Spatial Ecology of Large Herbivores in the Serengeti Ecosystem

MISSING IMAGE

Material Information

Title:
Spatial Ecology of Large Herbivores in the Serengeti Ecosystem
Physical Description:
1 online resource (167 p.)
Language:
english
Creator:
Bhotika, Smriti
Publisher:
University of Florida
Place of Publication:
Gainesville, Fla.
Publication Date:

Thesis/Dissertation Information

Degree:
Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Interdisciplinary Ecology
Committee Chair:
Holt, Robert D
Committee Members:
Binford, Michael W
Christman, Mary C
Palmer, Todd
Bolker, Benjamin M

Subjects

Subjects / Keywords:
abundance -- community -- distribution -- diversity
Interdisciplinary Ecology -- Dissertations, Academic -- UF
Genre:
Interdisciplinary Ecology thesis, Ph.D.
Electronic Thesis or Dissertation
bibliography   ( marcgt )
theses   ( marcgt )
government publication (state, provincial, terriorial, dependent)   ( marcgt )

Notes

Abstract:
With rapidly increasing human populations, ensuring thelong-term effectiveness of protected areas through wise management isincreasingly important.  To conservebiodiversity, species abundance and distribution patterns must be identifiedand underlying processes must be understood. This research examined how human activities, spatial processes, andspecies traits collectively influence the abundance, occupancy, andinterspecific associations of species in the Serengeti ecosystem in EastAfrica.  Thirteen large herbivore specieswere investigated using nine annual aerial surveys from 1988-2006.  Using spatial regression models, influencesof habitat characteristics on community distributions were assessed.  Results indicate efforts to manage forspecies richness would involve emphasizing habitat characteristics differentfrom those that would maximize total abundance, biomass, or metabolicrate.  Human activities could be managedto mitigate negative effects on wildlife habitat use (e.g., monitor roadusage).  It may also be important tomaintain the spatial and temporal heterogeneity of plant resources due to theirinfluence on the spatial distribution of the community.  Occupancy and average abundance of individualspecies differed.  Species with lowoccupancy and abundance tend to have distinct social behavior and specifichabitat associations, whereas species with high occupancy and abundance tend tobe migratory and smaller.  Species withstrong grouping behavior tend to deviate from these general patterns.  Rank occupancy-abundance profiles revealed thatthe overall shape of the distribution (straight, S-shaped, etc.) for mostspecies appears to be fairly consistent over time.  Clustering of species decreased in relationto body mass, with migratory species showing more variability inaggregation.  The structure of thecommunity, summarized using rank-abundance plots, indicated a few speciesnumerically dominate and overall community structure appears constant overtime.  Observed strong negativeassociations tend to be for species with large body sizes and which form largegroups, suggesting competition for resources and space.  In addition, negative interactions may berelated to habitat specificity.  Weaknegative associations are observed for migratory species.  The patterns observed provide an expansiveview of the large herbivore community for this study area and could potentiallybe applied to other systems or species and to predict effects of environmentalchanges and management strategies on communities.
Statement of Responsibility:
by Smriti Bhotika.
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Bibliography:
Includes bibliographical references.
General Note:
Description based on online resource; title from PDF title page.
General Note:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Thesis:
Thesis (Ph.D.)--University of Florida, 2012.
General Note:
Adviser: Holt, Robert D.

Record Information

Source Institution:
UFRGP
Rights Management:
Applicable rights reserved.
Classification:
lcc - LD1780 2012
System ID:
UFE0045013:00001


This item is only available as the following downloads:


Full Text

PAGE 1

1 SPATIAL ECOLOGY OF LARGE HERBIVORES IN THE SERENGETI ECOSYSTEM By SMRITI BHOTIKA 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 2012

PAGE 2

2 2012 Smriti Bhotika

PAGE 3

3 ACKNOWLEDGMENTS I thank my advisor, Robert Holt, and the members of my supervisory committee, Benjamin Bolker, Michael Binford, Todd Palmer, and Mary Christ man, for the insights, guidance, and suggestions they offered on various aspects of my research the contributions they made to my intellectual growth during my years of study, and the steadfast support and encouragement they provided throughout the proc es s of completing this degree. I th a nk the National Center for Ecological Analysis and Synthesis (NCEAS) Serengeti Biocomplexity Working Group for making d ata for this project available and T ony Sinclair, K ris Metzger, and G rant Hopcraft for facilitating acc ess to data sources. T he Tanzania Wildlife Research Institute (TAWIRI), the Tanzania Wildlife Conservation Monitoring Program, and the Frankfurt Zoological Society provid ed animal survey data and maps of roads and rivers for the study area; S.J. McNaughto n provided nutrient data; Rico Holdo provided nutrient maps; and J. Dempewolf provided fire maps. For awarding funds for this research, I thank the Harold Hume Presidential Fellowship John Olowo Memorial Research Award Department of Biology, School of N atural Resources and Environment, Center for African Studies, and the Office of Research and Graduate Programs of the University of Florida. I thank the faculty, staff, and graduate students in the department; their collegiality greatly influenced my journ ey in this degree program I n particular Mi ke Barfield Rico Holdo and Kristen Sauby provided input guidance, and encouragement during various stages of my research. I extend a special thanks to my dear friend Jame McCray, for her company during numer ous work sessions and for her support week after week during the final stages of my graduate program. Lastly, my foundation -I

PAGE 4

4 thank my family and friends for believing in me and my parents for support ing me however and whenever they could throughout my lifetime and especially during this endeavor.

PAGE 5

5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 3 LIST OF TABLES ................................ ................................ ................................ ............ 8 LIST OF FIGURES ................................ ................................ ................................ ........ 10 ABSTRACT ................................ ................................ ................................ ................... 12 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 14 General Introduction ................................ ................................ ............................... 14 Study Area ................................ ................................ ................................ .............. 16 Outline ................................ ................................ ................................ .................... 17 2 ON THE RELATIONSHIP OF A COMMUNITY OF LARGE HERBIVORES TO ENVIRONMENTAL AND ANTHROPOGENIC INFLUENCES IN THE SERENGETI ECOSYSTEM: A COMPARISON OF FOUR COMMUNITY MEASURES ................................ ................................ ................................ ............ 20 Methods ................................ ................................ ................................ .................. 23 Study Area ................................ ................................ ................................ ........ 23 Census Data ................................ ................................ ................................ ..... 24 Habitat Characteristics ................................ ................................ ..................... 26 Statistical Analyses ................................ ................................ .......................... 26 Community correlations ................................ ................................ ............. 26 Collinearit y among habitat variables ................................ .......................... 26 Spatial regression model selection ................................ ............................ 27 Results ................................ ................................ ................................ .................... 29 Community Correlation Patterns ................................ ................................ ...... 29 Community Distribution Patterns ................................ ................................ ...... 29 Spatial Regression Model Selection ................................ ................................ 30 Model Results ................................ ................................ ................................ ... 31 Discussion ................................ ................................ ................................ .............. 32 Community Correlation Patterns ................................ ................................ ...... 32 Community Distribution Patterns ................................ ................................ ...... 34 Human activity ................................ ................................ ........................... 34 Topography ................................ ................................ ................................ 36 Resources ................................ ................................ ................................ .. 36 Concluding Remarks ................................ ................................ ........................ 39 3 OCCUPANCY AND ABUNDANCE PATTER NS IN RELATION TO SPECIES TRAITS FOR LARGE HERBIVORES IN A SAVANNA ECOSYSTEM .................... 52

PAGE 6

6 Methods ................................ ................................ ................................ .................. 56 Study Area ................................ ................................ ................................ ........ 56 Census Data ................................ ................................ ................................ ..... 57 Occupancy and Abundance ................................ ................................ ............. 58 Aggregation ................................ ................................ ................................ ...... 59 Spatial Autocorrelation ................................ ................................ ..................... 59 Results ................................ ................................ ................................ .................... 60 Occupancy and Abundance ................................ ................................ ............. 60 Aggregation ................................ ................................ ................................ ...... 62 Spatial Autocorrelation ................................ ................................ ..................... 63 Discussion ................................ ................................ ................................ .............. 63 Occupancy and Abundance ................................ ................................ ............. 63 Aggregation ................................ ................................ ................................ ...... 67 Spatial Autocorrelation ................................ ................................ ..................... 68 Concluding Remarks ................................ ................................ ........................ 68 4 COMMUNITY STRUCTURE AND INTERSPECIFIC ASSOCIATIONS AMONG LARGE HERBIVORES IN THE SERENGETI ................................ ......................... 82 Methods ................................ ................................ ................................ .................. 84 Study Area ................................ ................................ ................................ ........ 84 Census Data ................................ ................................ ................................ ..... 85 Ra nk Abundance ................................ ................................ .............................. 86 ............................ 87 Results ................................ ................................ ................................ .................... 88 Rank Abundance ................................ ................................ .............................. 88 ............................ 89 Discussion ................................ ................................ ................................ .............. 90 Rank Abundance ................................ ................................ .............................. 90 ............................ 93 Concluding Remarks ................................ ................................ ........................ 94 5 CONCLUSIONS ................................ ................................ ................................ ... 103 APPENDIX A SUPPORTING INFORMATION FOR CHAPTER 2 ................................ ............... 108 Appendix A 1. Large Herbivore Species Surveyed in the Study System ............. 108 Appendix A 2. Distribution of Migratory Wildebeest ................................ ............. 111 Appendix A 3. Dates for Species and Habitat Data ................................ ............. 114 Appendix A 4. Spatial Distribution for Each Community Measure for Each Survey ................................ ................................ ................................ ............... 116 Appendix A 5. Habitat Characteristics Methods (with Figures). ........................... 121 Human Activity ................................ ................................ ............................... 121 Topography ................................ ................................ ................................ .... 122 Resources ................................ ................................ ................................ ...... 122

PAGE 7

7 Appendix A 6. Model Selection and Model Output for Subset of Data Consisting of Three Wet Season Surveys ................................ ......................... 137 Appendix A 7. Correlations Among the Community Measures ............................ 143 B SUPPORTING INFORMATION FOR CHAPTER 3 ................................ ............... 147 C SUPPORTING INFORMATION FOR CHAPTER 4 ................................ ............... 155 LIST OF REFERENCES ................................ ................................ ............................. 15 7 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 167

PAGE 8

8 LIST OF TABLES Table page 2 1 Species names and traits of the twelve herbivores in the study ......................... 41 2 2 Habitat characteristics considered in the study to incorporate the major human and environ mental influences in the system ................................ .......... 42 2 3 Model selection using AIC for models with species richness, total abundance, total biomass, and total BMR as the response variable for analysis of eight wet season surveys. ................................ ................................ ........................... 43 2 4 Model output for species richness for analysi s of eight wet s eason surveys ..... 45 2 5 Model output for total abundance for analysi s of eight wet season surveys ....... 46 2 6 Model output for total bi omass for analysi s of eight wet season surveys ........... 47 2 7 Model output for total basal metabolic rate (BMR) for analysi s of eight wet season surveys ................................ ................................ ................................ .. 48 3 1 Species names and traits of the th irteen herbivores in the study ........................ 70 4 1 Species names and traits of the twelve herbivores in the study ........................ 96 4 2 For each census, estimates of k and C k for the abundances of species modeled by a geometric series ................................ ................................ ........... 97 A 1 Population estimates for each herbivore specie s surveyed in the study system and included in the main analyses. ................................ ...................... 108 A 2 Population estimates for each herbivore species surveyed in the study system but not included in the main analyses. ................................ ................. 109 A 3 Traits for species recorded in the surveys but not included in the analysis ...... 110 A 4 Dates for static habitat data ................................ ................................ ............. 114 A 5 Dates for spe cies and dynamic habitat data ................................ .................... 115 A 6 Model selection using AIC for model s with species richness, total abundance total biomass, and total BMR as the response variable for analysis of th ree wet season survyes ................................ ................................ ......................... 138 A 7 Model output for species richness for analysi s of three wet season surveys ... 139 A 8 Model output for total abundance for analysi s of three wet season surveys .... 140

PAGE 9

9 A 9 Model output for total biomass for analysis of three wet season surveys. ........ 141 A 10 Model output for total basal metabolic rate for analysis of three wet season surveys ................................ ................................ ................................ ............. 142 A 11 Correlation s over time for each of the community measures ............................ 146

PAGE 10

10 LIST OF FIGURES Figure page 2 1 Serengeti National Park and neighboring conservatio n areas and game reserves in Tanzania and Kenya, East Africa ................................ .................... 49 2 2 Correlations among the four community measures for the 1996 wet season survey ................................ ................................ ................................ ................. 50 2 3 Spatial distribution for each community measure Values are the grid wise averages of eight wet season surveys between 1988 and 2006. ....................... 51 3 1 Serengeti National Park and neighboring conservation areas and game reserves in Tanzania and Kenya, East Africa ................................ ..................... 71 3 2 Abundance (average abundance of occupied sites) in relation to occupancy for a census. Nine cens uses are plotted for each species ................................ 72 3 3 Rank occupancy abundance profile (ROAP) for each species for wet and dry seasons ................................ ................................ ................................ .............. 73 3 4 Estimates for parameters m and k (indicating mean and clustering ... 77 3 5 ion) in relation to body mass of each herbivore ................................ ................................ ................................ ............. 80 4 1 Serengeti National Park and neighboring conservation areas and game reserves in Tanzania and Kenya, East Africa ................................ ..................... 98 4 2 Rank abundance plot for twelve large h erbivore species for each survey .......... 99 4 3 Abundance of each species in relation to total abundance of the rest of the community for each survey ................................ ................................ ............... 100 A 1 The distributions of migratory wildebeest for April 1999 ................................ ... 113 A 2 Spatial distribution for each community measure for eight wet season survey years between 1988 and 2006 ................................ ................................ ......... 116 A 3 Community distributions f or a dry season survey in 1996 ................................ 120 A 4 Indicators of human act ivity at each survey location in the study area ........... 126 A 5 Topography of the study area ................................ ................................ ........... 127 A 6 The distance from each survey location to the nearest river ............................ 128

PAGE 11

11 A 7 Plant nutrient maps for the study area ................................ .............................. 129 A 8 Tree cover in the study area ................................ ................................ ............ 130 A 9 NDVI from approximately the month of each survey for the study area ............ 131 A 10 Standard deviation of NDVI for four wet season surveys for the study area ..... 133 A 11 Average rainfall for the two months preceding each survey date in the study area ................................ ................................ ................................ ................. 134 A 12 Percent burn within the prior year for the stu dy area ................................ ........ 136 A 13 Correlation s among the community m easures for each survey ....................... 143 B 1 Abundance (average abu ndance of occupied sites) in relation to occupancy for a census including wildebeest ................................ ................................ ... 147 B 2 Average abundance at occupied sites in relation to occupancy for each species ................................ ................................ ................................ ............. 148 B 3 Estimates for parameters m and k (indicating mean and clustering respectively) of negative binomial fit to each speci including wildebeest ................................ ................................ ......................... 152 C 1 Rank abundance plot for twelve large herbivore species for each survey ....... 155

PAGE 12

12 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 SPATIAL ECOLOGY OF LARGE HERBIVORES IN THE SERENGETI ECOSYSTEM By Smriti Bhotika December 2012 Chair: Robert Holt Major: Interdisciplinary Ecology With rapidly increasing human populations, ensuring the long term effectiveness of protected areas through wise management is increasingly important. To cons erve biodiversity, species abundance and distribution patterns must be identified and underlying processes must be understood. This research examined how human activities, spatial processes, and species traits collectively influence the abundance, occupan cy, and interspecific associations of species in the Serengeti ecosystem in East Africa. Thirteen large herbivore species were investigated using nine annual aerial surveys from 1988 2006. Using spatial regression models, influences of habitat characteri stics on community distributions were assessed. Results indicate efforts to manage for species richness would involve emphasizing habitat characteristics different from those that would maximize total abundance, biomass, or metabolic rate. Human activiti es could be managed to mitigate negative effects on wildlife habitat use (e.g., monitor road usage). It may also be important to maintain the spatial and temporal heterogeneity of plant resources due to their influence on the spatial distribution of the c ommunity. Occupancy and average abundance of individual species differ ed Species with low occupancy and abundance tend to have distinct social behavior and specific

PAGE 13

13 habitat associations, whereas species with high occupancy and abundance tend to be migra tory and smaller. Species with strong grouping behavior tend to deviate from these general patterns. Rank occupancy abundance profiles revealed that the overall shape of the distribution (straight, S shaped, etc.) for most species appears to be fairly co nsistent ov er time. Clustering of species decreased in relation to body mass, with migratory species showing more variability in aggregation. The structure of the community, summarized using rank abundance plots, indicated a few species numerically domin ate and overall community structure appears constant over time. Observed strong negative associations tend to be for species with large body sizes and which form large groups, suggesting competition for resources and space. In addition, negative interact ions may be related to habitat specificity. Weak negative associations are observed for migratory species. The patterns observed provide an expansive view of the large herbivore community for this study area and could potentially be applied to other syst ems or species and to predict effects of environmental changes and management strategies on communities.

PAGE 14

14 CHAPTER 1 INTRODUCTION General Introduction With increasing human population and activity, the need to balance ecosystem and biodiversity conservati on with sustainable anthropogenic living practices has intensified. Parks, one of many approaches used to achieve biological conservation and preserve ecosystem processes, are coupled with human activities varying from wildlife tourism and game reserves t o agricultural and ranching intensification in neighboring areas (Sinclair et al. 2007, Homewood et al. 2001) Parks can be established in areas having little human population or may require integrat ing with local human communities Once established, pr otected areas must be monitored and managed to ensure the habitat remains adequate for conservation in the long term amid changes in environmental conditions, anthropogenic activities, and management practices within and across park boundaries. Knowledge of species responses to changes in an ecosystem is essential to establish effective long term conservation and wise management programs, including the interactions between wildlife and wildlife dependent human livelihoods and other human activities To c onserve biodiversity effectively within protected areas and predict how species within and outside protected areas will respond to such changes, species abundance and distribution patterns must be identified and the processes underlying the se patterns must be understood. To gain knowledge of the distribution, abundance, and composition of species within a community, multiple processes that function at a range of spatial and temporal scales must be taken into consideration. The spatial arrangement of a spec ies is in the first place the result of physiological constraints due to abiotic characteristics of the

PAGE 15

15 habitat (i.e., the fundamental niche of a species) (Hutchinson 1957, Mac Arthur et al. y affected by biotic interactions such as competition, predation, mutualisms, and disease (i.e., the realized niche of a species) as well as conspecific attraction and stochastic dispersal and disturbance events (Hutchinson 1957, Pulliam 2000, Hubbell 2001 Lichstein et al. 2002, Tilman 2004). The limitations imposed by abiotic and biotic factors in a system influence the traits of species that persist and coexist in a system and in turn species Wh en investigating populations and communities, ecological studies and theory have traditionally assumed spatial homogeneity for simplicity H owever, spatial heterogeneity is a central factor that influences processes such as resource availability, species dispersal, and physiology (Pickett and Cadenasso 1995). Ecological studies that conduct gradient analyses -investigating how populations of species change along gradients such as elevation, temperature, vegetation, or precipitation (Whittaker 1952) -m ay reveal important mechanisms regulating species populations and make apparent thresholds above and below which dramatic shifts in the populations occur (Hutchinson 1957, Turner 2005). Furthermore, due to the spatial nature of many ecological studies, sp atial autocorrelation is often present in ecological data and must be accounted for in statistical analyses (Legendre 1993). The presence of spatial autocorrelation can also be informative biologically for instance indicating the effects of species group ing behav ior, dispersal limitations, and other factors Ecological studies have historically also had a tendency to focus on pairwise interactions such as competition and pre dation, an approach that may be appropriate

PAGE 16

16 for low diversity systems (McGill et a l. 2006) but not sufficient for species rich communities To understand processes regulating populations in diverse communities, an approach that integrates the traits of species and how they relate to environmental heterogeneity may be preferable (Whitta ker 1952, Scott et al. 2002, McGill et al. 2006). Since species in a community that share a common trait may function similarly in responding to heterogeneity in ecological conditions analyses that incorporate species traits may identify processes that c ollectively regulate multiple species and may thus improve our ability to arrive at general principles for communities. General f unctional traits, such as metabolic rate and body mass, are of particular interest because they are traits that strongly affec t the performance of an organism (McGill et al. 2006). Study Area The Serengeti Mara ecosystem is renowned for human wildlife interactions, habitat heterogeneity, and a diverse species assemblage. The ecosystem is a network of protected areas in Tanzania and Kenya (~25,000 km 2 ), with Serengeti National Park (~15,000 km 2 ) at the core (Sinclair et al. 2008). The national park was first established in 1951; its boundaries were realigned over time and other areas were later given protection status and added t o the system. The park is almost entirely enclosed by conservation areas and game reserves that serve as a buffer from the effects of a rapidly growing human population in the surrounding region (Sinclair et al. 2008). The heterogeneity of the habitat is characterized by distinct wet (Dec Apr) and dry (May Nov) seasons and by an increasing rainfall gradient from the southeast (500 mm) to the northwest (1,200 mm) (Norton Griffiths et al. 1975). The vegetation shifts from shortgrass treeless plains in the southeastern area to taller grass and woodlands in the northern and western areas (Sinclair et al. 2007). In addition, fires (human caused)

