15, pp. 619 631.
141 Foody, G. M., Lucas, R. M., Curran, P. J. and Honzak, M. (1997). Non linear mixture modelling without endmembers using an artificial neural network. International Journal of Remote Sensing 18, pp. 937 953. Fuller, D. and Prince, S. (1996). Rainfall and foliar dynamics in tropical southern Africa: potential impacts of global climatic change on savanna vegetation. Climatic Change 33, pp. 69 96. Galvao, S., Formaggio, A. and Tisot, D. (2005). Discrimination of sugarcane varieties in southeastern Brazil with EO 1 Hyperion data. Remote Sensing of Environment 94, pp. 523 534. Galvin, K. and Reid, S. (2010). People in savanna ecosystems: Land use, Change and sustainability. In. Hill, M. and Hanan, N. (eds). Ecosystem Function in Savannas CRC Press, Boca Raton, Fl. pp. 481 495. Gamba, P. and Houshmand, B. (2002). Joint analysis of SAR, LIDAR and aerial imagery for simultaneous extraction of land cover, DTM and 3D shape of buildings. International Journal of Remote Sensin g 23 pp. 4439 4450. Gaughan, A. and Waylen, P. (under review). Spatial and temporal precipitation variability in the Okavango Kwando Zambezi catchment, southern Africa. Journal of Arid Environments. Geist, H. and Lambin, E. (2001). Proximate causes and underl ying forces of tropical deforestation. BioScience 52, pp. 143 150. Geoghegan, J., Pritchard, L., Ogneva Himmelberger, Y., Chowdhury, R., et al. (1998). Liverman, D., Moran, E., R indfuss, R. and Stern, P. (eds) People and pixels: linking remote sensing and social science. Committee on the Human Dimensions of Global Environmental Change, National Research Council. Washington, DC: National Academy of Science Press, pp. 51 69. Gibbes, C., Adhikari, S., Rostant, L., and Southworth J. (2010). Utility of object oriented classification and high resolution satellite imagery for semi arid savanna ecosystem analysis. Remote Sensing 2, pp. 2748 2772. Gibbes, C., and Keys, E. (2010). The illusi on of equity: Community based natural resource management in southern Africa. Geography Compass 6, pp. 1 15. Gibbes, C., Southwort h, J., Waylen, P. and Cui, X. (u nder review ) Measuring the effect of global climate shifts on semi arid vegetation using a t ime series analysis approach. Nature Climate Change Gillanders, S., Coops, N., Wulder, M., Gergel, S. and Nelson T. (2009). Multitemporal remote sensing of landscape dynamics and pattern change: describing natural and anthropogenic trends. Progr. Phys. Ge ogr 33, pp. 528 546.
142 Gluch, R., Quattrochi, D. and Luvall, J. C. (2006). A multi scale approach to urban thermal analysis. Remote Sensing of Environment 104, pp. 123 132. Goheen, J., Young, T., Keesing, F. and Palmer, T. (2007). Consequences of herbivory by native ungulates for the reproduction of a savanna tree. J. Ecol 95, pp. 129 138. Goward, S., Tucker, C. and Dye, D. (1985). North American vegetation patterns observed with the NOAA 7 advanced very high resolution radiometer. Vegetatio 64, pp. 3 14. G oward, S. and Prince, S. (1995). Transient effects of climate on vegetation dynamics: Satellite observations. J. Biogeogr 22, pp. 549 563. Graham, N.E. (1994). Decadal scale climate variability in the 1970s and 1980s: observations and model results. Clim. Dyn 10, pp. 135 159. Green, G.M., Schweik, C.M. and Randolf, J.C. (2005). Retrieving land cover change information from Landsat satellite images by minimizing other sources of reflectance variability. In: Moran, E.F., Ostrom, E.(eds) Seeing the Forest an d the Trees: Human Environment Interactions in Forest Ecosystems MIT Press: Cambridge, MA, USA. Groen, T., van Langevelde, F., van de Vijver, C., Govender, N. and Prins, H. (2008). Soil clay content and fire frequency affect clustering in trees in Souther n African savannas. J. Tropical Ecol 24, pp. 269 279. Guerschman, J.P., Hill, M.J., Renzullo, L.J., Barrett, D.J., Marks, A.S. and Botha, E.J., (2009). Estimating fractional cover of photosynthetic vegetation, non photosynthetic vegetation and bare soil in the Australian tropical savanna region upscaling the EO 1 Hyperion and MODIS sensors. Remote Sensing of Environment 113, pp. 928 945. Gutman, G., Janetos, A., Justice, C., Moran, E. (2004). Land change science. Observing, monitoring and understanding tr surface New York: Kluwer Academic Publishers, 459 pp. Haberl, H., Erb, K., Krausmann, F., Gaube, V., Bondeau, A., Plutzer, A., Gringrish, S., Lucht, W. and Fischer Kowalski, M. (2007). Quantifying and mapping the human appropriation of net primary production in earth's terrestrial ecosystems. Proc Natl Acad Sci USA 104:12942 12947. Hall, O. and Hay, G. (2003). A multiscale object specific approach to digital change detection. Int. J. Appl. Earth Obs. Geoinf 4, pp. 311 3 27. Hanan, N. and Lehmann, C. (2010) Tree grass interactions in savannas: Paradigms, contradictions, and conceptual models. In: Hill, M. and Hanan, N. (eds.) Ecosystem Function in Savannas. Taylor and Francis Group, Boca Raton, Fl.
