|UFDC Home||myUFDC Home | Help|
This item has the following downloads:
1 VARIABILITY AND CHANGE IN A DRYLAND SYSTEM: CLIMATE LAND INTERACTION IN THE OK A VANGO KWAN D O ZAMBEZI CATCHMENT OF SOUTHERN AFRICA By ANDREA E. GAUGHAN A DISSERTATION PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 201 1
2 201 1 Andrea E. Gaughan
3 To my parents, Ron and Karen Brown
4 ACKNOWLEDGMENTS I would like to thank my advisor, Dr. Michael W. Binford. I never fully appreciated the decision of choosing an advisor until I was well underway with my studies and I am thankful for the relationship and store of knowledge I acquired while working with Dr. Binford. I also recognize and thank the rest of my committee for providing support and filling in the gaps for other areas of expertise Dr. Jane Southworth provide d practical advice and spoke to the larger picture of the dissertation. Dr. Peter Waylen generously gave of his time to help me understand the more statistical and climatological aspects of my study. Dr. Brian Child first introduced me t o the study region back in 2006 through both the questions of savanna ecology and also the balance of conservation and development in rural southern Africa. Dr. Greg Kiker showed me the power of modeling to simultaneously answer scientific question s and provide information f or communities and agencies. Additionally, w ithout the help of colleagues and friends, in the field and in the lab, I would not have be en able to complete this study. While there are many to whom I am thank ful I would specifically like to acknowledge Luk e Rostant, Cerian Gibbes, Jg Collomb, Forrest Stevens and Caroline Staub. I would also like to say thank you to those colleagues from Africa who without their assistance, I would not have been able to do field work. R ichard Diggle and Simon Ma yes especial ly helped coordinate logistics with communities and the Ministry for Environment and Tourism. Alfons Mosimane and John Mfune from University of Namibia were very gracious with their time and expertise both in Caprivi and Windhoek. Toivo Uahengo was instrum ental in helping obtain research permits each field season. And to those at the Okavango Research Institute (especially Mike Murray Hudson, Lin Cassidy, and Piotr Wolski) who took the time to
5 introduce the study region both the physical and social components and, by doing so, contributed towards my overall understanding of the area Most importantly, I must recognize the one person who made much of the field work possible Bennety Likukela. Bennety opened doors into the world of savanna ecology and rural African livelihoods. The field seasons would have been lacking without his assistance for the environmental history interviews, his expert knowledge in savanna vegetation, and his knack to interact with the local peo ple. L astly I would like to thank my family, especially my husband, Patrick Gaughan, who has been nothing but support ive in my on going educational journey. This work has been supported by the following grants and organizations: NSF IGERT Adaptive Managem ent, Water, Wetlands, and Watersheds (No. 0504422) NSF study entitled Parks as Agents of Social and Environmental Change in Eastern and Southern Africa (No. 0624343) and a NASA Land Cover/Land Use Change grant (No. NNX09A125G), Center for African Studies and the NSF IGERT Working Forest in the Tropics.
6 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 8 LIST OF FIGURES ................................ ................................ ................................ .......... 9 ABSTRACT ................................ ................................ ................................ ................... 11 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 13 General Introduction ................................ ................................ ............................... 13 Overarching Research Question and Study Objectives ................................ .......... 14 Linking Objectives to Theory ................................ ................................ ................... 15 Climate land Interactions for Coupled Human Environment Systems .............. 15 Shifting Dynamics of Dryland Environments ................................ ..................... 16 Effects of Precipitation on Dryland Ecosystems ................................ ............... 17 Spatial and Temporal Precipitation Patterns, 1950 2005 ................................ ........ 17 Vegetation Response to Seasonal Precipitation ................................ ..................... 19 Detecting Seasonality in Dryland Savannas ................................ ........................... 20 Importance of the Study ................................ ................................ .......................... 22 2 SPATIAL AND TEMPORAL PRECIPITATION VARIABILITY IN THE OKAVANGO KWANDO ZAMBEZI CATCHMEN T, SOUTHERN AFRICA .............. 24 Materials and Methods ................................ ................................ ............................ 27 Study Region ................................ ................................ ................................ .... 27 Datasets ................................ ................................ ................................ ........... 28 Statistical Examination of Precipitation Patterns ................................ .............. 29 Results ................................ ................................ ................................ .................... 32 Climate Shift ................................ ................................ ................................ ..... 32 Spatial and Te mporal Patterns ................................ ................................ ......... 33 Basin Level Analysis ................................ ................................ ........................ 35 Cell by Cell Analysis ................................ ................................ ......................... 36 Local versus Regional Input ................................ ................................ ............. 38 Discussion ................................ ................................ ................................ .............. 39 Summary ................................ ................................ ................................ ................ 44 3 LINKING VEGETATION RESPONSE TO SEASONAL PRECIPITATION IN the OKAVANGO KWANDO ZAMBEZI CATCHMENT OF SOUTHERN AFRICA ......... 59 Materials and Methods ................................ ................................ ............................ 65 Study Region ................................ ................................ ................................ .... 65
7 Datasets ................................ ................................ ................................ ........... 66 Data Sampling and Standardization ................................ ................................ 67 Modeling Approach ................................ ................................ .......................... 68 Model Comparison across Vegetation Classes ................................ ................ 70 Results ................................ ................................ ................................ .................... 71 Spatial and Temporal Relationship between NDVI and Rainfall ....................... 71 Model Results ................................ ................................ ................................ ... 72 GWR Results by Land Cover ................................ ................................ ........... 73 Discussion ................................ ................................ ................................ .............. 75 Summary ................................ ................................ ................................ ................ 79 4 DETECTING LONG TERM VEGETATION TRENDS BY ADJUSTING FOR SEASONALITY IN A SAVANNA LANDSC APE ................................ ...................... 88 Materials and Methods ................................ ................................ ............................ 94 Study Region ................................ ................................ ................................ .... 94 Datasets ................................ ................................ ................................ ........... 95 Landsat TM d ata ................................ ................................ ........................ 95 Rainf all d at a ................................ ................................ ............................... 97 Precipitation corrected NDVI ................................ ................................ ............ 98 Long term Land Cover Change in the KCA ................................ .................... 100 Results ................................ ................................ ................................ .................. 101 PC NDVI Compared to Observed NDVI ................................ ......................... 101 Tempor al Clusters of PC NDVI ................................ ................................ ...... 102 PC NDVI Change ................................ ................................ ........................... 103 Discussion ................................ ................................ ................................ ............ 104 Summary ................................ ................................ ................................ .............. 109 5 CONCLUSION ................................ ................................ ................................ ...... 117 Overall Findings ................................ ................................ ................................ .... 117 Significance of Findings ................................ ................................ ........................ 119 LIST OF REFERENCES ................................ ................................ ............................. 120 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 138
8 LIST OF TABLES Table page 2 1 ENSO frequency across a) dry years from 1950 2005 and b) wet years from 1950 2005. ................................ ................................ ................................ ......... 54 2 2 ENSO frequency across a) dry years and b) wet years for P1 (1950 1975) and P2 (1980 2005). ................................ ................................ ........................... 54 2 3 Chi Square results for precipitation input within catchments. ............................. 57 2 4 Observed relative frequencies between local and regional precipitation input ... 58 3 1 The global OLS model diagnostics compared to those of the GWR. ................. 85
9 LIST OF FIGURES Figure page 2 1 Study region in southern Africa outlining the Okav ango Kwando Zambezi catchment. ................................ ................................ ................................ ......... 46 2 2 Cell by cell 10 year moving window analysis ................................ ..................... 47 2 3 Time series of in the OKZ catchments in southern Africa from 1950 2005 with ENSO warm and cold phases identified. ................................ ............................ 48 2 4 Spatial variability across the OKZ basin calculated as the coefficient of variation (CV). ................................ ................................ ................................ ..... 4 9 2 5 The tw o top images indicate an aggregated depic tion of median annual rainfall and t he bottom image shows the difference (P2 P1). .......................... 50 2 6 The two top images indicate areas of large change between E NSO phases (La Nia El Nio) and the bottomr image shows the difference (P2 P1) ....... 51 2 7 Box plot of average annual precipitation input for Period 1: 1950 1975 (white) compared to Period 2: 1980 2005 (gray) for each OKZ sub catchment. ............ 52 2 8 F requency of upper and lower climatological third of precipitation data indicating the number of a) dry and b) wet years for each time period ............... 53 2 9 Cells reporting si gnificant associations between warm phase ENSO events and "dry" years d uring a dry year for P1 and P 2 ................................ ............... 55 2 10 Cells reporting significant associations between warm phase (panels a and b) and cold .................... 56 3 1 Study region depicting the three catchment areas of interest ............................. 81 3 2 of different land covers in the OKZ catchment. ................................ ................... 82 3 3 total mean wet season precipitation respectively from 2000 2009. .................... 83 3 4 Time series of averaged April NDVI for the OKZ catchment plotted against the total average precipitation for each wet season across 2000 2009. ............. 84 3 5 Residuals from the estimates of the (a) global OLS and (b) GWR models. ........ 86 3 6 types. ................ 87
10 4 1 Study region in southern Africa outlining the larger Okavango Kwando Zambezi catchment and the local protected area of interest ............................ 111 4 2 Landsat TM RGB: 5,4,3 composite images and observed TM NDVI for each acquisition year in the study. ................................ ................................ ............ 112 4 3 Shows a) Landsat TM overall NDVI values and b) precipitation corrected NDVI values for the KCA in the form of residuals (NDVI obs NDVI pred ). ........... 113 4 4 A K means 10 class unsupervised classification of the KCA based on four discrete time steps of PC NDVI (1984 1990 2000 2007). ................................ 114 4 5 Trends PC NDVI over the four discrete time steps (1984 1990 2000 2007), defined as subsequent increase or decrease in PC_NDVI ............................... 115 4 6 Change trajectory for the three differenced PC NDVI images (1990 1984, 2000 1990, and 2007 2000) w ith a threshold of 1.96 standard deviations. ... 116
11 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy VARIABILITY AND CHANGE IN A DRYLAND SYSTEM: CLIMATE LAND INTERACTION IN THE OK A VANGO KWAN D O ZAMBEZI CATCHMENT OF SOUTHERN AFRICA By Andrea E. Gaughan M ay 201 1 Chair: M ichael W. Binford Major: Geography Southern Africa is one of the most uncertain regions in the world with respect to projections of climate change and the response of land use/land cover dynamics. How vegetation phenology, structure, and composition will change with increased climate variab ility, more or less rainfall, changes in seasonality of rainfall, or changes in temperature is complicated by local land use decisions for agriculture, ranching, and settlement. Before competing and interacting drivers of climate and land use may be disent angled, effects of water availability, arguably the most limiting factor in savanna environments, must be accounted for in landscape change. This dissertation asks the question of how spatial and temporal patterns of precipitation in part of southern Afric a affect the seasonal and long term response of different savanna vegetation types. I use geospatial analyses and field measurements to examine the relationship between vegetation productivity and rainfall variability at a regional catchment scale an d with in a local protected area to better understand long term change acros s different savanna vegetation types
12 The first part of this study describes and explains the long term pattern of total annual precipitation and changes from one period (1950 1975) to another (1980 2005) across sub catchments of three rivers, the Okavango, Kwando, system). R esults indicate decreasing precipitation patterns and increased dry years and warm phase ENSO in the last quarter of the twentieth century. Th e second part develops an empirical model of the relationship between seasonal rainfall and vegetation productivity response. Results show at the landscape level that intra annual rainfall across wet season months is strongly associated with beginning of dry season vegetation productivity as measured by the Normalized Difference Vegetation Index (NDVI) Th e relationship between wet season months of rainfall and NDVI varies by savanna vegetation type The model is also used to correct for seasonal precipit ation ef fects on vegetation productivity in order to identify areas of longer term land cover change attributed to other savanna processes This research contributes to savanna ecological theory through investigation of precipitation vegetation dynamics fo r a semi arid region The conclusions emphasize the changing nature of precipitation patterns over the second half of the 20 th century and the importance of inter annual variability in seasonal rainfall on savanna vegetation types
13 CHAPTER 1 INTRODUCTION General Introduction arid, semi arid, and dry sub humid regions (Reynolds et al. 200 7) important limiting resource is water availability (Walker et al. 1981) which interacts with other factors to control vegetation structure and other ecosystem characteristics, and the ability of humans to exploit the system (Scholes and Walker 1993) Water is not the only lim iting resource in controlling ecosystem dynamics but is a necessary component of every physiological process (Rodriguez Iturbe 2000) While precipitation is only a surrogate for water availability in drylands, (Archibald and Scholes 2007) the limited availability of water means any s ubtle shift in precipitation will affect all ecosystem aspects. In dryland systems, average rainfall and its variability influences the rates and dynamics of transitions among various tree grass states (Frost et al. 1986; Allen and Bre shears 1998; Sankaran et al. 2005; House et al. 2003) At tropical and subtropical latitudes, the seasonal rainfall amount is largely controlled by the movement of the Inter Tropical Convergence Zone (ITCZ), the global low pressure system that defines the alternating we t/dry seasons (Scholes and Archer 1997) Other parts of the water cycle, such as soil moisture, evapotranspiration, and flooding also contribute to dryland vegetation growth but are d ependent on the quantity, timing, and frequency of rainfall. In the same way, other biotic factors (ex. fire, herbivory) that contribute to driving landscape dynamics are also dependent on the underlying hydro metrological factors to maintain co existence of grasslands and woody vegetation in dryland areas (Sankaran
14 et al. 2005; Wang, Rich, and Price 2003; Breshears and Barnes 1999) The dependence of other ecosystem processes on the initial input of water into the system makes precipitation dynamics and their inter and intra annual effects on tree grass cover an essential component to understanding savanna environments. Overarching Research Question and Study Objectives The research question of this dissertation project is how does precipitation variation at several spatial and temporal scales influence vegetation response in the dryland system of the Okavango Kwando Zambezi River basins? Before examining other forcing fa ctors in dryland environments, one must first understand past and current precipitation patterns, and how their variability influences land cover dynamics. This is due to the dominating influence rainfall has in dryland water budgets (McCartney 2000) and how its timing and duration have important implications for agriculture and water resource management. Inter and intra annual precipitation patterns also strongly influence the continuum of tree grass cover in savanna ecosystems (Scanlon et al. 2005; Sankaran et al. 2005; Scholes et al. 2002) This dissertation inc ludes a set of complementary articles that address different spatial and temporal aspects of precipitation variability and how it correlates to land cover dynamics in three catchments in the Kavango Zambezi Transfrontier Conservation Area (KAZA), a rural b ut important socio ecological system in southern Africa. The area is both regionally important for conservation efforts that cross national boundaries and locally important for its perennial water sources and charismatic wildlife, the basis for increasing conservation based tourism that benefits local communities. To investigate seasonal and longer term precipitation patterns in the Okavango Kwando Zambezi (OKZ) catchment a spatio temporal, multi scale analysis is used to identify the effects of precipitati on on
15 vegetation patterns at both a regional catchment level (~693,000 km2) and within a local protected area (1,300 km2). The overall objective of this research is to contribute to the broader knowledge of how dynamics of human and environmental factors i nteract in dryland socio ecological systems by accounting for precipitation vegetation relationships in the OKZ catchment. Four questions comprise my focus: 1. What are the spatial and inter annual precipitation patterns in the Okavango Kwando Zambezi (OKZ) c atchment of southern Africa? 2. What is the variation in vegetation productivity response to seasonal precipitation patterns in the OKZ catchment? 3. What is the relative importance of seasonal precipitation in regulating savanna vegetation in the Kalahari sand woodlands of Caprivi? 4. After controlling for seasonal effects, what temporal and spatial trends are detectable in vegetation in the Kwando Core Area of Bwabwata National Park from 1984 to 2007? Linking Objectives to Theory Climate land I nteractions for C o upled Human Environment S ystems The ability to detect and monitor landscape dynamics is a fundamental task of global environmental change studies (Turner et al. 1990; Lambin and Geist 2006; GLP 2005) Land change science focuses on global environmental change through a coupled human environmental systems approach with emphasis on understanding linkages across ecosystems, climate, and social dimensions of land use and land cover change (Rindfuss et al. 2004; Turner, Lambin, and Reenberg 2007) Within this framework, the Global Land Project Science Plan and Implementation Strategy document emphasizes the need to understand climate land interactions better, especially directional trends in distribution, variability, and average rainfall at local and regional scales and its effects on vegetation (GLP 2005). Variability and shifts in
16 precipitation influence changes of ecosystem structure and function but adequate data on the response of ecosystems to changing rainfall patterns is still lackin g (GLP 2005; Goward and Prince 1995; IPCC 2007; Zhang 2005) Shifting Dynamics of Dryland Environments Specifically in a rid and semi arid southern Africa, land cover is characterized by a continuous gradient of tree shrub grass ratios that results from interaction of multiple ecological factors over space and time (Skarpe 1992) These dryland areas have a range of land cover types whic h include grasslands, shrublands, savannas, xerophytic woodlands, and hot and cold deserts (Puigdefabregas 1998) and lack of water is an important limitation for growth. A variety of terms describe this continuum of variable and diverse ecosystems comprised of different proportions of mixed woody and herbaceo us species (House et al. 2003; Breshears and Barnes 1999) I adopt the definition of a mixed woody herbaceous system from House et al (2003) which refers to and trees with range of density and canopy cover while gra ss includes grass, forbs, and sedges. Savanna systems exist along this continuum but a useful distinction can be productivity in dry savannas responding strongly to increased annu al rainfall whereas the relationship is much weaker in wet savannas (Scholes and Walker 1993) The transition where this relationship weakens occurs around 500 700 mm/yr rainfall although soil type decreases (sandy) or increases (clays) the exact leveling off of the curve between these two different savanna systems (Accatino et al. 2010; Scholes and Walker 1993) Vegetation type also changes along this savanna gradient with a higher
17 woody to grass cover ratio in areas with hig her mean annual rainfall (Sankaran, Hanan, and Scholes 2007) Effects of Precipitation on Dryland E cosystems Changes in climate will influence changes in land cover and significant shifts or changes in climate patterns (in this case, rainfall) can contribute to change in vegetation ratios along the tree grass co ntinuum. The shifts in average rainfall or frequency and timing of rainfall will alter competitive advantage of either woody or grass species (House et al. 2003) and changes in rainfall average and/or variation may affect system drivers, such as soil moisture or fuel load, in determining woody to grassland ratios in dryland ecosystems. The next few sections describe how I studied the changes in long term precipitation patterns for the past half century in the OKZ catchment in southern Africa and use a multi scale approach to identify and examine the effects of seasonal rainfall on vegetation dynamics for the OKZ catchment in southern Africa. The dissertation is arranged as a series of three stand alone but related articles that will be published in peer reviewed journals. Spatial and T emporal P recipitation P atterns, 1950 2005 The first article, presented in Chapter 2, describes and explains the long term pattern of total annual precipitation and how it has changed from one period (1950 1975) to the next (1980 2005) across and within sub catchments of three ri ver s, the Ok variable precipitation on vegetation responses underlie the modifying effects of disturbances such as fire, herbivory and limited resource availability (Nemani et al. 2003) Shifts in timing, distribution, and frequency of rainfall influence ecosystem function, composition, and structure (Huxman et al. 2005) In southern Africa wetter
18 conditions persisted throughout the mid 20th century but during the 1980s and into the 1990s there has been a drying trend across Africa (Nicholson 2001) This phenomenon corresponds to a global climatic regime driven by a shift in the Pacific atmosphere and ocean in the late 1970s (Hare and Mantua 2000; Chavez et al. 2003) which has been linked to increases in El Nio Southern Oscillation warm periods (El N io events) and changes in teleconnection patterns of southern African rainfall (Mason 2001; Fauchereau et al. 2003) This cha pter provides a spatio temporal descriptive analysis to determine if similar patterns of changing precipitation exist for the OKZ catchment pre/post the late 1970s climate shift. Using a monthly precipitation modeled dataset (0.5 o x 0.5 o ) to quantify rainf all patterns (Matsuura and Willmott 2007) I show that the late 1970s shift in rainfall is significant for a large portion of the OKZ basin and that precipitation and ENSO patterns differ before and after the climate shift. Annual precipitation totals to each ind ividual basin were calculated and number of wet (upper tercile) or dry (lower tercile) years experienced in two periods, 1950 75 and 1975 2005 are compared to those expected at random using the hypergeometric distribution. Rainfall correspondence to, and the frequency of, El Nio Southern Oscillation (ENSO) events within these periods was also investigated. Results from this chapter indicate decreasing precipitation patterns and increased dry years and warm phases of ENSO across all three sub catchments i n the last quarter of the twentieth century. Knowledge of the historical spatio temporal shifts in precipitation plays a direct role in decisions made on the ground regarding agriculture, wildlife, and resource management. An explanation of inter annual pr ecipitation patterns
19 provides important information for local collaboration between national parks and communities about decisions concerning water access and usage between wildlife and humans. Vegetation Response to Seasonal P recipitation Chapter three al so examines precipitation patterns at the regional OKZ catchment scale but at a monthly time step by investigating the response of vegetation productivity to the timing of wet season rainfall. The relationship between plant phenology and seasonality of pre cipitation in drylands is not straightforward (Archibald and Scholes 2007). In dryland regions a pulse system dictates the cycling of nutrients in which temporary water availability strongly drives biogeochemical processes (Scholes and Walker 1993). In sem i arid savannas, woody plant growth typically greens up just prior to first rainfall as storage of previous season nutrients and carbohydrates make them less dependent on the timing of rainfall (Sekhwela and Yates 2007; Shackleton 1999; Do et al. 2005) In contrast, grasslands are more limited by water availability and have a more tightly coupled response to rainfall (Prins 1988; Archibald and Scholes 2007) And although less attention has been given to beginning of dry season vegetation productivity, the status of begi nning of dry season vegetation influences the availability of water resources and other important forage materials for wildlife, therefore affecting wildlife movement and distribution throughout the dry season. Timing in seasonal rainfall will affect syst em drivers, such as soil moisture or fuel load, in turn influencing the tree grass ratios in semi arid dryland ecosystems. This is an important consideration for a region where large proportions of the human population and economy are dependent upon wildli fe tourism. This study uses MODIS NDVI as a proxy for vegetation gross primary productivity (GPP) to investigate the response of
20 GPP of different savanna vegetation types to month to month precipitation variability. I focus on the effects of monthly wet se ason rainfall on vegetation status at the beginning of the dry season across the OKZ catchment over the years 2000 2009. I estimate monthly precipitation using the Tropical Rainfall Monitoring Mission 3B43 dataset and use the MODIS 13A1 VI product. A time series model that estimates beginning of dry season vegetation productivity from the prior rainy season (October April) using geographically weighted regression (GWR) determines within wet season rainfall influence on beginning of dry season GPP. Result s show at the landscape level intra annual rainfall variability accounts for significant amounts of beginning dry season vegetation GPP variation and overall has a stronger effect for areas with greater grassland proportion and dry, deciduous woodlands. T he tighter association of grassland and open canopy woodlands (specifically Mopane woodland) to end of wet season monthly rainfall (February April) highlights the importance of both beginning and end of the rainy season for understanding the effects of f uture climate change. Detecting Seasonality in D ryland S avannas Detecting and accounting for inter annual variability in seasonal precipitation on vegetation with remotely sensed imagery is important for identifying long term trends in land cover change. T his is especially important for ecosystems with strong photosynthetic response to wet and dry seasons such as those in southern Africa. The seasonal response of vegetation productivity prevalent in African savannas may obscure detection of long term change that results from decadal climate shifts or from anthropogenic land use decisions.