PAGE 17

17 occur in the system in the dry season (Sinclair et al. 2007). The unique, largely intact diverse assemblage of lar ge herbivores inhabiting this system consists of twenty eight prey species that range in size from 5 kg (dikdik) to 3000 kg (elephant Loxodonta africana ) (Sinclair et al. 2003) and vary in migratory behavior (migratory, resident), feeding guild (grazer, b rowser, mixed feeder), and digestive strategy (ruminant, non ruminant) (Mduma and Hopcraft 2008, Sinclair et al. 2008). The animal census data provided for this study cover the park and some neighboring areas and spans an 18 year time span (1988 2006; thi s includes eight wet season surveys and one dry season survey). Outline Th e research reported in this dissertation investigates the large herbivore community of the Serengeti Mara The overall objective is to advance the understanding of how human activit ies, spatial processes, and species traits collectively influence at a landscape level, the abundance, occupancy, and interspecific associations of species. Chapter 2 assesses the influences of habitat characteristics (both natural and anthropogenic) on t he spatial distribution of a suite of aggregate community measures. Diversity metrics such as the number of species present (richness) and their relative abundances (evenness) (Magurran 2004), are often used to identify the relative conservation values of areas (Lindenmayer and Hunter 2010). Beyond diversity metrics, aggregate community properties such as total abundance, biomass, and energy use can be used to indicate resource availability to foraging guilds and total rates of community consumption (Rowe et al. 2011). In addition, species richness may not always be well correlated with other community attributes such as total abundance

PAGE 18

18 (Bock et al. 2007); therefore, when used alone, richness provides an incomplete picture of how a community responds to th e environment. This chapter explores the spatial distribution of the large herbivore community over time using species richness, total abundance, total biomass, and cumulative basal metabolic rate Moreover, correlations between these community measures are examined. The influences of seventeen habitat characteristics representing human activity, topography, and resources on the spatial distribution of each of the four community measures are assessed using spatial regression models. Chapter 3 relates dis tribution patterns of species over time to species traits. The distribution of species can be characterized both by the sites occupied by species and by the abundance of those species in occupied sites. Occupancy is generally positively correlated with abundance (Gaston et al. 2000); that is, widespread species are expected to be more abundant and vice versa. Combinations in levels of occupancy and abundance can be used to characterize species, for instance in depicting forms of rarity of species vulne rable to extinction (e.g., regionally and locally rare, regionally rare but locally abundant, and regionally common but locally rare) (Rabinowitz 1981, Collins et al. 2009). This chapter explores the occupancy and average abundance (of occupied sites) of each species over time (within the same season and seasonally). The abundance distribution patterns of each species over time are further investigated using rank occupancy abundance profile (ROAP) analyses (Collins et al. 2009). In addition, the clusteri spatial autocorrelation). Throughout this study, relationships between species body

PAGE 19

19 size, feeding guild, and behavior on spatial patterns of occupancy and abundance are investigate d. Chapter 4 characterizes patterns in community structure over time and patterns in interspecific asso ciations for each species Communities can vary in their structure due to differences in the richness and evenness of their species (Magurran 2004). Furthermore, the species in a community may have different roles in the system ( some species may be typically numerically dominant, for instance). Among species, positive or negative associations could arise. Negative associations may be due to species u tilizing different habitats (e.g., plains vs. woodland habitats) or from species interacting negatively due to competition for resources or apparent competition (Holt 1977). Note that, because species overlap in their diet, the total availability of high quality food for a species may to some degree depend on the rest of the community (i.e., diffuse competition) (MacArthur 1972). Positive associations could also be present from species being mutualists or indirect mutualists. This chapter first summarize s the overall community structure using rank abundance plots and compares patterns in species relative abundances over time. Next, the interspecific associations are in abundance to the aggregate community abun dance. Together, these chapters explore how anthropogenic activities, spatial processes, and species traits jointly influence community habitat use, species abundance distributions, and interspecific associations among species over time. The collective re sults and implications of these chapters are summarized in Chapter 5.

PAGE 20

20 CHAPTER 2 ON THE RELATIONSHIP OF A COMMUNITY OF LA RGE HERBIVORES TO ENVIRONMENTAL AND AN THROPOGENIC INFLUENC ES IN THE SERENGETI ECOSYSTEM: A COMPAR ISON OF FOUR COMMUNI TY MEASURES Afri can s avanna ecosystems are characterized by diverse assemblages of large, mammalian herbivores (Shorrocks 2007) Conserving species diversity in these systems is vital, because the range of body sizes and foraging strategies of ungulates shape s the struct ure and function of savannas, due to the effects of ungulates on vegetation diversity and nutrient cycling (du Toit and Cumming 1999). As t he se systems have important ecological as well as economic (Nelson and Agrawal 2008) value it is fortunate that th e number of protected areas designated to conserve African savannas has increased over time. Since 1970, the area of habitat dedicated to protected areas in Africa has nearly doubled to ~3 million km 2 with more than 1,100 national parks and reserves desi gnated in sub Saharan Africa (Newmark 2008). However, e ffects of anthropogenic activities such as agricultural development, hunt ing, and disease transfer from domesticated animals have also intensified and are increasingly threatening wildlife populations (Newmark 2008, Wittemyer et al. 2008) On a continent with pressing needs to accommodate a booming human population and with severely limited financial resources ensuring the long term effectiveness of protected areas through wise management is increasi ng ly important T o conserve biodiversity, processes underlying community responses to changes in the environment and in management practices must be understood, requiring species diversity to be monitored at appropriate spatial and temporal scale s (Cromsig t et al. 2009 a ). P rimary factors influencing ungulate habitat selection include nutrient or energy maximization and predation risk minimization (Wilmshurst et al. 2000, Sinclair et al.

PAGE 21

21 2003, Fryxell et al. 2004). Nutrient and energy maximization of herbi vores indirectly re flects the nutrients and water available to forage plants Plant quantity and quality can vary in opposing ways. Soil nutrients increase the nutrient content and productivity (biomass per unit time) of plants whereas moisture increase s productivity but reduces plant per unit biomass nutrient content. On a global scale, areas with high soil nutrients and intermediate moisture seem to have both sufficient plant biomass to sustain larger herbivores and plant quality high enough for utili zation by smaller herbivores ; such areas may contain more diversity because they can support a range of herbivore body sizes (Olff et al. 2002) In the Serengeti, hotspots of grazer richness tend to occur in areas that are above a rainfall threshold of 65 0 mm per year, relatively flat, farther from rivers, and have relatively low standing plant biomass (Anderson et al. 2010), presumably reflecting higher forage quality and /or lower predation risk. U ngulate abundances and spatial distributions have been inf luenced historically and at present by local anthropogenic and climatic perturbations (e.g., rinderpest, Sinclair et al. 2007) Regions surrounding the park have experienc ed a rapidly growing human population, particularly west of the park where higher ra infall is favorable for agricultur e (Sinclair et al. 2008 ). Conversion of land to cultivation has progressively disrupted the wildebeest ( Connochaetes taurinus ) m igration route and eliminat ed dry season refuges that historically have been outside the prot ected areas F luctuating poaching activities close to human settlements have resulted in the near extinction of black rhinoceros ( Diceros bicornis ) and have had transient but substantial effects on populations of African elephant ( Loxodonta africana ) and buff alo ( Syncerus caffer) ( Metzger et al. 2007, Sinclair et al. 2007) E xtreme weather events have had dramatic

PAGE 22

22 effec ts. For instance, a drought in 1984 caused a substantial decline in buffalo ; also, a drought in 1993 caused a 40% decline in wildebeest a nd a 70% decline in buffalo due to starvation (Sinclair et al. 2007). To identify the relative conservation values of specific locations, diversity metrics are often used (Lindenmayer and Hunter 2010) such as the number of species present (richness) and th eir relative abundances (evenness) (Magurran 2004). Going beyond diversity, aggregate community properties such as total abundance, biomass, and energy use can measure total rates of community consumption and indicate resource requirements o f foraging gui lds (Rowe et al. 2011). Such measures have not typically been employed in conservation, but may provide insights beyond traditional measures such as species richness. These aggregate measures can indicate the carrying capacity of a system for a given spe cies assembly (Fritz and Duncan 1994) and help monitor community responses to broad environmental changes. Energy availability is believed to be a main determinant of species richness, abundance, and biomass (Evans et al. 2005); therefore, positive relati onships might be hypothesized to exist among these measures. However, species richness may not always be well correlated with other community attributes such as total abundance (Bock et al. 2007); therefore, when used alone, richness provides an incomplet e picture of how a community responds to the environment. A decline in total abundance may not lead to an immediate reduction in species richness, but still provide a warning signal of detrimental environmental changes. In this study, a rich historic data set from the Serengeti National Park (SNP) is used to understand how community metrics (species richness, total abundance, total

PAGE 23

23 biomass, and total basal metabolic rate) for twelve large herbivore species are distributed across space in relation to habitat characteristics and anthropogenic factors. We assess trends in these relationships over an 18 year period and examine cross correlations of these aggregate community measures The dataset from the aerial surveys does not include wildebeest. Migratory w ildebeest are a well known and dominant species in this system ( Sinclair 2003, Sinclair et al. 2007) but because they are so abundant, their properties would tend to overwhelm any analysis of the community taken as a whole (Appendix A 1). We therefore ha ve deliberately put aside the wildebeest in our assessment of patterns in community metrics. This study addresses the hypothesis that although soil fertility, moisture, and predation risk should influence the spatial distribution of the community human ac tivity and extreme weather events will also have pronounced effects. Identifying the areas of the park and associated habitat characteristics that support the greatest diversity and abundance of large herbivores sharpens expectations about the kinds of ch anges that could affect this community M ethods Study Area The Serengeti Mara ecosystem is a long established network of protected area s straddling the border of Tanzania and Kenya in East Africa It spans approximately 25,000 km 2 most of which is SNP ( a pproximately 15,000 km 2 ), ( Sinclair et al. 2008 ) (Figure 2 1) The park and surrounding buffer areas do not permit livestock or agriculture; however some buffer areas do allow licensed hunting (in game reserves) and controlled pastoralism (i.e., ranc hing) (in conservation areas).

PAGE 24

24 The spatially and temporally heterogeneous habitat of the Serengeti Mara is characterized by an annual cycle of a wet season ( March May ) followed by a dry ( August October) season. Rainfall increases along a gradient from southeas t (500 mm /yr ) to northwest (1 200 mm /yr ) (Norton Griffiths et al. 1975, Sinclair et al. 2008). The vegetation transitions from treeless, short grass plains in the southeast to tall grass savannas and woodlands in the north and west (Sinclair et al. 2007). Dry season f ires generally human caused, are a vital factor in the system (Sinclair et al. 2007). The park contains a largely intact community of twenty eight large herbivore species and ten carnivorous large predator s pecies ( Mduma and Hopcraft 2008 Si nclair et al. 2008 ). Wildebeest are a dominant species and vital for maintaining the ecosystem in its current state ( Sinclair 2003, Sinclair et al. 2007, Holdo et al. 2011) Their annual migration (between the southern grasslands in the wet season and th e northern woodlands and savannas in the dry season) is driven by the seasonal rainfall gradient which governs vegetatio n growth and availability (Pennycuick 1975 Boone et al. 2006 Holdo et al. 2009) ( Appendix A 2 ). The broad question we address is how patterns in the species richness and abundance of the other complementary species in the system reflect major environmental gradients and anthropogenic influences. Census Data The census data consists of nine annual surveys from 1988 2006 (eight wet seaso n s and one dry season) (Appendix A 3 Table A 5 ) Data were collected by the Tanzania Wildlife Research Institute using Systematic Reconnaissance Flights (SRF) to estimate wildlife densities across a survey grid with a cell size of 5 x 5 km (Figure 2 1) f rom flights along east west transects across the middle of each grid cell. Herbivores were counted in subunits, defined as 30 seconds of flying time (approximately 2 km)

PAGE 25

25 with a strip width of 150 170 m on either side of the aircraft (thus, animals were co unted in approximately 6 7% of the survey grid cell). These data were used to calculate a density (number per km 2 ) for each 25 km 2 survey grid cell. Survey data were recorded using a Universal Transverse Mercator (UTM) coordinate system (easting and nort hing coordinate pair), with each location at the center of a grid cell (see Campbell and Borner (1995) for detailed methods). The annual surveys resulted in 730 sample location s ( covering the majority of the park and some neighboring areas) consistently sa mpled across surveys. Based on perceived reliability of the survey data (detectability) (Campbell and Borner 19 95, Mduma and Hopcraft 2008), twelve herbivore species (out of twenty eight observed) were selected for analys i s ( Table 2 1, Appendix A 1 Table A 1 and A 2 ) Migratory wildebeest were not counted during these aerial surveys. Future studies I conduct will incorporate this species, using other data sources. F our community measures were determined (number per km 2 ) for each grid cell : species rich ness, total abundance, total biomass, and total basal metabolic rate (BMR) (Appendix A 4) Species richness was de fined as the number of species observed in a grid cell; total abundance was found by summing the number of individuals across the twelve spec ies. T otal biomass (kg) was calculated by multiplying abundance by its mass derived from the literature ( Table 2 1), and summing values across species. Basal metabolic rate (measured in watts) was calculated for each species using an allom etric equation for ungulates: R (W) = 3.392 M (kg) 0.75 where R is basal metabolic rate and M is body mass (Coe et al. 1976, also see Savage et al.

PAGE 26

26 sum across species in a cell. This sum estimates total metabolic demand of this assemblage of large herbivores. Habitat C haracteristics Habitat characteristics were selected represent ing major environmental and human influences in the system based on prior studies (Olff et a l. 2002, Metzger et al. 2007, Holdo et al. 2009, Anderson et al. 2010) The 17 habitat characteristics (Table 2 2) include both natural and anthropogenic factors Data for each habitat characteristic were resolved to the same spatial scale as the survey data. In addition to habitat variables (see Appendix A 3 and Appendix A 5 for a detailed description), year and spatial coordinates (eastings and northings) were included in the set of potential independent variables. Statistical Analyses Community correl ations Correlations between each pair of aggregate community measures (i.e., species richness, total abundance, total biomass, and total BMR) within each year, and correlations between each pair of years for each measure, were determined at the grid cell l evel using Pearson correlation coefficients. Collinearity among habitat variables Collinearity among the independent variables (habitat variables, spatial coordinates, and date) was examined by calculating the variance inflation factor (VIF) using the car package (Fox et al. 2009) in R v2.10.1 (R Development Core Team 2009). A threshold of 3 was used for inclusion in the model (10 is commonly used, however, a lower threshold provides a more rigorous approach for weak ecological signals) (Zuur et al. 2010). The VIF was calculated for all covariates; if any VIFs were above the

PAGE 27

27 threshold, the covariate with highest VIF was excluded from the model. This process was repeated until all covariates had a VIF below the threshold. This analysis was completed using all eight wet season surveys combined into a single dataset with year ignored. Based on the VIF calculations, the parameters were not included in the analysis Distance to permanent river and plant N were stron gly correlated with each other (0.91) and with northings ( 0.92). They were also correlated with a variable included in the analysis, plant P (0.61 and 0.77, respectively). Spatial regression model selection For each of the four community metrics, s patial regression models were implemented t o determine the effects of habitat characteristics on the community Spatial correlation was examined in two ways First, spatial coordinates were incorporated as predictors to determine if there are broad scale spati al patterns in the species data not due to the measured habitat variables. Second, spatially correlated errors were implemented to capture fine scale autocorrelation due to factors such as species group behavior The first modeling approach used linear re gression (LR) to determine if incorporating spatial coordinates as predictors improved the model (i.e., trend surface analysis), following Lichstein et al. (2002). Three models were compared which used the habitat variables identified from the above VIF a nalysis and either (a) did not incorporate spatial coordinates, or incorporated spatial coordinates as a (b) first order relationship, or (c) second order relationship (Legendre and Legendre 1998). The spatial coordinates were centered to have a mean of 0 (corresponding to 696787.7 and

PAGE 28

28 9723808 in eastings and northings) to reduce the magnitude of values used in calculations. Model s were compared (Crawley 2007, Zuur et al. 2009), a measure of how much information is explained by A to indicate models that are essentially equivalent in performance ( following the recommendation of Burnham and Anderson (1998) and Bolke r (2008)) A second modeling approach incorporated spatially correlated errors with generalized least squares (GLS) models using the nlme package (Pinheiro et al. 2009) in R v2.10.1 (R Development Core Team 2009); this method allow s errors to be correlated and have unequal variances (Crawley 2007, Zuur et al. 2009). The GLS models used the habitat variables identified from the above VIF analysis and include spatial coordinates as predictors (none, linear, or quadratic) as determined by the best model of th e LR analysis above The five candidate GLS models differ ed in the implemented spatial correlation error stru cture (exponential, Gaussian, linear, rational quadratic or spherical ) Model selection for error structure for the GLS models was conducted usi ng AIC, as above. Then, all eight models (3 LR and 5 GLS) were compared using AIC for each community measure. The analysis used an interaction between each of the dynamic variables ( N ormalized Difference Vegetation Index ( NDVI ) heterogeneity of NDVI rai nfall, and percent burn) and year (categorical). The analysis was initially limited to the three years for which data for percent burn and heterogeneity of NDVI were available. These two habitat variables proved not to be significant ( Appendix A 6 ). The refore, the final

PAGE 29

29 analys e s us ed data from all eight wet seasons, with these two habitat variables excluded as predictors. R esults Community Correlation Patterns For each survey (i.e., within year), three aggregate measures ( total abundance, total biomass, and total BMR ) were strongly correlated (Figure 2 2; Appendix A 7, Figure A 1 3 ). In particular, total biomass and total basal metabolic rate were highly correlated. However, the values were not highly correlated between pairs of years for any of the four measures with two exceptions: the 2000 wet season survey and the 1996 dry season survey were highly correlated for each of the four aggregate community measures (e.g., biomass in 2000 is correlated with biomass i n 1996) (Appendix A 7, Table A 11 ). G eogr aphic proximity (e.g., due to habitat features, grouping behavior) could potentially be driving some of the correlations observed, therefore these results are descriptive and we do not make any statistical inferences. Community Distribution P atterns Specie s richness during the wet season appears to be consistently higher in two regions of the park: across a mid latitude band and in the southern plains along the southeastern boundary of the park (note, migratory wildebeest are also largely present in the la tter area, and so adding them to the dataset would not markedly alter these spatial patterns) (Figure 2 3; Appendix A 2, Appendix A 4, Figure A 2 a) Richness is lower in the center of the park and in game reserves Richness is often noticeably lower in cells near the SNP boundaries, particularly in the northern third of the park, and along the southwestern boundary. Lower richness was observed in the southeast in 1988 and 2000. The spatial pattern of richness in 2000 is unusual, compared to other wet s eason

PAGE 30

30 years, in that diverse assemblages were more common and distributed more widely spatially; this wet season distribution is more similar to the 1996 dry season survey (Appendix A 4, Figure A 2 a and Figure A 3 a). R ichness also shows considerable var iability at a local scale among surveys. Total abundance is prominently higher primarily in the southern plains along the southeastern boundary (migratory wildebeest are also largely present here) (Figure 2 3; Appendix A 2; Appendix A 4, Figure A 2 b ) Th e mid latitude band noted for high species richness is not apparent in total abundance. As observed with species richness, total abundance values in the southeast were not as high in 1988 and 2000, and total abundance is distributed more widely spatially in 2000 resembling the 1996 dry season survey (Appendix A 4, Figure A 2 b and Figure A 3 b). Total biomass and total metabolic rate show patterns similar to total abundance (Figure 2 3; Appendix A 4, Figure A 2 c d and Figure A 3 c d ) These two measures show lower values in the southeast in another year, 1991, and have higher values adjacent to the southern tip of the park, in the Maswa game reserve, in 2006. Spatial Regression Model Selection Of the three candidate linear regression models (independent error models), the model including spatial coordinates as a second order relationship performed best for each of the four aggregate community measures (Table 2 3). For species richness and total abundance, the linear regression model with spatial coordina tes as a second order relationship performed best ( AIC of the next best model was > 10 ). For total biomass and total BMR, all three linear regression models were within the threshold for best approximating model; however, the linear regression model with spatial coordinates as

PAGE 31

31 a second order relationship performed best (Table 2 3). The GLS models were therefore implemented with spatial coordinates as a second order relationship. When comparing the five candidate GLS models (which differ in the spatial co rrelation structure they incorporate) with the linear regression models, only the GLS models were within the threshold for best approximating models (Table 2 3). Of the GLS models, the best model correlation structure varied depending on the response vari able: species richness (exponential), total abundance (rational quadratic), total biomass (rational quadratic), and total BMR (exponential) (Table 2 3). The performance of the various correlation structures was at times not distinguishable for a particul ar community measure. Model Results The GLS model applied to the dataset consisting of eight wet seasons resulted in several significant habitat effects on species richness (Table 2 4). Species richness decreases with increasing distance west of the weste rn park boundary (p < 0.001). There is also a positive effect of road density (p < 0.01) and a negative effect of average elevation (p < 0.001). Of the resource variables, there is a positive effect of plant P (p < 0.05) and a negative effect of NDVI but the magnitude of the slope depends on year. There also appears to be a difference in the intercept for the year 1996 (p < 0.05); there are some positive spatial trend effects (eastings x northings (p < 0.001) and eastings 2 (p < 0.05)). Many of these eff ects are not so clear when looking at univariate relationships, as there is much scatter in the data. Total abundance, total biomass and total BMR showed similar patterns to each other (Table 2 5, 2 6, 2 7). For these three measures, there was a negative effect with increasing distance west of the western park boundary (abundance p < 0.001; biomass