143 Hansen, A. and DeFries, R. (2007). Ecological mechanisms linking protected areas to surrounding lands. Ecological Applications 17, pp. 974 988. Hare, S. and Mantua, N.J. (2000). Empirical evidence for North Pacific regime shifts 1977 and 1989. Progress in Oceanography 47, pp. 103 145 Hartter, J. and Southworth, J. (2009). Dwindling resources and fragmentation of landscapes around parks: Wetlands and forest patches around Kibale National Park, Uganda. Landscape Ecol ogy 24, pp. 643 656. Hay, G., Blaschke, T., Marceau, D. J. and Bo uchard, A. (2003). A comparison of three image object methods for the multiscale analysis of landscape structure. Photogrammetry and Remote Sensing 1253, pp. 1 19. Hay, G., Castilla, G., Wulder, M. and Ruiz, J. (2005). An automated object based approach fo r the multiscale image segmentation of forest scenes. International Journal of Applied Earth Observation and Geoinformation 7, pp. 339 359. Hay, G., Dube, P., Bouchard, A. and Marceau, D. (2002). A scale space primer for exploring and quantifying complex l andscapes. Ecol. Model 153, pp. 27 49. Hay, G., Marceau, D., Bouchard, A. and Dube, P. (2001). A multiscale framework for landscape analysis: object specific upscaling. Landscape Ecology 16, pp. 471 490. Hayes, D., Cohen, W., Sader, S. and Irwin, D. (2008 ). Estimating proportional change in forest cover as a continuous variable from multi year MODIS data. Remote Sensing of the Environment 112, pp. 735 749. Holdo, R. M., Holt, R. D. and Fryxell, J. M. (2009). Opposing rainfall and plant nutritional gradient s best explain the wildebeest migration in the Serengeti. Amer. Naturalist 173, pp. 431 445. Hudak, A. and Wessman, C. (1998). Textural analysis of historical aerial photography to characterize woody plant encroachment in South African savanna. Remote Sens ing of Environment 66, pp. 317 330. Hudak, A. and Wessman, C. (2001). Textural analysis of high resolution imagery to quantify bush encroachment in Madikwe Game Reserve, South Africa, 1955 1996. International Journal of Remote Sensing 22, pp. 2731 2740. Hu ete, A., Didan, K., Miura, T., Rodriguez, E (2002). Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sensing of the Environment 830, pp. 195 213. Huete, A. R., Jackson, R. D. and Post, D. F. (1985). Spectral response of a plant canopy with different soil backgrounds, Remote sensing of the Environment 17, pp. 37 53.
144 Huete, A., Justice, C. and Liu, H. (1994). Development of vegetation and soil indices for MODIS EOS. Remote sensing of the Environment 49, pp. 224 234. Huttich, C., Gessner, U., Herold, M., Strohbach, B., Schmidt, M., Keil, M. and Dech, S. (2009). On the suitability of MODIS time series metrics to map vegetation types in dry savanna ecosystems: A case study in the Kalahari of NE Namibia. Remote Sens ing 1, pp. 620 643. Im, J., Jensen, J. R. and Hodgson, M. E. (2008). Object based land cover classification using high posting density LiDAR data. Giscience & Remote Sensing 45, pp. 209 228. Intergovernmental Panel on Climate Change (2007) Climate Change 2007: The physical science basis. Summary for policymakers. Contribution of working group I to the fourth assessment report. The Intergovernmental Panel on Climate Change. Iverson, L. R., Graham, R. L. and Cook, E. A. (1989). Applications of satellite remo te sensing to forested ecosystems. Landscape E cology 3, pp. 131 143. Jansen, L. and Di Gregorio, A. (2002). Parametric land cover and land use classifications tools for environmental change detection. Agriculture Ecosystems and Environment 91, pp. 89 100. Jensen, J. (2005). Introductory digital image processing: a remote sensing perspective Upper Saddle River, NJ: Pearson Prentice Hall. Jiang, Z., et al. (2006). Analysis of NDVI and scaled difference vegetation index retrievals of vegetation fraction. Remo te Sensing of Environment 101, pp. 366 378. Johnson, K. (1994). Segment based land use classification from spot satellite data. Photogrammetric Engineering and Remote Sensing 60, pp. 47 53 Johnson, R. and Tothill, J. (1985). Definition and broad geographi c outline of savanna lands. In Tothill, J. and Mott, J. (eds). Ecology and Management of the World's Savannas. Canberra, ACT: Aust. Acad. Sci. pp. 1 13. Jolly, W. and Running, S. (2004). Effects of precipitation and soil water potential on drought deciduou s phenology in the Kalahari. Global Change Biology 10, pp. 303 308. Justice, C. and Townshend, J. (2002). Special issue on the moderate resolution imaging spectroradiometer (MODIS): a new generations of land surface monitoring. Remote Sensing of the Enviro nment 83, pp. 1 2. Kasperson, J., Kasperson, R. and Turner, B. (1999). Regions at risk: Comparisons of threatened environments. United Nations University Press, Tokyo.
145 Katerere, Y., Hill, R. and Moyo, S. (2001). A Critique of Transboundary Natural Resource Management in Southern Africa, The IUCN ROSA Series on Transboundary Natural Resources Management Paper 1, IUCN: Nairobi, Kenya. Kauth, R. and Thomas, G. (1976). The tasseled cap a graphic description of the spectral temporal development of agricultural c rops as seen by Landsat. In: Proceedings of the symposium on machine processing of remotely sensed data. West Lafayette, IN: Purdue University, pp. 41 51. Kepner, W., et al. (2000). Landscape approach for detecting and evaluating change in a semiarid envir onment. Environmental Monitoring and Assessment 64, pp. 179 195. Koetz, B. et al. (2008). Multi source land cover classification for forest fire management based on imaging spectrometry and LiDAR data. Forest Ecology and Management 256, pp. 263 271. Kuemme rle, T., Damm, A. and Hostert, P. (2008). A method to detect and correct single band missing pixels in Landsat TM and ETM plus data. Computers & Geosciences 34, pp. 445 455. Laliberte, A., Fredrickson, E. and Rango, A. (2007). Combining decision trees with hierarchical object oriented image analysis for mapping arid rangelands. Photogramm. Eng. Remote Sensing 73, pp. 197 207. Laliberte, A., Rango, A., Havstad, K., Paris, J. (2004). Object oriented image analysis for mapping shrub encroachment from 1937 to 2 003 in southern New Mexico. Remote Sensing of Environment 93, pp. 198 210. Lambin E. (2003). Linking socioeconomic and remote sensing data at the community or at the household level: two case studies from Africa. In: Fox J., Rindfuss R.R., Walsh S. and Mi shra V. (eds), People and the Environment: Approaches for Linking Household and Community Surveys to Remote Sensing and GIS Kluwer Academic Publishers, Boston, pp. 223 240. Lambin, E. F. and Ehrlich, D. (1997). Land cover changes in sub saharan Africa (19 82 1991): application of a change index based on remotely sensed surface temperature and vegetation indices at a continental scale. Remote Sensing of Environment 61, pp. 181 200. Lambin, E. and Geist, H. (2005). Focus: Land use and Land cover change. IHDP Update Newsletter of the international Human Dimensions Programme on Global Environmental Change, No. 3, August 2005. Lambin, E. and Geist, H. (2006). Land use and land cover change. Local processes and global impacts. IGBP Series, New York : Springer.