21 The fourth chapter applies the empirical model developed in Chapter 3 to examine local scale vegetation patterns in a core area (1,300 km2) of Bwabwata National Park in Ca privi, Namibia. The main focus of this chapter is to account for seasonal precipitation effects on gross primary productivity (GPP) at the beginning of dry season for the southern African savanna landscape, as estimated by the Normalized Difference Vegetat ion Index (NDVI). Beginning of dry season productivity is important to savanna ecosystem phenological patterns due to water availability and the importance of both grass and woody species for wildlife forage pattern and livestock feed (Chase and Griffin 2008; Paterson et al. 1998) Four discrete time steps across the study period that correspond to the Landsat TM derived NDVI provide a twenty three year period (1984 2007) to examine the vegetation change. To correct for seasonal precipitation effects, I estimate NDVI from the downscaled Geographically Weighted Regression (GWR) derived relationship between MODIS (or Moderate Resolution Imaging Spectroradiometer) NDVI and the Tropical Rainfall Monitoring Mission (TRMM) estimated precipitation. The estimat ed NDVI, calculated for each TM image year, is then subtracted from the observed TM derived NDVI resulting in a new, precipitation corrected (PC) NDVI value, or residual (Evans and Geerken 2004) The PC NDVI corrects for seasonal effects on the landscape and thus, any trend (positive or nega tive) through time present in the PC NDVI values indicates change in NDVI not due to seasonality. Findings emphasize the importance of inter annual variability of seasonal precipitation effects on land cover change in savanna ecosystems. The minimal change in PC NDVI across the four TM dates indicates the regional model, parameterized for a
22 savanna lan d cover using the prior wet season months, accounts for a large portion of variability in NDVI for the local study area. However, temporal clusters of PC NDVI and image differencing identify areas in which PC NDVI varies over space and time. The trends rem aining in vegetation productivity after accounting for seasonal effects are then attributed to longer term climate or anthropogenic drivers on the landscape. Importance of the Study The three papers complement one another by providing an in depth examinati on of spatio temporal precipitation changes and subsequent effects on savanna vegetation productivity in a southern African dryland catchment. The process of this dissertation project has been couched within an interdisciplinary context largely driven by a n U.S. National Science Foundation Integrative Graduate Education and Research Traineeship (IGERT) with a theme of adaptive management, water, watersheds, and wetlands. The IGERT program emphasizes collaboration and cross disciplinary efforts to tackle res earch questions larger than the scope of just one discipline. As such, this dissertation research builds upon an understanding of the larger socio ecological system with four field seasons working with academic, governmental, and local collaborators who ma de it possible to capture a more holistic understanding of the OKZ system. In addition, coursework that spanned physical and social subject matters, all related to the theme of water, allows for greater breadth of my own understanding in directing this res earch toward not only a theoretical contribution to knowledge but also an applied understanding of the OKZ catchment in which adaptive management is important to the process of land use and management decisions for conservation and development trade offs.
23 Successful management of this semi arid savanna system requires recognition and understanding of ecosystem processes that operate on multiple scales. While local level processes such as fire and grazing affect vegetation structure, the response of vegetati on productivity to climate variability also must be taken into account for conservation management decisions. Adaptive management incorporates a learning by doing approach that provides a flexibility necessary when managing natural resources such as water and wildlife that demand larger spatial extents than provided by protected or political boundaries alone (Agrawal 2000; Walters and Holling 1990) Addi tionally, incorporating the capacity to learn in managing this socio ecological system will provide a stronger management framework to deal with the uncertainly of climate change. While the adaptability of system dynamics is mostly a function of the social element (Walker et al. 2004) to respond to change and uncertainty people must first know how the physical environment responds to stress and change over space and time. This research contributes to savanna ecological theory by testing hypotheses that relate to precipitation vegetation dynamics for semi arid dryland regions but examines this relationship for the beginning of dry season, a lesser studied but still critically important to phenological patterns in savanna ecosystems. This study also contributes to land change science by providing a multi scale approach to understanding land use/land cover dynami cs for an area undergoing extensive and rapid changes in both the conservation and development sectors. Lastly, this study contributes an applied understanding to historical environmental change for three dryland catchments necessary to look at future proj ections of climate change and variability and its effect on semi arid dryland vegetation both at the local and regional scale.
24 CHAPTER 2 SPATIAL AND TEMPORAL PRECIPITATION VARIAB ILITY IN THE OKAVANG O KWANDO ZAMBEZI CATCHMENT, S OUTHERN AFRICA Changes in pr ecipitation patterns strongly influence ecologic, hydrologic, and socio economic components of a system making their quantification critical in water limited environments such as African savannas (Frost et al. 1986; Scholes and Archer 1997) Modeling and observational studies over the recent decades suggest that there has been a drying trend over much of the African continent (Giannini et al. 2008) Specific to southern Africa, multi decadal trends highlight d eclining mean annual precipitation (MAP), increasing variability and drier conditions, and an increased number of warm phase El Nio Southern Oscillation (ENSO) events (Mason 2001, 1996; Nicholson and Entekhabi 1987; Giannini et al. 2008) The relationship between observed precipitation patterns and associated climate drivers vary at different spatial and tempor al scales ( Nicholson and Kim 1997) which has important effects for savanna ecosystems where variable precipitation patterns influence both wildlife movements and agricultural land use decisions. This paper examines precipitation patterns from 1950 2005 and investi gates changes in rainfall patterns and associations to the El Nio Southern Oscillation (ENSO) pre/post the late 1970s climate shift for the Okavango Kwando Zambezi (OKZ) catchment in southern Africa. Regionally, southern Africa is one of the most sensitiv e areas to precipitation shifts and variability (Mason et al. 1996; Archibald and Scholes 2007; IPCC 2007) and a large degree of uncertainty exists about future rainfall distribution, frequency and variation (Flato 2001; Gordon et al. 2000) Along with ENSO, the most important underlying mechanisms driving rainfall patter ns in southern Africa are the migration of the inter tropical convergence zone (ITCZ), and sea surface temperatures (SSTs) in the Indian
25 and Atlantic Oceans (Nicholson 1997c; Nash and Endfield 2008; Mason 2001) The ITCZ controls seasonal rainfall but global climate teleconnections related to ENSO and SSTs of the Indian and Atlantic oceans will influence the inter annual and inter decadal variability in rainfall (Richard et al. 2001; Mason 2001; Nicholson 2001; Rouault et al. 2003; Hirst and Hastenr ath 1983; Reason and Rouault 2002) ENSO refers to the irregular shifts in Pacific sea surface temper atures and atmospheric conditions and is the major inter annual influence on rainfall in the tropics (C urtis 2008) Its effect on southern African rainfall patterns is modulated by SSTs from the Atlantic and Indian Oceans (Nicholson 1997a; Paeth and Friederichs 2004; Mason and Jury 1997; Jury, White, and Reason 2004) In recent years there has been an increase in the number of warm phase ENSO events (Mason 2001) This phase often is associated with drier conditions in southern Africa (Mason 2001; Mason and Jury 1997; Jury, White, and Reason 2004; Alemaw and Chaoka 2006) If a warm phase occurs during an already dry year conditions will be exacerbated and th e drought may persist into subsequent years (Nicholson, Leposo, and Grist 2001; Nash and Endfield 2008) However, the strength of association varies making an underst anding of the association between ENSO phase and rainfall patterns critical to natural resource management (Richard et al. 2001) O ne possible explanation to recent multi decadal changes in southern Africa rainfall patterns and potential changes in ENSO phase frequency is a shift in the background state of the coupled ocean atmosphere system in the tropical Pacific during the late 197 0s (Graham 1994; Meehl, Hu, and Santer 2009; Mason 2001) Main characteristics of the shift included altered patterns in sea surface temperatures
26 (increase in SSTs > 0.75 o C) an d an eastward shift in convection patterns in the near equatorial Pacific (Graham 1994) The slowly varying pattern of sea surface temperatures, known as the Interdecadal Pacific Oscillation (IPO) (Power et al. 1999) has a concurrent effect on the underlying processes of ENSO in the Pacific Ocean (Wang 1995) The 1970s shift to warmer SSTs in the tropical Pacific is coincident with a strengthening of winter circulation patterns in the North Pacific affecting climatological, hydrological and biological components of the system (Graham 1994; Francis et al. 1998; Nitta and Yamada 1989) and with altered relationships between ENSO and p recipitation patterns over Australia (Power et al. 1999) Similar changes in frequency, intensity, duration of ENSO, air temperatures and rainfall patterns since the 1970s have been noted by Mason (2001) and Nicholson (1997b) over southern Africa. This study exami nes the spatial and temporal trends in precipitation over the Okavango Kwando Zambezi (OKZ) catchment (693,000 km 2 ) which includes three inter connected yet distinct catchment areas. Included in the three catchment area is the future Kavango Zambezi Transboundary Conservation Area (KAZA) that spans five southern African nations attempting to achieve regional scale conservation through cooperative management of shared natural resources (van Aarde and Jackson 2007) The need to understand how southern African rainfall varies across large geographical but politically connected landscapes is even more critical for such transboundary conservation initiatives that are becoming more common in southern Africa (Wolmer 2003) As geographically extensive as these areas may be, their boundaries often exclude important hydro meteorological processes that operate at larger regional scales and contribute important in puts to the system from beyond conservation limits. The
27 KAZA region provides a vital wildlife corridor to which surface waters generated by precipitation upstream constitute an important water source in an otherwise water limited ecosystem. Wildlife patter ns are dictated by changes in both local rainfall patterns, and those throughout the basin, which influence availability and accessibility to forage and surface water (Chase and Griffin 2009) If there has been a change in rainfall average, frequency, and distribution over the last half century, then we expect to observe a corresponding increase in El Nio years, increased frequency in dry years, and decrease in MAP in the Okavango Kwando Zambezi catchment. Addit ionally, we expect that the association between local and regional input to be less strong in more recent years due to an increase in the variability of climate patterns across southern Africa. Materials and Methods Study Region The Okavango Kwando Zambez i (OKZ) catchment (693,000 km 2 ) encompasses the vast majority of KAZA, the politically defined conservation area that includes parts of Angola, Zambia, Namibia, Botswana, and Zimbabwe (Figure 2 1). The main KAZA areas in the OKZ basin are the Caprivi Regi on of Namibia, northern Botswana, and western Zambia. KAZA is set to become the largest transfrontier conservation area in Africa connecting multiple protected areas and thus has considerable potential to sustain increased wildlife densities and diversity (O'Connell Rodwell et al. 2000) The region possesses one of the largest free ranging populations of elepha nts left in Africa, to whom the three perennial rivers constitute a critical source of water in an otherwise water limited ecosystem (Craig 1997) Moreover, most human settlement also occurs along these water courses thereby increasing the potential conflicts as wildlife
28 congregates near these perennial water sources e specially during the dry season (Chase and Griffin 2009) The highest human populations are located in the lower OKZ basin particularly along rivers in the Caprivi Region (20,000 km 2 ) of Namibia (O'Connell Rodwell et al. 2000; Mendelsohn and Roberts 1997) The majority of the headwaters of all three catchments are located in Angola although a good portion of the Zambezi lies in western Zambia. The upper catchment is characterized by Miombo woodland in areas recei ving > 1000 mm/year of rainfall and with a canopy flush that occurs a few weeks before the first rains (Frost, 1996). The lower catchment is of a more mixed composition of tree shrub grass and a sub parallel system of drainage lines (Omirumbas) that run al ong a NW SE gradient (Thomas et al. 2000) The region expe riences natural variability in precipitation with a gradient of decreasing rainfall from north to south. Kalahari sandveld characterizes a large portion corresponds to t he N S rainfall regime (Dube and Pickup, 2001). However, variation in soil type exists with upland areas in Angola and Zambia defined by a range of ferralsols to sandy arenosols for mid elevation regions and the lower, flatter areas of the catchment conta in clay enriched gleysols (Batjes 2000) Datasets The boundaries of the OKZ catchment provide the limits within which comparisons of mean annual precipitation across and within three sub catchments are completed. Digital elevation model (DEM) data from the World Wildlife Fund HydroSHEDS project delineate the catchment (Lehner, Verdin, and Jarvis 2006) HydroSHEDS data products have a spatial resolution of 15 arcseconds and were processed using the ArcHydro data
29 model and toolset (Maidment 2002) to extract drainage networks and sub catchment extents. To examine long term precipitation patterns, we used a gridded monthly time se ries of modeled rainfall across the OKZ catchment (Matsuura and Willmott 2007) The Willmott Matsuura (WM) dataset was developed at the Department of Geography and the University of Delaware and is based on an earlier global mean monthly precipitation dataset (Legates and Willmott 1990) The WM dataset improves upon the Legates and Willmott dataset with a refined Shepard interpolation algorithm and an increased number of neighboring station points included in analysis (Fekete et al. 2004) The WM dataset uses a spatial interpolation of monthly total precipitation station values created a 0.5 o x 0.5 o degree latitude/longitude grid with g rid nodes centered on 0.25 degree which results in a total of 232 points for the OKZ catchment. We chose the Matsuura and Willmott dataset over other global modeled datasets (Kalnay et al. 1996) because of its finer spatial resolution and inclus ion of more recent years. Existing meteorological station data alone did not provide sufficient spatial or temporal extent and satellite based estimates (ex. Tropical Rainfall Monitoring Mission (TRMM)) did not provide the necessary longitudinal history fo r analysis. The WM model output accuracy for global precipitation patterns compares favorably to five other monthly precipitation datasets (Fekete et al. 2004) and the finer spatial resolution and inclusion of more recent years made it the most appropriate dataset for this study. Statistical E xamination of P recipitation P atterns To identify any temporal shifts in precipitation patterns across the OKZ catchment we use progressive windows comparing consecutive 5 and 10 year blocks of time from 1950 2005. Comparisons were conducted for the two sample blocks usin g a Mann U
30 Whitney test. The nonparametric test was applied cell by cell. In order to further test the presence of the postulated climate shift in the late 1970s (Mason and Jury 1997; Mason 2001; Chavez et al. 2003; Graham 1994) estimated basin wide precipitation input data from 1950 2005 are separated into two periods (P1: 1950 1975 and P2: 1980 2005) excluding the latter half of the 1970s Correlations comparing overall precipitation input (1950 2005) for each sub catchment spatially identify the strength of association between catchments in the OKZ correlation that assumes a linear relationship between the variables of interest, in this case the precipitation input for each sub catchment (Burt and Bar ber 1996) In addition, visual comparisons of patterns in P1 and P2 in the OKZ basin are made by investigating the number of years in each period experiencing above/below median rainfalls. Similar comparisons are possible for rainfalls in cold and warm phases of ENSO. After describing spatial and temporal change for the OKZ basin, we use the hypergeometric distribution (Agresti 2007; Mason and Goddard 2001) to determine the likelihood of experiencing any number of above or below median rainfall events i n P1 and P2 under the null hypothesis of no significant change in characteristics between the two time periods. The observed number of above and below median rainfall years within a given time period can then be tested against this null hypothesis of rando mness. The test which has no underlying assumption of normality has been used successfully to look at rainfall distributions and probabilities globally (Mason and Goddard 2001) and for other regions in Africa (Owusu, Waylen, and Qiu 2008) The hyper geometric distribution test can be stated as such:
31 Where N is the population of total annual precipitation inputs (N=55); the time period from 1950 2005), n is the subset of estimated available years for each time period (n=25), and k represents the n umber of years in the population that are considered a success (for example, the number of years experiencing below median rainfall in the N; N/2). The equation returns the probability, p(x) of experiencing x dry (wet) years at random. The same analysis i s performed for El Nio and La Nia years to determine the correspondence between phases of ENSO and wet and dry years respectively (Grimm et al., 2000). Years which correspond to warm phase and cold phase events are identified by the Center for Ocean Atmo spheric Prediction Studies (COAPS) at Florida State University ( http://www.coaps.fsu.edu/jma.shtml ). The statistical null hypothesis states that there is no significantly greater number of El Nio (La Ni a) years occurring in the driest (wettest) third of all years than one would expect at random. Similarly, we hypothesize that there is no significantly lesser number of El Nio (La Nia) years occurring in the wettest (driest) third of all years than one w ould expect at random. Significance is a discrete p value determined by the hypergeometric distribution probability of x or more (x or less) events occurring outside what is expected at random. Lastly, as water resources available to the KAZA conservation area include the local precipitation and regional waters imported from the respective upstream sub
32 catchments, it is important to focus on these different patterns of precipitation across and within the sub catchments and to examine the association between regional and local input (Figure 1). By comparing precipitation input between local and regional areas of each catchment we determine a general idea of how synchronized hydrological and meteorological droughts (floods) are in the area. A Chi Square test o f independence is used to examine the relationship between "wet", "normal", and "dry" years between the local and regional study areas. For examinations of Periods 1 and 2, where the expected sample size falling into each possible combination of categories is less than 5, (Fisher 1970) Associated probabilities for above and below median precipitation values are also analyzed in each catchment. We use regional sub catchment est imates of spatially averaged total precipitation as a substitute for discharge due to the lack of longitudinal discharge measurements across all three basins. The analysis is completed for the entire time period (1950 2005) and repeated for purposes of com parison between P1:1950 1975 and P2:1980 2005. Results Climate Shift Results of the application of the Mann Whitney test to compare differences in mean values of precipitation using a 5 year progressive window (water years 1950 2005) are shown in Figure 2 2. There were no significant differences in mean values for the 10 year progressive window. Figure 2 2 values highlighted in white and grey are grid points that had a p value < 0.05 and 0.10, respectively. The comparison between years 1975 79 and 1980 84 showed the largest number of grid points in which the mean value differed significantly from each other. Of the total 232 grid points that comprise the OKZ basin, 32% are signif icant (p< 0.05) for the 1975 79 and 1980 84 period. The next
33 largest number of grid points to have significant difference between time steps was 21% between 1965 69 and 1970 74. Other paired comparisons show no change (ex.1950 54 to 1955 59) or regional di fferences (ex. 1955 59 to 1960 64) but only the late 1960s and late 1970s time periods show a difference across all three catchment areas. Excluding the time period 1975 79, allows comparison of changing precipitation patterns before and after the shift wh ich clearly manifests itself in the study area. Spatial and T emporal P atterns Figure 2 3 indicates that the two more western catchments (the Okavango and the Kwando) experience similar magnitudes and variabilities of estimated annual basin inputs, while th e Zambezi exhibits a slightly out of phase temporal pattern and increased level of precipitation input. Correlations calculated for the entire time period between the three catchments showed a significant positive association for all comparisons. The posit ive association was strongest between the Kwando and Okavango catchments (64%) and weakest between the Okavango and Zambezi catchments (44%). Period 1 and Period 2 did not show statistically significant results suggesting that precipitation between time pe riods is not temporally associated. Warm and cold phase ENSO periods identified in Figure 2 3 also suggest a reduction in the number of cold phase years since 1980. Variability, calculated by the coefficient of variation, is relatively higher for more semi arid regions in the southern part of the OKZ catchment (Figure 2 4). The NE SW gradient of increasing variability (exception is a small area in upper northeastern area) in the OKZ basin is coupled with a decreasing gradient of overall less MAP input (Figu re 2 5). Breaking down the MAP input by P1 and P2 suggests, in general, more rainfall fell overall in the OKZ catchment during P1 (Figure 2 5a). A noticeable shift in P2 isolines
34 indicate less total annual rainfall occurred in the OKZ basin in recent decad es. The majority of the OKZ basin shows a decline (P2 P1) in total median rainfall. Areas experiencing higher rainfall in P2 are limited to the most northeasterly extreme of the Zambezi. The difference in median rainfalls between cold and wet phase of E NSO (La Nia El Nio) shown in Figure 2 6 illustrates a much larger range (more marked impact on variability) since 1980. Higher values identify areas with bigger extremes between amount of rainfall during an ENSO event (LN EN) while lower values are are as that do not differ as strongly in total median rainfall during El Nio (warm) and La Nia (cold) years. During P1 (1950 1975) the largest difference in La Nia and El Nio occurs in the lower catchment areas of the Okavango and Kwando, while the differe nce is much more pronounced across the entire area including the Zambezi in Period 2: 1980 2005 (Figure 2 6a and 2 6b). The largest difference of ENSO events across the two time periods (LN EN 1950 1975 minus LN EN 1980 2005) occurs in a broad arc stretchi ng across the uppermost reaches of the Okavango and Kwando catchments and through mid portions of the Zambezi catchment (Figure 2 6c) Positive values identify regions with increased variability in ENSO periods during P2 while negative values indicate are as with less variability in ENSO during P2. The KAZA area itself seems to be sensitive to changes in ENSO phase during both periods, but more so during the second. The change appears most marked over the Zambezi section and upper Okavango portions of the OKZ basin.
35 Basin Level A nalysis Figure 2 7 displays the differences in mean annual precipitation input for each sub catchment of the study area for P1 and P2. In all three sub catchments the mean annual precipitation was higher in P1 (1950 1975) when compa red to P2 (1980 2005). In addition to experiencing less rainfall overall, the more recent decades also exhibit an increase in the number of "dry" years (years in the long run lowest tercile), while the total number of wet years has decreased for each basin (Figure 8). Application of the hypergeometric distribution indicates that the number of dry and wet years in each basin over each separate time period is different than that expected at random. The opposite patterns in dry and wet year frequency for P1 an d P2 respectively complements findings on total annual precipitation input declining in each sub catchment. Observations of figures 2 3 and 2 8 prompt the related questions of whether the numbers of years experiencing various ENSO phases changed in the lat e 1970s, and also whether the severity of the association of ENSO events to wet and dry years. Table 2 1 shows the probability that the concordance between the historic (1950 2005) number of warm and cold phase events and dry and wet years is not random T he likelihood of experiencing a dry year during an El Nio event is significantly greater than random for the two most western catchments, the Okavango and the Kwando, but not in the Zambezi. Only in the Kwando catchment is La Nia events associated with s ignificantly fewer dry years than expected at random. Wet years are significantly more likely to occur simultaneously with cold phase (La Nia) events. The associations may be further broken down to see if similar associations are present pre/post mid 19 70s, although levels of significance may drop in association with
36 the smaller sample sizes in comparison to the previous analyses. Table 2 2a shows that the number of dry years increases (decreases) coincident with El Nio (La Nia) events in all three bas ins from P1 to P2. For the driest third of all years, in P1 the probability of experiencing an El Nio or La Nia event does not fall outside the number of events one would expect at random with the exception being La Nia events in the Kwando catchment ( Table 2 2a). In P2, there is a much stronger association than expected at random between dry years in the Okavango and Kwando catchments and El Nio events. However, there is no such relationship for the Zambezi and none of the three catchments seem to be associated with La Nia events aside from what we would expect at random. Considering the highest third of all years (Table 2 2b), the only association that falls outside that expected at random for P1 is that between the Kwando catchment and La Nia even ts indicating more wet years than random. In Period 2, a pattern emerges across all three catchments with El Nio events eliciting fewer wet years than expected at random. Again, no such relationship in P2 for any catchment seems to be associated with La N ia events aside from what we would expect at random. Cell by Cell A nalysis A cell by cell analysis is employed to determine how patterns of association are changing between phases of ENSO with dry (wet) years within the basins themselves and also provides a more refined spatial depiction of the relationship between ENSO and mean annual precipitation. Figure 2 9 shows the cell values with associations between warm phase and dry conditions that have significantly greater numbers of association than random. A significance level of 0.059 level is employed as this is the closest that the discrete hypergeometric distribution comes to providing the more usual
37 level of 0.05. The significance level corresponds to the likelihood of experiencing 4 or more warm phase E NSO events in the lowest historic tercile during one of the time periods. Only 4 grid cells in the Angolan Highlands are significant during Period 1 (Figure 2 9a). However, a more extensive and coherent pattern is present in Period 2 (figure 9b) over much of the lower catchment area including precipitation directly into the central region of KAZA itself. There is no statistically significant relationship (p < 0.059) for either time period for cold phase ENSO and dry years. The same analysis is done for the highest (wettest) climatological third of all comparable seasonal totals and correspondence to warm or cold phase ENSO across the two time periods. Figure 2 10 identifies the cell values with associations between warm phase and wet conditions (panels a an d b) that have a significantly lesser number of associations than random (p < 0.069). One grid cell is highlighted in P1 (Figure 2 10a), located in the Angolan highlands. However, a large portion of the southern half of the OKZ basin shows a relatively str ong response of less number of warm phase ENSO Okavango and Kwando catchment regions. The cell values with associations between cold phase and wet conditions (Figure 2 10c a nd 2 10d) that have significantly more numbers of association than random show P1 has a minimal association between cold phase ENSO and wet conditions. However, a strong geographic pattern exists in P2 covering much of t he central region of the OKZ basin. This area encompasses part of all three basins and emphasizes the importance of wet years coincident with cold phase ENSO.