PAGE 32

32 p < 0.001; BMR p < 0.01). In addition, there were positive effects of plant P (abundance p < 0.1; biomass p < 0.05; BMR p < 0.05), plant Na (abundance p < 0.05 ; biomass p < 0.05; BMR p < 0.05), and plant Ca (abundance p < 0.001; biomass p < 0.05; BMR p < 0.01). NDVI had a negative slope for years 1989 and 2006 for total abundance (p < 0.1 and p < 0.05, respectively); no significant effects were observed for the other two response variables. Also, rain had a negative effect for year 1996 for total abundance (p < 0.01) whereas it had a positive effect for years 1991, 2003, and 2006 for total biomass (p < 0.1, p < 0.1, and p < 0.05, respectively), and for years 19 89, 1991, and 2006 for total BMR (p < 0.05, p < 0.05, and p < 0.01, respectively). There also appears to be a difference in the intercept for the year 1996 (relative to 1988) for all three of these response variables (abundance p < 0.001 ; biomass p < 0.1 ; BMR p < 0.05) and also for the years 2000, 2003, and 2006 (again relative to 1988) for biomass (p < 0.1, p < 0.05, and p < 0.05, respectively) and BMR (p < 0.1, p < 0.1, and p < 0.1, respectively). These three response variables also showed some spatial trends: a negative effect of eastings for total abundance (p < 0.01) and total BMR (p < 0.1), a positive effect of northings for total abundance (p < 0.1), a positive effect of northings 2 for total biomass (p < 0.05) and total BMR (p < 0.1), and a positi ve effect of eastings northings for total abundance, total biomass and total BMR (abundance p < 0.01; biomass p < 0.05; BMR p < 0.05). D iscussion Community Correlation Patterns Species richness was initially expected to have a positive relationship with total abundance, as such a positive relationship in general is observed in other studies (e.g., Mittelbach 2001, Martinko et al. 2006, Bock et al. 2007). However, we found that

PAGE 33

33 herbivore richness is not positively correlated with total abundance. This is a surprising result and we are not aware of other studies reporting this clear lack of a relationship between overall abundance of a taxon and species richness. Indeed, the patterns in Figure 2 2 and Appendix A 7, Figure A 1 3 if anything suggest a hump s haped relationship between maximal richness and total abundance ( we caution that this may be influenced by the large number of zeros in the data); in most censuses, the maximal richness occurs at relatively low total abundance values. We note that a long a mid latitude band of the park species richness was high whereas total abundance was not ; the community distribution in this area may in particular be driving the overall relationship between these measures One can speculate that this pattern could aris e because of strong interspecific interactions. The presence of a numerically dominant species that interferes with other species may result in the lack of correlation. At high abundances of the dominant species (e.g., due to herding), not as many other species would be present; at low abundances or absence of the dominant species, more species are present. Note that the dataset does not include wildebeest. Survey grid locations where wildebeest are highly distributed in the southeast have lower species richness and total abundance; adding wildebeest to the analysis is therefore expected to make the overall relationship between total herbivore abundance and species richness even more negative. A high positive relationship between total abundance and tota l biomass (as observed in this study) suggests that as total abundance increases, either the increase in abundances of species across body sizes is fairly even or there is an increase in abundances of larger species. With an even increase in abundance, a high positive

PAGE 34

34 correlation between total biomass and total BMR is expected to be observed as well. Chapter 4 examines in more detail the species specific patterns underlying these aggregate community results. As these three aggregate community measures we re highly correlated in our dataset, the discussion hereafter refers primarily to total abundance. Community Distribution Patterns The specific regions of the park, and locations outside the park boundar y, that support higher richness and total abundance o f species potentially have more interactions and greater strength of interactions among species. These are areas of potentially greatest importance in management. The habitat characteristics identified by our analyses that influence species richness and total abundance are as follows. H uman activity Large herbivore species richness and abundance appear to be showing discernible negative effects from human activity, as seen by the effect of distance from and into unprotected areas and game reserves (the la tter is intended to serve as a buffer zone for the park). The unprotected areas, with their rapid human population growth, will not be available to wildlife in the future Without proper management, herbivores will lack viable refuges to the west, thus i ncreasing their overall vulnerability even within the park. Compared to the park area, impala ( Aepyceros melampus ) in neighboring partially protected areas not only have lower density, but also a sex ratio skewed towards females and more alert and flighty behavior, likely due to illegal hunting as well as unregulated legal hunting (Setsaas et al. 2007). Our results suggest comparable patterns should be expected for other species, but hunting may not be responsible for the results we found. Even though th e nearby human population has

PAGE 35

35 increased over the study period (1988 2006), poaching has declined from the era of 1977 1986 (when it was commonplace) to the present (Hilborn et al. 2006, Metzger et al. 2010). If poaching remains at low levels, t he park are a may be able to maintain the herbivore populations for the long term. The lower richness observed on the western side of the park may more subtly relate to a mid domain effect, where the existence of an edge itself can lead to lower richness (Colwell and Lees 2000, Colwell 2011), e.g., because opportunity for migration to the west is precluded even without deterministic processes such as poaching acting to depress abundance and richness near the edge. Unexpectedly, we found a positive effect of road densi ty on species richness. The correlation may not imply causation, of course: r oads inside the park may have been built where animal sightings would be more likely. In addition, human presence on roads may scare off predators or provide a warning that pre dators are near. Alternatively, locally cleared space could increase visibility, permitting herbivores to detect predators more effectively. There are some areas with high road density (e.g., Seronera in the central area of the park and the Maswa game res erve) that do appear to have lower s pecies richness, as initially expected. However, because there was not a negative effect of roads on total abundance or species richness, herbivores do not appear to be avoiding roads overall. The herbivore community might in the future be quite vulnerable to increases in poaching, if poachers enter the park via roads. We caution that a quite different effect would likely emerge, were the planned addition of a major new road through the northern Serengeti (linking the Lake Victoria area and eastern Tanzania) completed. The proposed road will likely disrupt movement, in particular preventing

PAGE 36

36 wildebeest migration (Dobson et al. 2010, Holdo et al. 2011), and will surely result in roadkills due to high traffic volume. Th e current roads have a low traffic density. Topography The increase in species richness in areas of lower elevation (which are largely towards the west) may be due to plant N rather than to a direct effect of elevation (elevation and plant N were correlate d, thus the latter was not included in the analysis); see the discussion of nutrients below. Although herbivores were expected to prefer flat areas due to lower energy demands for movement, heterogeneity of elevation did not influence species richness or total abundance. Our results contrast with those of Anderson et al. (2010), who did observe the expected relationship between species richness and elevation heterogeneity in the Serengeti. They suggested that the underlying cause was that animals are mor e susceptible to predation in topographically complex areas because lions ( Panthera leo ) use flat areas less often (Hopcraft et al. 2005). Our two datasets, however, differ quite considerably: our study includes four large herbivore species (elephant, gi raffe ( Giraffa camelopardalis ), buffalo, and eland ( Taurotragus oryx )) and impala, in addition to the seven species in the study by Anderson et al. (2010). Larger species are much less susceptible to predation (Sinclair et al. 2003), possibly explaining t he difference between our results and those of Anderson et al (2010) Resources The lack of effect of distance to river on species richness or total abundance is reasonable, as water sources are not as crucial to herbivore well being during the wet season (when the surveys were conducted). Anderson et al. (2010) observed that herbivore hotspots occur away from rivers, and argued this was due to lower risk of

PAGE 37

37 predation by lions (Hopcraft et al. 2005). By contrast, w e found no significant effects of distan ce to river on species richness or total abundance in this study. Again, this may reflect body size; some larger species that are not as susceptible to predation are located closer to rivers whereas smaller species are more distant from rivers (Hopcraft e t al. 2012). In addition, note that another measure of rivers, distance to permanent river, was not included in the analysis due to its strong positive correlation with plant P, which did have a significant positive effect on species richness (see discuss ion of effects of nutrients below); thus there may in fact be higher species richness farther from a permanent river, but the reason may have to do with nutrient supply rather than water availability Large herbivores respond to the patchy distribution of nutrient and vegetation resources. On a global scale, large mammalian herbivore diversity is higher in locations with high nutrients and intermediate moisture, because larger herbivore species accept lower plant nutrient content than do smaller species bu t also require greater plant abundance (Olff et al. 2002). Our results are partially consistent with this broad trend. There was a positive effect of plant P on species richness; however, note that plant P was strongly correlated both with distance to pe rmanent river and to plant N which thus were not included in the analysis. It is difficult to discuss which of these variables might actually matter. The importance of plant nutrients was noted by Anderson et al. (2010) (though the nutrients in that anal ysis found to predict species hotspots (high leaf concentrations of N, Na, and Mg) differed from those in our study). However, in contrast to the results of Olff et al. (2002), in our study NDVI had a negative effect on species richness (in 1989, 1991, 19 96, and 2006). The discrepancy may be because

PAGE 38

38 the majority of species in our study are grazers or smaller species for whom higher plant productivity (or leaf area index) is not as favorable. Likewise, Anderson et al. (2010) observed species hotspots in a reas with low standing biomass and concluded that these areas offer higher quality vegetation and less predation risk. On a regional and continental scale, herbivore density correlates positively with primary productivity (Petorelli et al. 2009) and herbi vore biomass correlates positively with rainfall and soil nutrients (Fritz and Duncan 1994). In this study, n utrients (plant P, Na, and Ca) had a positive effect on total abundance. However, in contrast to other studies NDVI had a negative effect on tota l abundance (in 1989 and 2006) and t he effects of rainfall were more varied and showed no clear pattern. This suggests that there may not be a straightforward relationship between richness and productivity. The result that species richness and total abund ance were not affected by fires from the prior year agrees with Anderson et al. (2010), who concluded that hotspots of grazing ungulates in SNP are not related to fire. A study of ungulates in Benoue National Park, Cameroon, Central Africa found that spec ies richness was not different on burned and unburned sites; however total species density was higher on burned sites due to vegetation regrowth (Klop and van Goethem 2008) In contrast, in a regional study of West Africa, fires, due to their effects on g rass quality and structure, were more important for species richness of grazers than climate or soil fertility (Klop and Prins 2008). While fire does not appear to affect habitat use in the Serengeti, there is thus currently no general consensus on the ef fects of fire on ungulate habitat use in savannas taken as a whole. These differences may be due to effects of temporal and spatial scales of study and the composition of the ungulate assembly. Moreover, the

PAGE 39

39 burn regime (the frequency, intensity, and area burned) and time elapsed since fire are expected to affect the productivity, quality, structure, and heterogeneity of vegetation and these factors may vary in a complex manner among sites (Anderson et al. 2007, Hassan et al. 2008). Although fire is an i mportant management tool in our system, it does not appear to have large scale influences, at least as assessed by the period of our dataset, on the aggregated distribution of these herbivores. There may however be a pronounced effect on a local scale or on particular species; unraveling such effects would require a different kind of analysis than presented here. There was not a consistent effect of floods and droughts on herbivore distributions (floods occurred in 1997 1998 and 2001 2003; droughts occurre d in 1993 and 1999 2000 ( Sinclair et al. 2007 Ogutu et al. 2008 )). The flood and drought episodes do not all coincide with the survey years, making it harder for us to discern any effects. However, we hypothesize the strong anomaly observed for the 2000 wet season distribution and its similarities to the 1996 dry season survey may reflect the 2000 drought. Conclu ding Remarks Identifying preferred habitat areas and their characteristics should help us predict how the Serengeti large herbivore community wi ll respond to changes in environmental conditions and management strategies M anaging for different community measures improves our ability to achieve desired outcomes. In this important suite of protected areas, s pecies richness and total abundance are vulnerable to encroachment from growing human populations in surrounding areas, as seen by the effect of distance from and into partially protected and unprotected areas on aggregate community measures Though the buffer areas present may reduce human imp acts within park boundaries,

PAGE 40

40 they do not appear to completely mitigate such effects for herbivores. Management needs to consider the landscape of the ecosystem, take as a whole, and in particular, understanding the mechanisms by which richness and abundan ce are depressed near the park boundary may identify key issues of management concern. Within the park, to our surprise, roads do not appear to be a negative influence on herbivores as commonly believed (Newmark 2008). This may reflect limits on the leve l of road usage by staff and visitors, a factor that could easily change over time if not carefully monitored. Resources explaining community distributions are patchy, dynamic properties of the system (e.g., nutrients and NDVI), emphasizing the importance of maintaining the spatial and temporal heterogeneity of the ecosystem. Fire, although a prominent management tool in this system, does not appear to be affecting community distributions on a landscape level at least at short (within annual cycle) time scales Managing for total abundance or biomass would lead one to emphasize quite different system attributes than managing for species richness. For instance, locations that support higher species richness do not necessarily support higher total abundanc e. These insights may be useful in continuing to effectively maintain a diversity of species across space and over an extended time in this globally important conservation area.

PAGE 41

41 Table 2 1 S pecies names and trait s of the twelve herbivores in the study (ordered by decreasing body mass) Abundance values, total biomass, and total BMR are from the average of the eight wet surveys (n = 730 grid cells in a survey ) between 1988 and 2006. Species n ame a Common n ame a Mass (kg) b Feeding g uild c Ruminant/ n on ruminant Behavior d Abundance Total b iomass (kg) Total BMR (W) Loxodonta africana African elephant 3000 Mixed Non ruminant Resident 4,338 13,012,500 5,963,981 Giraffa camelopardalis Giraffe 800 Browser Ruminant Resident 7,741 6,192,500 3,949,5 66 Syncerus caffer African buffalo 450 Grazer Ruminant Resident 64,094 28,842,188 21,241,291 Taurotragus oryx Eland 400 Mixed* Ruminant Migratory 13,519 5,407,500 4,101,450 Equus burchellii 250 Grazer Non ruminant Migratory 157,009 39,252,344 33,483,931 Kobus defassa Defassa waterbuck 180 Grazer Ruminant Resident 972 174,938 162,002 Alcelaphus buselaphus Kongoni 150 Grazer Ruminant Resident 11,391 1,708,594 1,656,044 Damaliscus korrigum Topi 120 Grazer Ruminant Resident 50,491 6,058,875 6,209,441 Phacochoerus aethiopicus Warthog 60 Grazer Non ruminant Resident 4,322 259,313 316,040 Aepyceros melampus Impala 50 Mixed Ruminant Resident 76,447 3,822,344 4,875,769 Gazella gr anti 50 Mixed Ruminant Migratory 45,663 2,283,125 2,912,346 Gazella thomsoni 20 Mixed Ruminant Migratory 166,191 3,323,813 5,331,321 Total: 602,175 110,338,031 90,203,182 Sources: a. Mduma and Hopcra ft (2008). b. Sinclair et al. (2003). c. Prez Barbera et al. (2001); items marked with from Kingdon (1997). d. Sinclair et al. (2008).

PAGE 42

42 Table 2 2. H abitat characteristics considered in the study to incorporate the ma jor human and environmental influences in the system (see Appendix A 3 and Appendix A 5 for a detailed description). Habitat category Habitat v ariable Units Human activity Distance from western boundary x Direction (east or west) a m Road d ensity km/km 2 Topography Elevation (average) m Elevation (standard deviation) m Resources Distance to river m Distance to permanent river d m Plant nutrients: Ca ppm Plant nutrients: Mg ppm Plant nutrients: N d percent Plant nutrients: Na ppm Plant nutrients: P ppm Tree cover (average) percent Tree cover (standard deviation) percent NDVI (average) b n/a NDVI (standard deviation) b, c n/a Rainfall (average) b mm/month Fire area b, c percent Date Year a n/a Spatial coordinates Eastings d m Northings d m Ea stings northings m 2 Eastings 2 m 2 Northings 2 m 2 a. Categorical variable. b. Dynamic variable used different data values over time. c. Variables that were not used in the full analysis of eight wet season surveys (as data for these variables were only availabl e for the analysis of three wet season surveys). d. Variables that were not used in the full analysis of eight wet season surveys because of strong correlation with other variables.

PAGE 43

43 Table 2 3. Model selection using AIC for models with species richness, to tal abundance, total biomass, and total BMR as the response variable. Candidate models were linear regression (LR) and generalized least squares (GLS) models incorporating spatial correlation. Data from eight wet season surveys were included in this anal ysis. Candidate m odel a Spatial c oordinates Spatial correlation s tructure K b Species r ichness GLS Second order Exponential 45 0.0 GLS Second order Rational quadratic 45 2.2 GLS Second order Spherical 45 26.8 GLS Second order Gaussian 45 28.0 GLS Second order Linear 45 176.4 LR Second order n/a 43 478.1 LR First order n/a 40 580.0 LR No spatial coordinates n/a 38 648.3 Total a bundance GLS Second order Rational quadratic 45 0.0 GLS Second order Exponential 45 10.7 GLS Second order Gaussian 45 28.6 GLS Second order Spherical 45 48.3 GLS Second order Linear 45 109.9 LR Second order n/a 43 470.0 LR First order n/a 40 483.9 LR No spatial coordinates n/a 38 494.3 Total b iomass GLS Second order Rational quadratic 45 0.0 GLS Second order Exponential 45 0.1 GLS Sec ond order Gaussian 45 2.4 GLS Second order Linear 45 18.4 GLS Second order Spherical 45 18.4 LR Second order n/a 43 124.2 LR No spatial coordinates n/a 38 130.0 LR First order n/a 40 131.4 Total BMR GLS Second order Exponential 45 0 GLS Second order Rational quadratic 45 1.8 GLS Second order Gaussian 45 1.9 GLS Second order Spherical 45 3.5 GLS Second order Linear 45 20.1 LR Second order n/a 43 190.8 LR First order n/a 40 194.9 LR No spatial coordinates n/a 38 19 6.7

PAGE 44

44 a. The l i near regression models include spatial coordinates as predictors ( none, linear, or second order ). The g eneralized least squares models implement second order spatial coordinates (as determined from LR model) and a spatial correlation error stru cture (exponential, Gaussian, linear, rational quadratic, or spherical) b. ranked model (lowest AIC value) as a baseline. models.