146 Lambin, E., Turner, B., Geist, H., Agbola, S., Angelsen, A., Bruce, J., et al. (2001). The causes of land use and land cover change: moving beyond the myths, Global Environmental Change 11, pp. 261 269. Lanfredi, M., Simoniello, T. and Macchiato, M. (2004) .Temporal persistence in vegetation cover changes observed from satellite: Development of an estimation procedure in the test site of the Mediterranean Italy. Remote Sensing of Environment 93, pp. 565 576. Lecke, D. (1990). Synergism of synthetic aperture forest type discrimination. Photogrammetric Engineering and Remote Sensing 56 (9), pp. 1237 1247. Lefsky, M., Harding, D., Cohen, W., Parker, G. et al. (1999). Surface lidar remote sensing of basal area and biomass in deciduous forests of eastern Maryland, USA. Remote Sensing of Environment 67, pp. 83 98. Levin, S.A. (1992). The problem of pattern and scale in ecology. Ecology 63, pp. 1943 1967. Leyk, S. and Zimmermann, N. (2007). Improving land change detection based o n uncertain survey maps using fuzzy sets. Landscape Ecology 22, pp. 257 272. Liu, H. and Weng, Q. H. (2009). Scaling effect on the relationship between landscape pattern and land surface temperature: a case study of Indianapolis, United States. Photogramme tric Engineering and Remote Sensing 75, pp. 291 304. Loveland, T. R., Merchant, J. W., Ohlen, D. O. and Brown, J. F. (1991). Development of a land cover characteristics data base for the conterminous U.S. Photogrammetric Engineering and Remote Sensing 57, pp. 1453 1463. Lu, D., Mausel, P., Brondizio, E. and Moran, E. (2004). Change detection techniques. International Journal of Remote Sensing 25, pp. 2365 2407. Ludwig, F., De Kroon, H. and Prins, H. (2008). Impacts of savanna trees on forage quality for a large African herbivore. Oecologia 155, pp. 487 496. Ludwig, J., Wiens, J. and Tongway, D. A (2000). Scaling rule for landscape patches and how it applies to conserving soil resources in savannas. Ecosystems 3, pp. 84 97. Magadza, C. (1994). Climate change : some likely multiple impacts in Southern Africa. Food Policy 19, pp. 165 191. Mantua, N., Hare, S., Zhang, Y., Wallace, J. and Francis, R. (1997). A Pacific interdecadal climate oscillation with impacts on salmon production. Bulletin of the American Met eorological Society 78, pp. 1069 1079.
147 Markham, B. L. and Barker. J. L. (1986). Landsat MSS and TM post calibration dynamic rangers, exoatmospheric reflectance and at satellite temperatures. EOSAT Landsat Technical Notes 1, pp. 3 8. Martineu, C., Caneill, J. and Sadourny, R. (1999). Potential Predictability of European Winters from the Analysis of Seasonal Simulations with an AGCM. J. Climate 12, pp. 3033 3061. Martiny, N., Camberlin, P., Richard, Y. and Philippon, N. (2006). Compared regimes of NDVI and ra infall in semi arid regions of Africa. International Journal of Remote Sensing 27, pp. 5201 5223. Masek, J. G. (2001). Stability of boreal forest stands during recent climate change: evidence from Landsat satellite imagery. Journal of Biogeography 28, pp. 967 976. Masek, J., Huang, C., Wolfe, R., Cohen, W., et al. (2008). North American forest disturbance mapped from a decadal Landsat record. Remote Sensing of the Environment 112, pp. 2914 2926. Mason, S. (2001). El Nio, climate change and Southern African climate. Environmetrics 12, pp.327 345. Mason, S. and Jury, M.R. (1997). Climate variability and change over southern Afrsica: A reflection on underlying processes. Progress in Physical Geography 21, pp. 23 50. Mason, S.J. and Joubert, A.M. (1997). Simula ted Changes in Extreme Rainfall over Southern Africa. Int. J. Climatol 17, pp. 291 301. Mayaux, P., et al. (2007). Remote sensing of land cover and land use dynamics. In: Chuvieco, E. (ed s .) Earth observation of global change. The Netherlands: Springer, p p. 85 108. Mayes, S. (2008). Special Edition: Bwabwata National Park. Newsletter for the Strengthening the Protected Area Network (SPAN) Project. Sandpaper 10, pp. 6 9. McGlynn, I. and Okin, G. (2006). Characterization of shrub distribution using high spa tial resolution remote sensing: Ecosystem implications for a former Chihuahuan Desert grassland. Remote Sensing of Environment 4, pp. 554 566. Mendelsohn, J. and Roberts, C. (1997). An Environmental Profile and Atlas of Caprivi, Directorate of Envionmental Affairs, Ministry of Environment and Tourism: Windhoek, Namibia. Mertens, B. and Lambin, E. (2000). Land cover change trajectories in southern Cameroon. Annals of the Association of American Geographers 90, pp. 467 494.