38 L ocal versus R egional I nput Table 2 3 suggests that precipitation input to the entirety of each sub catchment fluctuates in a similar fashion to that portion of the sub basin falling within the KAZA region itself. When the entire time period (1950 2005) is considered, this similarity of behavior of both local precipitation and exotic waters brought into the KAZA by the rivers could have severe consequences for the management of water resources in the parks. The indications are that the congruity of meteorologic and hydrologic droughts and flooding have increased in the latest decades, and that this may be particularly strong and persist ent as one moves eastwards in the KAZA area. Table 2 4 summarizes the observed relative frequencies with which combinations of above/below median catchment (horizontal) and local (vertical) are recorded in the three sub catchments over the time periods con sidered. All basins and time periods reflect the strong positive associations (large values in the diagonal top left bottom right) between local and regional conditions. The first column highlights the overall time period (1950 2005) and shows a positiv e association for all three catchments. The Okavango and Kwando are in the same state (above/below) 68% of the time, while the Zambezi evinces even greater similarities (88%). Examined by Period, two changes are apparent; 1) during period 2 there is a grea ter tendency for both local and regional inputs to be in the same state, and 2) the most common combination of states switch from simultaneous above median conditions during period 1 to simultaneous below median conditions in Period 2. Thus, while the ass ociation remains positive, there is a greater probability that each catchment will experience both below average rainfall as well as below average flow from 1980 2005. Again the Zambezi has the strongest
39 positive association with 96% of the years from 1980 2005 either having above average rainfall and flow or below average rainfall and flow. Discussion Spatial and temporal precipitation patterns in the OKZ basin shows a decrease in total mean annual rainfall between P1 (1950 1975) and P2 (1980 2005). The late 1970s shows the largest percent difference in precipitation input on a cell by bell basis (Figur e 2 2). The late 1960s is identified as another period of anomalous precipitation patterns but not to the same magnitude as the late 1970s (Nicholson 2000) The late 1970s difference in precipitation corresponds to the time period observed for the global climatic shift that resulted from climate shifts in the northern Pacific basin. The dec line between P1 and P2 manifests itself in an increased frequency of dry years and a rise in number of warm phase ENSO events, which are often coincident. Although there is a decline in wet year frequency from P1 to P2, when wetter than average years of ra infall occur they are often coincident with cold phase ENSO events. The distinction of cold and warm phase ENSO associated with either wet or dry years is more apparent in P2 compared to P1. This suggests that ENSO phases are more strongly associated in re cent decades with either dry (warm phase ENSO) or wet (cold phase ENSO) years. These findings support the strong relationship previously identified between inter annual rainfall variability and ENSO for southern Africa (Mason 2001; Mason and Jury 1997; Ropelewski and Halpert 1987) and also indicate that the global climate shift of the late 1970s is detectable within the OKZ catch ment. Both the overall basin calculations and the cell by cell analysis indicate increase in number of warm phase ENSO events associated to dry years in P2 for all three OKZ sub catchments (1980 2005), corresponding to previous findings for other regions i n
40 southern Africa and globally (Mason 2001; Nyenzi and Lefale 2006) The b asin scale correspondence is strongest for the increased frequency of warm phase ENSO events during P2 (1980 2005) and occurrence with dry years in the Okavango and Kwando basins. The cell by cell analysis also shows the strongest correspondence in P2 but emphasizes not only areas in the Okavango and Kwando, but the entire lower half of the OKZ basin (Figure 2 9b). In addition for P2, the cell by cell analysis indicates a decoupling in the strength of warm phase ENSO and wet years (Figure 2 1 0 b) and a stronger association of wet years to cold phase ENSO events (Figure 2 1 0 d) than expected at random. The change in frequency from P1 to P2 of ENSO phases (more warm phase versus cold phase) and the increased association of warm phase ENSO and dry years coincides with positive (negative) atmospheric temperature fluctuations of recent decades that trigger global warm (cold) phase ENSO events (Tsonis et al. 2005) Tropospheric warming in the latter half of the 20 th centur y could also be a factor in higher frequency of warm phase ENSO events (Flohn and Kapala 1989) Our results for these southern African catchments correlate to the global change in the ENS O signal (Chang et al. 2006) and warming of global ocean temperatures (Levitus et al. 2000) and may be related to a possible enhanced greenhouse effect or just be part of a normal, variable climate pattern (Mason 2001) Regardless of the underlying mechanisms, the varying strength of associations suggests ENSO phases will have a stronger influence in certain part of the OKZ catchment than others. ENSO appears to impact rainfall patterns across the entire OKZ basin, but the association of warm phase
41 ENSO and dry year appears particularly strong in the more westerly Okavango and Kwando catchments, in recent decades. Some of the main underlying factors that influence the spatial and temporal rainfall patterns across the OKZ catchment and its association to ENSO are lar gely related to Atlantic and Indian basin ocean atmospheric processes (Nicholson 1997a; Jury 2010; Vigaud et al. 2009; Rouault et al. 2003; Hirst and Hastenrath 1983; Nicholson and Entekhabi 1987) Specifically, the association of Atlantic ocean atmospheric processes on southern African rainfall variability is growing in its importance especially for western and central regions of southern Africa (Vigaud et al. 2009) The periodic appearance of anomalously warm coastal waters off the coast of Angola coincides with increased rainfall over regions of Namibia, Angola and Zambia (Hirst and Hastenrath 1983; Rouault et al. 2003; Nicholson and Entekhabi 1987) The phenomenon of warming waters, Benguela Nios, might be a response to ENSO like processes in the equatorial Atlantic. However no consistent link between the phenomena in the two ocean basins has been found (Binet, Gobert, and Maloueki 2001) In addition, variability in the intensity of moisture circulation across the South Atlant ic, the South Atlantic midlatitude mode, has been observed to influence rainfall patterns in parts of southern Africa (Vigaud et al. 2009) The positive and negative phases of this mode are linked to shifts in intensity of the ITCZ a nd Angolan low (a strong tropical trough) which in turn affects precipitation in the OKZ basin (Vigaud et al. 2009) There is also a seasonal association for North Atlantic ocean atmospheric circulation patterns and river flow for the Okavango basin (Jury 2010)
42 To the east, the Indian basin constitutes the major source of austral summer rainfall (Rouault et al. 2003; Reason 2002; Washington and Preston 2006) There is evidence that Indian Ocean sea surface temperatures influence rainfall patterns across eastern and central southern Africa (Goddard and Graham 1999) including large portions of the OKZ basin Surface moisture fluxes that develop over the OKZ basin therefore reflect a gradient of the influences of circulation in the Atlantic and Indian basins, both of which may act to amplify or dampen the strength and signal of ENSO events. The varying contri butions from these two ocean basins may help explain the difference in rainfall patterns and associations with ENSO witnessed between the western and southern regions and the more northeastern area of the study area. The shifts in global climate forcings s uch as ocean atmospheric processes and climate phenomenon such as ENSO coupled with local variability (Shongwe et al. 2009) creates inconsistency in rainfall patterns across the OKZ basin having implications for agricultural and wildlife management (Phillips, Cane, and Rosenzweig 1998; Chase and Griffin 2009) Another important geographic consideration is that of a transitional zone i n response to ENSO between the western and southern, and northeastern sections of the OKZ basin The ITCZ migrates along an asymmetrical loop in southern Africa due to differential heating from topographical changes in the landscape and also from warm ea stern coastal waters that encourage convective activity and cold western coastal waters that have the opposite effect (Marchant et al. 2007) Variation in the direction and strength of zonal trade winds also will affect the ITCZ seasonal rainfall distribution patterns (Marchant et al. 2007) Thus rainfall variability in the western and southern
43 portion of the OKZ basin will be more sensitive to seasonal fluctuations in ITCZ movement. The nature of the seasonal migration of the ITCZ, brings more rainfall for the northeastern section of the OKZ basin where rainfall is more consistent than southern parts of the basin (Figure 4). This is also the area of mixing between the ITCZ and another zone of convergence, the Congo Air Boundary (CAB) (Tyson and Preston Whyte 2000) The CAB is a complex union of convergin g air streams that originate from East Africa and the Indian and Atlantic oceans which frequently create low pressure systems and conditions favorable to rainfall (Tyson and Preston Whyte 2000; Hansingo and Reason 2009) Another layer of complexity involves the Angolan low located in the southern portion of the OKZ basin (Gasse et al. 2008) This local climate feature has also been shown to be influenced by processes occurring in the Atlantic and Indian ocean basins and appears to fluctuate in strength along with the ITCZ, (Vigaud et al. 2009) The limits of all these climatic features c oincide at about at about 15 17 o S producing a dominance of influence of the Angolan low in the central southern region of the OKZ catchment, and a region of strongest influence of the ITCZ across the northeastern portion of the catchment (Gasse et al. 2008) Inter annual f luctuations in the exact location of this boundary will affect variability in rainfall patterns across the region potentially creating differing precipitation patterns for various areas of the OKZ catchment. If long term change in global climate creates pe rmanent shifts in the operation of these local climate features then it is even more imperative that transboundary initiatives are cognizant of precipitation patterns and how such low frequency or global climate change may affect rainfall over the larger r egion.
44 Precipitation variability is common in semi arid environments but increased variability or persistently drier conditions demands effective coping and adaptation strategies (Vetter 2009; Thornton et al. 2004) Those strategies rely on knowledge of spatial patterns of past and present precipitation and how shifts in the underlying climate may influence the timing, distribution, or frequency of rainfall. While the state of local and regional inputs for each sub catchment remains relatively strong (Table 4), the switch from above median conditions in P1 to below median conditions in P2 suggests that land use and conservation decisions should be cognizant of the historical shifts in above/below average rainfall and flow in the OKZ basin. Multiple years of drought may contribute to vegetation changes (Ringrose et al. 2007) and negatively influence perennial vegetation which affects rangeland management decisions (Vetter 2009) Fisheries in the lower part of the OKZ basin are affected by spatial variation in rainfall patterns as lake levels may not reflect long term lo cal climate changes but rather (Shaw 1983) The decrease in mean annual rainfall for the OKZ basin and increase in dry years associated with ENSO warm phase events suggests spatial ranges of large mammal species may have decreased to adapt to drier conditions (Chase and Griffin 2009) In addition, effective measures should be developed in response to better understand the effects of ENSO on rainfall variability for crop production as ENSO may have an influence on fa vorable cropping seasons (Phillips, Cane, and Rosenzweig 1998) Summary Unpredictable or inconsistent rainfall patterns will negatively influence marginal landscapes such as savannas where the t iming and quantity of rainfall is critical for rural livelihoods. Thus, an understanding of the spatial and temporal patterns of rainfall in the
45 OKZ catchment is paramount for transboundary management initiatives, with fixed spatial limits, that involve de cisions for wildlife and people in such dryland environments. The more recent period (1980 2005) had less overall mean annual precipitation input into the OKZ catchment than the period (1950 1975) before the late 1970s global climate shift. In addition, th e number of dry years and frequency of dry years associated to warm phase ENSO events has increased. These patterns suggest short term changes to the shifts in the functioning and response of the catchment to hydro meteorological patterns from 1950 2005. H owever, since 2008, the OKZ basin has experienced higher than normal rainfall and increased flooding for all three sub catchments (Wolski 2010) More research is necessary to determine how short te rm variability in rainfall patterns fit within the longer term changes of climate. Climate change is a global phenomenon that impacts the abundance and distribution of flora and fauna across a range of ecological levels, from individual species to entire e cosystems (Walther et al. 2002) making it vital to recognize and incorporate an understanding of spatial and temporal precipitation patterns into management plans for the future KAZA region.
46 Figure 2 1. Study region in south ern Africa outlining the three sub catchment areas that make up the larger Okavango Kwando Zambezi catchment. Local study areas in the Caprivi Region are defined in white (Okavango), stripes (Kwando) and gray (Zambezi).
47 Figure 2 2. Cell by cell 10 year moving window analysis which shows the most statistically significant cells that indicate a difference in precipitation input from time 1 to time 2 occurs between the 1975 79 and 1980 84 periods.
48 Figure 2 3. Time series of in the Okavango, Kwando, and Zambezi catchments in southern Africa from 1950 2005 (based on water years, October September) with ENSO warm and cold phases identified in background.
49 Figure 2 4. Spatial variability across the OKZ basin calculated as the coeffi cient of variation (CV). The graph shows decreasing CV with increasing mean annual precipitation.
50 Figure 2 5. The two images on top indicate an aggregated spatial depiction of median annual rainfall from 1950 1975 and 1980 2005 respectively. The bottom image shows the differences between the two time periods (P2 P1). c a b.
51 Figure 2 6. The two images on top indicate areas of large change between ENSO phases (La Nia El Nio) for each respective period. Th e lower image shows the difference between the two periods. Areas with higher values indicate a larger difference between warm and cold phase ENSO periods in P2: 1980 2005.
52 Figure 2 7. Box plot of average annual precipitation input for Period 1: 1 950 1975 (white) compared to Period 2: 1980 2005 (gray) for each OKZ sub catchment.
53 Figure 2 8 Shows the frequency of upper and lower climatological third of precipitation data indicating the number of a) dry and b) wet years for each time period and the probability of experiencing more (less) than expected dry (wet) years in Period 1 (Period 2). Significant at the 0.1 level, ** Significant at the 0.05 level. b. a.
54 Table 2 1. ENSO frequency across a) dry years from 1950 2005 and b) wet years from 1950 2005. Dry Years: 1950 2005 Basin El Nio La Nia Okavango 7* 2 Kwando 7* 1** Zambezi 5 3 Significant at the 0.1 level, ** Significant at the 0.05 level Table 2 2. ENSO frequency across a) dry years and b) wet years for P1 ( 1950 1975 ) and P2 ( 1980 2005). Dry Years P1: 1950 1975 El Nio La Nia P2: 1980 2005 El Nio La Nia Okavango 2 2 5** 0 Kwando 2 1* 5** 0 Zambezi 2 2 3 1 Wet Years P1: 1950 1975 El Nio La Nia P2: 1980 2005 El Nio La Nia Okavango 2 4 1* 2 Kwando 1 6** 0* 2 Zambezi 4 4 0* 2 Significant at the 0.1 level, ** Significant at the 0.05 level Wet Years: 1950 2005 Basin El Nio La Nia Okavango 2 7** Kwando 2 9** Zambezi 4 7** a. b. a. b.
55 Figure 2 9. Cells reporting statistical significant associations between the concordance ( x or more number ) of warm phase ENSO events and "dry" years during a dry year for Period 1 (1950 1975) and Period 2 (1980 2005), figures a and b respectively. No such associations of significance are detected during cold phases of ENSO.
56 Figure 2 10. Cells reporting statistical significant associations between the concordance of warm phase (panels a and b) and cold phase (panels c and d) ENSO 1975) and Period 2 (1980 2005). Panels a and b show cells display x or less number of expected warm phase events coincident with wet years. Panels c and d show cells that are x or more number of expected cold phase events coincident with wet years.