PAGE 45

45 Table 2 4. Model output for species richness for analysis of eight wet season surveys. The best model determined was a GLS model including second order spatial coordinates as predictors and an exponential spatial correlation structure. p < 0.05 ; p < 0. 0 1 ; ** p < 0. 00 1. Parameter Coefficient SE t p Intercept 3.695 1.213 3.05 < 0.01 ** Human activity Direction w est 0.464 0.084 5.56 < 0.001 *** Distance from west boundary 3.133 E 06 5.214 E 06 0.60 0.55 Direction w est x d istance 0.0001 1.214 E 05 4.78 < 0.001 *** Road density 0.301 0. 101 2.99 < 0.01 ** Topography Elevation a verage 0.002 0.001 3.84 < 0.001 *** Elevation SD 0.001 0.002 0.79 0.43 Resources Distance to river 4.651 E 06 5.299 E 06 0.88 0.38 Plant P 0.0001 0.0001 2.30 0.02 Plant Na 2.900 E 05 2.690 E 05 1.08 0.28 Plant Mg 0.0002 0.0004 0.58 0.56 Plant Ca 0.0001 0.0001 0.93 0.35 Tree c over 0.013 0.010 1.32 0.19 Tree c over SD 0.003 0.017 0.19 0.85 NDVI 0.882 0.566 1.56 0.12 Rain 0.004 0.005 0.82 0.41 Date Date 1989 0.130 1.053 0.12 0.9 0 Date 1991 0.250 0.997 0.25 0.80 Date 1996 4.332 2.188 1.98 0.05 Date 2000 0.092 1.500 0.06 0.95 Date 2001 1.213 1.980 0.61 0.54 Date 2003 0.300 1.209 0.25 0.80 Date 2006 2.208 1.644 1.34 0.18 NDVI x 1989 2.174 0.865 2.51 0.01 NDVI x 1991 2.189 1.036 2.11 0.03 NDVI x 1996 2.305 1.080 2.13 0.03 NDVI x 2000 0.385 0.943 0.41 0.68 NDVI x 2001 0.760 1.069 0.71 0.48 NDVI x 2003 1.231 1.127 1.09 0.27 NDVI x 2006 3.295 1.277 2.58 < 0.01 ** Rain x 1989 0.009 0.007 1.20 0.23 Rain x 1991 0.007 0.008 0.86 0.39 Rain x 1996 0.020 0.016 1.29 0.20 Rain x 2000 0.005 0.025 0.20 0.85 Rain x 2001 0.013 0.021 0.64 0.53 Rain x 2003 0.004 0.016 0.25 0.81 Rain x 2006 0.001 0.016 0.07 0.95 Spat ial coordinates Eastings 2.014 E 06 3.831 E 06 0.53 0.60 Northings 1.217 E 06 1.810 E 06 0.67 0.50 Eastings n orthings 2.626 E 10 4.891 E 11 5.37 < 0.001 *** Eastings 2 8.593 E 11 3.527 E 11 2.44 0.01 Northings 2 1.786 E 11 1.999 E 11 0.89 0 .37

PAGE 46

46 Table 2 5. Model output for total abundance for analysis of eight wet season surveys. The best model determined was a GLS model including second order spatial coordinates as predictors and a rational quadratic spatial correlation structure. Sig p < 0.05 ; p < 0. 0 1 ; ** p < 0. 00 1. Parameter Coefficient SE t p Intercept 2.399 60.018 0.04 0.97 Human activity Direction w est 9.919 4.448 2.23 0.03 Distance from west boundary 0.001 0.0002 3.01 < 0.01 ** Direc tion w est x d istance 0.002 0.0006 3.35 < 0.001 *** Road density 2.795 5.359 0.52 0.60 Topography Elevation a verage 0.021 0.027 0.78 0.44 Elevation SD 0.103 0.083 1.24 0.21 Resources Distance to river 0.0003 0.0003 1.03 0.30 Plant P 0.00 5 0.003 1.80 0.07 Plant Na 0.003 0.001 2.28 0.02 Plant Mg 0.017 0.020 0.87 0.38 Plant Ca 0.021 0.005 4.19 < 0.001 *** Tree c over 0.534 0.508 1.05 0.29 Tree c over SD 0.955 0.924 1.03 0.30 NDVI 9.299 29.269 0.32 0.75 Rain 0.271 0.244 1.11 0.27 Date Date 1989 2.680 49.856 0.05 0.96 Date 1991 37.100 46.928 0.81 0.42 Date 1996 345.713 98.583 3.51 < 0.001 *** Date 2000 1.327 68.418 0.02 0.98 Date 2001 17.243 89.141 0.19 0.85 Date 2003 1.564 55.379 0.03 0.98 Da te 2006 0.449 74.227 0.01 0.99 NDVI x 1989 83.505 44.763 1.87 0.06 NDVI x 1991 23.405 55.263 0.42 0.67 NDVI x 1996 75.377 57.031 1.32 0.19 NDVI x 2000 3.940 48.003 0.08 0.93 NDVI x 2001 3.920 53.853 0.07 0.94 NDVI x 2003 11.374 57.102 0.20 0.84 NDVI x 2006 140.114 65.054 2.15 0.03 Rain x 1989 0.527 0.329 1.60 0.11 Rain x 1991 0.461 0.373 1.23 0.22 Rain x 1996 1.844 0.715 2.58 < 0.01 ** Rain x 2000 0.107 1.119 0.10 0.92 Rain x 2001 0.053 0.940 0.06 0.95 Rain x 2003 0.306 0.734 0.42 0.68 Rain x 2006 1.165 0.734 1.59 0.11 Spatial coordinates Eastings 0.001 0.0002 3.22 < 0.001 Northings 0.0002 0.0001 1.89 0.06 Eastings n orthings 7.041 E 09 2.425 E 09 2.90 < 0.001 Eastings 2 1.359 E 09 1.685 E 09 0.81 0.4200 Northings 2 1.364 E 09 9.392 E 10 1.45 0.1465

PAGE 47

47 Table 2 6. Model output for total biomass for analysis of eight wet season surveys. The best model determined was a GLS model including second order spatial coordinates as predictors and a rational quadratic spatial correlation structure. p < 0.05 ; p < 0. 0 1 ; ** p < 0. 00 1. Parameter Coefficient SE t p Intercept 1149.298 12860.920 0.09 0.93 Human activity Direction w est 1720.002 1050.2 34 1.64 0.10 Distance from west boundary 0.098 0.051 1.92 0.06 D irection w est x d istance 0.469 0.135 3.48 < 0.001 *** Road density 1014.552 1305.555 0.78 0.44 Topography Elevation a verage 4.660 5.891 0.79 0.43 Elevation SD 5.353 20.095 0.27 0.79 Resources Distance to river 0.051 0.060 0.85 0.40 Plant P 1.190 0.555 2.14 0.03 Plant Na 0.614 0.307 2.00 0.05 Plant Mg 1.625 4.177 0.39 0.70 Plant Ca 2.415 1.039 2.32 0.02 Tree c over 51.776 121.889 0.42 0.67 Tree c o ver SD 56.781 226.193 0.25 0.80 NDVI 4630.356 6537.306 0.71 0.48 Rain 64.449 47.749 1.35 0.18 Date Date 1989 2322.940 10351.020 0.22 0.82 Date 1991 11161.300 9702.311 1.15 0.25 Date 1996 34355.470 19541.430 1.76 0.08 Date 2000 23 800.320 13781.450 1.73 0.08 Date 2001 2633.034 17711.330 0.15 0.88 Date 2003 24753.670 11140.110 2.22 0.03 Date 2006 30866.670 14750.650 2.09 0.04 NDVI x 1989 10601.480 9988.222 1.06 0.29 NDVI x 1991 4375.184 12708.940 0.34 0.73 NDVI x 1996 18146.500 13050.980 1.39 0.16 NDVI x 2000 7359.634 10560.880 0.70 0.49 NDVI x 2001 2453.384 11704.970 0.21 0.83 NDVI x 2003 14741.630 12525.680 1.18 0.24 NDVI x 2006 785.999 14285.770 0.06 0.96 Rain x 1989 97.927 65.182 1.5 0 0.13 Rain x 1991 129.758 74.257 1.75 0.08 Rain x 1996 116.323 143.053 0.81 0.42 Rain x 2000 352.828 223.459 1.58 0.11 Rain x 2001 8.871 185.385 0.05 0.96 Rain x 2003 271.412 145.494 1.87 0.06 Rain x 2006 347.357 147.891 2.35 0.02 S patial coordinates Eastings 0.061 0.039 1.55 0.12 Northings 0.007 0.019 0.37 0.71 Eastings n orthings 1.146 E 06 5.266 E 07 2.18 0.03 Eastings 2 2.308 E 07 3.538 E 07 0.65 0.51 Northings 2 4.845 E 07 1.952 E 07 2.48 0.01

PAGE 48

48 Table 2 7. M odel output for total basal metabolic rate (BMR) for analysis of eight wet season surveys. The best model determined was a GLS model including second order spatial coordinates as predictors and an exponential spatial correlation structure. Significance c p < 0.05 ; p < 0. 0 1 ; ** p < 0. 00 1. Parameter Coefficient SE t p Intercept 1460.358 9499.271 0.15 0.88 Human activity Direction w est 1584.091 800.961 1.97 0.05 Distance from west boundary 0.064 0.038 1.69 0.09 D irection w es t x d istance 0.324 0.102 3.19 < 0.01 ** Road density 698.430 990.476 0.71 0.48 Topography Elevation a verage 3.314 4.435 0.75 0.45 Elevation SD 3.808 15.230 0.25 0.80 Resources Distance to river 0.022 0.046 0.48 0.63 Plant P 0.953 0.418 2.28 0.02 Plant Na 0.499 0.235 2.13 0.03 Plant Mg 2.021 3.136 0.64 0.52 Plant Ca 2.292 0.776 2.95 < 0.01 ** Tree c over 35.903 92.444 0.39 0.70 Tree c over SD 112.974 171.258 0.66 0.51 NDVI 4192.519 4831.727 0.87 0.39 Rain 56.8 63 31.857 1.78 0.07 Date Date 1989 2811.853 7256.815 0.39 0.70 Date 1991 8973.332 6817.707 1.32 0.19 Date 1996 30489.680 13316.210 2.29 0.02 Date 2000 17937.570 9567.181 1.87 0.06 Date 2001 2371.143 12111.990 0.20 0.84 Date 2003 13598.740 7657.129 1.78 0.08 Date 2006 19202.180 10062.600 1.91 0.06 NDVI x 1989 10467.380 7353.131 1.42 0.15 NDVI x 1991 4542.516 9622.679 0.47 0.64 NDVI x 1996 13392.480 9788.120 1.37 0.17 NDVI x 2000 7075.129 7738.593 0.91 0.36 NDVI x 2001 1002.697 8552.717 0.12 0.91 NDVI x 2003 9464.711 9210.085 1.03 0.30 NDVI x 2006 8618.160 10530.370 0.82 0.41 Rain x 1989 96.337 44.199 2.18 0.03 Rain x 1991 106.910 50.526 2.12 0.03 Rain x 1996 119.545 98.439 1.21 0.22 Rain x 2000 239.471 154.500 1.55 0.12 Rain x 2001 21.606 125.791 0.17 0.86 Rain x 2003 114.497 99.131 1.16 0.25 Rain x 2006 281.719 102.556 2.75 < 0.01 ** Spatial coordinates Eastings 0.052 0.029 1.79 0.07 Northings 0.010 0.015 0.72 0.47 Eastings n orthings 8.552 E 07 3.9513E 07 2.16 0.03 Eastings 2 1.854 E 07 2.606 E 07 0.71 0.48 Northings 2 2.734 E 07 1.413 E 07 1.94 0.05

PAGE 49

49 Figure 2 1. Serengeti National Park and neighboring conservation areas and game reserves in Tanz ania and Kenya East Africa. The animal census survey grid (730 cells of size 5 x 5 km) is shown CA = conservation area, GR = game reserve, NP = national park, RA = reserve area.

PAGE 50

50 Figure 2 2. C orrelation s among the four community measures for the 1 996 wet season survey (n = 730 grid cells in a survey ). Patterns are similar for the other surveys (Appendix A 7, Figure A 1 3 ).

PAGE 51

51 A ) B ) C ) D ) Figure 2 3. Spatial distribution for each community measure (for sample locations at 5 km intervals): A) spe cies richness, B) total abundance, C) total biomass (kg), D) total basal metabolic rate (W). Values are the grid wise averages of eight wet season surveys (n = 730 grid cells in a survey ) between 1988 and 2006.

PAGE 52

52 CHAPTER 3 O CCUPANCY AND ABUNDANCE PATTERNS IN RELATION TO SPECIES TRAITS FOR LARGE HERBIVORES IN A SAVANNA ECOSYSTEM The distribution of species can be characterized both by the sites occupied by a species and by the abundance in occupied sites. Distribution (i.e., occupancy) is generally positive ly correlated with abundance (Gaston et al. 2000); that is, widespread species are expected to be more abundant and vice versa. Various combinations of levels of occupancy and abundance can be used to characterize species, for instance in depicting forms of rarity for species vulnerable to extinction (e.g., regionally and locally rare, regionally rare but locally abundant, and regionally common but locally rare) (Rabinowitz 1981, Collins et al. 2009). This study investigates the occupancy and abundance p atterns of large mammalian herbivores in the Serengeti Mara ecosystem. The ungulates in the Serengeti ecosystem, which are mostly resident grazer species, exhibit a noticeable patchy distribution across the landscape (Seagle and McNaughton 1992). The dis tribution and abundance of each species is expected to be influenced by the spatial and temporal environmental variability across the landscape a s well as by species traits. This study compares patterns of abundance over time and among species and, in par ticular, addresses 1) whether the occupancy and abundance of each species changes over years (within the same season) and seasonally, within the Serengeti; and, 2) how species -which vary substantially in body size, feeding guild, and behavior -differ in their spatial patt erns of occupancy and abundance. Species preferences for vegetation relate to the influence of body size on foraging. Body size influences the spatial scale of perception and the grain of resource utilized ( Cromsigt et al. 2006) as w ell as the quantity and quality of vegetation needed by foraging ungulates. S maller species forage at a finer scale and require lower

PAGE 53

53 amounts of food, but with their higher per mass metabolic rates they tend to seek higher quality plant types and parts ( Demment and VanSoest 1985, Cromsigt et al. 2006, Hopcraft et al. 2012 ). For instance, smaller ungulates in the Serengeti such as impala ( Aepyceros melampus) a nd gazelle are often associated with short grasses and consume more leafy material, both of which are relatively high in quality relative to other potential food sources (McNaughton 1985, Wilsey 1996). Compared to migratory species, the resident ungulate species, which are often smaller and have more specialized diets, are generally in the tall grass areas, woodlands, and kopjes (large rocky outcrops) of the north and west (Dobson 2009). In addition, s ome areas are recognized as herbivore hotspots areas with mixed herds of resident grazers that are temporally stable (Anderson et al. 2010). On a m ore local level, fires which occur periodically during the dry season create new patches of green vegetation that are more nutritious and palatable and are preferentially consumed by herbivores (Dobson 2009), compared to unburned areas. On the other hand larger species forage at a coarser scale and require higher quantities of food, but due to their lower per mass metabolic rate they are able to use a wider range in quality of resources (Demment and VanSoest 1985, Cromsigt et al. 2006 ). For instance, l arger ungulates in the Serengeti such as zebra ( Equus burchellii) can consume abundant coarser material and often feed on more stemmy grass tissues (NcNaughton 1985, Wilsey 1996). T he distributions of large grazers, such as African buffalo ( Syncerus caffe r) are relatively unconstrained and principally driven by forage abundance (Hopcraft et al. 2012) rather than say predation D ue to their wider food

PAGE 54

54 quality tolerance, larger species are able to use a greater diversity of habitat types (Cromsigt et al. 2009). In addition to body size, other factors such as digestive strategy and migratory behavior should influence resource use. Non ruminants have a wider diet tolerance than ruminants (Cromsigt et al. 2009 b ). Compared to the expected diet based on body size, the diets of non ruminants resemble those of larger ruminant species (Kleynhans et al. 2011). M igratory and larger ungulates in the system are provided resources ultimately dependent on seasonal rainfall which creates large scale spatial and tempor al heterogeneity in vegetation (Dobson 2009). For instance, large herds of wildebeest ( Connochaetes taurinus ) migrate annually in seasonal movements that strongly correlate with rainfall and new vegetation availability (Pennycuick 1975, Musiega and Kazadi 2004, Boone et al. 2006). Nutritious forage that seasonally emerges for a few months in the short grass plains of the southeast attracts the larger migratory species (Dobson 2009). In the Serengeti, traits such as body size and migratory behavior also ha ve a strong influence on herbivore vulnerability to predation. The number of predators upon a given prey species decreases with prey body size (Sinclair 2003). Smaller herbivores such as impala are more vulnerable to predators and experience greater adul t mortality due to predation (Sinclair et al. 2003). Herbivores larger than the threshold body size of approximately 150 kg have few natural predators because predators have difficulty killing them (Sinclair 2003). For instance, the lion ( Panthera leo ) i s the sole predator of buffalo and giraffe ( Giraffa camelopardalis) (450 and 800 kg respectively), and the largest herbivores in the system, rhinoceros, hippopotamus, and elephant ( Loxodonta

PAGE 55

55 africana ) (1200, 2000, and 3000 kg, respectively), rarely experi ence predation (Sinclair 2003). In these larger species, food limitation is the principal cause of mortality (Sinclair 2003). In addition to larger body size, species can also escape predation limitation by movement through migration (Fryxell et al. 200 8 ) or by group formation. Due to their differences in susceptibility to predation, the distributions of larger species are expected to be relatively unimpeded by the presence of predators, whereas smaller species are expected to seek areas with greater visi bility (less vegetation biomass) to detect predators. Smaller species are also expected to avoid areas where predators are commonly found, such as near river confluences, where lions are more abundant as they strive to increase access to prey (Hopcraft et al. 2005). In support of these predictions, smaller grazers are observed to be more abundant in areas with less predation risk, that is in areas with short grasses, less woody cover, and which are farther from rivers (Hopcraft et al. 2012). The largest grazer, African buffalo, is observed to not be influenced in its habitat distribution by predation risk ( Hopcraft et al. 2012). Based on the processes discussed above, we propose the following hypotheses for the abundance and occupancy of the species relat ive to one another and over time: 1) Larger species, due to their requirements for greater vegetation quantity, are expected to be more widespread and less abundant, and to show more variation in occupancy and abundance over time than smaller species, 2) Migratory species, due to their mobility, are expected to be more widespread and more abundant, and to show more variation in occupancy and abundance over time than resident species, 3) Grazers, due to the less patchy distribution of their preferred vege tation, are expected to

PAGE 56

56 be more widespread than browsers, 4) Non ruminants, due to their flexibility for food quality, are expected to be more widespread than ruminants, and 5) Between seasons, higher abundance and lower occupancy are expected during the dry season due to decreased resource availability and larger and mobile species are expected to exhibit greater seasonal differences than smaller and more sedentary species. Methods Study A rea The Serengeti Mara ecosystem is a long established network of p rotected areas across the border of Tanzania and Kenya in East Africa. The system spans approximately 25,000 km 2 the majority of which is Seregneti National Park ( SNP approximately 15,000 km 2 ) (Sinclair et al. 2008) (Figure 3 1). The park was establish ed in 1951 S urrounding buffer areas do not permit livestock or agriculture; however some of the buffer areas allow licensed hunting (in game reserves) and controlled pastoralism (i.e., ranching) (in conservation areas). The spatially and temporally heter ogeneous habitat of the Serengeti Mara is characterized by a core wet (March May) and dry (August October) season and an increasing rainfall gradient from the southeast (500 mm/yr) to the northwest (1,200 mm/yr) (Norton Griffiths et al. 1975, Sinclair et a l. 2008). In relation to annual rainfall, the vegetation transitions from treeless, short grass plains in the southeast to tall grass savannas and woodlands in the north and west (Sinclair et al. 2007). Fires, which occur in the dry season and are usuall y human caused, are a vital factor in the system (Sinclair et al. 2007). The park contains a unique, largely intact community of twenty eight herbivores and ten carnivorous large predators ( Mduma and Hopcraft 2008, Sinclair et al. 2008).

PAGE 57

57 Wildebeest are a dominant species and vital for maintaining the ecosystem in its current state (Sinclair 2003, Sinclair et al. 2007) Their annual migration (between the southern grasslands in the wet season and the northern woodlands and savannas in the dry season) is dr vegetation growth and availability (Pennycuick 1975, Boone et al. 2006, Holdo et al. 2009) (Appendix A 2). Census D ata The census data for animals other than migratory wildebeest consist s of nine annual surveys from 1988 2006 (eight wet season and one dry season) (Appendix A 3 Table A 5 ). Data were collected by the Tanzania Wildlife Research Institute using Systematic Reconnaissance Flights (SRF) to estimate wildlife densities across a survey grid having a cell size of 5 x 5 km (Figure 3 1). A plane was flown along east west transects across the middle of the survey grid cells. Herbivores were counted in subunits, which were defined as 30 seconds of flying time (approximately 2 km) wit h a strip width of 150 170 m on either side of the aircraft (thus, animals were counted in approximately 6 7% of the survey grid cell). These data were used to calculate a density (number per km 2 ) value for each 5 x 5 km survey grid cell. Survey data wer e recorded using a Universal Transverse Mercator (UTM) coordinate system (easting and northing coordinate pair) with each survey location at the center of a s urvey grid cell. See Campbell and Borner (1995) for detailed methods. The annual surveys resulte d in 730 sample locations (covering the majority of the park as well as some neighboring areas) that were consistently sampled in each survey. Based on the perceived reliability of the survey data considering detectability (Appendix A 1), twelve herbivore species (out of twenty eight observed) were selected for the analyses (Table 3 1,

PAGE 58

58 Appendix A 1 Table A 1 and A 2 ) Abundance data used in this study represent abundance for a 1 km 2 area in a 5 x 5 km survey grid cell. The census data for migratory wilde beest consists of three annual wet season surveys (1994, 1998, and 1999). Data were collected by the Tanzania Wildlife Conservation Monitoring Program using Aerial Point Surveys (APS), which were timed to take place when the migratory wildebeest are most evenly distributed on the southern plains. The survey extent was determined by the boundary of the wildebeest distribution. A plane was flown along transects at 5 km spacing and photographs were taken at subunits along the transect. These data were used to calculate a density (number per km 2 ) for each 5 x 5 km 2 survey grid cell. Detailed methods can be obtained elsewhere (Norton Griffiths 1973, Borner et al 1987, Tanzania Wildlife Conservation Monitoring 1994). Wildebeest data were selected for the sa me survey extent (n = 730 grid cells) used for the other species, with zero abundance assigned to any cells for which wildebeest were not surveyed. Occupancy and A bundance For each species, the proportion of cells occupied in a census and the average abund ance of occupied cells were determined. The correlation between occupancy and average abundance across all censuses, ignoring species identity, was characterized with a Pearson correlation coefficient. This analysis was repeated both with and without wil debeest given that their abundances are much larger than those of the other species. Individual correlations between occupancy and average abundance across censuses were also determined for each species. In addition, distributions of species abundances w ere depicted using rank occupancy abundance profiles (ROAP), a technique developed by Collins et al. (2009)

PAGE 59

59 occupancy, maximum local abundance, and distribution of abundances ac ross occupied sites. A cell in the survey area was used as the unit of data. For a given species, a ROAP was created as follows: 1) the cells, including those with zero abundance, were ranked from highest (rank 1) to lowest abundance ; 2) the relative ra nk was determined by dividing the rank by the total number of cells in the survey; this standardizes ranks for comparison among surveys ; 3) relative rank was then plotted on the x axis and abundance was plotted on the y axis. Zero abundance values were no t plotted to improve the readability of the graph s Aggregation To further quantify the pattern of abundance, a negative binomial probability distribution was fit to the abundance distribution of each species for each census. The negative binomial is ofte n used to describe aggregation (overdispersion). The negative binomial parameters (the mean number of counts in a sample) and k (the overdispersion parameter; smaller values of k indicate greater amounts of clustering, aggregation, or heterogeneity) were estimated using the method of moments with = and k = 2 /( s 2 ) (Bolker 20 08). For each species, the deviations from the spec average parameter value were examined in relation to census year, for and for k This analysis was repeated both with and without wildebeest inasmuch as their abundances are much larger than those of the other species. Spatial Autocorrelation autocorrelation is not detected ( i.e., the distribution is random); positive and negative