148 Mertens, K. C., Verbeke, L. P. C., Ducheyne, E. I. and De Wulf, R. R. (2003). Using genetic algorithms in subpixel mapping. International Journal of Remote Sensing 24, pp. 4241 4247. Metzger, M., Rounsevell, M., Acosta Michlik, L., Leemans, R. and Schrter, D. (2006). The vulnerability of ecosystem services to land use change. Agric. Ecosyst Environ. 114, pp. 69 85. Meyer, K., Wiegand, K., Wards, D. and Moustakas, A. (2007). The rhythm of savanna patch dynamics. Journal of Ecology 95, pp. 1306 1315. Milton, S. and Dean, W. (1995). South Af they changing and can they be restored? Environ. Monitor. Assess 1995, 37, 245 264. Mitra, P., Shankar, B. and Pal, S. (2004). Segmentation of multispectral remote sensing images using active support vector mac hines. Pattern Recog. Lett 25, pp. 1067 1074. Mitri, G. H. and Gitas, I. Z. (2004). A performance evaluation of a burned area object based classification model when applied to topographically and non topographically corrected TM imagery. International Jou rnal of Remote Sensing 25, pp. 2863 2870. Mittermeier, R., Goettsch Mittermeier, C., Gil, P., Pilgrim, J., Fonseca, G., Brooks, T. and Konstant, W. (2003). University of Chicago Press: Chicago, IL, USA. Moleele, N., Ri ngrose, S., Arnberg, W., Lunden, B. and Vanderpost, C. (2001). Assessment of vegetation indexes useful for browse (forage) prediction in semi arid rangelands. Int. J. Remote Sens 22, pp. 741 756. Moran, E. and Ostrom, E. (2005). Seeing the forest and the trees: Human environment interactions in forest ecosystems MIT Press, Cambridge, Massachusetts, USA. Morisette, J. Privette, J., Justice, C. (1999). The EOS Land Validation Core Sites: background information and current status. Earth Observer 11, pp. 21 2 5. Moustakas, A., Guenther, M., Wiegand. K., Mueller, K., Ward, D., Meyer, K. and Jeltsch, F. (2006). Long term mortality patterns of the deep rooted Acacia erioloba: The middle class shall die. J. Veg. Sci 17, pp. 473 480. Munroe, D., Nagendra, H. and So uthworth, J. (2007). Monitoring landscape fragmentation in an inaccessible mountain area: Celaque National Park, Western Honduras. Landscape Urban Plan 83, pp. 154 167.
149 Myint, S. W. and Okin, G. (2009). Modelling land cover types using multiple endmember spectral mixture analysis in a desert city. International Journal of Remote Sensing 30, pp. 2237 2257. Nagendra, H. (2008). Do parks work? Impact of protected areas on land cover clearing. Journal of Human Environment 37, pp. 330 337. Nagendra, H., Rocchin i, D., Ghate, R., Sharma, B. and Pareeth, S. (2010). Assessing plant diversity in a dry tropical forest: Comparing the utility of Landsat and IKONOS satellite images. Remote Sens ing 2, pp. 478 496. Nagendra, H., Tucker, C., Carlson, L., Southworth, J., Kar macharya, M. and Karna, B. (2007). Monitoring parks through remote sensing: Studies in Nepal and Honduras. Environ. Manag 34, pp. 748 760. Namias, J. (1978). Multiple Causes of the North American Abnormal Winter 1976 77. Mon. Wea. Rev ., 106, pp. 279 295. Nicholson S E (2000). The nature of rainfall variability over Africa on time scales of decades to millenia. Global Planet Change Letters 26, pp.137 15 Nicholson, S.E. and Farrar, T.J. (1994). The influence of soil type on the relationships between NDVI, rainfall and soil moisture in semiarid Botswana I. NDVI response to rainfall. Remote Sensing of the Environ ment 50, pp. 107 120. Nicholson, S.E., Leposo, D. and Grist, J. (2001). The relationship between El Nio and drought over Botswana. Journal of Clima te 14 pp. 323 335. Ochoa Gaona, S. and Gonzalez Espinosa, M. (2000). Land use and deforestation in the highlands of Chiapas, Mexico. Applied Geography 20, pp. 17 42. Ord, J. K. and Getis, A. (1995). Local spatial autocorrelation statistics: Distributional issues and an application. Geogr. Anal 27, pp. 286 306. Patenaude, G., Hill, R., Milne, R., Gaveau, D., et al. (2004). Quantifying forest above ground carbon content using LiDAR remote sensing. Remote Sensing of the Environment 93, pp. 368 380. Peddle, D. Johnson, R., Cihlar, J. and Latifovic, R. (2004). Large area forest classification and biophysical parameter estimation using the 5 Scale canopy reflectance model in Multiple Forward Mode. Remote Sensing of the Environment 89, pp. 252 263. Peddle, D. R. and Franklin, S. E. (1991). Image texture processing and data integration for surface pattern discrimination. Photogrammetric Engineering and Remote Sensing 57, pp. 413 420.