57 Table 2 3 Chi Square results suggest precipitation input within catchments operates similarly for all three catchments across the entire time period but there is less agreement in the Okavango for both P1 and P2, and less agreement in the Kwando for P1. Basin n Chi Sq Value DF Probability Fisher's Value Okavango Median P1: 1950 1975 25 3.59 1 0.087 P2: 1980 2005 25 3.59 1 0.087 Total: 1950 2005 55 8.04 1 ** Kwando Median P1: 1950 1975 25 0.44 1 0.671 P2: 1980 2005 25 10 1 ** 0.00365 Total: 1950 2005 55 8.04 1 ** Zambezi Median P1: 1950 1975 25 11.99 1 ** 0.00217 P2: 1980 2005 25 20.3 1 ** 0.00004 Total: 1950 2005 55 36.87 1 **
58 Table 2 4. Observed relative frequencies between local and regional precipitat i on input Local versus Regional Input by Catchment Okavango Catchment Overall: Period 1: Period 2: 1950 2005 1950 1975 1980 2005 Regional Input Regional Input Regional Input Local Input Above Below Above Below Above Below Above 0.34 0.17 Local Input Above 0.54 0.17 Local Input Above 0.17 0.12 Below 0.15 0.34 Below 0.12 0.17 Below 0.17 0.54 Kwando Catchment Overall: Period 1: Period 2: 1950 2005 1950 1975 1980 2005 Regional Input Regional Input Regional Input Local Input Above Below Above Below Above Below Above 0.34 0.17 Local Input Above 0.44 0.2 Local Input Above 0.2 0.09 Below 0.15 0.34 Below 0.2 0.16 Below 0.09 0.61 Zambezi Catchment Overall: Period 1: Period 2: 1950 2005 1950 1975 1980 2005 Regional Input Regional Input Regional Input Local Input Above Below Above Below Above Below Above 0.46 0.04 Local Input Above 0.71 0.04 Local Input Above 0.24 0 Below 0.05 0.42 Below 0.08 0.21 Below 0.04 0.72
59 CHAPTER 3 LINKING VEGETATION R ESPONSE TO SEASONAL PRECIPITATION IN THE OKAVANGO KWANDO ZAMBEZI CATCHMENT OF SOUTHERN AFRICA Variability in inter and intra annual precipitation affects ecosystem structure and function in all climates (Scanlon et al. 2005; Me rbold et al. 2009; Huxman et al. 2004) In dryland ecosystems water availability is the most important climate constraint on plant growth (Prince, Goetz, and Goward 1995; Nemani et al. 2003) Dryland areas, including arid, semi arid, and dry sub surface, including about half of the African continent (Reynolds et al. 2007) Under different scenarios of future climate change, many dryland regions may become even drier and precipitation timing and amount may shift (IPCC 2007) The underlying dryland mechanism s that govern atmosphere plant soil processes are strongly influenced by water availability and any subtle shift in precipitation will influence the ability of plants to respond to such change (Scholes and Walker 1993) The relationships between vegetation productivity and precipitation for such regions are determined n ot only by total rainfall, but also by precipitation timing and variability. An understanding of relationships between current climate variability, especially within seasons, and vegetation processes is necessary before projecting likely impacts of future climate change. All dryland areas in southern Africa are water limited for part of the year and the often unpredictable pattern and timing of water distribution plays an important role in the structure and function of savanna vegetation (Scholes and Walker 1993) These ecosystems consist of a mix of tree shrub grass vegetation that first appeared with a global expansion of C 4 vegetation about 5 7 million years ago (Cerling et al. 1997) and today exist along the 30 o N and 30 o S latitudinal belts within a wide range of mean
60 annual precipitation (MAP) regimes (~200 to ~3000 mm MAP) (Scholes and Archer 1997; Bond 2008) The mixture of continuous grass interspersed with woody vegetation results from interaction of multiple ecological factors (precipitation, soil nutrients, fire and herbivory) over space and time (Skarpe 1992) Projected anthropogenic climate change, especially in southern Africa, most likely includes drier conditions and increased inter annual variability (Giannini et al. 2008) which will probably alter tree grass biomass and cover ratios and subsequently affect other ecosystem processes such as water and nutrient cycles (Archibald and Scholes 2007) Water is not the only limiting resource controlling ecosystem dynamics but is a necessary component for every physiological process (Rodriguez Iturbe 2000) Across much of southern Africa, there is a strong seasonal response to wet and dry periods (Scholes and Walker 1993) Seasonal rainfall amount in the subtropical drylands is largel y controlled by the movement of the Inter Tropical Convergence Zone (ITCZ) (Scholes and Archer, 1997). Other parts of the water cycle, such as soil moisture, evapotranspiration, and flooding also contribute to dryland vegetation growth but are dependent on the quantity, timing, and frequency of rainfall. In the same way, other biotic factors (ex. fire, herbivory) that contribute to driving landscape change also depend on the underlying hydro metrological factors to maintain co existence of grasslands and wo ody vegetation in dryland areas (Sankaran et al. 2005; Wang 2003; Breshears and Barnes 1999) A common approach to examine the relationship between precipitation and savanna gross primary productivity (GPP) is remotely sensed estimates of vegetation productivit y (Sjostrom et al. 2009; Sims et al. 2006; Sims et al. 2008; Tucker and Sellers 1986) The Normalized Difference Vegetation Index is ideally suited for semi arid
61 regions where the index does not saturate at high foliage biomass (Nicholson and Farrar 1994; Richard and Poccard 1998) NDVI is the ratio of the difference between the NIR and Red band reflectances divided by their sum (Los et al. 2000; Justice et al. 1985; Huete et al. 2002; Cohen and Goward 2004; Tucker 1979) Despite the fact NDVI measurements are not a pure value of leaf chlorophyll, trends in GPP measured by NDVI are still a valuable proxy for vegetation productivity of different vegetation types (Archibald and Scholes 2007; Chapin, Matson, and Mooney 2002) While other vegetation indices may have advant ages over NDVI in dryland areas (Gobron et al. 2000; Leprieur, Kerr, and Pichon 1996) NDVI is one of the few long term vegetation records available across multiple sensors over many years making it useful for longitudinal vegetation studies (Fisher and Mustard 2007; Brown et al. 2006) and also has a good correspondence with vegetation productivity in semi arid regions (Martiny et al. 2006) Furthermore, it is well known and so allows us to compare our research findings to a large and significant literature and make useful comparison to similar studies. Many studies of the precipitation vegetation relationship in Africa have used NDVI to identify responses of vegetation productivity to inter and intra annual rainfall patterns (Martiny, Richard, and Camberlin 2005; Fuller and Prince 1996; Ji and Peters 2005; Richard et al. 2008; Richard and Poccard 1998; Martiny et al. 2006; Nicholson, Davenport, and Malo 1990; Camberlin et al. 2007; Goward and Prince 1995) At the yearly time step, the association be tween annual NDVI and precipitation for large portions of the African continent shows a correlation with increasing NDVI along a latitudinal precipitation gradient (Martiny et al. 2006; Martiny, Richard, and Camberlin
62 2005; Camberlin et al. 2007; Richard and Poccard 1998) The strongest association of the NDVI precipitation relationship exists for semi arid regions (200 600 mm yr 1 ) (Martiny et al. 2006; Fuller and Prince 1996) When the relationship is examined on a month by month basis, vegetation responds to both increases and decreases of rainfall with a greening or senescen ce within 0 2 months (Richard and Poccard 1998; Martiny et al. 2006; Nicholson, Davenport, and Malo 1990; Fuller and Pri nce 1996) Fuller and Prince (1996) used the Global Inventory Modeling and Mapping Studies (GIMMS) Advanced Very High Resolution Radiometer (AVHRR) monthl y maximum value composite (MVC) NDVI and found the strongest positive correlation in vegetation response to rainfall to be a 1 month lag in southern Africa. Martiny et al (2006) also use the AVHRR GIMMS but conclude that the lag of vegetation response to r ainfall varies depending on rainfall regime with less importance placed on other co varying factors. Using the peak of the rainy season (February) as the rainfall measurement, they find a one month lag in the East Kalahari (higher annual rainfall) while th ere is a two month lag in the Karoo and West Kalahari (lower annual rainfall). Richard and Poccard (1998) examined Africa south of 15 o latitude at a spatial resolution of 1x1 o identify a 1 2 month lag of AVHRR NDVI to monthly precipitation. MODIS NDVI is often used to track ecosystem phenology (Archibald and Scholes 2007; Knight et al. 2006; Choler et al. 2010; Huemmr ich et al. 2005) Phenology refers to timing of plant production and its response to climate characteristics (White et al. 2009) and differentiati ng the phenology of tree and grass savanna species is useful for studying biological (Owen Smith and Cooper 1989) ecological (Shackleton 1999) and hydrological (Borchert 1994) ecosystem processes. However, the relationship between
63 plant phenology and precipitation seasonality in drylands is not straightforward, as different species respond to timing, frequency, and intensity of rainfall in diffe rent ways (Archibald and Scholes 2007) The use of an overall NDVI estimate for savanna vegetation is not sufficient when representing this ecosystem in global climate models (Field et al. 1998) because the level of complexity that results from the mix of different functional groups (grass, trees, shrubs, etc.) is not ad equately captured for phenological cycles incorporated into global climate models (Chase et al. 1996) This is especially relevant for unique tree grass ecosystems where potential for different growth strategies exists year round due to tree and grass response to timing of water availability (Archibald and Scholes 2007) A wide range of temporal and spatial scales of analysis of NDVI response to rainfall exists (Nicholson, Davenport, and Malo 1990; Richard et al. 2008; Mart iny et al. 2006; Fuller and Prince 1996; Richard and Poccard 1998; Fuller 1999; Archibald and Scholes 2007; Zhang 2005) Peak NDVI lags more than 1.5 months after the peak of the rainy season (February) in southern Africa (Martiny et al. 2006) However, the response of savanna vegetation to the onset of the rainy season varies depending on the tree grass cover ratio (Fuller and Prince 1996) Grassland response to the start of the rainy season (Oc tober) is more tightly coupled with precipitation with minimal lag while timing of woody vegetation leaf expansion occurs a couple months prior to the first rains (Fuller and Prince 1996; Archibald and Scholes 2007) In addition, the timing of leaf flush peak at the beginning of the dry season, and leaf fall at the end have large inter annual variation (Do et al. 2005) with vegetatio n dormancy occurring with a lag of 3 months after the end of the rainy season (Zhang 2005)
64 We examine the response of GPP at the beginning of th e dry season to monthly variation in wet season rainfall. The identified 0 2 month vegetation response to rainfall suggest that February, March, and April rainfall should have the strongest association to beginning of dry season GPP measured in April for s avanna vegetation covers with increased amount of grass cover. Vegetation status at the beginning of the dry season influences the availability of water resources and other important forage materials for wildlife, therefore affecting wildlife species throu ghout the dry season. This is important for a region where large proportions of the human population and economy are dependent upon wildlife tourism (Barnes, MacGregor, and Weaver 2002 ) In addition, the structure of the rainy season may influence savanna vegetation covers differently which, in turn, will affect other savanna ecosystem processes (Sankaran et al. 2005; Martiny et al. 2006) This study investigates the response of photosynthetic activity at the beginning of the dry season to month by month precipitation variability in the Okavango Kwan do Zambezi (OKZ) catchment of southern Africa, using two remotely sensed data sets, MODIS (Moderate Resolution Imaging Spectroradiometer) 13A1 (vegetation indices) and mean monthly precipitation data from the Tropical Rainfall Monitoring Mission (TRMM) (0. 25 o x 0.25 o ). We hypothesize that savanna land cover with more grass will show a stronger positive response to precipitation later in the rainy season than vegetation with proportionally more woody plants. The main focus is the regional relationship betwee n mean monthly rainfall and vegetation GPP at the beginning of the dry season across the OKZ catchment. We address the following questions: What is the relationship of GPP measured at the beginning of the dry season to several earlier
65 ? How much variation in GPP is explained by relationships between vegetation and different months of wet season rainfall? And what coupling relationships exist between different vegetation response groups to seasonal precipitation from 2000 2009? Current c onsensus in the literature emphasizes the complex and distinctly different phenological responses tree and grass species have to inter and intra annual precipitation for savanna ecosystems. Being able to differentiate the influence of wet season months of rainfall on vegetation productivity leading into the dry season provides a crucial link to better understanding precipitation vegetation interaction in savanna ecosystems. Materials and Methods Study Region The study region comprises the Okavango, Kwando, and Zambezi catchments, situated in sub tropical southern Africa, and cover a total area with an annual precipitation range of 400 2200 mm/yr (Figure 3 1). High variability of intra annual and inter annual rainfall across the three catchments varies due to the underlying influence of climatic controls such as the Inter tropical Convergence Zone (ITCZ) and atmospheric circulation patterns (McCarthy et al. 2000) .The lower part of the study region is semi arid, defined by scarce and typically unpredictable patterns of precipitation while the upper part of the region has a higher average annual precipitation and lower inter annual var iability. The majority of the Okavango and Kwando catchments and all three headwaters are located in Angola although a good portion of the Zambezi catchment is in western Zambia. Soils typically range from ferralsols in upland areas of Angola and Zambia to sandy arenosols in mid elevation regions, and clay containing gleysols throughout lower, flatter areas of the catchments (Batjes 2000)
66 The mid elevation Kalahari sandveld, which makes up the majority of the region, consists of deep Aeolian sand deposits, longitudinal dunes, pans and fossil valleys with the variations in vegetation type along the north south, h igh low rainfall gradient (D ube and Pickup 2001) The OKZ catchment is located in the fu ture Kavango Zambezi Transboundary Conservation Area (KAZA). KAZA is a politically defined conservation area that spans Angola, Zambia, Namibia, Botswana, and Zimbabwe. The transboundary conservation area is designed to provide vital wildlife corridors amo ng multiple protected areas. Topography is very flat across the southern portion of the region. The topography of the lower area (especially the Caprivi Strip and northern Botswana) makes it difficult to clearly separate these different catchments because hydrological flows are inconsistent across the flat terrain. Datasets The study region was delineated hydrologically into three neighboring watersheds with void filled Shuttle Radar Topography Mission (SRTM) DEM data as well as derived flow direction and accumulation grids at a spatial resolution of 15 arcseconds from the World Wildlife Fund HydroSHEDS project (Lehner, Verdin, and Jarvis 2006) These data were processed using the ArcHydro data model and toolset (Maidment 2002) to extract drainage networks and basin delineations, which form the basic units for further analyses. We examined the long and short run precipitation patterns across the region using the TRMM 3B43 dataset (version 6). TRMM 3B43 data are best estimates of average daily rainfall rate by month (Kummerow et al. (2000) combining the TRMM instrument rain calibration algorithm estimates (3B42) and several rain gauge data
67 sources using Huffman et al. (1997) method. Daily instrument data are severely limited in terms of availability and quality in this study region. At a relatively coarse scale of 0.25 x 0.25 TRMM 3B43 data are comparable to rain gauge estimates and show very little bias in West A frica (Nicholson et al. 2003) The TRMM 3B43 data product has the least bias of any data across both season and reg ion and shows a high degree of association with regional rain gauge estimates (Adeyewa and Nakamura 2003) Average daily rainfall rates by month w ere converted to total monthly rainfall in millimeters for further analyses. We use NDVI from MODIS 13A1 data (version 5) with 500 x 500 m spatial resolution to approximate GPP at the beginning of the dry season (April). The delineated study region includ es two full MODIS 16 day composite footprints downloaded from Land Processes Distributed Active Archive Center (LP DAAC) which is a component of NASAs Earth Observing System (EOS) Data and Information System (EOSDIS). Data were acquired for each year 2000 2009 and encompass dates between April 23 rd and May 8 th for each composite (see Leeuwen et al. 1999 for detailed description of the compositing algorithm). Data Sampling and Standardization A temporal and spatial scale mismatch exists between the monthly T RMM precipitation pixels (predictor variables) measured at 0.25 o x 0.25 o and the MODIS NDVI pixels (response variable) measured at 500 x 500 meters. For any point in time a single TRMM pixel value will overlap many MODIS pixels (~3600). To minimize bias one MODIS pixel per water year was randomly sampled within a single TRMM pi xel area, yielding 17,900 data points
68 Since this research is primarily concerned with terrestrial, semi arid vegetation we used a 5 kilometer buffer around permanent and seasonally flooded areas to exclude samples that may contain water and riparian veget ation, resulting in a final sample of 15,466 observation points. Although a smaller buffer may be have been sufficient, the 5 km buffer was selected based on field observations regarding the spatial extent of seasonal water presence and flooding affected v egetation. The buffer safely excluded known perennial water sources and many seasonally flooded areas. These areas become inundated after rainfall events at various lags and were removed from the analysis because of the influence of semi /permanent water o n NDVI estimates. Modeling Approach We use geographically weighted regression (GWR) to account for spatial non stationarity in the relationship between rainfall and beginning of dry season vegetation productivity to minimize the effect of unmeasured, spatially varying covariates (Brunsdon, Fotheringham, and Charlton 1998; Foody 2003) GWR extends the traditional regression framework by allowing local model parameters to be estimated at any point within the study area by incorporating a spatial weighting of observations and allowing the estimated models to vary across space (Fotheringham et al. 2000). Observations are weighted by their proximity to the point at which the r elationship is estimated so that the weighting of an observation is not constant but varies with each location GWR models assume that observations nearest the estimation point are more informative than those further away (Charlton et al. 20 0 6). Sampled N DVI and TRMM data were standardized prior to modeling by subtracting the mean and dividing by the standard deviation. The unitless measures of within month variation were used to assess relative explanatory power of variability
69 instead of absolute effects of monthly rainfall. The relationship between April MODIS NDVI (dependent variable) and standardized TRMM values for seven months of preceding rainfall (7 independent variables) varies across space, minimizing the effects of any unmeasured, important covar iates while enabling us to compare the relative vegetation productivity. The GWR model that for April NDVI is: NDVI Apr ( u ) = 0 ( u ) + 1 ( u ) Rain Oct + 2 ( u ) Rain Nov 7 ( u ) Rain Apr + e ( u ) Where NDVI Apr (u) represents modeled NDVI for April at location u and 0 7 (u) Rain Apr are the intercept and regression coefficients specific for the model at u estimated from distance weighted observations nearest to th at location and e ( u ) is the random error term Locations need not be coincident with observation points, and from the stratified sample across space and time regression surfaces for all eight estimated coefficients were used to predict April NDVI on a 500 x 500m grid for each year 2000 2009. During estimation GWR requires the choice of a kernel and its associated distance weighting function. The kernel function and associated bandwidth parameter, which may be thought of as a smoothing parameter, (Fotheringh am et al. 2002) may be chosen subjectively if a strong theoretical base exists or through more objective quantitative methods that have either a fixed or variable kernel function and bandwidth parameter (Brunsdon, Fotheringham, and Charlton 1998; Foody 2003) We used a Gaussian, fixed kernel with an optimal bandwidth chosen to minimize Akaike
70 Information Criterion ( AIC c ) of the estimated model. The Gaussian, fixed kernel incorporates all observations into each local model calculation but weights each observation by the distance from each regression po int according to a Gaussian decay model. To determine the bandwidth parameter used in the weighting decay function, the AIC c were used as diagnostics. These were calculated fo r the model at bandwidths across a range of 18 to 90 km at equal one km steps. The results were used to identify the bandwidth parameter value with the lowest model AIC c and that locally minimized spatial autocorrelation. Using this strategy an ideal bandw idth parameter value was identified (39,356 m, see Table 3 1) to use in the weighting decay function. We calculated the final GWR model in ArcGIS 9.3 using the standardized NDVI to compare the influence of prior monthly rainfall on April NDVI. We conducte d post hoc examinations of local coefficient estimates and determined at the optimal bandwidth that some multicollinearity persisted at the extreme coefficient values, but overall trends showed little correlation and coefficient interpretations are valid. Model Comparison across Vegetation Classes The coupling between precipitation and vegetation varies by vegetation type (Archibald & Scholes, 2007). The results of the GWR model create a continuous gradient of coefficients for each rainfall predictor in th e model. We extracted and compiled coefficient estimates across the study region and for each class of the Vegetation Map (White 1983) (Figure 3 2). The vegetation map, although dated, was tested for accuracy by comparison with 1) training sample data collected during 2006 2009 field seasons and 2) a coarser classification produced by the IGBP classification (Friedl et al. 2010) The land cover groups identified by Whi
71 with ground cover observed during ground data collection in 2006 2009. The White and IGBP maps were found to be consistent with each other. We used the White map e detailed breakdown of woodland groups, and defines a transition area between upland Miombo woodland and dry deciduous and secondary grasslands in the lower part of the OKZ basin. We extracted sample points of NDVI and precipitation estimates within six v egetation classes: wetter Zambezian Miombo woodland, Zambezian evergreen forest, Brachystegia thicket and edaphic grassland, dry deciduous and secondary grassland, edaphic and secondary grasslands, and Colophospermum mopane woodland. Herbaceous swamp and se mi aquatic vegetation were excluded from analysis We compared the coefficient values of each predictor month by land cover type for each across different vegetation covers. Analyses were conducted in R (2.10.1) and ENVI 4.3. Results Spatial and Temporal Relationship between NDVI and R ainfall Average pixel by pixel calculation indicates higher NDVI values in the upper catchment and a gradient of increasing precipitation from s outhwest to northeast (Figure 3 3). The darker band of values that stretches down the center of the basin is low lying higher NDVI values correspond approximately to Mio mbo woodland and dry evergreen also receive the highest amount of total precipitation during the rainy season (Oct Apr). While the Okavango Delta area located at the bottom of the OKZ basin also shows
72 areas with high NDVI, we exclude this area, along with other wetland and riparian areas in the basin, since NDVI values within these areas are highly influenced by basin wide precipitation and flooding rather than the loc al precipitation patterns. Mean April NDVI for each land cover and mean total wet season precipitation (October April) for the 9 years covered by the MODIS dataset (2000 2009) is shown in Figure 3 4. Miombo woodland and dry Evergreen forests have the hig hest, most consistent average April NDVI values (ranges are 0.67 0.71 and 0.68 0.65 respectively). In contrast, mean April NDVI in Mopane woodland increases and decreases with average wet season precipitation, ranging from 0.39 to 0.56. Similarly, the dr y deciduous and secondary grassland land cover also tends to increase or decrease with corresponding change in precipitation though over a smaller range than mopane woodlands (0.48 0.58). The edaphic and secondary grassland maintains an April NDVI value ar ound 0.55 without as large a range as dry deciduous and secondary grassland. The Brachystegia thicket maintains a relatively consistent April NDVI around 0.62 ( 0.015 Std. Dev.) and provides a transition between the wetter Miombo woodland and the drier, m ore deciduous mix of woody and secondary grassland areas in the OKZ catchment The different April NDVI patterns across the 10 years indicates that varying amounts of precipitation influence the overall productivity for the beginning of the dry season diff erently for specific savanna vegetation types. Model Results A comparison of the results from a global OLS linear regression with the GWR regression is presented in Table 3 1 and Figure 3 5 to illustrate the effects of spatially co varying factors such as soil type and dominant vegetation cover. Results of the global ordinary least squares (OLS) linear regression indicate previous months of wet
73 season rainfall explain 37.7% (adjusted R 2 ) of the variation in April NDVI. The strength of the relationship between NDVI and rainfall increases substantially with the local GWR model, with rainfall explaining 78.2% (adjusted R 2 ) of the variation in April NDVI. Table 3 1 shows that the difference between AIC c values for the global OLS and local GWR model is very large indicating the two models are not equivalent in their explanatory power. Residuals from subtracting the modeled standardized April NDVI predictions from the observed, standardized April NDVI results suggest the GWR model does a better job at reducing spatial autocorrelation (spatial pattern in the residuals) than the OLS model (although it does not completely eliminate it) (Figure 3 5). While not all spatial autocorrelation is accounted for in the GWR model the spatial dependence in model errors is g reatly reduced resulting in better predictions of April NDVI than the OLS model while using the same number of wet season precipitation months as predictors in the model. GWR Results by Land Cover We sampled the standardized coefficient surfaces for each p oint of the original ure 3 2) and (Kampstra 2008) (Figure 3 6). The standardized monthly rainfall coefficients represent the effects of lower or higher than average monthly rainfall on April NDVI. The cou pling between rainy season months and vegetation varies by vegetation type. Model estimates across land cover type (Figure 3 6) suggest vegetation classes with woodland components (Miombo woodland, dry evergreen, and Brachystegia thicket) and consequently higher NDVI (Figure 3 4) are less affected by variability in monthly precipitation (i.e. the medians of most predictor coefficients on the bean plots are near zero and the distribution of coefficients indicated by the bean plot
74 histograms straddle zero ev enly) In this study, February precipitation was the most common month associated with April NDVI for all land covers in the OKZ basin although the strength of association varied by vegetation class. Regardless of where along the savanna continuum each lan d cover is situated, there is a positive effect of February precipitation on April NDVI. The month of February is the only month that shows a positive association for the two higher NDVI classes, Miombo woodland and dry evergreen forest. October precipita tion has a slight positive association on the dry evergreen forest but the beginning of the rainy season is not associated with Miombo woodland productivity in April. The Brachystegia thicket and edaphic grassland land cover, which provides a transition b etween the upland Miombo woodland and drier, more deciduous and secondary grassland, shows a slight positive association between April NDVI and February, March and April rainfall. The dry deciduous and secondary grassland land cover shows a disproportiona tely large positive association with dry season productivity for the same three months when compared to rainfall in the beginning of the wet season. Colophospermum mopane woodland is also positively associated with late wet season precipitation, although t his particular woodland type is positively influenced by December precipitation and negatively influenced by October precipitation. The edaphic and secondary grasslands on Kalahari sands do not show strong associations between any of the wet season months and April NDVI, except for a slight positive association with February rainfall. This land cover behaves more similarly to the denser woodland and forest covers of Miombo and Evergreen land covers. The strongest association in its April NDVI productivity and precipitation exists with February
75 precipitation and, to a lesser degree, slight positive associations for all other months minus October and April. Discussion Our results indicate that precipitation patterns within the wet season, not just annual or seasonal totals, drive vegetation productivity at the beginning of dry season, with different rainy season months more strongly associated for some vegetation types than others. Specifically, those vegetation types with more mixed tree grass cover composit ion rather than woodland dominated covers tend to have an April NDVI more strongly influenced by February April months of rainfall. An exception to this was Colophospermum mopane woodland which showed more variation in April NDVI response than other woodla nd dominated vegetation covers like Miombo woodland and Zambezian evergreen forest. The end of the rainy season occurs relatively uniformly across the region in April (Zhang et al 2005). The positive and negative model coefficient patterns for each month for respective vegetation classes (Figure 3 6) identify lags in April NDVI response to the different months of rainfall in the wet season. For example, while most coefficient values for rainfall months are zero for Miombo woodland and Zambezian evergreen forest, February has a positive association with April NDVI. This suggests there is a two month lag in response of April NDVI to February rainfall. The 0 2 month lag in vegetation response to rainfall agrees in general to the vegetation response to rainfal l measured for other parts of the growing season in southern Africa (Martiny et al. 2006; Fuller and Prince 1996; Goward and Prince 1995; Richard and Poccard 1998) This two month lag after the height of the rainy season most likely results from a set of ecosystem processes that differ at the level of vegetation type (Williams et al. 2009)
76 Different types of savanna woodland species adapt different coping mechanisms and strategies to deal with the highly variable and seasonal savanna environment (Shackleton 1999; Fuller 1999) For Miombo woodland and Zambezian evergreen forest the minimal effect in February may suggest a short term storage of water and carbohydrate reserves th at are accessed when the rainy season ends and provides an indication to how sensitive the ecosystem may be to within wet season rainfall (Schwinning et al. 2004) However, the effect may be minimal with a stronger, inter annual effect more dominant on vegetation response depending on number of previous dry years or high water stress prior to rainfall (Richard et al. 2008; Martiny, Richard, and Camberlin 2005) The longer inter annual lag that may be present depends on ecological processes that influence nutrient and water cycling (Williams et al. 2009; Martiny, Richard, and Camberlin 2005) although further investigation is necessary to determine how vegetation types respond to inter annual variation within the OKZ basin. The specific responses of different woodland species to months of wet season rainfall also helps explain the much stronger association in the response of Colophospermum mopane April NDVI to wet season rainfall. The strong seasonal response of Colophospermu m mopane corresponds to previous studies which use field and remotely sensed data to show the intra annual rainfall effect on these woodlands (Fuller 1999; Fuller and Prince 1996) Colophospermum mopane woodlands in the OKZ basin mostly are found on im pervious clay soils that limit the depth of water penetration and are not conducive to water storage (White 1983) The shallow root system of Colophospermum mopane woodlands also means this woodland species will compete more with grass species for soil moisture in the top 25 cm of soil. The transit ion zone of
77 Brachystegia thicket and edaphic grassland shows precipitation in months during the latter half of the rainy season (Feb Apr) having a slight association with April NDVI although the higher percent of mixed land cover may dampen the effect obse rved in more open, less woody land cover classes. For those vegetation types with higher grass cover (edaphic and secondary grasslands and dry, deciduous and secondary grassland) or more open canopy areas (such as Colophospermum mopane woodland), March an d April precipitation have similar if not more influence on April NDVI values than does February. Colophospermum mopane woodlands also have a negative October coefficient suggesting rainfall at the outset of the wet season suppresses April GPP. However, t he rest of the rainy season months show a positive association to Colophospermum mopane April GPP. The difference in association may relate to the highly seasonal nature of Colophospermum mopane woodland and their response to variable timing of the onset o f the rainy season (Fuller 1999; Veenendaal, Kolle, and Lloyd 2004) For all vegetation types with higher grass cover, the strong one month and concurrent effect of April and M arch precipitation on April NDVI may be due to the higher dependence grasslands have on rainfall for seasonal productivity patterns (Scanlon, 2002). The shallow root depth of grass roots suggests these species will be more sensitive to temporal and spatial variations in water availability making the timing of rainfall, rather than total amount, and its effects important on plant productivity ( Sher, Goldberg, and Novoplansky 2004) This study identifies the importance of the precipitation vegetation relationship for GPP leading into the dry season and also shows that the strength of association varies
78 depending on the ratio of tree to grass c over. The variation in response across the OKZ basin corresponds to previous findings that study differences in response to rainfall across savanna vegetation types (Fuller and Prince 1996; Archibald and Sch oles 2007) Our study complements previous studies that identify the importance of vegetation response to different rainfall regimes (Martiny et al. 2006; Nicholson, Davenport, and Malo 1990) by emphasizing this relationship is further broken down by savanna vegetation type. Timing of wet season rainfall on GPP for beginning of the dry season is a vital component to understanding shifting dynamics of dryland ecosystems (Archibald and Scholes 2007; Scanlon et al. 2005; Fuller and Prince 1996) Our study complements but differentiates from these previous studies by investigating the NDVI precipitation relationship leading into the dry season using a GWR model. The GWR model provides a useful statistical tool to minimize the effects of unmeasured, spatially varying factors and the negative impacts of autocorrelation on conventional models. The focus on April vegetation productivity provides insight to vegetation status leading into the dry season which will affect ecosystem processes such as nutri ent cycling, water energy budgets, and forage availability throughout the dry season. Our findings support the idea that vegetation with shallower root systems utilizes soil water more quickly than vegetation with deeper root systems, more commonly found i n more humid regions, which can access sub surface water months after the initial precipitation event (Porporato et al. 2003) These proces ses operate at different spatial and temporal scales depending on the composition and structure of the savanna ecosystem. Therefore quantifying the timing of the prior month rains is an
79 important control of April GPP, therefore dry season vegetation proces ses for different savanna vegetation covers. Summary Findings from this study support our hypothesis that April GPP for savanna vegetation types with higher grass cover positively respond with a tighter coupling to late wet season rainfall. Month by month variability affects system drivers, such as soil moisture or fuel load, in turn influencing April GPP differently for various savanna vegetation types. If less rainfall occurs at the end of the rainy season it will most strongly influence GPP for savanna v egetation covers of grassland and open canopy woodlands. As a result, it may cause more frequent and intense fires early in the dry season due to a drier fuel load. The designation of average monthly precipitation is an arbitrary means to identify the effe cts of rainfall on beginning of dry season productivity and a finer break down in rainfall events may highlight other important interactions between wet season rainfall and vegetation productivity. However, expressing wet season precipitation by monthly to tals shows how months in the beginning, middle and end of the rainy season influence NDVI, and by proxy GPP, at the beginning of dry season. This study highlights the importance of within season precipitation variability to vegetation growth for the begi nning of the dry season. While a landscape level approach makes detailed analysis of growth patterns of different vegetation types difficult, the analysis of finer taxonomic scale classes of savanna vegetation provides shows how timing of rainfall during the wet season influences vegetation types with different tree grass ratios along the savanna continuum. Rainfall is only one environmental variable that influences the growth and function of vegetation but is the
80 most limiting factor in savanna ecosystem s. In these water limited systems, the response of different savanna vegetation types April GPP to the various months of wet season rainfall depends on multiple ecosystem processes that relate to how grassland and woodland species differ in their timing an d uptake of available water through the season. We show that intra annual variability explains a large amount of observed variation of the productivity of vegetation at the beginning of the dry season and that the relative importance of different months on savanna vegetation type varies throughout the wet season. Other studies suggest that altering the seasonal timing of rain will strongly influence its role in ecosystem processes (Schwinning et al. 2004) making the quantification of both beginning and end of the rainy season important for understanding the effects of future climate change.