PAGE 60

60 values up to 1 and 1 indicate positive and negative spatial autocorrelation, respectively (i.e., clustering and spatial over for each survey using the R ape package v2.6 (Paradis et al. 2004). The mean and = 8). To investigate a pattern in spatial aggregation in relation to body size, a logarithmic regression was per formed for the wet season data and also for the dry season survey. Analyses were completed using R v2.10.1 (R Development Core Team 2009). Results Occupancy and Abundance Figure 3 2 shows, for each sample period and for each species, the average abundance per cell within occupied sites. Ignoring species identity and excluding wildebeest, there is still a modest correlation between occupancy and abundance (i.e., species with higher occupancy have higher average local abundance) (r = 0.29, Figure 3 2). Howe ver, with wildebeest included, occupancy and average local abundance is not correlated (r = 0.09) (Figure B 1). For a given species, occupancy and abundance most often showed a positive trend over time (Figure 3 2). Elephant (r = 0.55), kongoni ( Alcel aphus buselaphus ) (r = ( Gazella granti) ( r = 0.43) had a moderate positive correlation whereas eland ( Taurotragus oryx ) (r = 0.37), top i ( Damaliscus korrigum ) ( r = 0.34), and giraffe (r = 0.25) had a weak positive correlation (F igure B 2). Zebra (r = ( Gazella thomsoni) ( r = 0.44), and buffalo (r = 0.32) exhibited negative correlations (Figure B 2). Warthog ( Phacochoerus aethiopicus) (r =

PAGE 61

61 0.17), impala (r = 0.05), wildebeest (r = 0.1), and waterbuck ( Kobus Defassa) ( r = 0.14) showed little evidence of an obvious trend (Figure B 2). Individual species showed discernible differences in levels of occupancy and average abundance and in the amount of variation in occupancy and average abundance over time. Some species always have relatively low occupancy and abundance over time such as elephant (oval 1 in Figure 3 2 ) giraffe, eland, kongoni, waterbuck, and warthog (Figure 3 2). Similarly, some species always have relatively high occupancy and abundance over time these include (oval 2 ) topi, and zebra (Figure 3 2) Two species, buffalo (oval 3) and wildebeest (Figure 3 2 and B 2), have high average abundance but low occupancy and also show more variation in a bundance but do not vary as much in occupancy. Some species do n o t vary much in abundance over time but do vary in occupancy such as kongoni (oval 4) and warthog (Figure 3 2) gazelle (in particular) showed greater variation in both occupancy and abundance compared to other species (Figure 3 2 and B 1). When comparing the distribution of abundances of each species among surveys qualitatively, by inspection of the ROAP graphs, species show some variability in occupan cy and abundances ; however the overall shape s of the distribution s (straight, S shaped, etc.) in general did not change for most species over time (Figure 3 3). Species showing noticeable change in the shape of the abundance distribution over time includ e elephant, buffalo (in particular, 2006 is distinct), wildebeest (in particular, season survey did not appear uniquely different from the wet season surveys for any of

PAGE 62

62 the species. The moderate variation in occupancy and abundance observed for each species appear s to be more due to where they are scarce (x axis) rather than due to shifts in maximum abundance. In addition, the species show differences among themselves in the shape of the distributions of their abundances compared to one another (Figure 3 3). For instance, elephant, waterbuck, and wildebeest have relatively steeper ROAP curves compared to other species whereas giraffe, buffalo, eland, kongoni, and wart hog have relatively it more curvature in their ROAPs Aggregation The parameters of the negative binomial were used to summarize the shape of tion for each census by describing the mean abundance and amount of clumping (aggregation). The relationship between and k (Figure 3 4 a and B 3 a) shows similarity to the relationship between average abundance and occupancy (Figure 3 2 and B 1). Notab le exceptions are that the mean abundance ( ) for buffalo is lower relative to other species (Figure 3 4 a) compared to the pattern for its average abundance (Figure 3 2) and also that the amount of clumping ( k) for topi is lower relative to other specie s (Figure 3 4 a) compared to the pattern for its occupancy (Figure 3 2). The pattern for over time indicates certain years have noticeable deviations from the average (Figure 3 4 b). In particular, higher values are observed in the surveys in 1989 (for (buffalo and zebra). For wildebeest, a higher value is observed in the 1999 survey

PAGE 63

63 (Figure B 3 b). Lower val ues are observed in 1988 (for buffalo and zebra), 1991 and re is no obvious pattern for variation in k over time (Figure 3 4 c and B 3 c ). Spa tial Autocorrelation The average Mora the wet season censuses and the standard deviation was used to measure how clustering varies across time. The species exhibited at most, modest clustering as differed in the amount of spatial autocorrelation they exhibited in the wet season and in the amount of variation in their clustering across time (Figure 3 5 a). Clustering of species, based on abundance, was found to decrease in relation to body mass in the wet season (y = 0.0212log(x) + 0.0683, r 2 = 0.4936). wet seasons compared to the other species and giraffe, wildebeest, waterbuck, and buffalo showed the least variation in M in the amount of spatial autocorre lation they exhibited; however, the strength of the relationship with body mass declined in the dry season compared to the wet season (y = 0.0212log(x) + 0.0683, r 2 = 0 .4936, Figure 3 5 b). Discussion Occupancy and Abundance The positive correlation between average local abundance and occupancy observed in general supports the expected relationship for both interspecific and intraspecific abundance occupancy relationship s (Gaston et al. 2000). Each species seems to fall into a particular slice of the occupancy abundance plane, with considerable differences apparent among species suggesting different governing

PAGE 64

64 forces These interspecific differences in occupancy abundan ce patterns should reflect a combination of factors such as differing social behavior and group formation among the species, specific habitat associations, and effects of heterogeneity across the landscape. Of the species with relatively lower abundance and occupancy, elephant and giraffe have forag ing behavior (for browse) distinct from other species in the study (Baskerville et al. 2011) Although low occupancy was observed for elephants, they are distributed throughout the landscape but may be limited to areas with ready drinking water access (Olff and Hopcraft 2008). Also, waterbuck is restricted to grasslands near larger rivers (Sinclair et al. 2008). Warthog (a small, non ruminant species ) is concentrated in locations with specific soil and moistu re levels, potentially seeking sites with particular resource qualit ies (Olff and Hopcraft 2008). It is intriguing that there is very little temporal variation in warthog average abundance within occupied cells, despite moderate variation in occupancy ove r time (Figure 3 2). This suggests some kind of strong density dependence may be operating at local scales. Kongoni likewise show relative constancy in abundance. Of the species with relatively higher abundance and occupancy, several of the species are m exception of zebra, they are of relatively smaller body mass (< 150 kg). Migratory behavior, of wildebeest for instance, can free a species from constraints which otherwise limit resid ent species, including relaxing the constraints of resource availability and vulnerability to predation (Fryxell et al. 200 8) The small body size of species also potentially allows them to be less limited by resources in the system,

PAGE 65

65 although they are at the same time more vulnerable to predation ( Demment and VanSoest 1985, Sinclair 2003, Sinclair et al. 2003, Hopcraft et al. 2012 ). Buffalo and wildebeest, both larger species (> 150 kg) had high abundance but low occupancy; these two species form very lar ge herds ( Bell 1971, Jarman 1974) and thus their grouping behavior may influence their distinct distribution patterns compared to the some of gazelle, also exhibited greater variation in occ upancy and/or abundance than did the other species, as well as negative intraspecific correlations over time. The migratory nature and grouping behavior of these species likely underly these patterns as the amount of space occupied over time may show grea ter fluctuations for species which have greater movement. Also, species which form large herds and smaller species are more likely to have greater sampling error due to effects o f detectability (Campbell and Borner 1995, Mduma and Hopcraft 2008). The fair ly consistent overall occupancy abundance patterns observed in the ROAPs for some species over multiple time points is unexpected. For example, note that in giraffe, the ROAP is always steep, at low rank, then becomes more shallow, for each year. By cont rast, for buffalo the ROAP is linear for most years but exhibits curvature in a few years. The variability that was observed may be due to changes in vegetation and rainfall from year to year. However, as the overall shape of the ROAP (i.e., the distribu tion of abundanc es) was in general similar for many species over time, it suggests that although there is variation in abundance at a very local level (Campbell and Borner 1995) it does not affect important qualitative features of the aggregate pattern. The absence of a distinctive distribution of abundance for each species for the

PAGE 66

66 1996 dry season was unexpected; species were expected to decrease in occupancy and increase in local abundance due to resources being more limited in the dry season (i.e., reso urces are available in fewer locations). Bu t instead, the dry season patterns appear to be bracketed by the variation among years in the patterns of wet season surveys. The shape of the ROAP for each species provides insight into how aggregated a species is spatially (e.g., a ROAP with a steeper slope indicates a species is clustered whereas a flatter slope indicates a species is more widely distributed) (Collins et al. 2009). Species were expected to have different shapes for their ROAPs due to their var ying traits. The species with steeper curves wildebeest, elephant, and waterbuck form large herds or have specific habitat associations as discussed above for the relationship between average local abundance and occupancy. Some of the species with f latter curves, indicating that the abundances among g rid locations are more uniform appear to be less specialized in the range of habitats they use; for instance buffalo is a large grazer distributed throughout the study area and eland is a drought toler ant migratory species (Olff and Hopcraft 2008). Several species have greater positive curvature in their ROAPs, indicating they have grid locations with similar moderate abundances but also have occupied locations with considerably higher abundances (hots pots) and lower abundances (rarity) include three migratory species which all prefer low rainfall areas in the wet season tolerant) (Olff and Hopcraft 2008). Similarly, topi and im pala prefer specific spectrums of the rainfall gradient (topi are found in wet, fertile areas; impala are found in dry, fertile areas) (Olff and Hopcraft

PAGE 67

67 2008). Overall, species show lower temporal var iability in their maximal abundance. Understanding these patterns will require one to understand how density dependent and density independent processes vary over space and through time. Aggregation The mean abundance ( ) and clumping ( k ) parameters of th e negative binomial for each species abundance distribution overall was comparable to the patterns observed for average abundance and occupancy discussed above. Buffalo, which exhibited a difference in the pattern of its abundance relative to other species (Figure 3 4 a) had lower abundance using the negative binomial estimate compared to the average abundance for occupied cells, indicating that the abundance pattern is more affected by the inclusion of grid cells with zero abundance compared to other spec ies. The other exception, topi, exhibited a difference in the pattern of its aggregation relative to other species; however, the reason is less clear. The deviations of the mean abundance parameter from the species mean across year shows that particula r years have considerably larger differences from the mean suggesting that species may be responding demographically to changes in resource availability (Figure 3 4 b) The species with large deviations correspond ing to these years are all migratory (zeb wildebeest) or a large grazer (buffalo). The aggregation pattern, which appeared random over time, suggests occupancy across the landscape is not as affected by fluctuating resources compared to local abundance.

PAGE 68

68 Spa tial Autocorrelation The clustering patterns observed for species in relation to body size support the results of Olff and Hopcraft (2008) for resident herbivores during a wet season survey of the Serengeti in 2001. This study expands on their results by providing variation in I for a dry season in 1996, including migratory species. The migratory species do not appear to deviate from the overall pattern with the exceptio which exhibits higher clustering than expected for its mass. These results contrast with those of Ogutu and Dublin (2004), who observed that the biomass of ungulate herbivores is spatially autocorrelated within 4.1 km when consider ing resident species only and within 3.3 km for residents and migrants (thus, migratory species reduce the amount of spatial autocorrelation observed for the community). The difference in the species measure used (abundance compared to biomass) may accoun t for the inconsistency. The two species with the most variation in spatial gazelle and eland, are both migratory (also zebra). Wildebeest, although migratory, have surprisingly little variation in autocorrelation; however, they are comparatively very abundant, form very large herds, and concentrate in the plains during the wet season (Appendix A 2) autocorrelation in abundances. Conclu ding Remarks Understanding the occu pancy and abundance patterns of species can improve our understanding of how species relate to the environment and interact with each other. S pecies may show similar relationships based on their traits thereby providing a

PAGE 69

69 broader way in which to describe a diverse community. The large herbivore species investigated in this savanna ecosystem exhibit several distribution patterns in relation to species traits. The species overall display a positive relationship between average local abundance and occupanc y This fits broad patterns reported in other systems (Gaston et al. 2000), but there are a number of intriguing exceptions. I ndividual species show discernible differences in levels of abundance and occupancy which can be related to their traits. Spec ies with low occupancy and abundance tend to have distinct social behavior and specific habitat associations whereas species with high occupancy and abundance tend to be migratory and smaller species. Species with strong grouping behavior tend to deviate from these general patterns. Over time, the overall shape of the occupancy abundance patterns for each species appear to be fairly consistent, suggesting that year to year variations in abundance at the local level do not appear at the aggregate level. Spatial autocorrelation of species declines with body mass; migratory species appear to follow this broad trend also, although they in general show more variability in aggregation compared to other species. In comparison to the wet season, species decreas e in clustering in the dry season, suggesting species are dispersing spatially due to more limited resources and potentially reducing intraspecific competition. The patterns observed in this system for occupancy and abundance of large herbivore species in relation to traits such as body size and migratory behavior provide insights into large herbivore dynamics for this study area and could potentially be applied to other systems or species.

PAGE 70

70 Table 3 1 Species names and traits of the t hirteen herbivores in the study (ordered by decreasing body mass) Species n ame a Common n ame a Mass (kg) b Feeding g uild c Ruminant/ n on ruminant Behavior d Loxodonta africana African elephant 3000 Mixed Non ruminant Resident Giraffa camelopardalis Giraffe 800 Brows er Ruminant Resident Syncerus caffer African buffalo 450 Grazer Ruminant Resident Taurotragus oryx Eland 400 Mixed* Ruminant Migratory Equus burchellii 250 Grazer Non ruminant Migratory Kobus defassa Defassa waterbuck 18 0 Grazer Ruminant Resident Connochaetes taurinus Wildebeest 170 Grazer Ruminant Migratory Alcelaphus buselaphus hartebeest) 150 Grazer Ruminant Resident Damaliscus korrigum Topi 120 Grazer Ruminant Resident Phacochoerus aethiop icus Warthog 60 Grazer* Non ruminant Resident Aepyceros melampus Impala 50 Mixed Ruminant Resident Gazella granti 50 Mixed Ruminant Migratory Gazella thomsoni 20 Mixed Ruminant Migratory Sources: a. Mduma and Hopcraft (2008). b. Sinclair et al. (2003). c. Prez Barbera et al. (2001); items marked with from Kingdon (1997). d. Sinclair et al. (2008).

PAGE 71

71 Figure 3 1. Serengeti National Park and neighboring conservation areas and game reserves in Tanzania and Kenya East Africa. The animal census survey grid (730 cells of size 5 x 5 km) is shown CA = conservation area, GR = game reserve, NP = national park, RA = reserve area.

PAGE 72

72 Figure 3 2. A bundance (average abundance of occupied sites ) in relation to o ccupancy for a census ( n = 730 grid cells in a census ) Nine censuses are plotted for each species (eight wet season, one dry season). The ovals refer to: 1) 1 4 3 2

PAGE 73

73 Figure 3 3. Rank occupancy abund ance profile ( ROAP ) for each species for wet and dry seasons Each point on the ROAP represents a grid cell in which the species was observed (cells with zero abundance are not shown for clarity). Local abundance is the log of the number of organisms per km 2 for each 5x5 km 2 grid cell. Relative rank is the rank orde r of the cell divided by the total number of cells in the census (n = 730 grid cells in a census). Note that the scale of the axes varies among species graphs and graphs are ordered by decrea sing body mass A) elephant, B) giraffe, C) buffalo, D) eland, E) zebra, F) waterbuck, G) wildebeest, H) kongoni, I) topi, J) warthog, K) impala, L) A) B) C) D)

PAGE 74

74 E) F) G) H)

PAGE 75

75 I) J) K) L)

PAGE 76

76 M)

PAGE 77

77 Figure 3 4 Estimates for parameters m and k (indicating mean and clustering (n = 730 grid cells in a census). Nine censuses are plotted for each species (eight wet season, one dry seas on ) A) estimates of mu in relation to k, B) A)

PAGE 78

78 B)

PAGE 79

79 C)

PAGE 80

80 Figure 3 5 of each herbivore for A) the mean of eight wet seasons (error bars are +/ 1 standard deviation) and, B) a dry season ( n = 730 grid cells in a census ) BU = buffalo, = kongoni, T warthog, WT = wildebeest, ZB = zebra. Note the log scale on the x axis. A)

PAGE 81

81 B)

PAGE 82

82 CHAPTER 4 COMMUNITY STRUCTURE AND INTERSPECIFIC ASSOCIATIONS AMONG LARGE HERBIVORES IN THE SERENGETI Communit ies and foraging guilds typically have some species which are very abundant, several which are moderately abundant, and many species which are rather (to very) rare (Magurran 2004). The coexistence of species in diverse systems is intriguing because theor y predicts that dominant competitors will competitively exclude other species (Gause 1934, Hardin 1960), yet we observe diverse assemblages of ecologically similar species in many systems. Species coexistence is likely depende nt on the interplay of many f actors (Holt 2001), including physiological constraints due to abiotic characteristics of the habitat ( determining the fundamental niche of a species) (Hutchinson 1957, Mac Arthur et al. 1966, Brown et al. 1995), biotic interactions such as competition, pr edation, mutualisms, and disease ( determining the realized niche of a species), and conspecific attraction and stochastic dispersal and disturbance events (Hutchinson 1957, Pulliam 2000, Hubbell 2001, Lichstein et al. 2002, Tilman 2004). T he realized nich e of a species resulting from interspecific competitive interactions among organisms in a community, involves strong primary interactions, such as the direct effect of one competing species on another, and aggregate effects of weak competitive interactions among species, known as diffuse competition (MacArthur 1972). I t is difficult to infer process from pattern but characterizing patterns of coexistence provides a first step towards a deeper mechanistic understanding of the deterministic community struct ure The mechanisms that permit species coexistence in the Serengeti is the subject of intense investigation. Current theories explain these mechanisms on the basis of variation in body sizes of the different herbivore species, spatial and temporal variat ion

PAGE 83

83 of resources in the system, and the behavioral responses of the species (Fryxell et al. 2008). The body size theory posits that a broad range of body sizes increases species diversity through body size effects on resource use and predation susceptibil ity. While smaller species are more vulnerable to predation, the principal cause of mortality in larger species is resource limitation (Sinclair 2003, Sinclair et al. 2003). In addition, body size relates to partitioning of resources in relation to facto rs such as vegetation height, food quantity, and food quality (Murray and Brown 1993, Wilmshurst et al. 2000, Fryxell et al. 2008). Smaller species have better access to short vegetation and require less food due to their higher per mass metabolic rates. Therefore, they tend to seek low growing, higher quality plant types and parts such as short grasses and leafy material (Demment and VanSoest 1985, McNaughton 1985, Wilsey 1996). Conversely, larger species require higher quantities of food, and can utili ze taller and coarser vegetation (Demment and VanSoest 1985, McNaughton 1985, Wilsey 1996) due to their lower per mass metabolic rate. Such species can in fact at times facilitate smaller species by altering vegetation accessibility and palatability such that it is more suited to the higher quality resource demands of the latter (Fryxell et al. 2008). The spatial and temporal variation of resources and the behavioral response of species may also play a role in species coexistence in the system through enha ncing the scope for niche differentiation. Environmental variability across space and time in the Serengeti (see study area description in methods) include s seasonal differences in resource availability (e.g., green grass) and gradients in vegetation distr ibution across space. Herbivores in the Serengeti can be partitioned into different feeding guilds i.e., grazers, browsers and mixed feeders, and this variation in feeding habit couples them to

PAGE 84

84 the environment because the three guilds preferentially feed on grass, woody vegetation, or both vegetation types respectively. T he behavioral response of the migratory wildebeest ( Connochaetes taurinus ) is regarded as being particularly influential in the Serengeti ecosystem. The annual migration of large herds of wildebeest potentially benefits their survival by allowing them to escape predators and access transient resources through their movement (Fryxell et al. 2008). Given their large numbers, the migration of the wildebeest can potentially promote the surv ival of other species through a process of relaxed competition and predator satiation (Fryxell et al. 2008). The mechanisms underlying species coexistence in the Serengeti ecosystem thus are likely to involve the interplay of abiotic and biotic factors. A s a step towards elucidating these mechanisms, it is useful to have in hand a quantitative description of community structure, and to detect signals of co association (or repulsion) in abundance. The objectives of this study are to 1) compare patterns in community structure over time, and 2) identify interspecific associations which may be influencing the abundance of different herbivore species. This study addresses the hypotheses that species in this system will either exhibit 1) no associations (i.e., null model), 2) positive associations due to factors such as grouping or co migration, or 3) negative associations due to factors such as habitat differentiation and competition. Methods Study A rea The Serengeti Mara ecosystem, across the border of Tanzan ia and Kenya in East Africa, is a long established network of protected areas. The system covers approximately 25,000 km 2 most of which is Serengeti National Park ( SNP,