150 I remote sensing based methods for the retrieval of land surface energy fluxes and soil surface moisture. Progress in Physical Geography 33, pp. 224 250. Pickup, G. and Foran B. D (1987). The use of spectral and spatial variability to monitor cover chang e on inert landscapes. Remote Sensing of Environment 23, pp.361 363. Pimm, S. and Raven, P. (2000). Biodiversity: Extinction by numbers. Nature 403, pp. 843 845. Pirot, J., Meynell, P. and Elder, D. (2000). Ecosystem based management: Principles. In : Piro t, J., Meynell, P., Elder, D. (eds) Ecosystems Management Lesson from Around the World: A Guide for Development and Conservation Practitioners IUCN Publications Services Unit: Gland, Switzerland and Cambridge, UK. Platt, R. and Rapoza, L. (2008). An evalu ation of an object oriented paradigm for land The Professional Geographer 60, pp. 87 100. Pohl, C. and Van Genderen, J. (1998). Multisensor image fusion in remote sensing: concepts, methods and applications. International J ournal of Remote Sensing 19 (5), pp. 823 854. participatory learning and interaction. IIED Training Series (No. 2). International Institute for Environment and Development, Lo ndon, UK. Qin, Z., Karnieli, A. and Berliner, P. (2001). A mono window algorithm for retrieving land surface temperature from Landsat TM data and its application to the Israel Egypt border region. International Journal of Remote Sensing 22, pp. 3719 3746. Quattrochi, D. A. and Ridd, M. K. (1994). Measurement and analysis of thermal energy responses from discrete Urban surfaces using remote sensing data. International Journal of Remote Sensing 15, pp. 1991 2022. Ramankutty, N. and Foley, J. (1999). Estimatin g historical changes in global land cover: Croplands from 1700 to 1992. Global Biogeochemical Cycles 13, pp. 997 1027. Ramankutty, N., Graumlich, L., Achard, F., Alves, D., Chhabra, A., et al. (2006). Global land cover change: Recent progress, remaining ch allenges. In Lambin, E. F. and Geist, H. J. (eds). Land use and land cover change. Local processes and global impacts. IGBP Series, New York: Springer pp. 9 12. Ramsey, E., Lu, Z., Rangoonwala, A. and Rykhus, R. (2006). Multiple baseline radar interferome try applied to coastal land cover classification and change analyses. GIScience & Remote Sensing 43, pp. 283 309.
151 Reason, C. J. and Mulenga, H. M. (1999). Relationships between South African rainfall and SST anomalies in the southwest Indian Ocean. Interna tional Journal of Climatology 19, 1651 1673 Reynolds, J., Smith, D., Lambin, E., Turner, B. (2007). Global desertification: building a science for dryland development. Science 316, pp. 847 851. Rice, M. (1997). Integrated Rural Development and Nature Con servation CBNRM Project in Caprivi Region of Namibia, Unpublished Report. Richard, Y. and Poccard, I. (1998). A statistical study of NDVI sensitivity to seasonal and interannual rainfall variations in Southern Africa. International Journal of Remote Sensin g 19, pp. 2907 2920. Rignot, E., Salas, W. and Skole, D. (1997). Mapping deforestation and secondary growth in Rondonia, Brazil, using imaging radar and thematic mapper data. Remote Sensing of Environment 59 pp. 167 179. Ringrose, S., Kampunzu, A., Vink, B ., Matheson, W. and Downey, W. (2002). Origin and palaeo environments of calcareous sediments in the Moshaweng dry valley, southeast Botswana. Earth Surf. Process. Landf 27, pp. 591 611. Ringrose, S. and Matheson, W. (1987). Spectral assessment of indicat ors of range degradation in the Botswana hardveld environment. Remote Sensing of Environment 23, pp. 379 386. Ringrose, S., Matheson, W., Mogosti, B. and Tempest, F. (1989). Nature of the darkening effect in drought affected savanna woodland environments r elative to soil reflectance in Landsat and SPOT wavebands. Remote Sensing of Environmen t 25, pp. 519 524. Ringrose, S., Matheson W. and Vanderpost, C. (1998). Analysis of soil organic carbon and vegetation cover trends along the Botswana Kalahari Transect J. Arid Environ 38, 379 396. Ritchie, J. C. (1996). Remote sensing applications to hydrology: airborne laser altimeters. Hydrological Sciences Journal Journal Des Sciences Hydrologiques 41, pp. 625 636. Rocchini, D. (2007). Effects of spatial and spect r diversity by satellite imagery. Remote Sensing of Environment 111, pp. 423 434. Roughgarden, J., Running, S. and Matson, P. (1991). What does remote sensing do for ecology? Ecology 72, pp. 1918 1922. Rouse, J. W., H aas, R. H., Schell, J. A. and Deering, D. (1974). Monitoring vegetation Systems in the Great Plains with ERTS. Proceedings of the Third Earth Resources Technology Satellite 1 Symposium, NASA SP 351 pp. 3010 317.
152 Roy, D., Ju, J., Lewis, P., Schaaf, C., et al. (2008). Multi temporal MODIS Landsat data fusion for relative radiometric normalization, gap filling, and prediction of landsat data. Remote Sensing of Environment 112, pp. 3112 3130. Rozemeijer, N. and Van der Jagt, C. (2000). Community based Natural Resource Management in Botswana: In: Shack leton, S., Campbell, B.M. (eds). Empowering Communities to Manage Natural Resources: Case Studies from Southern Africa SADC Wildlife Sector Natural Resources Management Programme and CSIR: Lilongwe, Malawi. Rudel, T. (2008). Meta analyses of case studies: a method for studying regional and global environmental change. Global Environmental Change Human Policy and Dimensions 18, pp. 18 25. Ruiz Luna, A. and Berlanga Robles, C. (2003). Land use, land cover changes a nd coastal lagoon surface reduction associated with urban growth in northwest Mexico. Landscape Ecology 18, pp. 159 171. Sankaran, M., Hanan, N., Scholes, R., et al. (2005). Determinants of woody cover in African savannas. Nature 438, pp. 846 849. Sarkar, S., et al. (2009). A dynamic graph automata approach to modeling landscape change in the Andes and the Amazon. Environment and Planning B Planning & Design 36, pp. 300 318. Scanlon, T., Caylor, K., Manfreda, S., Levin, S. and Rodriguez Iturbe, I. (2005). Dynamic response of grass cover to rainfall variability: Implications for the function and persistence of savanna ecosystems. Adv. Water Resour. 28, pp. 291 302. Scanlon, T., Caylor, K., Levin, S. and Rodreiguez Iterbe, I. (2007). Positive feedbacks promo te power law clustering of Kalahari vegetation. Nature 449, pp. 209 212. Scarpa, G., Gaetano, R., Haindl, M. and Zerubia, J. (2009). Hierarchical multiple markov chain model for unsupervised texture segmentation. IEEE Transactions on Image Processing 18, p p. 1830 1843. Schmidt, H. and Karnieli A. (2000). Remote Sensing of the seasonal variability of vegetation in a semi arid environment. J. Arid Environ. 45, pp. 43 59. Schneider, W. and Steinwender, J. (1999). Landcover mapping by interrelated segmentation and classification of satellite images. Int. Arch. Photogramm. Remote Sens. 32, Part 7 4 3. Scholes, R.J. and Archer, S. R. (1997).Tree grass interactions in savanna. Ann. Rev. Ecol. Systemat. 28 pp. 517 544.