81 Figure 3 1. S tudy region depicting the three catchment areas of interest with elevation from a Digital Elevation Model (source: World Wildlife Fund HydroSHEDS proect) with a spatial resolution of 15 arcseconds.
82 Figure 3 that shows a detailed classification of different land covers in the OKZ catchment.
83 Figure 3 series of (a) mean April NDVI and (b) total mean wet season precipitation respectiv ely from 2000 2009. The spatial distribution highlights the highest mean precipitation in the northeast of the OKZ basin and the higher April NDVI values for areas of predominant woodland.
84 Figure 3 4. Time series of averaged April NDVI for the O KZ catchment plotted against the total average precipitation for each wet season (October April) across 2000 2009.
8 5 Table 3 1. This table compares the global OLS model diagnostics to those of the GWR. GWR fits the data across the catchment much better as indicated by the delta AIC and increase in R2. Model Type RSS R 2 Adj. R 2 c Parameters Estimated Bandwidth Global OLS 12007.54 0.329 0.329 16646.95 8 N/A GWR 3656.42 0.795 0.754 0 3015.40* 39,356 m* contributing to local model estimates. The bandwidth parameter (here fixed, in m) controls at what distance a gaussian spatial weighting is applied (Fotheringham, 2002).
86 Figure 3 5. Residuals from the estimates of the (a) global OLS and (b) GWR models. Spatial autocorrelation still exists in high/low residuals in the GWR, but is significantly minimized when compared to the OLS.
87 Figure 3 sample point has its own GWR model. The median coefficient for the Y intercepts and each month is indicated by the long horizontal bar above each month (independent variable) Each individual local model is represented by the spread of short horizontal bars above and below the med ian value and the distribution for each predictor variable shown with vertical histograms.
88 CHAPTER 4 DETECTING LONG TERM VEGETATION TRENDS BY ADJUSTING FOR SEASONALITY IN A SAVANNA LANDSCAPE Savanna ecosystems cover approximately 1/5 th surface and over half of the African continent (Sankaran et al. 2005, Scholes & Archer, 1997) are third only to tropical and temperate forests in terms of global terrestrial carbon sequestration (Field et al. 1998) and woody vegetation in these semi arid regions are increasingly recognized as a vital component of the global climate system (Dewees et al. 2010, Rotenberg & Yakir, 2010) Many rural communities in these marginal landscapes depend on local natural resources and shifts in the ratio of tree to grass cover across different savanna vegetation types influence local livelihood sustainability (Shackleton et al. 2007) Specifically, if savanna composition and structure changes, the kind of resources available to rural communitie s will be affected because different savanna woody and grass species are preferred for building materials, firewood, thatching, and medicines (Dewees et al. 2010, Shackleton & Shackleton, 2004) Determining how best to measure changes in savanna vegetation over time is critical considering the tight dependence on local natural resources and potential contributions of carbon gases to global climate patterns that may result as a consequence of long term changes in the tree grass ratio. Vegetation productivity in arid and semi arid regions of Africa is influenced by inter and intra annual precipitation variation (Hulme et al. 2001, Nicholson, 1986, Nicholson, 2001) Global climate phenomena such as the El Nio Southern Oscillation (ENSO) and sea surface temperatures in the Atlantic and Indian Oceans regulate the variability in longer term mean annual precipitation across Africa (Mason, 2001, Nicholson, 1997, Nicholson, 2000) Historical shifts in distribution and frequency of
89 r ainfall influenced the expansion and contraction of desert regions, presence of marshes in the western Sahara, and lake sizes in the Rift Valley (Hulme et al. 2001, Nicholson, 2001) Recent climatic changes include a drying trend over large parts of Africa that, in part, is attributable to anthropogenic climate change (Giannini et al. 2008) These changes include semi arid savanna ecosystems in southern Africa although separating the influence of variation in rainfall from human induced changes on vegetation productivity is difficult (Wessels et al. 2007) As such, the strong seasonal response of savanna vegetation to wet and dry periods may obscure detection of long term chang e that results from decadal scale climate shifts or from land use decisions. Separating the seasonal signal from other factors influencing savanna vegetation dynamics is necessary for identifying long term vegetation change on the landscape. This study ex amines vegetation status at the beginning of the dry season (BDS) in terms of remotely sensed gross primary production (GPP). BDS vegetation status is the basis for support of ecosystem processes through months of no rainfall. Different savanna vegetation types (evergreen and deciduous woodland, thickets, grasslands, etc) behave differently with respect to ecosystem processes making it important to quantify long term changes in the ratio of trees to grass cover of savanna landscapes (Martiny et al. 2006, Sankaran et al. 2005) The variation in behavior is due to growth strategies that differ between the responses of trees and grasses to timing of water availability (Archibald & Scholes, 2007) Total precipitation during the wet season and the timing of that rainfall affect biological processes of BDS GPP (Schwinning et al. 2004) and must be accounted for when determining long term landscape change.
90 Remotely sensed Normalized Difference Vegetation Index NDVI ((NIR Red)/(NIR+Red)) was originally designed as an index of GPP (Tucker & Sellers, 1986) Previous work using remotely sensed estimates of vegetation productivity describe the strong correlation between precipitation and gross primary production (GPP) especially in dryland regions (Chapin et al. 2002, Nemani et al. 2003, Prin ce et al. 1995, Tucker & Sellers, 1986) NDVI is ideally suited for semi arid regions where the index does not saturate at high foliage biomass or leaf area index (Nicholson & Farrar, 1994, Richard & Poccard, 1998) NDVI is also one of the f ew long term vegetation records available across multiple sensors over many years making it useful for longitudinal vegetation studies (Chapin et al. 2002, Cohen & Goward, 2004, Huete et al. 2002, Justice et al. 1985, Los et al. 2000, Tucker, 1979) A relatively strong, albeit slightly non linear relationship between mean annual NDVI and mean annual rainfall exists for much of southern Africa (Goward & Prince, 1995, Martiny et al. 2006) This relationship also corresponds to an increase in fractional tree cover with increasing mean wet season rainfall along the Kalahari Transect, stretching from central Botswana northward into Angola and western Zambia (Scanlon et al. 2002) It is important to note that the coupling between GPP and precipitation varies by vegetation type (Gaughan et al., in prep Scholes and Archibald, 2007). Change in NDVI over multiple decades may indicate shifts along the tree grass continuum that, in turn, influence storages and flows of water, carbon, and nutrients affecting spatial and temporal patterns of ecosystem production (Sankaran et al. 2008) However, to detect long term landscap e change with NDVI, the variation in NDVI due to the effect of seasonal precipitation must be controlled.
91 Various approaches to correct for precipitation effects on NDVI attempt to separate the climate signal from other drivers of land cover change (Archer, 2004, Evans & Geerken, 2004, Groen eveld & Baugh, 2007, Ji & Peters, 2005, Omuto et al. 2010, Wessels et al. 2007) Evans and Geerken (2004) calculated numerous linear regressions between 8 km AVHRR annual NDVI (1981 1996) and different periods of accumulated rainfall to identify the precipitation vegetation productivity relationship for a dryland area in Sy ria. They found the best correlation for March/April maximum NDVI was the accumulation of the previous months of wet season rainfall (September through mid April) while absolute maximum NDVI best correlated to the preceding four months of rainfall (Evans & Geerken, 2004) Evans and Geerken (2004) also show that stronger correlations exist at the pixel scale compared to a dryland average. Wessels et al. (2007) used AVHRR 1 km, 10 the entire growing season (Oct Apr) for years 1985 2003 and cal culated the relationship to the sum of rainfall during the same period. Then using of a residual trend analysis, Wessels et al. (2007) found that a residual trends approach was more robust than a rain use efficiency model in detecting trends for degraded a reas. Archer (2004) used expert opinion to determine a 2 month lag of average rainfall as the best correlated rainfall predictor to predict monthly AVHRR 1 km NDVI data from 1984 1997. After establishing a relationship between measures of accumulated rain fall and annual precipitation signal, as defined by a seasonal measure of rainfall from other factors that influence change in ND VI, the difference of the predicted NDVI from observed NDVI (NDVI obs NDVI pred ) produces a
92 residual value that corrects for the precipitation signal. The resulting positive and negative residual trends over time indicate change in NDVI attributed to somet hing other than inter annual variability in seasonal rainfall (Archer, 2004, Evans & Geerken, 2004, Wessels et al. 2007) The approach, derived in previous studies from a linear regression, assumes that the measure for rainfall is representative of the climate signal, and removes the effect of inter annual variability on vegetation productivity, isolating other fac tors driving landscape changes (Archer, 2004) The residual trends analysis is a useful method to account for precipitation effects on NDVI but a more robust, spatially explicit regression approach may be more appropriate for highly heterogeneous landscapes such as semi arid savannas (Foody, 2003, Omuto et al. 2010) This study introduces a new, spatially explicit approach using a downscaled Geographically Weighted Regression (GWR) model (Fotheringham, 2002) that is parameterized by the statistical relationship between wet season months of precipitation and vegetation at a catchment level. The GWR model extends the traditional regression framework, creating local model parameters estimated for any point within the study area in which each observation point has its own specific set of coefficient values. The implicit inclusion of spatially co varying biophysical factors (ex. soil type, elevation, etc) in the GWR model is important due to differences in the NDVI precipitation relationship across different vegetation and soil types. (Nicholson et al. 1990, Wang et al. 20 01) The GWR model includes estimates of local parameters based on a weighting decay function that places more influence on data closer to the observation point at location i than data farther away (Harris et al. 2010) In addition, the non stationarity incorporated within the GWR f ramework helps minimize spatial autocorrelation in the residuals
93 (Brunsdon et al. 1998) By allowing spatial non stationarity in the relationship between rainfall and vegetation productivity the effects of any unmeasured, important covariates are mi explaining variation in vegetation productivity for a given location. This is important for relationships between the response and predictor variables in environments such as semi arid savannas where rainfall is highly variable across space and time. The objective of this study is to account for and remove inter annual variability in seasonal precipitation that influences the signal of gross primary productivity (GPP) at the begin ning of dry season, as measured by NDVI in order to identify longer term land cover changes. The new method estimates NDVI from the GWR derived relationship between MODIS (or Moderate Resolution Imaging Spectroradiometer) NDVI and precipitation estimated b y the Tropical Rainfall Monitoring Mission (TRMM) (Gaughan et al., in prep ). The GWR model represents the vegetation precipitation relationship for the larger catchment region (~700,000 km 2 ) and extends the traditional regression framework by creating loca l model estimates that vary across space (Fotheringham, 2002) We apply the GWR regression model within the Kwando Core Area (KCA) o f Bwabwata National Park (~1,300 km 2 ) to determine how well the regionally parameterized model controls for seasonal precipitation dynamics on NDVI. The KCA provides an ideal landscape that minimizes the effect of human activities on the landscape as the c ore area management has excluded people since the early 1970s. Our specific questions are: 1) How well does the regional GWR model, parameterized with prior wet season months of rainfall, predict NDVI for the Kwando Core Area? 2) What NDVI changes, not cau sed by inter annual precipitation variation, occurred during
94 the period from 1984 to 2007? 3) Do the GPP trends vary across different vegetation types (indicated by different ranges of NDVI) in the KCA? We expect there to be variability around the mean pr edicted NDVI values for each year. However, if the prior wet season rainfall is the sole determinant of NDVI for that pixel. If an individual s (increase or decrease) across time, that represents a shift in NDVI. The variation in NDVI unaccounted for by seasonal rainfall, as measured by positive or negative changes in the residuals, represents changes due to longer term climate or anthropogenic drivers on the landscape. Materials and Methods Study Region The regional GWR model includes the larger Okavango Kwando Zambezi catchment (~700,000 km 2 ) (Figure 1). The semi arid savanna that comprises the Kwando Core Area of Bwabwata National Park (BNP) in Caprivi, Namibia is located in the lower, central portion of the OKZ catchment. The protected area is characterized by a predominantly deciduous woodland on the relatively homogenous Kalahari Sand soil (Wang et al. 2007) The larger park area, Bwabwata National Park, was originall y known as the Caprivi Game Reserve. The Reserve was established in 1966 and upgraded to a park status in 1968 (Mayes, 2008) The Kwando Core Area (KCA) is one of three areas zoned for conservation and tourism and covers the eastern end of BNP (1,300 km 2 ). A parallel system of ancient drainage lines, called omirambas cut across the KCA in a west north west to east south east direction visible on Landsat imagery (Figure 2) (Thomas et al. 2000)
95 This region has experienced large precipitation fluctuations over the past century. The beginning of the 20 th century started with low rainfall, increased precipitation chara cterized the mid 20 th century, while the last quarter of the 20 th century experienced annual rainfall values lower than normal (Nicholson, 2001) Statements collected from local informants during the 2007 and 2008 f environmental history correspond to the above description of southern African rainfall variability described by Nicholson (2001). The wetter conditions mid 20 th century followed by drier conditions in the last couple decades were visually apparent to local inhabitants with less than normal mean annual rainfall and less flooding of the Kwando River since the late 1970s. The decrease in rainfall observed at the local level follows a global climatic shift identified in the late 1970s (Chavez et al. 2003) at least until 2009 2010 when very high rainfall has generated severe flooding. Datasets Landsat TM d ata We used four Landsat TM derived Normalized Difference Vegetation Index (NDVI) images (April 22, 2007, April 10, 2000, May 9, 1990, and June 10, 1984) for the Kwando Core Area (Figure 2). Image acquisition dates correspond to the beginning of the dry season, with a seven week differential between the earliest and latest acquired images. Images were georectified to a 1991 Land sat TM image from the NASA Global Land Cover project ( http://glcf.umiacs.umd.edu/index.shtml ), with a nearest neighbor resampling algorithm using the Autosync function in Erdas Imagine 9.2 (RMSE of < 0 .5 pixels, or < 15 m). Images were also calibrated to convert raw digital numbers to at sensor radiance and surface reflectance estimates using the ENVI 4.3 software calibration tools to control for bias due to difference in acquisition time, sun angle an d
96 sensor geometry. The atmospheric correction algorithm Fast Line of Sight Atmospheric Analysis of Spectral Hypercubes (FLAASH) was used to generate corrected estimates of surface reflectance (Berk et al. 1999, Cooley et al. 2002) Using a RGB 7,5,4 composite, we masked fire scars visibly apparent on the TM images from analysis. Aquatic vegetation and water pixels were masked to minimize the influence of riparian vegetation and water inundation from the Kwando River. From the final processed TM images, NDVI images were used to conduct the residual trends analysis with modeled estimates of NDVI. While we are not directly comparing the MODIS and TM NDVI values, the derived coefficients from the GWR model were generated using the MODIS NDVI 13A1 product (500 meter, 16 day composite), acquired from GLOVIS ( http://glovis.usgs.gov/ ), as the response variable. The MODIS NDVI product applies a bidirectional reflectance distribution function (BRDF) model to account for anisotropy in surface reflectance values by controlling for view and zenith angles (Leeuwen et al. 1999, Walthall et al. 1985) The final NDVI 16 day composite minimizes cloud cover and uses quality control flags to e (Leeuwen et al. 1999) The April 23, 2007 MODIS 13a1 NDVI composite was used to check spectral compatibility between MODIS NDVI and TM derived NDVI for April 2007. Landsat TM de rived NDVI pixels were resampled to match the 500 x 500 meter spatial resolution of the MODIS 13a1 composite NDVI values using a cubic convolution algorithm. Ten percent of the data was randomly sampled and extracted from the two layer stack of MODIS and T M NDVI bands. The Wilcoxon Signed Ranks non parametric test was used as a statistical measure to compare the NDVI distributions of Landsat
97 TM derived NDVI and the MODIS 13a1 NDVI composite. Equivalent to a paired samples t test, the Wilcoxon signed rank s tatistic tests the hypothesis of no difference between measurements taken on the same subject (in this case, pixel value). The test showed non significance (p < 0.632, df = 609) between NDVI estimates on a pixel by pixel basis for the April 23, 2007 MODIS 13a1 composite image and the April 22, 2007 Landsat TM derived NDVI image indicating no statistical significance was detectable between the two images. Pre processing methods differ between the two derived NDVI images which could potentially influence a di fference in final NDVI products. However, the use of a paired samples test has been shown to be a useful diagnostic tool to evaluate the comparability of MODIS and Landsat imagery (Melendez Pastor et al. 2010) Further analysis in this study uses the GWR de rived coefficient values generated with MODIS 13a1 NDVI to estimate NDVI and compare to observed TM derived NDVI for the KCA. Rainfall d ata The regional GWR model is parameterized using the TRMM 3B43 dataset (version 6) which includes best estimated rainf all calculations from a combined instrument rain calibration algorithm (3B42) and rain gauge data sources (Huffman et al. 1997) However, to calculate the GWR model for each TM image acquisition year (1984, 1990, 2000, and 2007) we used the gridded monthly precipitation time series (0.5 o x 0.5 o grid c ell resolution) produced by Willmott and Matsuura (hereafter known as the WM dataset) (Matsuura & Willmott, 2007) Ideally we would use the same precipitation data source at both the regional and local scale but data collection from the TRMM instrument dates bac k only to 1998. Meteorological station data in the region are sparse and lack of reliable existing records makes the modeled precipitation dataset necessary
98 for longitudinal analysis. The WM model output accuracy for global precipitation patterns compares favorably to five other monthly precipitation datasets (F ekete et al. 2004) and inclusion of all Landsat time steps made it an appropriate dataset for this study. Precipitation corrected NDVI The GWR derived relationship between NDVI and seasonal precipitation developed at the regional catchment scale (~600 ,000 km 2 ) (Gaughan et al., in prep ) identifies the strength of association between different wet season months and BDS GPP. The GWR creates location specific observation estimates that are weighted by a distance decay function and these estimates will vary depending on which point within the study region the model is being estimated (Fotheringham, 2002) The regional model uses TRMM ra infall estimates across ten years (2000 2009) for each of the seven wet season months (October April) to predict April NDVI. The years 2000 2009 include wet, normal, and dry years. By stratifying our sampling of MODIS NDVI response and TRMM predictor acr oss years we account for temporal variation in rainfall providing a representative relationship between savanna vegetation and prior wet season rainfall. The derived beta coefficients from the GWR model are used with WM monthly precipitation data to estima te NDVI for each image year (1984, 1990, 2000, and 2007). The coarseness of the WM dataset means only one grid cell of precipitation was included within the boundaries of the KCA. We compared the one point estimate of monthly precipitation to an inverse di stance weighted (IDW) interpolation approach for estimating NDVI. Minimal difference was evident in the mean and standard deviation when comparing the residual of sum squares (RSS) value between the two precipitation
99 estimation approaches. Thus we determin ed that the point estimates of monthly rainfall provided adequate predictors for each year of estimated NDVI. Before estimating NDVI, the GWR model was re calculated with a spatial resolution of 30 meters to be comparable with the observed TM NDVI images. The estimated model that predicts NDVI is: NDVI i = 0 (u i v i ) + 1 (u i v i ) Rain Oct 7 (u i v i ) Rain Apr + e i Where NDVI i represents predicted BDS NDVI for each Landsat pixel i and its spatial location is represented by coordinates (u i v i ) in the KCA. 0 (u i v i ), 1 (u i v i ) Rain Oct 7 (u i v i ) Rain Apr are realizations of the continuous, estimated coefficient functions of 0 7 (u, v) at location i and e i is a Gaussian random error term. The intercept and regression coefficients specific for the model at each point were generated from the regional GWR model coefficients multiplied by each WM monthly predictor value The total number of months included in t he model may vary depending on the TM image acquisition date. For the June 10 th 1984 image we re estimated the regional GWR model with May and June rainfall as predictors extending the model to include 9 total predictor variables. The model, parameterized with the additional two months of rainfall, provides a better fit to observed June NDVI. However, we did not include May rainfall for the May 9 th 1990 image. While the time difference between the June 1984 image and the other three image dates is not ide al, we assume that by adjusting the GWR model we capture the inter annual seasonal effects on the landscape and thus control for the seasonal differences one might observe in this environment. The coefficient
100 values extracted from the GWR model for the KCA correspond broadly to savanna land cover based on the International Global Biosphere Programme (IGBP) global land cover classification map, which defines savanna as lands with herbaceous and other mixed understory cover and a forest canopy cover of 10 30% with height greater than 2 meters (Friedl et al. 2002) We use this designation of savanna to determine how the KCA vegetation cover may vary over time after controlling for seasonal rainfall. After NDVI was estimated from the down scaled GWR model, we subtract estimated NDVI (NDVI est ) from the observed TM derived NDVI (NDVI obs ) resulting in a new, precipitation corrected (PC) NDVI value, or residual (Evans & Geerken, 2004) The PC NDVI corrects for seasonal effects on the landscape by removing the variation in NDVI explained by the prior wet season months of rainfall in the mode l for a given year. As a result, any trend (positive or negative) through time present in the PC NDVI values indicates change in NDVI not due to seasonal precipitation. If each subsequent year sees an increase in positive PC NDVI then it suggests a trend t owards increased vegetation productivity at the start of the dry season. Likewise, consecutive time steps of increasing negative PC NDVI suggest a landscape with less vegetation productivity over time at the start of the dry season. Long term L and C over C h ange in the KCA Temporal clusters of the raw PC NDVI were created using a K means unsupervised classification method on the stacked multi temporal PC NDVI layers. These clusters were then used to identify spatial variation in PC NDVI change over time by id entifying areas on the landscape with similar PC NDVI trajectories. By clustering similar PC NDVI values together across the four image dates we examined how spatial patterns of vegetation productivity varied since 1984. Calculations also were completed