PAGE 85

85 approximately 15,000 km 2 ) (Sinclair et al. 2008) (Figure 4 1). The park, established in 1951, and surrounding buffer areas do not permit livestock or agriculture; some of the buffer areas do allow licensed hunting (in game reserves) and controlled pastoralism (i.e., ranching) (in conservation areas), however. The habitat of the Serengeti Mara is spatially and temporally heterogeneous and is characterized by a core wet (March May) and dry (August October) season and an increasing rainfall gradient from the southeast (500 mm/yr) to the northwest (1,200 mm/yr) (Norton Griffiths et al. 1975, S inclair et al. 2008). The vegetation transitions from treeless, short grass plains in the southeast to tall grass savannas and woodlands in the north and west, in relation to annual rainfall (Sinclair et al. 2007). Fires, which occur in the dry season an d are at present usually caused by human s are a vital factor in the system (Sinclair et al. 2007). The park has a unique, largely intact community of twenty eight large herbivores and ten carnivorous large predators (Mduma and Hopcraft 2008, Sinclair et a l. 2008). Of these species, wildebeest is a dominant species and vital for maintaining the ecosystem in its state (Sinclair 2003, Sinclair et al. 2007). The annual migration of wildebeest (from the southern grasslands in the wet season and the northern w oodlands and savannas in the dry season) is driven by the seasonal rainfall gradient of the ecosystem and its effects on vegetation growth and availability (Pennycuick 1975, Boone et al. 2006, Holdo et al. 2009) (Appendix A 2). Census D ata The census data for herbivore species consists of nine annual surveys from 1988 2006 (eight wet season and one dry season) (Appendix A 3 Table A 5 ). Census data were collected by the Tanzania Wildlife Research Institute using Systematic

PAGE 86

86 Reconnaissance Flights (SRF) and used to estimate wildlife densities across a survey grid having a cell size of 5 x 5 km (Figure 4 1). To census the animals, a plane was flown along east west transects across the middle of the survey grid cells. Large herbivores were counted in subunits which were defined as 30 seconds of flying time (approximately 2 km) with a strip width of 150 170 m on either side of the aircraft (thus, animals were counted in approximately 6 7% of the survey grid cell). These data were used to calculate a density ( number per km 2 ) value for each 5 x 5 km survey grid cell. A Universal Transverse Mercator (UTM) coordinate system (easting and northing coordinate pair) was used to record survey data, with each survey location at the center of a survey grid cell. Detail ed methods are provided in Campbell and Borner (1995). The annual surveys resulted in 730 sample locations (covering the majority of the park as well as some neighboring areas) that were consistently sampled in each survey. Based on the perceived reliab ility of the survey data considering detectability (Appendix A 1), t welve large herbivore species (out of twenty eight observed) were selected for the analyses (Table 4 1, Appendix A 1 Table A 1 and A 2 ). Abundance data used in this study represent abund ance for a 1 km 2 area in a 5 x 5 km survey grid cell. Note that migratory wildebeest were not counted during these aerial surveys (see Appendix A 1 and A 2) Rank A bundance One classic way of describing a community is with a rank abundance relationship (M agurran 2004). These plots show the abundance of a species on the y axis and display the species ordered from most to least abundant on the x axis. Thus, the distribution of species abundances is displayed, simultaneously depicting the species richness ( number of species present ) and species evenness (level of similarity in

PAGE 87

87 abundances of species). A log 10 format was used for the abundances on the y axis due to the large range in magnitude of abundances among species. In addition, we used a modification of the conventional rank abundance, the Whittaker plot, which displays relative abundance on the y axis (i.e., the abundance of each species as a proportion of the total abundance of all the species) (Whittaker 1965). The overall community structure was s ummarized using rank abundance plots for each of the nine surveys. T he species rank abundance plots only provide information about a subset of twelve species in the system. A species abundance model was fit to the data for each survey. Upon visual inspec tion of the shape of the rank abundance plots (i.e., Whittaker plots), the geometric series model was selected due to the steep slope observed (Magurran 2004). With a geometric series (Motomura 1932, May 1975), the abundances of species ordered from most to least abundant is given by n i = NC k k (1 k ) i 1 where k = the proportion of available niche space or resource that each species occupies; n i = the number of individuals in the i th species; N = the total number of individuals; C k = (1 (1 k ) S ) 1 a constant n i = N ; and S = the total number of species. Estimates of k and C k were obtained using an iterative approach with the following equation (May 1975, Magurran 2004): = where N min = the nu mber of individuals in the least abundant species. Because species overlap in their diet, the total availability of high quality food for a species may to some degree depend on the abundances of the rest of the community.

PAGE 88

88 abundances (SRACA) were explored graphically for each survey by plotting the abundance of a species within a grid cell in relation to the total abundance of the rest of the community within th at grid cell (n = 730 grid cells). Graphs were created with each of the twelve species in the study as the focal species and displayed in order of decreasing body mass of the focal species. As the plots for each survey app eared similar, a single plot for each species was created with all of the surveys combined. To provide a null model which ignores species identify the abundance of each species and corresponding total abundance of the rest of the species in the grid cell was plotted, with a ll species together on one plot As we will return to in the discussion, negative associations could arise because of two reasons. First, species utilize different habitats (e.g., plains vs. woodland habitats). Second, species could n egatively interact due to competition for resources or apparent competition. Also, in some circumstances, species could be mutualists or indirect mutualists; hence, they might show positive rather than negative associations. Results Rank A bundance The ran k abundance plots for the community indicate substantive unevenness in the abundances of the species (as demonstrated by a steep slope), which is consistently observed over time (across wet season surveys and the dry season survey) (Figure 4 2 and C 1). T he community does show some variation in species abundances over time, however the overall community structure is very constant over time (Figure 4 2). In particular, the curves are very similar with respect to the more abundant species over time compared to slightly more temporal variability for rarer species (the variability is in part likely due to sampling error from small abundances).

PAGE 89

89 The abundances of species in the single dry season survey does not appear different from the wet season surveys. For the geometric series model fit to the species abundance data for each survey, the estimates of k which measures numerical dominance, range between 0.374 and 0.450 (Table 4 2). The dry season survey does not have a notably different value of k compared t o the other surveys (Table 4 2). Migratory wildebeest (not depicted on the rank abundance plots, but see Appendix A 1 Table A 2 ) are far more abundant than are any other species in the ( Gazella th omsoni) and zebra ( Equus burchellii) are consistently observed to be more abundant than the remaining species (Figure C 1). The species of intermediate abundance consistently consisted of impala ( Aepyceros melampus) topi ( Damaliscus korrigum ) buffalo ( Syncerus caffer) and G ( Gazella granti) ( although the order of these ) The species which were least abundant also consistently consisted of a particular set of speci es kongoni ( Alcelaphus buselaphus ) e land ( Taurotragus oryx) g iraffe ( Giraffa camelopardalis) w artho g ( Phacochoerus aethiopicus) e lephant ( Loxodonta africana ) and waterbuck ( Kobus Defassa) whose order of abundances relative to one another does show some variation over time. The two most abundant individuals, zebra and (between 41.9% and 58.7% for the nine surveys) of the twelve species in the study. sponses to Aggregate Community Abundances Responses of a species relative to the rest of the community indicate the presence of potential strong negative associations and suggest the presence of some potential weak negative associations (Figure 4 3). Abun dance patterns of certain

PAGE 90

90 species demonstrate a strong negative association (el ephant, buffalo, and waterbuck) -that is when the focal species is common within a sample unit, the sum of the remaining species is low. In general, the specie s showing strong negative associations have a large body mass (> 150 kg). No association is apparent for other species (giraffe, kongoni, warthog, and impala). The s pecies showing no association have both large (> 150 kg) and intermediate (50 150 kg) bod y mass. gazelle) show some indication of a weak negative association between the abundance of the focal species and the abundance of the rest of the community (Figure 4 3). Of the se species, the topi demonstrates more tendency towards a negative association compared to the others. The null model (Figure 4 3 m), is most similar to the weak negative association observed for these species. The species showing evidence for weak negat ive associations have large (> 150 kg), intermediate (50 150 kg), and small (< 50 kg) body mass. The species in this category are all migratory, with the exception of topi (note, the only other migratory species in the study system is wildebeest, which ar to aggregate community abundance and feeding guild (i.e., grazer, browser, mixed) or digestive strategy of the focal species (i.e., ruminant or non ruminant). Discussion Rank Abundance A community is considered to be more diverse if it exhibits higher species richness or higher species evenness (i.e., lower dominance) (Magurran 2004). In characterizing the diversity of this system, the steep slope of the rank abundance p lot indicates much of the diversity is generated by the high species richness of the

PAGE 91

91 community rather than by species evenness. One mechanistic interpretation of t he geometric series, which was used to describe this community based on the steep slope and straight shape of the log rank abundance plot, is a model in which the abundance of each species is associated with the proportion of resources it uses (Motomura 1932). The model is generally applied to species poor assemblages (such as those in harsh env ironments or early succession) (Magurran 2004). While the Serengeti ecosystem certainly does not fit this description, the presence of dominance preemption of resources is still possible (for instance, by modification of vegetation, in which herbivory by one species alters the vegetation in a way that makes it less profitable to competing species (Murray and Illius 2000)). A similar shape in species abundance distribution is intriguingly observed for ungulates in the savanna system of Tarangire also in T anzania (Shorrocks 2007). Interestingly, the slope of the rank abundance plots observed for th e Serengeti is consistent over time (including wet seasons and dry season), indicating there is some stability in the community structure which is keeping the re lative distribution of abundances unchanged. This may relate to the food wed structure of the Serengeti, which exhibits functionally distinct groups of carnivores, herbivores, and plants which are spatially coupled across trophic levels, an aspect which m ay be important for persistence of species and the stability of the system (Baskerville et al. 2011). Species having similar feeding guilds, behavior, and body size may be expected to have similar abundances due to the association between these traits and processes regulating survival and reproduction. The numerical dominance in abundance of a select few species may suggest they have a greater role in the dynamics of this

PAGE 92

92 community. T he abundance of a species is to some extent indicative of its ability to compete for limiting resources (Magurran 2004), thus the more abundant species may be considered to be better competitors. The three species which are more abundant in in that they all participate in an established annual migration across the landscape (Shorrocks 2007). Since migratory species, such as wildebeest, are not limited to a particular area over time, they may be able to escape from constraints that affect res ident species (e.g., migratory species can access new resources and escape predation) (Fryxell et al. 2008). The migratory wildebeest is much more abundant than s a smaller body size and thus may be more vulnerable to predation (Sinclair 2003). This migratory syndrome thus may underlie the strong dominance by a few species in the Serengeti. Zebra, as non ruminant, has greater flexibility for food quality and wider diet tolerance than ruminants (Cromsigt et al. 2009a); however, c ontrary to expectations, it is less abundant than the other two of these migratory species which are ruminant. The order in which the species migrate, by decreasing body size, may relate to this pattern of lower abundance of the non ruminant species. Zeb ra migrate first, consuming stems grass and herbs respectively (Bell 1971). This succession of species increases access to preferred vegetation for the next species in the o rder. Thus, the foraging of the zebra, as the first species in the succession, is not being facilitated in the way it potentially is for the other two species. An a lternative hypothesis is that compared to wildebeest, zebra have low first year survival and high predation on adults (Sinclair and

PAGE 93

93 Norton Griffiths 1982, Grange et al. 2004). The other migratory species in the system, which are less understood (Sinclair et al 2008), however their abundances are considerably lower than either wildebeest or ruminant species in the system, elephant and warthog, are non migratory and have considerably lower abundances than zebra, but may also hav e social behaviors and habitat associations which limit their abundances. For instance, elephant have foraging behavior (for browse) distinct from other species in the study (Baskerville et al. 2011) and may be limited to areas with ready drinking water a ccess (Olff and Hopcraft 2008). Warthog are concentrated in locations with specific soil and moisture levels, potentially seeking resource quality (Olff and Hopcraft 2008). In this system, species richn ess is not correlated with total abundance (see Chapter 2), contrary to general expectations for the relationship between richness and abundance (Evans et al. 2005). The lack of relationship may be a reflection of interspecific interactions and/or the sam e collection of species may not occur in areas of low and high community abundance. Positive association between species arises due to factors such as similar environmental requirements, similar responses to disturbance, and mechanistic dependencies betwe en species such as predation or symbiosis whereas negative association between species arises due to factors such as differing environmental requirements, competition, spatial exclusion, and interspecific antagonisms (McCune and Grace 2002). The strong n egative association patterns observed for buffalo in this system, which form large herds (Jarman 1974), may reflect its presence as a dominant

PAGE 94

94 competitor that interferes with other species when it is more abundant. For waterbuck and elephant, the negative association patterns observed may be effects of specific habitat associations, with the species being confined to locations along rivers (Sinclair et al. 2008) and foraging for browse respectively (Baskerville et al. 2011). As these species are of larger body mass, they consume larger quantities of vegetation, potentially presenting competition for resources with some other species. The species demonstrating a potential weak negative association, are all migratory, thus indicating movement patterns may a ct to reduce competition with other species in the system (Fryxell et al. 2008). This includes the grazing succession among zebra, wildebeest, above Conclu ding Remarks The composition and abundance of species in a communit y are the result of a combination of abiotic and biotic processes and interactions. In this savanna system, the overall community has a few species which dominate numerically and numerous other species which are less abundant and contribute to the species richness of the system. The rank abundances observed show relative consistency over time, suggesting that in the midst of a complex spatially and temporally heterogeneous environment, community structure can be quite stable. Weak negative associations a re observed for certain species, which are migratory, and migratory behavior also appears to be a key factor in allowing certain species to achieve greater abundances. There is also evidence for strong negative associations between certain species and the rest of their communities in particular species with large body sizes which form large groups, suggesting the potential for competition for resources and space. In addition, what appears to be negative interactions for some species may in fact be relate d to habitat

PAGE 95

95 specificity. This system thus demonstrates a range of interspecific associations but a surprising measure of overall stability in community structure over time.

PAGE 96

96 Table 4 1 Species names and traits of the twelve herbivores in the study (o rdered by decreasing body mass) Species n ame a Common n ame a Mass (kg) b Feeding g uild c Ruminant/ n on ruminant Behavior d Loxodonta africana African elephant 3000 Mixed Non ruminant Resident Giraffa camelopardalis Giraffe 800 Browser Ruminant Res ident Syncerus caffer African buffalo 450 Grazer Ruminant Resident Taurotragus oryx Eland 400 Mixed* Ruminant Migratory Equus burchellii 250 Grazer Non ruminant Migratory Kobus defassa Defassa waterbuck 180 Grazer Rumin ant Resident Alcelaphus buselaphus hartebeest) 150 Grazer Ruminant Resident Damaliscus korrigum Topi 120 Grazer Ruminant Resident Phacochoerus aethiopicus Warthog 60 Grazer* Non ruminant Resident Aepyceros melampus Impala 5 0 Mixed Ruminant Resident Gazella granti 50 Mixed Ruminant Migratory Gazella thomsoni 20 Mixed Ruminant Migratory Sources: a. Mduma and Hopcraft (2008). b. Sinclair et al. (2003). c. Prez Barbera et al. (2001); items marked with from Kingdon (1997). d. Sinclair et al. (2008).

PAGE 97

97 Table 4 2. For each census, estimates of k and C k for the abundances of species modeled by a geometric series n i = NC k k (1 k ) i 1 (where k = the proportion of available niche space or resource t hat each species occupies; n i = the number of individuals in the i th species; N = the total number of individuals; C k = (1 (1 k ) S ) 1 n i = N ; and S = the total number of species). Census k C k 1988 0.387 1.0046 1989 0.445 1.0015 1991 0.394 1.0040 1996 0.450 1.0008 1996 (dry season) 0.412 1.0017 2000 0.374 1.0037 2001 0.427 1.0013 2003 0.420 1.0014 2006 0.393 1.0025

PAGE 98

98 Figure 4 1. Serengeti National Park and neighboring conservation areas and game reserves in Tanzania and Kenya East Africa. The animal census survey grid (730 cells of size 5 x 5 km) is shown CA = conservation area, GR = game reser ve, NP = national park, RA = reserve area.

PAGE 99

99 Figure 4 2. R ank abundance plot (the species relative abundance is depicted on the y axis, with a log 10 scale, and the rank order of species from most to least abundant is depicted on the x axis) for twelve large herbivore species for each survey (eight wet season, one dry season).

PAGE 100

100 Figure 4 3. Abundance of each species in relation to total abundance of the rest of the community for each survey (n = 730 grid cells in a survey ). Graphs for each survey are o rdered by mass of the focal species. Note that the scale of the y axi s differ s among plots. A) elephant, B) giraffe, C) buffalo, D) eland, E) ll model (ignores species identity). A) B) C) D)

PAGE 101

101 E) F) G) H) I) J)

PAGE 102

102 K) L) M)

PAGE 103

103 CHAPTER 5 CONCLUSION S This research examined how anthropogeni c activities, spatial processes and species traits jointly influence on a landscape leve l, the occupancy, abundance, and interspecific associations of large herbivores in the Serengeti Mara. With rapidly increasing human populations, ensuring the long term effectiveness of protected areas through wise management is increasingly important. T o conserve biodiversity effectively within protected areas and to predict how species within and outside protected areas could respond to altered environmental conditions and management strategies, species abundance and distribution patterns must be identi fied and the processes underlying them must be understood. In exploring species distributions, species may show similar relationships based on their traits thereby providing a more general way in which to describe a diverse community. Chapter 2 explored the spatial distribution of the community over time using a suite of aggregate community measures (species richness, total abundance, total biomass, and cumulative basal metabolic rate). The influences of natural and anthropogenic habitat characteristics (e.g., road density, distance to river, plant nutrients, tree cover) on each of these community measures were assessed using spatial regression models. Total abundance, total biomass, and total basal metabolic rate were strongly cross correlated in each s urvey. There were similarities, as well as differences, in the spatial pattern of species richness compared to the other metrics; species richness, however, was often greater at lower values of the aggregate measures. The herbivore community appears vuln erable to the effects of humans in surrounding areas -these measures tend to be depressed near edges of the park

PAGE 104

104 adjacent to human activity. Roads within the protected areas do not negatively affect these herbivores. The resources most strongly influen cing the community distributions are plant nutrients and NDVI (i.e., plant productivity). Fire, although prominent in savannas, does not markedly affect herbivore community distributio ns at a broad landscape level at least in the short term (within year) These results indicate that efforts to manage for species richness would involve emphasizing habitat characteristics different from those that would maximize total abundance, biomass, or metabolic rate. Also, future management practices might strive to identify the mechanisms responsible for negative effects emanating from human populations in neighboring areas. In addition, roads overall currently do not appear to have a negative influence on herbivores suggesting road activity from ecotourism is not problematic for these species at the moment However, l evels of road usage should be monitored in the future and additional studies investigating the effects of road use intensity should be conducted to provide a more thorough assessment of the effects of roads on large herbivores. It may also be important to maintain t he spatial and temporal heterogeneity of the ecosystem since resources influencing herbivore distributions ( plant nutrients and productivity) are patchy and dynamic. Chapter 3 compared sin gle species occupancy and abundance patterns over time in relation to species traits Even i gnoring species identity, there is still a modest correlation between oc cupancy and average abundance In addition, i ndividual species showed discernible differen ces in levels of occupancy and average abundance and in the amount of variation in occupancy and average abundance they showed over time Species with low occupancy and abundance tend to have distinct social behavior and

PAGE 105

105 specific habitat associations, whe reas species with high occupancy and abundance tend to be migratory and smaller bodied species. Many individual species display a positive relationship between average local abundance and occupancy, but there are a number of intriguing exceptions. Specie s with strong grouping behavior tend to deviat e from these general patterns. ROAP analyses revealed that s pecies show some variability in occupancy and abundances over time. However, in general, t he overall shape of the distribution (straight, S shaped, e tc.) a ppear s to be fairly consistent for most species over time suggesting that year to year variation in abundance at the local level do not appear to a ffect qualitative features of the aggregate pattern In addition, s pecies show differences among them selves in the shape of the distributions of their abund ances, compared to one another. Species exhibiting steeper curves tend to form large herds or have specific habitat associations and species exhibiting flatter curves tend to be less specialized in h abitat. Species exhibiting greater curvature in their curves tend to be migratory and/or prefer specific levels of rainfall, which may lead to areas of high local abundance. The clustering of species, examined using was found to decrease in re lation to body mass. M igratory species appear to follow this broad trend also, although they in general show more variability in aggregation over time compared to other species. Clustering of species in general declined in the dry season compared to the wet season s Chapter 4 investigated patterns in community st r ucture and interspecific associations The structure of the community as summarized using rank abundance plots, indicate unevenness in the abundances of the species (as demonstrated by a

PAGE 106

106 steep slope). The community has a few species which dominate numerically and numerous other species which are less abundant. The community shows some variation in species abundances over time; however the overall community struc ture is very constant over time ( across wet season surve ys and the dry season survey) suggesting that in the midst of a complex spatially and temporally heterogeneous environment, community structure can be quite stable. There is evidence for strong negative associations among species, that is when the focal species is abundant within a sample unit, the sum med abundance of the remaining species is low Species showing negative associations tend to have large body sizes and form large groups, suggesting potential competition for resour ces and space. In addition, negative interactions present for some species may be related to habitat specificity. Weak negative associations are by contrast observed for some migratory species. M igratory behavior also may be a key factor in allowing cer tain species to achieve greater abundances. The results of these chapters contribute to our understanding of the combined influence of human activities, spatial processes, and species traits on the abundance, occupancy, and interspecific associations of sp ecies. Human activities could be managed to mitigate negative effects on wildlife habitat use (monitor road usage, maintain integrity of buffer zones) In addition, it may be important to maintain the s patial and temporal heterogeneity of resources in th e system due to the influence on the spatial distribution of the community. Interestingly and perhaps related, the overall over time. The traits of the species, in parti cular body size and migratory behavior,

PAGE 107

107 show patterns in relation to abundance occupancy and interspecific associations Increased body mass has a negative relat ionship with species clustering, with migratory species show ing more variability in aggregat ion compared to other species. Large body size and grouping behavior appear to contribute to strong nega tive associations among species, whereas m igratory behavior may contribute to weak negative associations among species. Also, m igratory behavior and s mall body size may be key factor s in allowing certain species to achieve greater abundances and high er occupancy. The p atterns observed in this system provide an expansive view for this study area and could potentially be applied to ot her systems or spec ies These results can also help predict effects of environmental changes and management strategies on communities. Future studies including the remaining large herbivore species present in the system (in particular, wildebeest ( Connochaetes taurinus ) ) w ould provide a more thorough assessment of the community dynamics. In addition, given that the presence of a diverse intact community of herbivores in this netwo rk of protected areas helps attract larg e volumes of tourists, policies sh ould be implemented to integrate local people in the ecotourism industry, generating a potentially sustainable source of income for local communities and providing incentives for maintaining the integrity of the conservation area The results of this dissertation provide use ful insights about the structure and function of an important group of species, insights which should inform wise management of the Serengeti.