153 Scholes, R. J. and Walker, B. H. (1993). An A frican Savanna: Synthesis of the Nylsvley Study, Cambridge University Press: Cambridge, UK, pp. 264 267. Schowengerdt, R. A. (1996). On the estimation of spatial spectral mixing with classifier likelihood functions. Pattern Recognition Letters 17, pp. 1379 1387. Schweik, C. and Green, G. M. (1999). The use of spectral mixture analysis to study human incentives, actions and environmental outcomes. Social Science Computer Review 17, pp. 40 63. Schweik, C., Nagendra, H. and Sinha, D. (2003). Using satellite im agery to locate innovative forest management practices in Nepal. Ambio 32, pp. 312 319. Sekhwela, M. and Yates, D. (2007). A phenological study of dominant acacia tree species in areas with different rainfall regimes in the Kalahari of Botswana. J. Arid En viron. 70, pp. 1 17. Sellers, P. J. (1985). Canopy reflectance, photosynthesis and transpiration. International Journal of Remote Sensing 6, pp. 1335 1372. Sha, Z., Bai, Y., Xie, Y., Yu, M., et al. (2008). Using a hybrid fuzzy classifier (HFC) to map typic al grassland vegetation in Xilin River Basin, Inner Mongolia, China. International Journal of Remote Sensing 29, pp. 2317 2337. Shackelford, A. and Davis, C. (2003). A combined fuzzy pixel based and object based approach for classification of high resoluti on multispectral data over urban areas. IEEE Transactions on Geoscience and Remote Sensing 41, pp. 2354 2363. Shackleton, C. M. (1999). Rainfall and topo edaphic influences on woody community phenology in South African savannas. Glob. Ecol. Biogeogr 8, pp 125 136. Shi, G., Ribbe, J., Cai, W. and Cowan, T. (2007). Multidecadal variability in the transmission of ENSO signals to the Indian Ocean. Geophysical Research Letters 34, L09706, doi10.1029/2007GL29528 Simard, M., Saatchi, S. and Grandi, G. D. (2000) The use of decision tree and multiscale texture for classification of JERS 1 SAR data over tropical forest. IEEE Transaction on Geoscience and Remote Sensing 38, pp. 2310 2321. Smith, G. M. and Fuller, R. M. (2001). An integrated approach to land cover c lassification: an example in the Island of Jersey. International Journal of Remote Sensing 22, pp. 3123 3142. Sobrino, J. A., Jimenez Munoz, J. C. and Paolini, L. (2004). Land surface temperature retrieval from LANDSAT TM 5. Remote Sensing of Environment 9 0, pp. 434 440.
154 Soudani, K., Francois, C., le Maire, G., Le Dantec, V., et al. (2006). Comparative analysis of IKONOS, SPOT, and ETM+ data leaf area index estimation in temperature coniferous and deciduous forest stands. Remote Sensing of Environment 102, pp. 161 175. South Dakota School of Mines and Technology MODIS Team. (2004). MODIS reprojection tool user's manual release 3.2a, July 2004, pp. 1 58. Available at http://lpdaac.usgs.gov/l anddaac/tools/modis/index.asp Southworth, J. (2004). An assessment of Landsat TM band 6 thermal data for analysing land cover in tropical dry forest regions. International Journal of Remote Sensing 25, pp. 689 706. Southworth, J. and Gibbes, C. (2010) Di gital remote sensing within the field of land changce science: Past, present and future directions. Geography Compass 4, pp. 1695 1712. Southworth, J., Nagendra, H., Carlson, L. and Tucker, C. (2004). Assessing the impact of Celaque National Park on forest fragmentation in western Honduras. Applied Geography 24, pp. 303 322. Stefanov, W. L., Ramsey, M. S. and Christensen, P. R. (2001). Monitoring urban land cover change: an expert system approach to land cover classification of semiarid to arid urban center s. Remote Sensing of Environment 77, pp. 173 185. Stige, L., Stave, J., Chan, K., Ciannelli, L., Pettorelli, N., Glantz, M. and Herren, H. (2006). The effect of climate variation on agro pastoral production in Africa. Proc Natl Acad Sci USA 103, pp. 3049 3053. Stickler, C. and Southworth, J. (2008). Application of multi scale spatial and spectral analysis for predicting primate occurrence and habitat associations in Kibale National Park, Uganda. Remote Sensing of Environment 112, pp. 2170 2186. Still, C. J ., Berry, J. A., Collatz, G. J. and DeFries, R. S. (2003). Global distribution of C3 and C4 vegetation: Carbon cycle implications. Glob. Biogeochem. Cycl. 17, pp.1006 1021 Swanson, K. L. and Tsonis, A. (2009). Has the climate recently shifted? Geophysical Research Letters 36, L06711, doi:1029/2008GL037022 Tang, X., Kainz, W. and Fang, Y. (2005). Reasoning about changes of land covers with fuzzy settings. International Journal of Remote Sensing 26, pp. 3025 3046. Thenkabail, P. (2004). Inter sensor relation ships between IKONOS and Landsat 7 ETM+ NDVI data in three ecoregions of Africa. International Journal of Remote Sensing 25, pp. 389 408.