101 to determine which areas in the KCA had a continual increase or decrease in PC NDVI over the four time steps (1984 1990 2000 2007). These areas in the KCA highlight a one directional (either positive or negative respectively) trend in vegetation productivity over the 23 year period. However, a continual increasing or decreasing trend in PC NDVI does not identify pixels that oscillate over time. We also calculate image differences between each PC NDVI year to identify trajectories of change within the KCA for 1984 1990, 1990 2000, and 2000 2007. Extreme difference in PC NDVI for two dates is identified by applying a 2 standard deviation threshold. Other cutoff values were investigated but the 2 standard deviations, while arbitrary, provides a conservative es identifies important biological shifts on the landscape. We then conduct a post classification analysis on the differenced PC NDVI images to determine the different pixel trajectories over the three differenced time steps. We examin e only those trajectories that are > 1% of the landscape We use ground data along with semi structured interviews collected within the study area in 2007 and 2008 to help explain the observed changes. Before applying the threshold value the main paved road was masked out of the analysis. We assume that the tar surface would not respond to seasonal precipitation and we wanted to exclude those pixel values from influencing the mean and standard deviation used in the threshold calculation. Results PC NDVI C ompared to Observed NDVI Figure 3a shows the observed TM derived NDVI and Figure 3b shows the precipitation corrected NDVI (PC NDVI) plotted as bean plots (Kampstra, 2008) For each year, ~6,000 sample points are plotted that are equivalent to 30 x 30 meter pixels
102 in the KCA. Every sample point has a unique GWR model shown by individual black lines in Figure 3a and 3b. The spread of the distribution around those points is indicated by the gray, vertical histograms. The PC NDVI (residual value) for each year centers on zero suggesting that the pattern of increase increase decrease shown for observed NDVI (Figure 3a) is minimized once seasonal effects are accounted for by the previous months of wet season rainfall (i.e. the bean plot medians of each y ear are near zero and the distribution of PC NDVI indicated by the bean plot histograms straddle zero evenly) Variation exists around zero as the parameter estimates used in the model describe the relationship of NDVI to rainfall at the larger catchment s cale. The distributions around zero, shown by the gray histograms, represent other factors (dominant vegetation type, soil cover, etc) that may influence change in NDVI. This figure indicates that the GWR model controls for the prior wet season rainfall ac ross image dates. Temporal Clusters of PC NDVI The comparison with 3a and 3b shows the importance of seasonality on inter annual variation in NDVI, but the spread of individual points (black lines) suggests variation remains unaccounted for in the landscap e. A ten class unsupervised classification (Figure 4), calculated on the four time steps stacked together as a single dataset, identifies similar PC NDVI trajectories over time and groups those areas on the landscape together. Classes that do not vary for the four years (eg. Classes 4) indicate minimal change in PC NDVI. Ecologically, the classes (groups of similar pixel trajectories) that increase and/or decrease in PC NDVI over time represent shifts in total GPP, as measured by NDVI. These classes that ex hibit change in PC NDVI (ex. Class 1 or Class 6) suggests factors other than seasonal precipitation influence shifts in vegetation cover. The ridge and dune system, natural to the region, is noticeable in the
103 clustering of PC NDVI. This pattern identifies with the natural variation of more woody vegetation on top of the ridges (ex. Class 5) and grassier areas more prevalent in the troughs (ex. Class 3). Additionally, there seems to be more change in PC NDVI, both positive and negative, that occurs through time in areas closer to the river (ex. Class 1, 6 or 10). PC NDVI Change Figure 5 highlights areas on the landscape that showed incremental increase or decrease in PC NDVI from 1984 to 2007. Areas with increasing PC NDVI (1984 < 1990 < 2000 < 2007) make u p 2.4% of the KCA. Continual increase in PC NDVI is more prevalent in the northwest region and along parts of the eastern side of the protected area. In contrast, continually decreasing PC NDVI (1984 > 1990 > 2000 > 2007) is concentrated in the southern ha lf of the protected area with smaller areas scattered through the KCA and makes up 3.3% of the total area in the KCA. The areas of directional change (continually increasing or decreasing) make up a relatively small percentage of the landscape (~ 5.7%). To identify patterns of PC NDVI that do not exhibit a unidirectional change in PC NDVI, a change trajectory of differenced PC NDVI across the three differenced time steps (1990 1984, 2000 1990, and 2007 2000) provides a more detailed view of a single p A 2 standard deviation threshold applied to the differenced images highlights areas on the landscape of extreme change in PC NDVI not accountable by prior wet season rainfall. The majority of the landscape remains stable (8 8.1%). The largest area of change occurs in the NE portion of the protected area. A decrease in PC NDVI occurs from 2000 2007 and covers 2.1% of the landscape. The second largest percent of change
104 occurs from 1990 2000 with an increase of PC NDVI coverin g 1.9% of the KCA also concentrated in the NE section of the protected area. Together the two trajectories make up the largest change observed in the NE. The northeastern increase from 1990 2000 corresponds in general to Class 1 and 6 in Figure 4 although the Class 6 cluster also corresponds to pixels that decreased from 2000 2007. For other parts of the KCA, the recent decrease (2000 2007) is prevalent in the lower SE part of the park and along the omirambas and corresponds generally to Classes 8 and 10, along with Class 6 from Figure 4. In contrast, over the same time period (2000 2007), an increase (1.4%) of PC NDVI occurs largely in the NW portion of the KCA corresponding to the increase in pixel values for Classes 2 and 4 in Figure 4. The other shif ts of extreme PC NDVI that is >1% of the landscape occurs from 1984 1990 with a 1.6% decrease mostly in the SW area and a 1.5% increase scattered across the KCA. Pictures a and b in Figure 6 show photographs of areas corresponding to pixel values that ex hibit a + 2 standard unit change in PC NDVI from 2000 to 20007 (Figure 6a) and a 2 standard deviation of change in PC NDVI from 1984 to 1990 (Figure 6b). Discussion Results show minimal change over time in PC NDVI (residuals) since 1984 in the KCA, which is expected as the KCA is a protected area with no human inhabitants. This indicates that the regional GWR model, parameterized for a savanna land cover, provides a robust approach to control for prior wet season precipitation on BDS GPP, as measured by ND VI. The strong seasonal influence emphasizes the importance of using precipitation corrected NDVI when examining biomass trends for regions with high inter annual precipitation variation (Evans & Geerken, 2004) No widespread shift towards
105 an increase or decrease in GPP since 1984 is detect ed due to the minimal human presence over the past couple decades but also because there is not demonstrable climatic trend detectable during the time period (Gaughan and Waylen, in prep ). Figure 6 shows no extreme change in NDVI for the majority of the pr otected area after controlling for prior wet season rainfall. However, a continual increase in PC NDVI is prevalent in the upper NW of the KCA and there is an apparent decrease in PC NDVI in the lower SW portion of the study area (Figure 5). These areas r epresent trends in different parts of the KCA in which pixel values move in a continual direction away from the overall mean NDVI for each time step. The decreasing trend of PC NDVI concentrated in the lower SW and a continual increase in PC NDVI in the NW indicate decreased and increased vegetative cover on the landscape since 1984, respectively. The increase concentrated in the NW corresponds to personal communication with park management of increased woody vegetation in recent decades. PC NDVI changes in other areas of the KCA are more spatially and temporally variable. The temporal clustering of PC NDVI (Figure 4) and the change trajectory of PC NDVI image differences (Figure 6) identifies these distinct patterns of increased and decreased regions of PC NDVI for different time steps. These changes identify the pattern of GPP shifts not due to seasonal effects but to some other factor(s) operating on the landscape. These other factors most likely correspond to a variety of different policy and land use decisions implemented over the past few decades. Throughout the 1980s, conflict continued in the Caprivi Strip due to the border war with Angola (Stanley, 2002) A decline in wildlife populations is attributable to the presence of the South African Defense Force (SADF) in the Caprivi Strip (Bruchmann, 2007). Additionally,
106 local communities along the Kwando River had access to grazing areas and natural resources on both sides of the river. Since Namibian independence in 1990 (Rodwell et al. 1995) much of the area in Caprivi previously occup ied by SADF has not been as heavily used (Rodwell et al. 1995) and with the implementation of Community Based Natural Resource Management initiatives in Caprivi (mid 1990s), local communities have had restricted access to the west side of the Kwando River (S tuart Hill et al. 2005) The northeastern portion of the KCA was one area that has undergone multiple changes since Namibian independence. The high variability observed in vegetation shifts within the northeastern portion of the KCA (Figure 6) probably relates to the exit of the SADF in 1989 and policy implementation prohibiting grazing rights along the river inside the park (Mayes, 2008) of land extends southward along the Kwando River, nar rowing towards the Botswana border (Mayes, 2008) While this area was not originally part of the protected area, people and their cattle have been excluded from this area since the mid 1990s due to the heightened focus on Caprivi as a wildlife conservati on area ( Rodwell et al. 1995) While this study does not explicitly identify the other potential drivers (ex. fire, herbivory, frost) that affect BDS GPP, anecdotal information from game guards and park wardens suggest that the timing of fire may play a signif icant role in dictating vegetation patterns on the landscape. Fuel moisture depends on how much relative humidity exists throughout various parts of the dry season while the fuel load depends on multiple factors such as tree cover, rainfall, soil type, and grazing pressure (Archibald et al. 2009, Higgins et al. 2000) The interaction of these different factors can have both a positive and negative effect on fuel load. While rainfall and soil
107 nutrients contribute to increased fuel load on the one hand, they also provide tree canopy cover and optimal grazing conditions which has a negative effect on fuel load (Archibald et al. 2009) The frequency and extent of fire will also be influenced by soil m oisture availability (Rodriguez Iturbe et al. 1999) Figure 6 identifies a more recent decrease in extreme PC NDVI along the edge of the ridges of the omirambas which may be a result of soil moisture dynamics negatively influencing the strength of BDS vegetation productivity from 2000 to 2007. Underlying the effects of these physical system drivers are management and land use decisions that have changed over recent decades. Until recently, a fire suppression policy was strongly supported for the entire Caprivi Region. Starting in 2006 an early burning regime was implemented on a yearly basis with the management intent to prevent the hot ter, more intense late season fires (pers. comm. Robin Beatty, IRDNC). Future research will focus on areas within the KCA that respond to other potential drivers of long term change such as the influence of fire, herbivory and changes in soil moisture. Our study builds on previous studies ( Archer, 2004, Evans & Geerken, 2004, Wessels et al., 2007 ) that apply a correction factor for the influence of climate on NDVI by applying a model with a more spatially detailed examination of the NDVI precipitation rel ationship within the KCA. The use of the GWR model allows detection of more subtle shifts in PC NDVI that may be obscured with a global regression model (ex. ordinary least squares) (Bini et al. 2009) In addition, the GWR model is a more robust regre ssion technique to separate the inter annual seasonal signal from other factors controlling BDS GPP over time. The use of GWR minimizes the effects of unmeasured, spatially varying factors and the negative impacts of autocorrelation (Fotheringham,
108 2002) In addition, unlike global regression models, a GWR model weights observation points that are closer to the estimation point more he avily than those far away. This provides a spatially explicit estimate of NDVI specific to a certain location on the landscape which minimizes spatial autocorrelation and provides a more representative relationship between wet season rainfall and vegetatio n across space. The use of the down scaled GWR model to calculate the PC NDVI for the KCA was specifically developed to control for the effects of seasonal precipitation. The GWR model incorporates the prior months of wet season rainfall and can be adjuste d to include shorter or longer term rainfall periods as needed. The down scaled GWR model correction for seasonal precipitation will decrease NDVI for years with higher than normal rainfall and increase NDVI for years of less than normal rainfall. The dete ction of subtle changes on the landscape is important for such heterogeneous areas such as savanna drylands where a shift within the mix tree shrub grass composition has implication for local livelihood land use decisions (Shackleton et al. 2007) and global environmental policies (Rotenberg & Yakir, 2010) In this study, areas in the protected area with the most change were those that had the strongest human presence ove r the past few decades. Prior to Namibian independence and the implementation of more rigorously applied conservation policies, the northeast corner of the KCA was accessible to communities and was also a base location for the SADF. The more dynamic vegeta tion change identified for this area emphasizes the importance land use and policy decisions have on landscape changes. The use of GWR to characterize the regional relationship of precipitation to savanna vegetation and apply it at a local scale provides a n important first step to
109 disentangling the complex interactions of various drivers of land cover change. Controlling for seasonal precipitation effects on vegetation productivity is a critical component necessary in order to detect other factors driving l ong term land cover change. These other competing drivers may be human induced or related to longer term climate patterns (Dube & Pickup, 2001, Wessels et al. 2007) PC NDVI accounts for wet season months of rainfall on BDS GPP but does not account for potential longer term lags in the system. Response of GPP to antecedent rainfall effects of previous years is not captured in the seasonal model and thus the response of certain vegetation types that are influenced by longer term rainfall will not be captured in the model (Richard et al. 2008, Wiegand et al. 2004) However, the purpose of the method is to account for the effects of inter annual variabil ity in seasonal rainfall to then detect these longer term trends in vegetation cover. The down scaled GWR model corrects for inter annual variability in seasonal rainfall and is a useful first step in determining long term change in a savanna landscape. Summary This study suggest a new approach to scale down a regional relationship between precipitation and vegetation to identify change in long term vegetation patterns after controlling for inter annual precipitation variation in seasonal rainfall. The us e of GWR as a statistical technique provides a spatially explicit estimation of the precipitation vegetation relationship across the regional catchment. Local model estimations provide a precipitation vegetation relationship specific for certain areas in t he study region. The structure of the model weights the importance of each wet season month accordingly rather than using a total value to represent the entire seven month period. This is important for determining the strength of association for different months of wet season
110 rainfall on BDS GPP for savanna vegetation (Gaughan et al., in prep ). By controlling for prior wet season monthly rainfall, we are then able to identify longer term changes in vegetation cover. The residual trends analysis provides a valuable technique to identify and remove precipitation effects on NDVI. Used in conjunction with other means of monitoring land cover changes, this approach helps identify changes in vegetation productivity due to factors other than prior wet season rainf all. This study shows that no large scale changes were detected in the protected area although spatial and temporal variation exists in certain parts of the KCA. These areas experienced change in vegetation productivity most likely due to factors operating on the landscape such as herbivory and fire. Their respective influences were dictated by changes in policy and land use decisions over the past few decades. More intensive investigation is needed to for areas that experienced extreme change in PC NDVI in order to determine other drivers that contribute to observed vegetation change. The GWR method to account for prior wet season rainfall effects on NDVI can also be applied to a larger management area which includes communal lands. Detecting how land use d ecisions affect vegetation productivity without an underlying climate signal biasing the detected changes is important for semi arid savanna regions in which management of wildlife, livestock and fire is critical to conservation and development initiatives
111 Figure 4 1. Study region in southern Africa outlining the larger Okavango Kwando Zambezi catchment and the local protected area of interest Kwando Core Area in Bwabwata NP, Caprivi, Namibia.
112 Figure 4 2. Landsat TM RGB: 5,4,3 composite images and observed TM NDVI for each natural vegetated linear system of dune ridges constructed from sediments mostly likely transported by fluvial processes during the Quaternary perio d (Thomas et al., 2000). Fire scars and aquatic vegetation are masked from images.
113 Figure 4 3 Shows a) Landsat TM overall NDVI values for the Kwando Core Area (KCA) plotted across four years showing a declining trend for 1984 1990 2000 and then an increase of overall NDVI in 2007 and b) precipitation corrected NDVI values for the KCA in the form of residuals (NDVI obs NDVI pred ). a. b.
114 Figure 4 4 Shows a K means 10 class unsupervised classification of the Kwando Core Area based on four discrete time steps of PC NDVI (1984 1990 2000 2007). Each class represents similar PC NDVI values over time.
115 Figure 4 5 Trends PC NDVI over the four discrete time steps (1984 1990 2000 2007), defined as subsequent increase or decrease in PC_NDVI, is highlighted for a) positive change and b) negative change in the KCA.
116 Figure 4 6 Change trajectory for the three differenced PC NDVI images (1990 1984, 2000 1990, and 2007 2000) with a threshold of 1.96 standard deviations. D = Decrease 1.96 SD, I = Increase +1.96 SD, and N = No Change. Percentages > 1% of the KCA region shown. Picture a shows an area of +2 SD Increase (1984 1990) and picture b shows an area of 2 SD Decrease (2000 2007).
117 CHAPTER 5 CO NCLUSION Overall Findings Climate change and climate variability will continue to influence the rate and timing of biological and ecological processes in savanna ecosystems (IPCC 2007) This research specifically focused on the precipitation aspect of climate to answer the quest ion of how precipitation variation, over different spatial and temporal scales, influences vegetation response for a dryland catchment in southern Africa. The study incorporated multiple scales across time and space to identify the climate land interactio n specific to precipitation patterns and savanna vegetation at both a regional catchment level (~693,000 km 2 ) and within a local protected area (1,300 km 2 ). The first chapter of this dissertation outlined the rationale for detecting and monitoring savanna landscape dynamics and identifies the importance of directional trends of timing, frequency, and distribution of precipitation on different savanna vegetation covers. To answer the overarching research problem set forth in Chapter 1, subsequent chapters ar e made up of individual papers with separate research questions, hypotheses, tests, and methods (Chapters 2, 3, and 4). The first paper showed that mean annual precipitation (MAP) patterns have changed before and after a late 1970s global climate shift wit hin the Okavango Kwando Zambezi catchment. More recent years (1980 2005) have seen a decrease in MAP across all three catchments. In addition, the number of dry years and frequency of dry years concurrent with warm phase ENSO events has increased. These pr ecipitation patterns suggest short term changes in the functioning and response of the OKZ catchment to hydro meteorological patterns from 1950 2005. The findings complement
118 existing studies which identify changes in southern African rainfall and teleconne ctions patterns post the late 1970s (Mason 2001; Fauchereau et al. 2003) And while the state of local and regional inputs to each sub catchment remains relatively strong across the study period (1950 2005), the historical shift from above median conditions in Period 1 (1950 1975) to below median conditions in Period 2 (1980 2005) is important for land use and conservation manage ment decisions. The second paper identified the strength of association between wet season (Oct Apr) precipitation on beginning of dry season (April) vegetation productivity as estimated by NDVI. A large proportion of April NDVI is explained by seasonal ra infall. However, the relationship of beginning of dry season NDVI response to timing of rainfall during the wet season differs across savanna land covers. A stronger association exists for end of the wet season monthly rainfall (February April) and savanna land covers such as grasslands and open woodlands compared to rainfall at the outset of the rainy season (October November). In addition, some woodland systems, such as Miombo woodland or evergreen forest, do not show a significant association between b eginning of dry season NDVI and the prior seasonal rainfall. The approach used in paper 2 created an empirical model of the precipitation vegetation relationship parameterized for savanna vegetation. The third paper applied this model at a local scale t o identify the effects of seasonality on remotely sensed imagery and, in turn, isolated areas on the landscape that have changed due to other potential forcing factors (ex. fire, herbivory, people). The model corrects for seasonality on the remotely sense d data and creates a new, precipitation corrected estimate of vegetation productivity. The model showed overall seasonality accounts for a large
119 percent of variation in vegetation productivity, as estimated by NDVI. There was no widespread change in vegeta tion cover detected from 1984 2007. However, after controlling for seasonal precipitation effects, finer spatial and temporal patterns suggest distinct pattern s of increased and decreased precipitation corrected NDVI exist within the local protected area These areas will be further investigated in future research as other potential drivers of savanna vegetation change, such as fire or herbivory, may contribute to the changes observed. Significance of Findings The conclusions drawn from this study emphasize the variable nature of climate especially common in southern African dryland regions. Land use decisions and adaptive management of natural resources and development initiatives must incorporate the knowledge of how patterns of precipitation change and va riability will affect savanna vegetation dynamics. The Okavango Kwando Zambezi catchment encompasses the future Kavango Zambezi Transfrontier Conservation area (KAZA) that is projected to become the largest transboundary conservation area in Africa. The ti ming, frequency, and amount of precipitation input across these three basins will play an important and critical role to the distribution of wildlife across wet and dry seasons, agricultural land use decisions, effects and influence of fire, and transbound ary decisions regarding flows of water to balance between ecosystem processes and sustainable livelihood needs. This dissertation identifies that precipitation has shifted across the OKZ basin over the latter half of the twentieth century. The shift may be short term but the nature of the oscillating pattern may not be and an understanding of such patterns is necessary to making conservation and development decisions in such a variable environment.