PAGE 108

108 APPENDIX A SUPPORTING INFORMATI ON FOR CHAPTER 2 Appendix A 1. Large Herbivore S pecies S urveyed in the Study S ystem Table A 1. Population estimates for each herbivore species surveyed in the study system and included in the main analyses Abundance values indicate the estimated number of individuals (obtained by several methods) in the Serengeti ecosystem (Mduma and Hopcraft 2008). Total biomass and total BMR were calculated using methods described in the main text. The herbivores in the core analyzable dataset. Comparisons between observed ab undances for the data set and estimated abundances for the system indicated below demonstrate population values are within 30% with the exception of buffalo ( Syncerus caffer) and elephant ( Loxodonta africana ) which are overestimated by the data, and Thoms ( Gazella thomsoni) which are underestimated by the data (see Table 2 1 in the main text for observed abundance values). Common name Abundance Total biomass (kg) Total BMR (W) Percent abundance Percent biomass Percent BMR African elephant 2, 360 7,080,000 3,244,956 0.13 2.45 1.23 Giraffe 10,460 8,368,000 5,337,096 0.56 2.89 2.02 African buffalo 30,276 13,624,200 10,033,760 1.62 4.71 3.79 Eland 15,773 6,309,200 4,785,366 0.84 2.18 1.81 183,815 45,953,750 39,200,517 9.82 15.89 14.81 Defassa waterbuck 1,186 213,480 197,695 0.06 0.07 0.07 hartebeest) 16,043 2,406,450 2,332,437 0.86 0.83 0.88 Topi 38,990 4,678,800 4,795,071 2.08 1.62 1.81 Warthog 3,737 224,220 273,270 0.20 0.08 0.10 Impala 90,692 4, 534,600 5,784,320 4.84 1.57 2.19 54,628 2,731,400 3,484,164 2.92 0.94 1.32 328,620 6,572,400 10,541,983 17.56 2.27 3.98 Subtotal: 776,580 102,696,500 90,010,636 41.49 35.50 34.01

PAGE 109

109 Table A 2. Population estimates f or each herbivore species surveyed in the study system but not included in the main analyses. Abundance values indicate the estimated number of individuals (obtained by several methods) in the Serengeti ecosystem (Mduma and Hopcraft 2008). Total biomass and total BMR were calculated using methods described in the main text. The herbivores that were not included in the core analyzable dataset (see Table A 3 for additional information) Co mmon name Abundance Total biomass (kg) Total BMR (W) Percent abundance Percent biomass Percent BMR Hippopotamus < 808 1,616,000 819,671 0.04 0.56 0.31 Rhinoceros 13 15,600 8,991 0.00 0.01 0.00 Roan antelope 180 48,600 40,668 0.01 0.02 0.02 Greater kudu 53 11,289 10,023 0.00 0.00 0.00 Fringe eared oryx < 100 16,900 15,899 0.01 0.01 0.01 Wildebeest 1,086,754 18,474,8180 173,549,666 58.06 63.86 65.57 Bushpig ------------Lesser kudu ------------Bohor reedbuck ------------Bushbuck ------------Mountain reedbuck < 200 6,000 8,696 0.01 0.00 0.00 Oribi 7,000 126,000 207,495 0.37 0.04 0.08 Klipspringer < 200 2,400 4,374 0.01 0.00 0.00 Grey duiker ------------Steinbuck ------------K ------------Subtotal: 1,095,308 186,590,969 174,665,483 58.51 64.50 65.99 Total: 1,871,888 289,287,469 264,676,119 100 100 100

PAGE 110

110 Table A 3 Traits for s pecies (ordered by decreasing body mass) recorded in the surveys ( see main text for details) but not included in the analysis. S maller species and/or species that were encountered infrequently due to very low population sizes or very restricted distributions w ere excluded from the analyse s because the existence of visib ility bias es leads to reduced detectability which inflates sampling error s (Campbell and Borner 1995, Mduma and Hopcraft 2008). Migratory wildebeest ( Connochaetes taurinus ) were not counted during these particular surveys. Traits for species included in the analysis are provided in Table 1 1 of the main text. Species n ame a Common n ame b Mass (kg) b Feeding g uild c Ruminant/ n on ruminant Behavior d Hippopotamus amphibius Hippopotamus 2000 Grazer Non ruminant Resident Diceros bicornis Rhinoceros 1200 Browser Non ruminant Resident Hippotragus equines Roan antelope 270 Grazer Ruminant Resident Tragelaphus strepsiceros Greater kudu 213 Browser Ruminant Resident Oryx beisa Fringe eared oryx 169 Mixed* Ruminant Resident Connochaetes taurinus Wild ebeest 170 Grazer Ruminant Migratory Potamochoerus porcus Bushpig 70 Browser* Non ruminant Resident Tragelaphus imberbis Lesser kudu 80 Mixed Ruminant Resident Redunca redunca Bohor reedbuck 60 Grazer* Ruminant Resident Tragelaphus scriptus Bushb uck 51 Browser Ruminant Resident Redunca fulvorufula Mountain reedbuck 30 Grazer Ruminant Resident Ourebia ourebi Oribi 18 Grazer Ruminant Resident Oreotragus oreotragus Klipspringer 12 Browser Ruminant Resident Sylvicapra grimmia Grey duiker 15 Browser Ruminant Resident Raphicerus campestris Steinbuck 10 Browser Ruminant Resident Madoqua kirkii 5 Browser Ruminant Resident Sources: a. Mduma and Hopcraft (2008). b. Sinclair et al. (2003); items marked with from Prins and Olff (1998 ). c. Prez Barbera et al. (2001); items marked with from Kingdon (1997). d. Sinclair et al. (2008).

PAGE 111

111 Appendix A 2. Distribution of M igratory W ildebeest The seasonal migration of the wildebeest is driven by the rainfall gradient and its effects on vegetati on growth and availability (Pennycuick, 1975, Boone et al. 2006, Holdo et al. 2009) During the wet season, when grass production is high in the short grass plains, the wildebeest migrate south (Figure A 1 ; see also Holdo et al. 2009 Figure 4 ) At the st art of the dry season, they move to the Western corridor and then migrate to the north where green grass is still available in the woodlands and tall grass savannas where they then stay during the dry season. With the onset of rain at the start of the we t season, the wildebeest return to the southern grasslands The distributions of migratory wildebeest are shown below for April 1999 (Figure A 1) Abundance data were collected by the Tanzania Wildlife Conservation Monitoring Program using Aerial Point Su rveys (APS), which were timed to take place when the migratory wildebeest are most evenly distributed on the plains. The survey extent was determined by the boundary of the wildebeest distribution. A plane was flown along transects at 5 km spacing and ph otographs were taken at subunits along the transect. These data were used to calculate a density (number per km 2 ) for each 25 km 2 survey grid cell. Detailed metho ds can be obtained elsewhere (Norton Griffiths 1973 Borner et al 1987 Tanzania Wildlife C onservation Monitoring 1994). Total biomass and total BMR were calculated using methods described in the main text. Grid cells where the wildebeest are highly distributed in the southeast appear to have lower species richness and abundances of the twelve species in the analysis, as discussed in the main text. There are also areas where wildebeest are not present in the southeast which appear to have higher species richness and abundances of the other species. It is unclear whether the omission of migrato ry wildebeest from the

PAGE 112

112 analysis greatly altered conclusions from this study. This will be the focus of future inquiry.

PAGE 113

113 A) B) C) D) Figure A 1. The distributions of migratory wildebeest for April 1999: A) presence/absence, B) abundance, C) biomass, D ) basal metabolic rate. CA = conservation area, GR = game reserve, NP = national park, RA = reserve area.

PAGE 114

114 Appendix A 3. Dates for Species and Habitat D ata Table A 4 Dates for static habitat data. Details for these data are discussed in the main text and in Appendix A 5. Road density (km/km 2 ) Elevation (m) River distance (m) Plant nutrients (ppm or Percent) Tree cover (Percent) 2001 2008 2001 1980s Nov 2000 Nov 2001

PAGE 115

115 Table A 5. Dates for species and dynamic habitat data. Details for these data are discussed in the main text and in Appendix A 5. Animal survey year Animal survey dates NDVI Rainfall Average (mm/month) Fire area (Percent) 1988 Mar 9 15 Mar 15, 1988 Jan, Feb 1988 n/a 1989 May 1 Apr 15, 1989 Mar, Apr 1989 n/a 1991 Apr 6 23 Apr 15, 1991 Feb, Mar 1991 n/a 1996 Apr 18 21 Apr 15, 1996 Jan, Feb 1996 n/a 1996 a Nov 18 22 Nov 15, 1996 Sep, Oct 1996 n/a 2000 Mar 15 16 Mar 5, 2000 Jan, Feb 2000 n/a 2001 Apr 1 4 Mar 22, 2001 Feb, Mar 2001 May 2000 April 2001 2003 Apr 26 30 Feb 2, 200 3 Feb, Mar 2003 May 2002 April 2003 2006 Apr 6 9 Apr 7, 2006 Feb, Mar 2006 May 2005 April 2006 a. Dry season survey

PAGE 116

116 Appendix A 4. Spatial Distribution for E ach Community M easure for E ach S urvey Figure A 2 Spatial distribution for each community me asure (for sample locations at 5 km intervals) for eight wet season survey years between 1988 and 2006 : A) species richness, B) total abundance, C) total biomass (kg), D) total basal metabolic rate (W). Patterns are described in the main text. A)

PAGE 117

1 17 B)

PAGE 118

118 C)

PAGE 119

119 D)

PAGE 120

120 A) B) C) D) Figure A 3 Community distributions (for sample locations at 5 km intervals) for a dry season survey in 1996: A) species richness, B) total abundance, C) total b iomass (kg), D) total basal metabolic rate (W)

PAGE 121

121 Appendix A 5. Habitat Characteristics Methods (with F igures). Dates for habitat data are provided in Table A 4 and A 5. Geographic Information System (GIS) analyses to prepare the habitat variables were completed in ArcGIS 9.3.1 (ESRI 2009). Human A ctivity Potential impacts of human activity were represented in the analysis using the following variables: distance (m) from the park western boundary, direction (east or west) from the park western boundar y, their interaction (distance x direction), and road density (km road/km 2 area). The protected area boundaries and distribution of roads are not expected to have changed over the duration of the study; however, human activities may have changed over time and are addressed in the Discussion in the main text. T he distance (m) from the western park boundary was examined to investigate the influence of human activity into the interior of the park. Such influences could be because humans (e.g., poachers) or l ivestock wander into the park. Alternatively, wildlife could wander across the boundary and experience different conditions and the existence of the boundary could constrain movement possibilities. The western boundary of the park has distinct land use a ctivities to its east and west (Figure A 4 a). To the west, there are unprotected areas, which are highly populated, and game reserves which are intended to serve as buffer areas for the park (but do allow hunting). To the east of this boundary, there is the national park, which does not permit livestock or agriculture, and further east there are conservation areas, which allow controlled pastoralism (i.e., ranching). Poacher density is likely to decrease in relation to distance from human population and poaching risk shows a general decrease from west to east

PAGE 122

122 within the park (Metzger et al. 2007). Consequently, the interaction between distance from park western boundary and direction (west or east) was incorporated in the analysis. The distance (m) fro m the western park boundary was calculated for each survey grid cell by finding the distance from the center of the cell to the nearest edge on the western park boundary (Figure A 4 b). Roads, which are associated with activities such as hunting and tourism were mapped by the Tanzania Wildlife Conservation Monitoring Program and Frankfurt Zoological Society The road types are as follows: main ; all weather ; seasonal; tourist ; restricted roads ; and tracks and trails. Road density (km road/km 2 area) was ca lculated for each survey grid cell as the length of roads within a circle of radius 2.5 km (centered 4 c). Topography Topography, which can influence vegetation type (and thus for age quantity and quality) and animal movement more directly, was incorporated in the study as average elevation (m) and standard deviation of elevation (m) (to represent local heterogeneity in terrain) (Figure A 5 ). The average and the standard deviation of the elevation values within each 5 x 5 km survey grid cell were calculated using a digital elevation model with a 90 m 2 resolution obtained from the Shuttle Radar Topography Mission (SRTM) (Jarvis et al. 2008) Resources Resources and disturbances hypot hesized to be important to large herbivores were represented using the following surrogate measures: distance to river (m) distance to permanent river (m), plant nutrients (ppm Ca, ppm Mg, percent N, ppm Na, ppm P) percent tree cover (average and standa rd deviation), N ormalized Difference

PAGE 123

123 Vegetation I ndex ( NDVI) (average and standard deviation), rainfall (mm/month) and percent burn area. The analysis used an interaction between each of the dynamic variables (NDVI, rainfall, and burn area) and date (tr eated as a categorical variable). Proximity to rivers (m), which indicates water availability, was determined using a digital map of r ivers prepared by the Tanzania Wildlife Conservation Monitoring Program and Frankfurt Zoological Society. The distributio n of rivers is not expected to have changed over the study period. D istance to the nearest river (seasonally and permanently flowing rivers) as well as distance to the nearest permanent river were calculated using the center of each 5 x 5 km survey grid c ell (Figure A 6 ). Plant nutrients (i.e., plant biomass nutrient concentrations), which are associated with vegetation quality, were determined for calcium, magnesium, nitrogen, phosphorus, and sodium. Methods are explained in Holdo et al. (2009); the plan t nutrient data were collected by S.J. McNaughton (unpublished data) at 115 sites and interpolated by R.D. Holdo to create plant nutrient maps, which he kindly provided for this study (Figure A 7 ). Holdo et al. (2009) suggest the data were collected in th e 1980s. Based on comparisons with subsequent independent nutrient studies, they conclude that general spatial patterns of nutrients have not changed considerably over the approximate period of 1980 2005. Percent tree cover, which indicates the amount of woody vegetation available and thus forage for browsers (and cover from predators), was estimated using the Vegetation Continuous Fields (VCF), Collection 3 original, a product derived from Moderate Resolution Imaging Spectroradiometer (MODIS) Terra satell ite data (Hansen et al. 2003). The VCF data indicates the percent of each 500 x 500 m pixel covered by

PAGE 124

124 woody vegetation as an annual representation for the period Nov 2000 Nov 2001. The percent tree cover was determined for each 5 x 5 km survey grid by finding the mean of the 100 VCF pixel values within each survey grid. To represent local heterogeneity in woody cover, the standard deviation of the VCF values within each 5 x 5 km survey gird was also determined (Figure A 8 ). Although general patterns of tree cover have not changed appreciably during the duration of the study, tree densities have increased (Sinclair et al. 2007); such changes in tree cover may be captured by another variable in the study, NDVI. The Normalized D ifference Vegetation I ndex (NDVI), an indicator of plant productivity, was used to represent an index of food availability for herbivores. NDVI data were obtained from two sources. For survey years prior to 2000, imagery was obtained from Advanced Very High Resolution Radiometer Global Inventory Modeling and Mapping Studies ( AVHRR GIMMS ) on the NOAA satellite series 7, 9, 11, 14, 16 and 17 at an 8 x 8 km resolution (Tucker et al. 2004). The center of each 5 x 5 km survey grid cell was used to extract a corresponding NDVI value (F igure A 9 ). For survey years 2000 and onwards, MODIS data were obtained as 16 day composite measures at a 250 x 250 m resolution (NASA n.d. ). For these years, the average of the NDVI values within each 5 x 5 km survey grid cell was calculated (Figure A 9 ) as well as the standard deviation of the NDVI values (to represent local heterogeneity) (Figure A 10 ). For both sources, the NDVI monthly time range selected was for the dates closest to that of the survey dates (Appendix A 2). NDVI values range betwee n 1.0 to 1.0, where increasing positive values indicate increasing amounts of green vegetation.

PAGE 125

125 Rainfall, which influences vegetation type, was acquired as maps from The University of Delaware Terrestrial Precipitation: 1900 2008 Gridded Monthly Time Seri es Version 2.01 (Matsuura and Willmott 2009). The rainfall data are monthly total precipitation interpolated to a 0.5 x 0.5 degree grid. A subset of the data was selected for the spatial extent of the study region. NDVI/vegetation has a 1 2 month lag in response to rainfall in savanna systems (Chamaille Jammes et al. 2006, Shinoda 1995); thus, the average rainfall of the two months preceding the month of the survey data was determined (Appendix A 2). These values were interpolated using universal krigin g with a first order trend to a spatial resolution of approximately 1 x 1 km. The rainfall (mm/month) for each 5 x 5 km survey grid cell was selected using the corresponding value at the center of the cell (Figure A 1 1 ). Fire, which has a prominent effect on savanna vegetation structure, was represented using fire maps for 2000 2006 produced by Dempewolf et al. (2007), who identified burned areas by applying an algorithm that uses red and near infrared reflectance values at a 250 x 250 m spatial resolution acquired by MODIS sensor. For the surveys in years 2001, 2003, and 2006 (which occurred in the wet season), maps were created for fire occurrence within the previous year (i.e., May of the prior year through April of the survey year) (Appendix A 2) As t here are few fires in the wet season the data essentially indicate fire oc currence in the prior dry season For each 5 x 5 km survey grid cell, the proportion of 250 x 250 m cells with fire present was used as a proxy for percent burn (i.e., area of eac h 5 x 5 km survey grid cell burned wit hin the prior year) (Figure A 12 ).

PAGE 126

126 A) B) C) Figure A 4 Serengeti National Park (~15,000 km 2 ) and its neighboring conservation areas and game reserves located in Tanzania and Kenya East Africa. Indicators of human activity at each survey location: A ) p rotectio n class, B ) distance (m) from western park boundary (white line is the boundary) and C) r oad density (km/km 2 ) CA = conservation area, GR = game reserve, NP = national park, RA = reserve area.

PAGE 127

127 A) B) Figure A 5 Topography of the study area measured by A) average elevation (m) and B) standard deviation of elevation (m) CA = conservation area, GR = game reserve, NP = national park, RA = reserve area.

PAGE 128

128 A) B) Figure A 6 The distance (m) from each survey location to A) the nearest river and B) to the nearest permanent river. CA = conservation area, GR = game reserve, NP = national park, RA = reserve area.

PAGE 129

129 A) B) C) D) E) Figure A 7 Plant nutrient maps of: A) percent N, B) ppm Ca, C) ppm Mg, D) ppm Na, E) ppm P. CA = conservation area, GR = game reserve, NP = national park, RA = reserve area.

PAGE 130

130 A) B) Figure A 8 Tree cover: A) mean percent tree cover and B standard deviation of percent tree cover within each 5 x 5 km grid cell. C A = conservation area, GR = game reserve, NP = national park, RA = reserve area.

PAGE 131

131 Figure A 9 NDVI from approximately the month of each survey. A) 1988, B) 1989, C) 1991, D) 1996, E) 1996 (dry season), F) 2000, G) 2001, H) 2003, I) 2006. CA = conserva tion area, GR = game reserve, NP = national park, RA = reserve area. A) B) C)

PAGE 132

132 D) E) F) G) H) I)

PAGE 133

133 A) B) C) D) Figure A 10 Standard deviation of NDVI values within each 5 x 5 km grid cell for four wet season surveys. A) 2000, B) 2001, C) 2003, D) 2006. CA = conservation area, GR = game reserve, NP = national park, RA = reserve area.

PAGE 134

134 Figure A 1 1 Average rainfall (mm/month) for the two months preceding each survey date. A) 1988, B) 1989, C) 1991, D) 1996, E) 1996 (dry season), F) 2000, G) 200 1, H) 2003, I) 2006. CA = conservation area, GR = game reserve, NP = national park, RA = reserve area. A) B) C)

PAGE 135

135 D) E) F) G) H) I)

PAGE 136

136 A) B) C) Figure A 1 2 Percent burn (area within each grid cell burned) within the prior year: A) May 2000 April 2001, B ) May 2002 April 2003, C ) May 2005 April 2006. CA = conservation area, GR = game reserve, NP = national park, RA = reserve area.