155 Thenkabail, P., Enclona, E., Ashton, M., Legg, C., et al. (2004). Hyperion, IKONOS, ALI, and ETM plus sensors in the s tudy of African rainforests. Remote Sensing of Environment 90, pp. 23 43. Thompson, M. (1996). A standard land cover classification scheme for remote sensing applications in South Africa. South African Journal of Science 92, pp. 34 42. Topp, C. and Mitchel l, M. (2003). Forecasting the environmental and socioeconomic consequences of changes in the Common Agricultural Policy. Agricultural Systems 76, pp. 227 252. Torres Vera, M., Prol Ledesma, R. and Garcia Lopez, D. (2009). Three decades of land use variatio ns in Mexico City. International Journal of Remote Sensing 30, pp. 117 138. Townshend, J. and Justice, C.O. (1986). Analysis of the dynamics of African vegetation using the normalized difference vegetation index. International Journal of Remote Sensing 7, pp. 1435 1445. Trenberth, K. E. (1990). Recent Observed Interdecadal Climate Changes in the Northern Hemisphere', Bull. Amer. Meteor. Soc 71, 988 993. Trenberth, K. E. and Hurrell, J. (1994). Decadal Atmosphere Ocean Variations in the Pacific', Clim. Dyn 9, pp. 303 319. Tucker, C. (1979). Red and photographic infrared linear combinations for monitoring vegetation, Remote Sensing of Environment 8, pp. 127 150. Tucker, C., Pinzon, J. E., Brown, M. E., Slayback, D., et al. (2005). An extended AVHRR 8km NDVI dataset compatible with MODIS and SPOT Vegetation NDVI data. International Journal of Remote Sensing 26 (20), pp. 4485 4498. Turner B., Kasperson R., Matson, P., McCarthy, J. et al. (2003). A framework for vulnerability analysis in sustainability science Proc Natl Acad Sci USA 100, pp. 8074 8079. Turner, B., Lambin, E. and Reenberg, A. (2007). The emergence of land change science for global environmental change and sustainability. Proc Natl Acad Sci USA 104 (52), pp. 20666 20671. Turner, M. (1989). Land scape ecology: the effect of pattern on process. Ann. Rev. Ecol Systemat. 20, pp. 171 197. Turner, M. (2003). Methodological reflections on the use of remote sensing and GIS in human ecological research. Human Ecology 31, pp. 255 279.
156 Turner, W., Spector, S., Gardiner, N., Fladeland, M., et al. (2003). Remote sensing for biodiversity science and conservation. Trends in Ecology and Evolution 18, pp. 306 314. United Nations Environment Programme (2002). Africa Environment Outlook: Past Present and Future Pe rspectives, United Nations Environment Programme (UNEP)/GRID Arendal, EarthPrint: England, UK. van Aarde, R. and Jackson, T. (2007). Megaparks for metapopulations: Addressing the causes of locally high elephant numbers in southern Africa. Biol Conserv 13 4, pp. 289 297. Vanacker, V., Linderman, M., Lupo, F., Flasse, S. and Lambin, E. (2005). Impact of shortterm rainfall fluctuation on interannual land cover change in sub Saharan Africa. Global Ecology and Biogeography 14, pp. 123 135. Verburg, P. H., Soepb oer, W., Limpiada, R., Espaldon, M. V. O., Sharifa, M. and Veldkamp, A. (2002). Land use change modelling at the regional scale: the CLUE S model. Environmental Management 30, pp. 391 405. Verburg, P., van de Steeg, J., Veldkamp, A. and Willemen, L. (2009) From land cover change to land function dynamics: A major challenge to improve land characterization. Journal of Environmental Management 90, pp. 1327 1335. Vieira, I. C. G., de Almeida, A. S., Davidson, E. A., Stone, T. A., de Carvalho, C. J. R. and Gue rrero, J. B. (2003). Classifying successional forests using Landsat spectral properties and ecological characteristics in Eastern Amazonia. Remote Sensing of the Environment 87, pp. 470 481. Vitousek, P., Mooney, H., Lubchenco, J.and Melillo J. (1997). Hum an domination of Science 277, pp. 494 500. Wald, L., Ranchin, T. and Mangolini, M. (1997). Fusion of satellite images of different spatial resolution: assessing the quality of resulting images. Photogrammetric Engineering and Remote Sen sing 63, pp. 691 699. Walker, B. H. (1976). An approach to the monitoring of changes in the composition and utilization of woodland and savanna vegetation. South. Afr. J. Wildlife Res 6, pp. 1 32. Walsh, S., Messin, J., Mena, C., Malanson, G. and Page, P. (2008). Complexity theory, spatial simulation models, and land use dynamics in the Northern Ecuadorian Amazon. Geoforum 39, pp. 867 878. Walter, V. (2004). Object based classification of remote sensing data for change detection. ISPRS Journal of Photogram metry and Remote Sensing 58, pp. 225 238.