120 LIST OF REFERENCES Accatino, F., C. De M ichele, R. Vezzoli, D. Donzelli, and R. J. Scholes. 2010. Tree grass co existence in savanna: Interactions of rain and fire. Journal of Theoretical Biology 267 (2):235 242. Adeyewa, Z. D., and K. Nakamura. 2003. Validation of TRMM radar rainfall data over major climatic regions in Africa. Journal of Applied Meteorology 42 (2):331 347. Agrawal, A. 2000. Adaptive management in transboundary protected areas: The Bialowieza Nati onal Park and Biosphere Reserve as a case study. Environmental Conservation 27 (4):326 333. Agresti, A. 2007. An Introduction to Categorical Data Analysis 2nd ed: John Wiley. Alemaw, B. F., and T. R. Chaoka. 2006. The 1950 1998 warm ENSO events and regi onal implications to river flow variability in Southern Africa. Water Sa 32 (4):459 463. Allen, C. D., and D. D. Breshears. 1998. Drought induced shift of a forest woodland ecotone: Rapid landscape response to climate variation. Proceedings of the Nationa l Academy of Sciences of the United States of America 95 (25):14839 14842. Archer, E. R. M. 2004. Beyond the "climate versus grazing" impasse: using remote sensing to investigate the effects of grazing system choice on vegetation cover in the eastern Karo o. Journal of Arid Environments 57 (3):381 408. Archibald, S., D. P. Roy, B. W. van Wilgen, and R. J. Scholes. 2009. What limits fire? An examination of drivers of burnt area in Southern Africa. Global Change Biology 15 (3):613 630. Archibald, S., and R. J. Scholes. 2007. Leaf green up in a semi arid African savanna separating tree and grass responses to environmental cues. Journal of Vegetation Science 18 (4):583 594. Barnes, J. I., J. MacGregor, and L. C. Weaver. 2002. Economic efficiency and incentives for change within Namibia's community wildlife use initiatives. World Development 30 (4):667 681. Batjes, N. H. ed. 2000. Global Soil Profile Data (ISRIC WISE). [Globa l Soil Profile Data (International Soil Reference and Information Centre World Inventory of Soil Emission Potentials)]. Oak Ridge, Tennessee, U.S.A. : Oak Ridge National Laboratory.
121 Berk, A., G. P. Anderson, L. S. Bernstein, P. K. Acharya, H. Dothe, M. W. Matthew, S. M. Adler Golden, J. H. Chetwynd, S. C. Richtsmeier, B. Pukall, C. Allred, L. S. Jeong, and M. L. Hoke. 1999. MODTRAN4 radiative transfer modeling for atmospheric correction. Paper read at Proceedings of SPIE Optical Spectroscopic Techniques and Instrumentation for Atmospheric and Space Research at Denver, CO, USA. Binet, D., B. Gobert, and L. Maloueki. 2001. El Nino like warm events in the Eastern Atlantic (6 degrees N, 20 degrees S) and fish availability from Congo to Angola (1964 1999). A quatic Living Resources 14 (2):99 113. Bini, L. M., J. A. F. Diniz, T. F. L. V. B. Rangel, T. S. B. Akre, R. G. Albaladejo, F. S. Albuquerque, A. Aparicio, M. B. Araujo, A. Baselga, J. Beck, M. I. Bellocq, K. Bohning Gaese, P. A. V. Borges, I. Castro Parg a, V. K. Chey, S. L. Chown, P. de Marco, D. S. Dobkin, D. Ferrer Castan, R. Field, J. Filloy, E. Fleishman, J. F. Gomez, J. Hortal, J. B. Iverson, J. T. Kerr, W. D. Kissling, I. J. Kitching, J. L. Leon Cortes, J. M. Lobo, D. Montoya, I. Morales Castilla, J C. Moreno, T. Oberdorff, M. A. Olalla Tarraga, J. G. Pausas, H. Qian, C. Rahbek, M. A. Rodriguez, M. Rueda, A. Ruggiero, P. Sackmann, N. J. Sanders, L. C. Terribile, O. R. Vetaas, and B. A. Hawkins. 2009. Coefficient shifts in geographical ecology: an em pirical evaluation of spatial and non spatial regression. E cography 32 (2):193 204. Bond, W. J. 2008. What limits trees in C4 Grasslands and Savannas? Annual Review of Ecology and Systematics 39:641 59. Borchert, R. 1994. Soil and Stem Water Storage Dete rmine Phenology and Distribution of Tropical Dry Forest Trees. Ecology 75 (5):1437 1449. Breshears, D. D., and F. J. Barnes. 1999. Interrelationships between plant functional types and soil moisture heterogeneity for semiarid landscapes within the grassla nd/forest continuum: a unified conceptual model. Landscape Ecology 14 (5):465 478. Brown, M. E., J. E. Pinzon, K. Didan, J. T. Morisette, and C. J. Tucker. 2006. Evaluation of the consistency of long term NDVI time series derived from AVHRR, SPOT Vegetati on, SeaWiFS, MODIS, and Landsat ETM+ sensors. Ieee Transactions on Geoscience and Remote Sensing 44 (7):1787 1793. Brunsdon, C., S. Fotheringham, and M. Charlton. 1998. Geographically weighted regression modelling spatial non stationarity. Journal of t he Royal Statistical Society Series D the Statistician 47:431 443. Burt, J. E., and G. M. Barber. 1996. Elementary Statistics for Geographers 2nd ed. New York: The Guilford Press.
122 Camberlin, P., N. Martiny, N. Philippon, and Y. Richard. 2007. Determinan ts of the interannual relationships between remote sensed photosynthetic activity and rainfall in tropical Africa. Remote Sensing of Environment 106 (2):199 216. Cerling, T. E., J. M. Harris, B. J. MacFadden, M. G. Leakey, J. Quade, V. Eisenmann, and J. R Ehleringer. 1997. Global vegetation change through the Miocene/Pliocene boundary. Nature 389 (6647):153 158. Chang, P., T. Yamagata, P. Schopf, S. K. Behera, J. Carton, W. S. Kessler, G. Meyers, T. Qu, F. Schott, S. Shetye, and S. P. Xie. 2006. Climate fluctuations of tropical coupled systems The role of ocean dynamics. Journal of Climate 19 (20):5122 5174. Chapin, F. S., P. A. Matson, and H. A. Mooney. 2002. Principles of Terrestial Ecosystem Ecology New York: Springer Verlag. Chase, M., and C. Griffin. 2008. Seasonal Abundance and Distribution of Elephants in Sioma Ngwezi National Park, southwest Zambia. Pachyderm (45):88 97. Chase, M. J., and C. R. Griffin. 2009. Elephants caught in the middle: impacts of war, fences and people on elephant distribution and abundance in the Caprivi Strip, Namibia. African Journal of Ecology 47 (2):223 233. Chase, T. N., R. A. Pielke, T. G. F. Kittel, R. Nemani, and S. W. Running. 1996. Sensitivity of a general circulation model to global changes in leaf area index. Journal of Geophysical Research Atmospheres 101 (D3):7393 7408. Chavez, F. P., J. Ryan, S. E. Lluch Cota, a nd M. Niquen. 2003. From anchovies to sardines and back: Multidecadal change in the Pacific Ocean. SCIENCE 299 (5604):217 221. Chavez, P. S. 1996. Image based atmospheric corrections revisited and improved. Photogrammetric Engineering And Remote Sensing 6 2 (9):1025 1036. Choler, P., W. Sea, P. Briggs, M. Raupach, and R. Leuning. 2010. A simple ecohydrological model captures essentials of seasonal leaf dynamics in semi arid tropical grasslands. Biogeosciences 7 (3):907 920. Cohen, W. B., and S. N. Goward. 2004. Landsat's role in ecological applications of remote sensing. Bioscience 54 (6):535 545.
123 Cooley, T., G. P. Anderson, G. W. Felde, M. L. Hoke, A. J. Ratkowski, J. H. Chetwynd, J. A. Gardner, S. M. Adler Golden, M. W. Matthew, A. Berk, L. S. Berns tein, P. K. Acharya, D. Miller, and P. Lewis. 2002. FLAASH, a MODTRAN4 based atmospheric correction algorithm, its application and validation In Geoscience and Remote Sensing Symposium, 2002. IGARSS '02 Air Force Res. Lab, Arizona University, Tucson, AZ, U SA: IEEE International. Crai g, C. 1997. The ELESMAP project : Namibia Nature Foundation, Namibia. Curtis, S. 2008. The El Nino Southern Oscillation and Global Precipitation. Geography Compass 2/3:600 619. DeFries, R., A. Hansen, A. C. Newton, and M. C. H ansen. 2005. Increasing isolation of protected areas in tropical forests over the past twenty years. Ecological Applications 15 (1):19 26. Do, F. C., V. A. Goudiaby, O. Gimenez, A. L. Diagne, M. Diouf, A. Rocheteau, and L. E. Akpo. 2005. Environmental influence on canopy phenology in the dry tropics. Forest Ecology and Management 215 (1 3):319 328. Dube, O. P., and G. Pickup. 2001. Effects o f rainfall variability and communal and semi commercial grazing on land cover in southern African rangelands. Climate Research 17 (2):195 208. Evans, J., and R. Geerken. 2004. Discrimination between climate and human induced dryland degradation. Journal o f Arid Environments 57 (4):535 554. Fauchereau, N., S. Trzaska, M. Rouault, and Y. Richard. 2003. Rainfall variability and changes in Southern Africa during the 20th century in the global warming context. Natural Hazards 29 (2):139 154. Fekete, B. M., C. J. Vorosmarty, J. O. Roads, and C. J. Willmott. 2004. Uncertainties in precipitation and their impacts on runoff estimates. Journal of Climate 17 (2):294 304. Field, C. B., M. J. Behrenfeld, J. T. Randerson, and P. Falkowski. 1998. Primary production of the biosphere: Integrating terrestrial and oceanic components. S cience 281 (5374):237 240. Fisher, J. I., and J. F. Mustard. 2007. Cross scalar satellite phenology from ground, Landsat, and MODIS data. Remote Sensing of Environment 109 (3):261 273. Fishe r, R. A. 1970. Statistical Methods for Research Workers : Oliver and Boyd. Flato, G. M., Boer, G.J. 2001. Warming asymmetry in climate change simulations. Geophysical Research Letters 28:195 198.
124 Flohn, H., and A. Kapala. 1989. Changes of Tropical Sea Air Interaction Processes over a 30 Year Period. Nature 338 (6212):244 246. Foody, G. M. 2003. Geographical weighting as a further refinement to regression modelling: An example focused on the NDVI r ainfall relationship. Remote Sensing of Environment 88 (3):283 293. Fotheringham, A. S., Brundsdon, C., Charlton, M. 2002. Geographically Weighted Regression. The Analysis of Spatially Varying Relationships. Chichester: Wiley Francis, R. C., S. R. Hare A. B. Hollowed, and W. S. Wooster. 1998. Effects of interdecadal climate variability on the oceanic ecosystems of the NE Pacific. Fisheries Oceanography 7 (1):1 21. Franz, T. E., K. K. Caylor, J. M. Nordbotten, I. Rodriguez Iturbe, and M. A. Celia. 2010 An ecohydrological approach to predicting regional woody species distribution patterns in dryland ecosystems. Advances in Water Resources 33 (2):215 230. Friedl, M. A., D. K. McIver, J. C. F. Hodges, X. Y. Zhang, D. Muchoney, A. H. Strahler, C. E. Woodcock, S. Gopal, A. Schneider, A. Cooper, A. Baccini, F. Gao, and C. Schaaf. 2002. Global land cover mapping from MODIS: algorithms and early results. Remote S ensing of Environment 83 (1 2):287 302. Friedl, M. A., D. Sulla Menashe, B. Tan, A. Schneider, N. Ramankutty, A. Sibley, and X. M. Huang. 2010. MODIS Collection 5 global land cover: Algorithm refinements and characterization of new datasets. Remote Sensin g of Environment 114 (1):168 182. Friedman, M. 1939. The use of ranks to avoid the assumption of normality implicit in the analysis of variance, (vol 32, pg 675, 1937). Journal of the American Statistical Association 34 (205):109 109. Frost, P., E. Medin a, J. C. Menaut, O. Solbrig, M. Swift, and B. Walker. 1986. Responses of Savannas to Stress and Disturbance: Proposal for a collaborative Programme of Research. Biology International (Special Issue 10):1 82. Fuller, D. O. 1999. Canopy phenology of some mo pane and miombo woodlands in eastern Zambia. G lobal Ecology and Biogeography 8:199 209. Fuller, D. O., and S. D. Prince. 1996. Rainfall and foliar dynamics in tropical southern Africa: Potential impacts of global climatic change on savanna vegetation. Cli matic Change 33 (1):69 96.
125 Gasse, F., F. Chalie, A. Vincens, M. A. J. Williams, and D. Williamson. 2008. Climatic patterns in equatorial and southern Africa from 30,000 to 10,000 years ago reconstructed from terrestrial and near shore proxy data. Quater nary Science Reviews 27 (25 26):2316 2340. Gaughan, A. E., M. W. Binford, and J. Southworth. 2009. Tourism, forest conversion, and land transformations in the Angkor basin, Cambodia. Applied Geography 29 (2):212 223. Giannini, A., M. Biasutti, I. M. Held and A. H. Sobel. 2008. A global perspective on African climate. Climatic Change 90 (4):359 383. GLP. 2005. Science Plan and Implementation Strategy. IGBP Report No. 53/IHDP Report No. 19, ed. I. Secretariat. Stockholm. Gobron, N., B. Pinty, M. M. Verst raete, and J. L. Widlowski. 2000. Advanced vegetation indices optimized for up coming sensors: Design, performance, and applications. Ieee Transactions on Geoscience and Remote Sensing 38 (6):2489 2505. Goddard, L., and N. E. Graham. 1999. Importance of t he Indian Ocean for simulating rainfall anomalies over eastern and southern Africa. Journal of Geophysical Research Atmospheres 104 (D16):19099 19116. Gordon, C., C. Cooper, C. A. Senior, H. Banks, J. M. Gregory, T. C. Johns, J. F. B. Mitchell, and R. A. Wood. 2000. The simulation of SST, sea ice extents and ocean heat transports in a version of the Hadley Centre coupled model without flux adjustments. Climate Dynamics 16 (2 3):147 168. Goward, S. M., and S. D. Prince. 1995. Transient effects of climate on vegetation dynamics: satellite observations. Journal of Biogeography 22:549 563. Graham, N. E. 1994. Decadal Scale Climate Variability in the Tropical and North Pacific during the 1970 s and 1980s Observations and Model Results. Climate Dynamics 10 (3):135 162. Groeneveld, D. P., and W. M. Baugh. 2007. Correcting satellite data to detect vegetation signal for eco hydrologic analyses. Journal of Hydrology 344 (1 2):135 145. Hansingo, K., and C. J. C. Reason. 2009. Modelling the atmospheric response over southern Africa to SST forcing in the southeast tropical Atlantic and southwest subtropical Indian Oceans. International Journal of Climatology 29 (7):1001 1012.
126 Hare, S. R. and N. J. Mantua. 2000. Empirical evidence for North Pacific regime shifts in 1977 and 1989. Progress in Oceanography :103 145. Harris, P., A. S. Fotheringham, and S. Juggins. 2010. Robust Geographically Weighted Regression: A Technique for Quantifying S patial Relationships Between Freshwater Acidification Critical Loads and Catchment Attributes. Annals of the Association of American Geographers 100 (2):286 306. Higgins, S. I., W. J. Bond, and W. S. W. Trollope. 2000. Fire, resprouting and variability: a recipe for grass tree coexistence in savanna. Journal of Ecology 88 (2):213 229. Hirst, A. C., and S. Hastenrath. 1983. Atmosphere Ocean Mechanisms of Climate Anomalies in the Angola Tropical Atlantic Sector. Journal of Physical Oceanography 13 (7):1146 1157. House, J. I., S. Archer, D. D. Breshears, and R. J. Scholes. 2003. Conundrums in mixed woody herbaceous plant systems. Journal of Biogeography 30 (11):1763 1777. Huemmrich, K. F., J. L. Privette, M. Mukelabai, R. B. Myneni, and Y. Knyazikhin. 2005. Time series validation of MODIS land biophysical products in a Kalahari woodland, Africa. International Journal of Remote Sensing 26 (19):4381 4398. Huete, A., K. Didan, T. Miura, E. P. Rodriguez, X. Gao, and L. G. Ferreira. 2002. Overview of the radiome tric and biophysical performance of the MODIS vegetation indices. Remote Sensing of Environment 83 (1 2):195 213. Huffman, G. J., R. F. Adler, P. Arkin, A. Chang, R. Ferraro, A. Gruber, J. Janowiak, A. McNab, B. Rudolf, and U. Schneider. 1997. The Global Precipitation Climatology Project (GPCP) Combined Precipitation Dataset. Bulletin of the American Meteorological Society 78 (1):5 20. Hulme, M., R. Doherty, T. Ngara, M. New, and D. Lister. 2001. African climate change: 1900 2100. Climate Research 17 (2): 145 168. Huxman, T. E., M. D. Smith, P. A. Fay, A. K. Knapp, M. R. Shaw, M. E. Loik, S. D. Smith, D. T. Tissue, J. C. Zak, J. F. Weltzin, W. T. Pockman, O. E. Sala, B. M. Haddad, J. Harte, G. W. Koch, S. Schwinning, E. E. Small, and D. G. Williams. 2004. Convergence across biomes to a common rain use efficiency. Nature 429 (6992):651 654. Huxman, T. E., B. P. Wilcox, D. D. Breshears, R. L. Scott, K. A. Snyder, E. E. Small, K. Hultine, W. T. Pockman, and R. B. Jackson. 2005. Ecohydrological implications of woody plant encroachment. Ecology 86 (2):308 319.
127 IPCC. 2007. Climate Change 2007: Th e Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change., eds. S. D. Solomon, D. Qin, M. Manning, Z. Chen, M. Marquis, K.B. Averyt, M. Tignor and H. L. Miller, 996. Cambridg e, UK. Ji, L., and A. J. Peters. 2005. Lag and seasonality considerations in evaluating AVHRR NDVI response to precipitation. Photogrammetric Engineering And Remote Sensing 71 (9):1053 1061. Jury, M. 2010. Climate and weather factors modulating river flo ws in southern Angola. International Journal of Climatology 30 (6):901 908. Jury, M. R., W. B. White, and C. J. C. Reason. 2004. Modelling the dominant climate signals around southern Africa. Climate Dynamics 23 (7 8):717 726. Justice, C. O., J. R. G. To wnshend, B. N. Holben, and C. J. Tucker. 1985. Analysis of the Phenology of Global Vegetation Using Meteorological Satellite Data. International Journal of Remote Sensing 6 (8):1271 1318. Kalnay, E., M. Kanamitsu, R. Kistler, W. collins, D. Deaven, L. Gan din, M. Iredell, S. Saha, G. White, J. Woollen, Y. Zhu, M. Chelliah, W. Ebisuzaki, W. Higgins, J. Janowiak, K. C. Mo, C. Ropelewski, J. Wang, A. Leetmaa, R. L. Reynolds, R. Jenne, and D. Jospeh. 1996. The NCEP/NCAR 40 Year Reanalysis Project. Bulletin of t he American Meteorological Society 77 (3):437 471. Kampstra, P. 2008. Beanplot: A Boxplot Alternative for Visual Comparison of Distributions. Journal of Statistical Software 28 (Code Snippet 1). Knight, J. F., R. S. Lunetta, J. Ediriwickrema, and S. Khor rarn. 2006. Regional scale land cover characterization using MODIS NDVI 250 m multi temporal imagery: A phenology based approach. Giscience & Remote Sensing 43 (1):1 23. Kummerow, C., J. Simpson, O. Thiele, W. Barnes, A. T. C. Chang, E. Stocker, R. F. Adler, A. Hou, R. Kakar, F. Wentz, P. Ashcroft, T. Kozu, Y. Hong, K. Okamoto, T. Iguchi, H. Kuroiwa, E. Im, Z. Haddad, G. Huffman, B. Ferrier, W. S. Olson, E. Zipser, E. A. S mith, T. T. Wilheit, G. North, T. Krishnamurti, and K. Nakamura. 2000. The status of the Tropical Rainfall Measuring Mission (TRMM) after two years in orbit. Journal of Applied Meteorology 39 (12):1965 1982. Lambin, E. F., and H. Geist. 2006. Land Use and Land Cover Change: Local Processes and Global Impacts. Land Use and Land Cover Change: Local Processes and Global Impacts Legates, D. R., and C. J. Willmott. 1990. Mean Seasonal and Spatial Variability in Gauge Corrected, Global Precipitation. Internati onal Journal of Climatology 10 (2):111 127.
128 Lehner, B., K. Verdin, and A. Jarvis. 2006. HydroSHEDS Technical Documentation, 27: World Wildlife Fund. Leprieur, C., Y. H. Kerr, and J. M. Pichon. 1996. Critical assessment of vegetation indices from AVHRR in a semi arid environment. I nternational Journal of Remote Sensing 17 (13):2549 2563. Levitus, S., J. I. Antonov, T. P. Boyer, and C. Stephens. 2000. Warming of the world ocean. S cience 287 (5461):2225 2229. Los, S. O., G. J. Collatz, P. J. Sellers, C. M. Malmstrom, N. H. Pollack, R. S. DeFries, L. Bounoua, M. T. Parris, C. J. Tucker, and D. A. Dazlich. 2000. A global 9 yr biophysical land surface dataset from NOAA AVHRR data. Journal of Hydrometeorology 1 (2) :183 199. Love, D., S. Twomlow, W. Mupangwa, P. van der Zaag, and B. Gumbo. 2006. Implementing the millennium development food security goals Challenges of the southern African context. Physics and Chemistry of the Earth 31 (15 16):731 737. Maidment, D R. ed. 2002. Arc Hydro GIS for Water Resources 2nd ed. Redlands, CA: ESRI Press. Marchant, R., C. Mumbi, S. Behera, and T. Yamagata. 2007. The Indian Ocean dipole the unsung driver of climatic variability in East Africa. African Journal of Ecology 45 (1):4 16. Martiny, N., P. Camberlin, Y. Richard, and N. Philippon. 2006. Compared regimes of NDVI and rainfall in semi arid regions of Africa. I nternational Journal of Remote Sensing 27 (23 24):5201 5223. Martiny, N., Y. Richard, and P. Camberlin. 2005. Interannual persistence effects in vegetation dynamics of semi arid Africa. Geophysical Research Letters 32 (24): Mason, S. J. 1996. Climatic change over the Lowveld of South Africa. Climatic Change 32 (1):35 54. 2001. El Nino, climate change, an d Southern African climate. Environmetrics 12 (4):327 345. Mason, S. J., and L. Goddard. 2001. Probabilistic precipitation anomalies associated with ENSO. Bulletin of the American Meteorological Society 82 (4):619 638. Mason, S. J., A. M. Joubert, C. Cos ijn, and S. J. Crimp. 1996. Review of seasonal forecasting techniques and their applicability to Southern Africa. Water Sa 22 (3):203 209.
129 Mason, S. J., and M. R. Jury. 1997. Climatic variability and change over southern Africa: A reflection on underlying processes. Progress in Physical Geography 21 (1):23 50. Matsuura, K., and C. J. Willmott. Terrestrial Precipitation: 1900 2006 Gridded Monthly Time Series Center for Climatic Research 2007 [cited February 18, 2009. Available from http://climate.geog.udel.edu/~climate/html_pages/Global_ts_2007/README.glob al.p_ts_2007.html Mayes, S. 2008. Park Profile: Bwabwata National Park. Sandpaper: Smoothing t he way forward for protected area management in Namibia. Bwabwata National Park. Newsletter fo the Strengthening the Protected Area Network (SPAN) Project 6 9. McCarthy, T. S., G. R. J. Cooper, P. D. Tyson, and W. N. Ellery. 2000. Seasonal flooding in the Okavango Delta, Botswana recent history and future prospects. South African Journal of Science 96 (1):25 33. McCartney, M. P. 2000. The water budget of a headwater catchment containing a dambo. Physics and Chemistry of the Earth Part B Hydrology Oceans and Atmosphere 25 (7 8):611 616. Meehl, G. A., A. X. Hu, and B. D. Santer. 2009. The Mid 1970s Climate Shift in the Pacific and the Relative Roles of Forced versus Inherent Decadal Variability. Journal of Climate 22 (3):780 792. Mendelsohn, J., and C. Roberts. 1997. An Environmental Profile and Atlas of Caprivi Windhoek: Directorate of Environmental Affairs, Mini stry of Environment and Tourism. Merbold, L., J. Ardo, A. Arneth, R. J. Scholes, Y. Nouvellon, A. de Grandcourt, S. Archibald, J. M. Bonnefond, N. Boulain, N. Brueggemann, C. Bruemmer, B. Cappelaere, E. Ceschia, H. A. M. El Khidir, B. A. El Tahir, U. Falk J. Lloyd, L. Kergoat, V. Le Dantec, E. Mougin, M. Muchinda, M. M. Mukelabai, D. Ramier, O. Roupsard, F. Timouk, E. M. Veenendaal, and W. L. Kutsch. 2009. Precipitation as driver of carbon fluxes in 11 African ecosystems. Biogeosciences 6 (6):1027 1041. Nash, D. J., and G. H. Endfield. 2008. 'Splendid rains have fallen': links between El Nino and rainfall variability in the Kalahari, 1840 1900. Climatic Change 86 (3 4):257 290. Nemani, R. R., C. D. Keeling, H. Hashimoto, W. M. Jolly, S. C. Piper, C. J. T ucker, R. B. Myneni, and S. W. Running. 2003. Climate driven increases in global terrestrial net primary production from 1982 to 1999. S cience 300 (5625):1560 1563.
130 Nicholson, S. E. 1986. The Nature of Rainfall Variability in Africa South of the Equator. J ournal of Climatology 6 (5):515 530. 1997a. An analysis of the ENSO signal in the tropical Atlantic and western Indian Oceans. International Journal of Climatology 17 (4):345 375. 1997b. An analysis of the ENSO signal in the tropical Atlantic a nd western Indian Oceans (vol 17, pg 345, 1997). International Journal of Climatology 17 (9):1008 1008. 2000. The nature of rainfall variability over Africa on time scales of decades to millenia. Global and Planetary Change 26 (1 3):137 158. 20 01. Climatic and environmental change in Africa during the last two centuries. Climate Research 17 (2):123 144. Nicholson, S. E., M. L. Davenport, and A. R. Malo. 1990. A Comparison of the Vegetation Response to Rainfall in the Sahel and East Africa, Usin g Normalized Difference Vegetation Index from Noaa Avhrr. Climatic Change 17 (2 3):209 241. Nicholson, S. E., and D. Entekhabi. 1987. Rainfall Variability in Equatorial and Southern Africa Relationships with Sea Surface Temperatures Along the Southweste rn Coast of Africa. Journal of Climate and Applied Meteorology 26 (5):561 578. Nicholson, S. E., and T. J. Farrar. 1994. The Influence of Soil Type on the Relationships between Ndvi, Rainfall, and Soil Moisture in Semiarid Botswana .1. Ndvi Response to Ra infall. Remote Sensing of Environment 50 (2):107 120. Nicholson, S. E., and E. Kim. 1997. The relationship of the El Nino Southern oscillation to African rainfall. International Journal of Climatology 17 (2):117 135. Nicholson, S. E., Kim, J. 1997c. The relationship of the El Nio Southern Oscillation to African rainfall. International Journal of Climatology 17:123 144. Nicholson, S. E., D. Leposo, and J. Grist. 2001. The relationship between El Nino and drought over Botswana. Journal of Climate 14 (3):323 335. Nicholson, S. E., B. Some, J. McCollum, E. Nelkin, D. Klotter, Y. Berte, B. M. Diallo, I. Gaye, G. Kpabeba, O. Ndiaye, J. N. Noukpozounkou, M. M. Tanu, A. Thiam, A. A. Toure, and A. K. Traore. 2003. Validation of TRMM and other rainfall estimates with a high density gauge dataset for West Africa. Part I: Validation of GPCC rainfall product and pre TRMM satellite and blended products. Journal of Applied Meteorology 42 (10):1337 1354.
131 Nitta, T., and S. Yamada. 198 9. Recent Warming of Tropical Sea Surface Temperature and Its Relationship to the Northern Hemisphere Circulation. Journal of the Meteorological Society of Japan 67 (3):375 383. Nyenzi, B., and P. F. Lefale. 2006. El Nino Southern Oscillation (ENSO) and g lobal warming. Advances in Geosciences 6:95 101. O'Connell Rodwell, C. E., T. Rodwell, M. Rice, and L. A. Hart. 2000. Living with the modern conservation paradigm: can agricultural communities co exist with elephants? A five year case study in East Capriv i, Namibia. Biological Conservation 93 (3):381 391. Omuto, C. T., R. R. Vargas, M. S. Alim, and P. Paron. 2010. Mixed effects modelling of time series NDVI rainfall relationship for detecting human induced loss of vegetation cover in drylands. Journal of Arid Environments 74 (11):1552 1563. Owen Smith, N., and S. Cooper. 1989. Nutritional ecology of a browsing ruminant, the kudu ( Tragelaphus strepsiceros ) through the seasonal cycle. Journal of Zoology 219 (1):29 43. Owusu, K., P. Waylen, and Y. Qiu. 2008 Changing rainfall inputs in the Volta basin: implications for water sharing in Ghana. GeoJournal 71:210 210. Paeth, H., and P. Friederichs. 2004. Seasonality and time scales in the relationship between global SST and African rainfall. Climate Dynamics 2 3 (7 8):815 837. Paterson, R. T., G. M. Karanja, R. L. Roothaert, O. Z. Nyaata, and I. W. Kariuki. 1998. A review of tree fodder production and utilization within smallholder agroforestry systems in Kenya. Agroforestry Systems 41 (2):181 199. Peters, A. J., E. A. Walter Shea, L. Ji, A. Vina, M. Hayes, and M. D. Svoboda. 2002. Drought monitoring with NDVI based standardized vegetation index. Photogrammetric Engineering And Remote Sensing 68 (1):71 75. Phillips, J. G., M. A. Cane, and C. Rosenzweig. 1998. ENSO, seasonal rainfall patterns and simulated maize yield variability in Zimbabwe. Agricultural and Forest Meteorology 90 (1 2):39 50. Porporato, A., F. Laio, L. Ridolfi, K. K. Caylor, and I. Rodriguez Iturbe. 2003. Soil moisture and plant stress dynamic s along the Kalahari precipitation gradient. Journal of Geophysical Research Atmospheres 108 (D3). Power, S., T. Casey, C. Folland, A. Colman, and V. Mehta. 1999. Inter decadal modulation of the impact of ENSO on Australia. Climate Dynamics 15 (5):319 324.
132 Prince, S. D., S. J. Goetz, and S. N. Goward. 1995. Monitoring Primary Production from Earth Observing Satellites. Water Air and Soil Pollution 82 (1 2):509 522. Prins, H. H. T. 1988. Plant Phenology Patterns in Lake Manyara National Park Tanzania. Journal of Biogeography 15 (3):465 480. Pu, R. L., P. Gong, Y. Tian, X. Miao, R. I. Carruthers, and G. L. Anderson. 2008. Using classification and NDVI differencing methods for monitoring sparse vegetation coverage: a case study of saltcedar i n Nevada, USA. International Journal of Remote Sensing 29 (14):3987 4011. Puigdefabregas, J. 1998. Ecological impacts of global change on drylands and their implications for desertification. Land Degradation & Development 9 (5):393 406. Reason, C. J. C. 2002. Sensitivity of the southern African circulation to dipole sea surface temperature patterns in the south Indian Ocean. International Journal of Climatology 22 (4):377 393. Reason, C. J. C., and M. Rouault. 2002. ENSO like decadal variability and Sout h African rainfall. Geophysical Research Letters 29 (13): Reynolds, J. F., D. M. Stafford Smith, E. F. Lambin, B. L. Turner, M. Mortimore, S. P. J. Batterbury, T. E. Downing, H. Dowlatabadi, R. J. Fernandez, J. E. Herrick, E. Huber Sannwald, H. Jiang, R Leemans, T. Lynam, F. T. Maestre, M. Ayarza, and B. Walker. 2007. Global desertification: Building a science for dryland development. S cience 316:847 851. Richard, Y., N. Fauchereau, I. Poccard, M. Rouault, and S. Trzaska. 2001. 20th century droughts in southern Africa: Spatial and temporal variability, teleconnections with oceanic and atmospheric conditions. International Journal of Climatology 21 (7):873 885. Richard, Y., N. Martiny, N. Fauchereau, C. Reason, M. Rouault, N. Vigaud, and Y. Tracol. 2008 Interannual memory effects for spring NDVI in semi arid South Africa. Geophysical Research Letters 35 (13): Richard, Y., and I. Poccard. 1998. A statistical study of NDVI sensitivity to seasonal and interannual rainfall variations in Southern Africa. International Journal of Remote Sensing 19 (15):2907 2920. Rindfuss, R. R., S. J. Walsh, B. L. Turner, J. Fox, and V. Mishra. 2004. Developing a science of land change: Challenges and methodological issues. Proceedings of the National Academy of Sciences of the United States of America 101 (39):13976 13981.
133 Ringrose, S., C. Vanderpost, W. Matheson, P. Wolski, P. Huntsman Mapila, M. Murray Hudson, and A. Jellema. 2007. Indicators of desiccation driven change in the distal Okavango Delta, Botswana. Journal o f Arid Environments 68 (1):88 112. Rodriguez Iturbe, I. 2000. Ecohydrology: A hydrologic perspective of climate soil vegetation dynamics. Water Resources Research 36 (1):3 9. Ropelewski, C. F., and M. S. Halpert. 1987. Global and Regional Scale Precipitation Patterns Associated with the El Nino Southern Oscillation. Monthly Weather Review 115 (8):1606 1626. Rotenberg, E., and D. Yakir. 2010. Contribution of Semi Arid Forests to the Climate System. S cience 327 (5964):451 454. Rouault, M., P. Florenchie, N. Fauchereau, and C. J. C. Reason. 2003. South East tropical Atlantic warm events and southern African rai nfall. Geophysical Research Letters 30 (5): Sankaran, M., N. Hanan, and R. J. Scholes. Characteristics of African savanna biomes for determining woody cover 2007 [cited. Available from http://www.daac.ornl.gov Sankaran, M., N. P. Hanan, R. J. Scholes, J. Ratnam, D. J. Augustine, B. S. Cade, J. Gignoux, S. I. Higgins, X. Le Roux, F. Ludwig, J. Ardo, F. Banyikwa, A. Bronn, G. Bucini, K. K. Caylor, M. B. Coughenour, A. Diouf, W. Ekaya, C. J. Feral, E. C. February, P. G. H. Frost, P. Hiernaux, H. Hrabar, K. L. Metzger, H. H. T. Prins, S. Ringrose, W. Sea, J. Tews, J. Worden, and N. Zambatis. 2005. Determinants of woody cover in African savannas. Nature 438 (7069):846 849. Sankaran, M., J. Ratnam, and N. Hanan. 2008 Woody cover in African savannas: the role of resources, fire and herbivory. G lobal Ecology and Biogeography 17 (2):236 245. Scanlon, T. M., K. K. Caylor, S. Manfreda, S. A. Levin, and I. Rodriguez Iturbe. 2005. Dynamic response of grass cover to rainfa ll variability: implications for the function and persistence of savanna ecosystems. Advances in Water Resources 28 (3):291 302. Scholes, R. J., and S. R. Archer. 1997. Tree grass interactions in savannas. Annual Review of Ecology and Systematics 28:517 5 44. Scholes, R. J., P. R. Dowty, K. Caylor, D. A. B. Parsons, P. G. H. Frost, and H. H. Shugart. 2002. Trends in savanna structure and composition along an aridity gradient in the Kalahari. Journal of Vegetation Science 13 (3):419 428.
134 Scholes, R. J., an d B. H. Walker. 1993. An African Savanna. Synthesis of the Nylsvley Study : Cambridge University Press. Schwinning, S., O. E. Sala, M. E. Loik, and J. R. Ehleringer. 2004. Thresholds, memory, and seasonality: understanding pulse dynamics in arid/semi arid ecosystems. Oecologia 141 (2):191 193. Sekhwela, M. B. M., and D. J. Yates. 2007. A phenological study of dominant acacia tree species in areas with different rainfall regimes in the Kalahari of Botswana. Journal of Arid Environments 70 (1):1 17. Shackle ton, C. M. 1999. Rainfall and topo edaphic influences on woody community phenology in South African savannas. G lobal Ecology and Biogeography 8 (2):125 136. Shackleton, C. M., S. E. Shackleton, E. Buiten, and N. Bird. 2007. The importance of dry woodlands and forests in rural livelihoods and poverty alleviation in South Africa. Forest Policy and Economics 9 (5):558 577. Shaw, P. 1983. Fluctuations in the level of Lake Ngami: the historical evidence. Botswana Notes and Records 15:79 84. Sher, A. A., D. E. Goldberg, and A. Novoplansky. 2004. The effect of mean and variance in resource supply on survival of annuals from Mediterranean and desert environments. Oecologia 141 (2):353 362. Shongwe, M. E., G. J. van Oldenborgh, B. J. J. M. van d en Hurk, B. de Boer, C. A. S. Coelho, and M. K. van Aalst. 2009. Projected Changes in Mean and Extreme Precipitation in Africa under Global Warming. Part I: Southern Africa. Journal of Climate 22 (13):3819 3837. Sims, D. A., H. Y. Luo, S. Hastings, W. C. Oechel, A. F. Rahman, and J. A. Gamon. 2006. Parallel adjustments in vegetation greenness and ecosystem CO2 exchange in response to drought in a Southern California chaparral ecosystem. Remote Sensing of Environment 103 (3):289 303. Sims, D. A., A. F. Rahman, V. D. Cordova, B. Z. El Masri, D. D. Baldocchi, P. V. Bolstad, L. B. Flanagan, A. H. Goldstein, D. Y. Hollinger, L. Misson, R. K. Monson, W. C. Oechel, H. P. Schmid, S. C. Wofsy, and L. Xu. 2008. A new model of gross primary prod uctivity for North American ecosystems based solely on the enhanced vegetation index and land surface temperature from MODIS. Remote Sensing of Environment 112 (4):1633 1646.
135 Sjostrom, M., J. Ardo, L. Eklundh, B. A. El Tahir, H. A. M. El Khidir, M. Hell strom, P. Pilesjo, and J. Seaquist. 2009. Evaluation of satellite based indices for gross primary production estimates in a sparse savanna in the Sudan. Biogeosciences 6 (1):129 138. Skarpe, C. 1992. Dynamics of Savanna Ecosystems. Journal of Vegetation S cience 3 (3):293 300. Stevens, F. R. 2009. Briding the Landsat data gap: Evaluating ASTER as an alternative, Geography, University of Florida, Gainesville. Thomas, D. S. G., P. W. O'Connor, M. D. Bateman, P. A. Shaw, S. Stokes, and D. J. Nash. 2000. Dune activity as a record of late Quaternary aridity in the Northern Kalahari: new evidence from northern Namibia interpreted in the context of regional arid and humid chronologies. Palaeogeography Palaeoclimatology Palaeoecology 156 (3 4):243 259. Thornton, P. K., R. H. Fawcett, K. A. Galvin, R. B. Boone, J. W. Hudson, and C. H. Vogel. 2004. Evaluating management options that use climate forecasts: modelling livestock production systems in the semi arid zone of South Africa. Climate Research 26 (1):33 42. Ts onis, A. A., J. B. Elsner, A. G. Hunt, and T. H. Jagger. 2005. Unfolding the relation between global temperature and ENSO. Geophysical Research Letters 32 (9): Tucker, C. J. 1979. Red and Photographic Infrared Linear Combinations for Monitoring Vegetati on. Remote Sensing of Environment 8 (2):127 150. Tucker, C. J., and P. J. Sellers. 1986. Satellite Remote Sensing of Primary Production. I nternational Journal of Remote Sensing 7 (11):1395 1416. Turner, B. L., W. C. Clark, R. W. Kates, J. F. Richards, J. T. Mathews, and W. B. Meyer. 1990. The Earth as Transformed by Human Action: Global and Regional Changes in the Biosphere over the Past 300 Years Cambridge: Cambridge University Press. Turner, B. L., E. F. Lambin, and A. Reenberg. 2007. The emergence of land change science for global environmental change and sustainability. Proceedings of the National Academy of Sciences of the United States of America 104 (52):20666 20671. Tyson, P. D. and R. A. Preston Whyte. 2000. The Weather and Climate of Southern Africa Oxford: Oxford University Press.
136 van Aarde, R. J., and T. P. Jackson. 2007. Megaparks for metapopulations: Addressing the causes of locally high elephant numbers in southern Afri ca. Biological Conservation 134 (3):289 297. Veenendaal, E. M., O. Kolle, and J. Lloyd. 2004. Seasonal variation in energy fluxes and carbon dioxide exchange for a broad leaved semi arid savanna (Mopane woodland) in Southern Africa. Global Change Biology 10 (3):318 328. Vetter, S. 2009. Drought, change and resilience in South Africa's arid and semi arid rangelands. South African Journal of Science 105 (1 2):29 33. Vigaud, N., Y. Richard, M. Rouault, and N. Fauchereau. 2009. Moisture transport between the South Atlantic Ocean and southern Africa: relationships with summer rainfall and associated dynamics. Climate Dynamics 32 (1):113 123. Walker, B., C. S. Hollin, S. R. Carpenter, and A. Kinzig. 2004. Resilience, adaptability and transformability in social ecological systems. Ecology and Society 9 (2): Walker, B. H., D. Ludwig, C. S. Holling, and R. M. Peterman. 1981. Stability of Semi Arid Savanna Grazing Systems. Journal of Ecology 69 (2):473 498. Walters, C. J., and C. S. Holling. 1990. Large Scale M anagement Experiments and Learning by Doing. Ecology 71 (6):2060 2068. Walther, G. R., E. Post, P. Convey, A. Menzel, C. Parmesan, T. J. C. Beebee, J. M. Fromentin, O. Hoegh Guldberg, and F. Bairlein. 2002. Ecological responses to recent climate change. N ature 416 (6879):389 395. Wang, B. 1995. Interdecadal Changes in El Nino Onset in the Last 4 Decades. Journal of Climate 8 (2):267 285. Wang, G. L. 2003. Reassessing the impact of North Atlantic Oscillation on the sub Saharan vegetation productivity. Global Change Biology 9 (4):493 499. Wang, J., K. P. Price, and P. M. Rich. 2001. Spatial patterns of NDVI in response to precipitation and temperature in the central Great Plains. International Journal of Remote Sensing 22 (18):3827 3844. Wang, J., P. M Rich, and K. P. Price. 2003. Temporal responses of NDVI to precipitation and temperature in the central Great Plains, USA. International Journal of Remote Sensing 24 (11):2345 2364.
137 Wang, L., P. D'Odorico, S. Ringrose, S. Coetzee, and S. A. Macka. 2007. Biogeochemistry of Kalahari sands. Journal of Arid Environments 71:259 279. Washington, R., and A. Preston. 2006. Extreme wet years over southern Africa: Role of Indian Ocean sea surf ace temperatures. Journal of Geophysical Research Atmospheres 111 (D15): Wessels, K. J., S. D. Prince, J. Malherbe, J. Small, P. E. Frost, and D. VanZyl. 2007. Can human induced land degradation be distinguished from the effects of rainfall variability? A case study in South Africa. Journal of Arid Environments 68 (2):271 297. White, F. 1983. The vegetation of Africa, a descriptive memoir to accompany the UNESCO/AETFAT/UNSO vegetation map of Africa Paris. White, M. A., K. M. de Beurs, K. Didan, D. W. Inouye, A. D. Richardson, O. P. Jensen, J. O'Keefe, G. Zhang, R. R. Nemani, W. J. D. van Leeuwen, J. F. Brown, A. de Wit, M. Schaepman, X. M. Lin, M. Dettinger, A. S. Bailey, J. Kimball, M. D. Schwartz, D. D. Baldocchi, J. T. Lee, and W. K. Lauenroth. 2009 Intercomparison, interpretation, and assessment of spring phenology in North America estimated from remote sensing for 1982 2006. Global Change Biology 15 (10):2335 2359. Wiegand, T., H. A. Snyman, K. Kellner, and J. M. Paruelo. 2004. Do grasslands have a memory: Modeling phytomass production of a semiarid South African grassland. Ecosystems 7 (3):243 258. Williams, C. A., N. Hanan, R. J. Scholes, and W. Kutsch. 2009. Complexity in water and carbon dioxide fluxes following rain pulses in an African sava nna. Oecologia 161 (3):469 480. Wolmer, W. 2003. Transboundary conservation: the politics of ecological integrity in the Great Limpopo Transfrontier Park. Journal of Southern African Studies 29 (1):261 278. Wolski, P. 2010. Mohembo Daily Hydrograph ( http://www.orc.ub.bw/datacat/mohembo.php?id=1&datatype=) Harry Oppenheimer Okavango Research Center and Department of Water Affairs, Botswana 2010 [cited 21, September 2010 2010]. Availa ble from http://www.orc.ub.bw/datacat/mohembo.php?id=1&datatype= Zhang, X., Friedl, M.A., Schaaf, C.B., and A.H. Strahler. 2005. Monitoring the response of vegetation phenology to pr ecipitation in Africa by coupling MODIS and TRMM instruments. Journal of Geophysical Research 110 (D12103).
138 BIOGRAPHICAL SKETCH Andrea E. Gaughan was born in Dallas, TX, and grew up in Texas, Southern California, and Tennessee. In May of 2003, she received a Bachelor of Arts in E nglish and a concentration in environmental studies from Furman University. During her time at Furman, A ndrea also spent a term in Chile studying environmental and community health and another term in Hawaii researching effects of engine noise on behaviors of humpback whales. In the year between undergrad and graduate school, Andrea worked at the Newfound Ma rine Harbor Institute teaching coastal and near shore ecology and also traveled in the South Pacific. Andrea began the M.S. in geography at the University of Florida in August of 2004 and completed the degree in December of 2006. She focused on land use an d land cover change in a tropical watershed in Siem Reap, Cambodia. In January 2007 she began the doctorate program in g eography at the University of Florida. Her dissertation was on climate land interactions in a southern African catchment, specifically t he response of savanna vegetation to spatial and temporal precipitation patterns. She also participated in a NSF Integrative Graduate Education and Research Traineeship (IGERT) focused on Adaptive Management, Water, W atersheds, and Wetlands (AM:W3).