157 Wang, F. (1990). Fuzzy supervised classification of remote sensing images. IEEE Trans. Geosci. Remote Sens 28 pp. 194 201. Wang, J., Rich, P., Price, K. and Kettle, W. (2005). Relations between NDVI, grassland p roduction, and crop yield in the central Great Plains. Geocarta International 20, pp. 5 11. Wang, L., Sousa, W. and Gong, P. (2004). Integration of object based and pixel based classification for mapping mangroves with IKONOS imagery. International Journal of Remote Sensing 25, pp. 5655 5668. Washington Allen, R. A., Ramsey, R. and West, N. E. (2004a). Spatiotemporal mapping of the dry season vegetation response of sagebrush steppe. Community Ecology 5, pp. 69 79. Washington Allen, R. A. Ramsey, R., West, N E. and B. E. Norton, B.E. (2008) Quantification of the ecological resilience of drylands using digital remote sensing. Ecology and Society 13, pp. 33. Washington Allen, R. A., Van Niel, T. G., Ramsey, R. D. and West, N. E. (2004b). Remote sensing based p iosphere analysis. GIScience and Remote Sensing 41, pp. 136 154. Washington Allen, R. A., West, N.E. and Ramsey, R. (2003). Remote sensing based dynamical systems analysis of sagebrush steppe vegetation in rangelands. In Allsopp, N., Palmer, A., Milton, S. et al. (eds). Proceedings of the VIIth International Rangelands Congress. Durban, South Africa. pp. 416 418. Washington Allen, R. A., West, N. E., Ramsey, R and Efroymson, R. A. (2006). A protocol for retrospective remote sensing based ecological monitor ing of rangelands. Rangeland Ecology and Management 59, pp. 19 29. Watanachaturaporn, P., Varshney, P. K. and Arora, M. K. (2004). Evaluation of factors affecting support vector machines for hyperspectral classification. American Society for Photogrammetry and Remote Sensing (ASPRS), 2004 Annual Conference, Denver, CO. Weisberg, P. J., Lingua, E. and Pillai, R. B. (2007). Spatial patterns of pinyon Juniper woodland expansion in Central Nevada. Rangel. Ecol. Manag 60, pp. 115 124. Weltzin, J. and Coughenour M. (1990). Savanna tree influence on understory vegetation and soil nutrients in northwestern Kenya. J. Veg. Sci 1, pp. 325 332. Weng, Q. (2002). Land use change analysis in the Zhujiang Delta of China using satellite remote sensing, GIS and stochastic modeling. Journal of Environmental Management 64, pp. 273 284.
158 Wessels, K., De Fries, R., Dempewell, J., Anderson, L., et al. (2004). Mapping regional land cover with MODIS data for biological conservation: examples from the Greater Yellowstone Ecosystem, USA and Para State, Brazil. Remote Sensing of Environment 92, pp. 67 83. Western, D. (1989). The ecological role of elephants in Africa. Pachyderm 12, pp. 42 45. We J. (1986). Measures of resilience: the response of coastal sage sc rub to fire. Vegetatio 65, pp. 179 189. Whyte, I. J., van Aarde, R. J. and Pimm, S. size and consequences for ecosystem heterogeneity. In: du Toit, J.T., Rogers, K.H., Biggs, H.C. (eds) The Kruger Experience: Ec ology and Management of Savanna Heterogeneity. Island Press: Washington, DC, USA, pp. 332 348 Wiens, J. A. Spatial scaling in ecology. Funct. Ecol. 1989, 3, 385 397. Williams, W., Hanan, N., Neff, J., Scholes, R., Berry, J., Denning, A. and Baker, D. (2007 ). Africa and the global carbon cycle. Carbon Balance Manag 2, pp.1 13. Wirbelaeur, C., Mosimane, A., Mabunda, R., Makota, C., Khumalo, A. and Nanchengwa, M. (2007). A preliminary assessment of the natural resources management capacity of community based organizations in Southern Africa. Cases from Botswana, Mozambique, Namibia, Zambia, and Zimbabwe. WWF SARPO Occasional Paper 17. Harare, Zimbabwe: WWF SARPO. Wulder, M. (1998). Optical remote sensing techniques for the assessment of forest inventory and bi ophysical parameters. Progress in Physical Geography 22, pp. 449 476. Wulder, M., Hall, R., Coops, N. and Franklin, S. (2004). High spatial resolution remotely sensed data for ecosystem characterization. BioScience 54, pp. 511 521. Xie, Y., Sha, Z. and Yu, M. (2008). Remote sensing imagery in vegetation mapping: a review. Journal of Plant Ecology Uk 1, pp. 9 23. Xu, H. Q. and Chen, B. (2004). Remote sensing of the urban heat island and its changes in Xiamen City of SE China. Journal of Environmental Scienc es China 16, pp. 276 281. Yang, L., Xian, G., Klaver, J. M. and Deal, B. (2003). Urban land cover change detection through sub pixel imperviousness mapping using remotely sensed data. Photogrammetric Engineering and Remote Sensing 69, pp. 1003 1010. Zen, N and Neelin, D. (2000). The role of vegetation climate interaction and interannual variability in shaping the African savanna. Journal of Climate 13, pp. 2665 2670.
159 Zhang, J. Wang, Y., Wang, Z. (2007). Change analysis of land surface temperature based on robust statistics in the estuarine area of Pe arl River (China) from 1990 Journal of Remote Sensing 28, pp. 2383 2390.
160 BIOGRAPHICAL SKETCH Cerian Gibbes graduated from the Bolles School in Jacksonville Florida in 2000. During h er undergraduate career (20 00 2004) at the University of Florida curiosity about the environment and the interactions between society and the physical landscape shifted her focus to geographic studies. Upon graduation Cerian began her graduate career at the University of Florida, working with Dr. Jane Southworth on land use land cover change topics in the Caribbean. Her M S c. research was supported by funding from US AID, a Ruth McQuown Scholarship, the Fik Summer Research Award, and the Dunkle Fellowship. Acquiring her M. S c degree in geography with a minor in wildlife ecology in December 2006, she continued her graduate career at the University of Florida p urs u ing her Ph .D As a Ph D student Cerian taught physical g eography and was a teaching assistant for Remote Sensing. Her doctoral research focused on land use land cover change in southern African and was funded by monies from the National Aeronautics and Space Administration (NASA), the Social Science Research Council (SSRC), College of Liberal Arts and Science, and th e Center for African Studies. Cerian completed her Ph D in Spring 2011 with a minor in wildlife e cology and a concentration in tropical conservation and m anagement.