PAGE 137

137 Appendix A 6. Model S election and Model O utput for Subset of D ata C onsisting of T hree W et Season S urve ys Detailed methods and results are discussed in the main text. The analysis was initially limited to three survey years (2001, 2003, and 2006), since data for percent burn and heterogeneity of NDVI (i.e., NDVI standard deviation) were only available for t hese years. Note that in addition to the correlated variables specified in the methods, rain was also not included in this initial analysis due to collinearity within the three year dataset. Percent burn and NDVI SD were not found to be significant for a ny of the aggregate community measures (Table A 6 A 10 ). Therefore, the analysis was continued using data from all eight wet season surveys, with these two variables excluded as predictors.

PAGE 138

138 Table A 6 Model selection using AIC for model s with specie s richness, total abundance, total biomass, and total BMR as the response variable. Candidate models were linear regression (LR) and generalized least squares (GLS) models incorporating spatial correlation. Data from three wet season surveys were include d in this analysis. Candidate model a Spatial coordinates Spatial correlation structure K b Species richness GLS Second order Exponential 33 0.0 GLS Second order Rational quadratic 33 0.1 GLS Second order Gaussian 33 6.4 GLS Second order Spherical 33 8.2 GLS Second order Linear 33 33.6 LR Second order n/a 31 127.7 LR First order n/a 28 149.2 LR No spatial coordinates n/a 26 212.3 Total abundance GLS Second order Gaussian 33 0.0 GLS Second order Spherical 33 1.0 GLS Second order Exponential 33 11.8 GLS Second order Rational quadratic 33 17. 9 GLS Second order Linear 33 19.8 LR Second order n/a 31 188.7 LR First order n/a 28 201.6 LR No spatial coordinates n/a 26 211.8 Total biomass GLS Second order Exponential 33 0.0 GLS Second order Gaussian 33 0.9 GLS Second ord er Linear 33 1.8 GLS Second order Spherical 33 1.8 GLS Second order Rational quadratic 33 4.3 LR Second order n/a 31 21.9 LR First order n/a 28 57.9 LR No spatial coordinates n/a 26 63.5 Total BMR GLS Second order Gaussian 33 0.0 GLS Second order Spherical 33 0.1 GLS Second order Linear 33 1.8 GLS Second order Exponential 33 3.3 GLS Second order Rational quadratic 33 10.7 LR Second order n/a 31 51.5 LR First order n/a 28 79.0 LR No spatial coordinates n/a 26 84.8 a. T he linear regression models include spatial coordinates as predictors (none, linear, or second order). The generalized least squares models implement second order spatial coordinates (as determined from LR model) and a spatial correlation error structure (exponential, Gaussian, linear, rational quadratic, or spherical). b. considered the threshold to be included in the set of best approximating models.

PAGE 139

139 Table A 7 Model output for species richness for analysis of three wet season s urveys. The best model determined was a GLS model including second order spatial coordinates as predictors and an exponential spatial correlation structure. p < 0.05 ; p < 0. 0 1 ; ** p < 0. 00 1. Parameter Coefficient SE t p Intercept 2.788 1.562 1.78 0.07 Human activity Direction w est 0.318 0.131 2.43 0.02 Distance from west boundary 1.408 E 06 7.015 E 06 0.20 0.84 Direction w est x d istance 4.039 E 05 1.768 E 05 2.29 0.02 Road density 0.176 0.160 1.10 0.2 7 Topography Elevation a verage 0.003 0.0008 3.40 < 0.001 *** Elevation SD 0.004 0.002 1.50 0.13 Resources Distance to river 4.387 E 06 7.976 E 06 0.55 0.58 Plant P 0.0003 0.0001 3.24 < 0.01 ** Plant Na 1.530 E 05 4.108 E 05 0.37 0.71 Plant Mg 0.0007 0.0006 1.31 0.19 Plant Ca 0.0001 0.0001 1.02 0.32 Tree c over 0.025 0.015 1.67 0.10 Tree c over SD 0.052 0.028 1.86 0.06 Percent burn 0.049 0.238 0.20 0.84 NDVI 0.571 0.972 0.59 0.56 NDVI SD 0.500 3.157 0.16 0.87 Date Dat e 2001 1.200 0.814 1.47 0.14 Date 2006 2.341 0.956 2.45 0.01 Percent burn x 2001 0.477 0.379 1.26 0.21 Percent burn x 2006 0.400 0.336 1.19 0.23 NDVI x 2001 1.642 1.241 1.32 0.19 NDVI x 2006 2.203 1.335 1.65 0.10 NDVI SD x 2001 7 .417 4.249 1.75 0.08 NDVI SD x 2006 7.159 4.263 1.68 0.09 Spatial coordinates Eastings 9.413 E 06 5.246 E 06 1.79 0.07 Northings 2.990 E 06 2.551 E 06 1.17 0.24 Eastings n orthings 1.579 E 10 6.986 E 11 2.26 0.02 Eastings 2 1.019 E 10 4.7 51 E 11 2.14 0.03 Northings 2 1.420 E 11 2.416 E 11 0.59 0.56

PAGE 140

140 Table A 8 Model output for total abundance for analysis of three wet season surveys. The best model determined was a GLS model including second order spatial coordinates as predict ors and a Gaussian spatial correlation structure. p < 0.05 ; p < 0. 0 1 ; ** p < 0. 00 1. Parameter Coefficient SE t p Intercept 87.790 81.797 1.07 0.28 Human activity Direction w est 2.348 7.172 0.33 0.74 Distance fr om west boundary 0.001 0.0003 3.50 < 0.001 *** Direction west x d istance 0.003 0.0009 3.13 < 0.01 ** Road density 11.842 8.724 1.36 0.17 Topography Elevation a verage 0.053 0.042 1.27 0.21 Elevation SD 0.013 0.134 0.10 0.92 Resources Dist ance to river 0.0002 0.0004 0.53 0.60 Plant P 0.015 0.004 3.79 < 0.001 *** Plant Na 0.0001 0.002 0.07 0.95 Plant Mg 0.020 0.029 0.68 0.50 Plant Ca 0.022 0.007 2.94 < 0.01 ** Tree c over 1.470 0.835 1.76 0.08 Tree c over SD 0.649 1.510 0.43 0.67 Percent burn 2.390 12.959 0.18 0.85 NDVI 30.906 51.759 0.60 0.55 NDVI SD 175.091 172.408 1.02 0.31 Date Date 2001 32.856 42.699 0.77 0.44 Date 2006 66.240 49.174 1.35 0.18 Percent burn x 2001 0.156 20.625 0.01 0.99 Perce nt burn x 2006 27.256 18.265 1.49 0.14 NDVI x 2001 21.643 64.964 0.33 0.74 NDVI x 2006 68.262 68.981 0.99 0.32 NDVI SD x 2001 225.267 232.503 0.97 0.33 NDVI SD x 2006 123.959 233.335 0.53 0.60 Spatial coordinates Eastings 0.0006 0.0003 2.28 0.02 Northings 0.0003 0.0001 2.20 0.03 Eastings n orthings 6.281 E 09 3.661 E 09 1.72 0.09 Eastings 2 1.720 E 09 2.396 E 09 0.72 0.47 Northings 2 3.420 E 09 1.202 E 09 2.84 < 0.01 **

PAGE 141

141 Table A 9 Model output for total biomass for analysis of three wet season surveys. The best model determined was a GLS model including second order spatial coordinates as predictors and an exponential spatial correlation structure. p < 0.05 ; p < 0. 0 1 ; ** p < 0. 00 1. Parameter Coefficient SE t p Intercept 27491.920 19680.470 1.40 0.16 Human activity Direction w est 2224.134 1856.909 1.20 0.23 Distance from west boundary 0.329 0.082 4.00 < 0.001 *** Direction w est x d istance 0.959 0.230 4.18 < 0.001 *** Road density 3935.020 2362.829 1.67 0.10 Topography Elev ation a verage 20.215 10.039 2.01 0.04 Elevation SD 13.037 35.851 0.36 0.72 Resources Distance to river 0.031 0.102 0.30 0.76 Plant P 3.695 0.930 3.97 < 0.001 *** Plant Na 0.053 0.526 0.10 0.92 Plant Mg 8.520 6.847 1.24 0.21 Plant Ca 2.110 1.702 1.24 0.22 Tree c over 298.776 220.457 1.36 0.18 Tree c over SD 5.035 408.942 0.01 0.99 Percent burn 1181.020 3360.571 0.35 0.73 NDVI 22503.970 12881.050 1.75 0.08 NDVI SD 30536.040 46327.440 0.66 0.51 Date Date 2001 9276.993 10620.630 0.87 0.38 Date 2006 2362.920 12072.320 0.20 0.84 Percent burn x 2001 2391.930 5275.664 0.45 0.65 Percent burn x 2006 5861.108 4673.416 1.25 0.21 NDVI x 2001 17191.600 1 5986.070 1.08 0.28 NDVI x 2006 5296.906 16879.150 0.31 0.76 NDVI SD x 2001 12582.800 62160.720 0.20 0.84 NDVI SD x 2006 3460.328 62579.820 0.06 0.96 Spatial coordinates Eastings 0.103 0.064 1.63 0.10 Northings 0.030 0.032 0.95 0.34 Ea stings n orthings 2.0004 E 06 8.875 E 07 2.25 0.02 Eastings 2 1.622 E 07 5.653 E 07 0.29 0.77 Northings 2 1.542 E 06 2.847 E 07 5.42 < 0.001 ***

PAGE 142

142 Table A 10 Model output for total basal metabolic rate for analysis of three wet season surveys. The best model determined was a GLS model including second order spatial coordinates as predictors and a Gaussian spatial correlation structure. p < 0.05 ; p < 0. 0 1 ; ** p < 0. 00 1. Parameter Coefficient SE t p Inter cept 22128.500 14448.330 1.53 0.13 Human activity Direction w est 1339.873 1353.763 0.99 0.32 Distance from west boundary 0.228 0.060 3.79 < 0.001 *** Direction w est x d istance 0.637 0.168 3.79 < 0.001 *** Road density 2696.205 1707.673 1.58 0.11 Topography Elevation a verage 12.228 7.371 1.66 0.10 Elevation SD 10.836 26.034 0.42 0.68 Resources Distance to river 0.019 0.075 0.25 0.80 Plant P 2.872 0.685 4.19 < 0.001 *** Plant Na 0.014 0.387 0.04 0.97 Plant Mg 6.005 5.042 1.19 0.23 Plant Ca 2.045 1.253 1.63 0.10 Tree c over 255.786 160.137 1.60 0.11 Tree c over SD 69.193 295.811 0.23 0.82 Percent burn 724.784 2447.511 0.30 0.77 NDVI 12785.790 9426.384 1.36 0.18 NDVI SD 4033.690 33548.960 0.12 0.90 Date Date 2001 4621.566 7762.390 0.60 0.55 Date 2006 841.541 8827.841 0.10 0.92 Percent burn x 2001 1383.450 3851.145 0.36 0.72 Percent burn x 2006 4526.173 3411.712 1.33 0.18 NDVI x 2001 11263.460 11703.040 0.96 0.34 NDVI x 2006 1840.03 9 12354.910 0.15 0.88 NDVI SD x 2001 17883.550 45077.090 0.40 0.69 NDVI SD x 2006 31241.350 45368.190 0.69 0.49 Spatial coordinates Eastings 0.084 0.047 1.80 0.07 Northings 0.030 0.023 1.28 0.20 Eastings n orthings 1.191 E 06 6.510 E 07 1. 83 0.07 Eastings 2 8.079 E 08 4.153 E 07 0.19 0.85 Northings 2 9.901 E 07 2.088 E 07 4.74 < 0.001 ***

PAGE 143

143 Appendix A 7. Co rrelations Among the Community M easures Figure A 1 3 C orrelation s among the community measures for each survey. A) 1988, B) 19 89, C) 1991, D) 1996, E) 1996 (dry season), F) 2000, G) 2001, H) 2003, I) 2006. Patterns are described in the main text. A) B) C) D)

PAGE 144

144 E) F) G) H)

PAGE 145

145 I)

PAGE 146

146 Table A 1 1 C orrelation s over time for each of the community measures. Community Measure 1989 1991 1996 1996 (dry) 2000 2001 2003 2006 Species Richness 1988 0.26 0.19 0.23 0.16 0.21 0.3 0.16 0.19 1989 0.25 0.33 0.07 0.12 0.31 0.22 0.34 1991 0.27 0.08 0.14 0.24 0.18 0.29 1996 0.12 0.13 0.31 0.24 0.4 1996 (dry) 0.77 0.08 0 0.0 8 2000 0.13 0.05 0.08 2001 0.19 0.33 2003 0.24 Total Abundance 1988 0.15 0.11 0.08 0.07 0.08 0.12 0.03 0.05 1989 0.33 0.28 0.01 0.02 0.31 0.26 0.18 1991 0.19 0.01 0 0.28 0.18 0.13 1996 0.03 0.05 0.29 0.2 5 0.31 1996 (dry) 0.89 0 0.04 0.05 2000 0 0.06 0.05 2001 0.19 0.18 2003 0.17 Total Biomass 1988 0.03 0.03 0.02 0 0 0.09 0.03 0.03 1989 0.05 0.11 0.03 0.04 0.05 0.02 0.02 1991 0.05 0.01 0.03 0.11 0.0 4 0.01 1996 0.02 0.02 0.1 0.02 0.04 1996 (dry) 0.9 0.02 0.02 0.01 2000 0.01 0.01 0 2001 0.01 0.04 2003 0.06 Total BMR 1988 0.05 0.05 0.02 0 0.01 0.07 0.02 0.04 1989 0.33 0.28 0.01 0.02 0.31 0.26 0. 18 1991 0.19 0.01 0 0.28 0.18 0.13 1996 0.03 0.05 0.29 0.25 0.31 1996 (dry) 0.89 0 0.04 0.05 2000 0 0.06 0.05 2001 0.19 0.18 2003 0.17

PAGE 147

147 APPENDIX B SUPPORTING INFORMATI ON FOR CHAPTER 3 Figure B 1. A bundance (average abundance of occupied sites ) in relation to occupancy for a census ( n = 730 grid cells in a census ) Nine censuses are plotted for each species other than wildebeest (eight wet season, one dry season). Three censuses are plotted for wildebeest (three wet season).

PAGE 148

148 Figure B 2. Average abundance at occupied sites in relation to occupancy for each species. Each point on the graph is a census (number of grid cells in a census: n = 730). Nine censuses are plotted for each species other than wil debeest (eight wet season, one dry season). Three censuses are plotted for wildebeest (three wet season). A) buffalo, B) eland, C) elephant, D) topi, J) warthog, K) waterbuck, L ) wildebeest, M) zebra. A) B) C) D)

PAGE 149

149 E) F) G) H)

PAGE 150

150 I) J) K) L)

PAGE 151

151 M)

PAGE 152

152 Figure B 3. Estimates for parameters m and k (indicating mean and clustering respectively) of negative binomial fit to (n = 730 grid cells in a census ) Nine censuses are plotted for each species other than wildebeest (eight wet season, one dry season). Three censuses are plotted for wildebeest (three wet season). A) estimates of m u in relation A)

PAGE 153

153 B)

PAGE 154

154 C)

PAGE 155

155 APPENDIX C SUPPORTING INFORMATI ON FOR CHAPTER 4 Figure C 1. R ank abundance plot (the species abundance is depicted on the y axis and the rank order of species from most to least abundant is depicted on the x axis) for twelve large herbivore species for each survey. A) 1988, B) 1989, C) 1991 D) 1996 E) 1996 (dry season) F ) 200 0 G ) 2001 H ) 200 3, I) 2006 A) B) C) D)

PAGE 156

156 E) F) G ) H) I )

PAGE 157

157 LIST OF REFERENCES Anderson, T. M., J. G. C. Hopcraft, S. Eby, M. Ritchie, J. B. Grace, and H. Olff. 2010. Landscape scale analyses suggest both nutrient and antipredator advantages to Serengeti herbivore hotspots. Ecology 91 :1519 29. Anderson, T. M., M. E. Ritchie, E. Mayemba, S. Eby, J. B. Grace, and S. J. McNaughton. 2007. Forage nutritive quality in the Serengeti ecosystem: the roles of fire and herbivory. The American Naturalist 170 :343 57. Anjos, L. D., C. D. C ollins, R. D. Holt, G. H. Volpato, L. B. Mendona, E. V. Lopes, R. Boon, M. V. Bisheimer, P. P. Serafini, and J. Carvalho. 2011. Bird species abundance occupancy patterns and sensitivity to forest fragmentation: Implications for conservation in the Brazil ian Atlantic forest. Biological Conservation 144 :2213 2222. Baskerville, E. B., A. P. Dobson, T. Bedford, S. Allesina, T. M. Anderson, and M. Pascual. 2011. Spatial guilds in the Serengeti food web revealed by a Bayesian group model. PLoS Computational Bio logy 7 :e1002321. Bell, R. H. V. 1971. A grazing ecosystem in the Serengeti. Scientific American 225 :86 93. Bock, C. E., Z. F. Jones, and J. H. Bock. 2007. Relationships between species richness, evenness, and abundance in a southwestern savanna. Ecology 88 :1322 7. Bolker, B. M. 2008. Ecological models and data in R Princeton University Press, Princeton New Jersey, USA Boone, R. B., S. J. Thirgood, and J. G. C. Hopcraft. 2006. Serengeti wildebeest migratory patterns modeled from rainfall and new vegetat ion growth. Ecology 87 :1987 94. Borner, M., C. D. FitzGibbon, M. Borner, T. M. Caro, W. K. Lindsay, D. A. Collins, and M. E. Holt. 1987. The decline of the Serengeti Thomson's gazelle population. Oecologia 73 : 32 40. Brown, J. H., D. W. Mehlman, and G. C Stevens. 1995. Spati a l variation in abundance. Ecology 76 :2028 2043. Burnham, K.P., and D.R. Anderson. 1998. Model selection and inference: a practical information theoretic approach. Springer Verlag, New York, New York, USA. Campbell, K., and M. Borner. 1995. Population trends and distribution of Serengeti herbivores: implications for management. Pages 117 145 in A. R. E. Sinclair and P. Arcese, editors. Serengeti II: Dynamics, Management and Conservation of an Ecosystem. University of Chicago Press, Chi cago Illinois, USA

PAGE 158

158 Chamaille Jammes, S., H. Fritz, and F. Murindagomo. 2006. Spatial patterns of the NDVI rainfall relationship at the seasonal and interannual time scales in an African savanna. International Journal of Remote Sensing 27 :5185 5200. Coe, M. J., D. H. Cumming, and J. Phillipson. 1976. Biomass and production of large African herbivores in relation to rainfall and primary production. Oecologia 22 :341 354. Collins, C. D., R. D. Holt, and B. L. Foster. 2009. Patch size effects on plant species decline in an experimentally fragmented landscape. Ecology 90 :2577 88. Colwell, R.K. 2011. Biogeographical gradient theory. The t heory of e cology (ed. by S.M. Scheiner and M.R. Willig), pp. 309 330. University of Chicago Press, Chicago Illinois, USA C olwell, R. K., and D. C. Lees. 2000. The mid domain effect: geometric constraints on the geography of species richness. Trends in Ecology and Evolution 15 :70 76. Crawley, M. J. 2007. The R book. Wiley, Hoboken New Jersey, USA Cromsigt, J. P. G. M., and H Olff. 2006. Resource partitioning among savanna grazers mediated by local heterogeneity: an experimental approach. Ecology 87 :1532 41. Cromsigt, J. P. G. M ., S. J. van Rensburg, R. S. Etienne, and H. Olff. 2009 a Monitoring large herbivore diversity at d ifferent scales: comparing direct and indirect methods. Biodiversity and Conservation 18 :1219 1231. Cromsigt, J. P. G. M., H. H. T. Prins, and H. Olff. 2009 b Habitat heterogeneity as a driver of ungulate diversity and distribution patterns: interaction of body mass and digestive strategy. Diversity and Distributions 15 :513 522. Demment, M. W., and P. J. Van Soest. 1985. A nutritional explanation for body size patterns of ruminant and nonruminant herbivores. The American Naturalist 125 :641 672. Dempewolf, J ., S. Trigg, R. S. DeFries, and S. Eby. 2007. Burned a rea mapping of the Serengeti Mara r egion using MODIS r eflectance d ata. IEEE Geoscience and Remote Sensing Letters 4 :312 316. Dobson, A. 2009. Food web structure and ecosystem services: insights from the Serengeti. Philosophical transactions of the Royal Society of London. Series B, Biological sciences 364 :1665 82.

PAGE 159

159 Dobson, A. P., M. Borner, A. R. E. Sinclair, P. J. Hudson, T. M. Anderson, G. Bigurube, T. B. B. Davenport, J. Deutsch, S. M. Durant, R. D. Estes, A. B. Estes, J. Fryxell, C. Foley, M. E. Gadd, D. Haydon, R. Holdo, R. D. Holt, K. Homewood, J. G. C. Hopcraft, R. Hilborn, G. L. K. Jambiya, M. K. Laurenson, L. Melamari, A. O. Morindat, J. O. Ogutu, G. Schaller, and E. Wolanski. 2010. Road will ru in Serengeti. Nature 467 :272 3. du Toit, J., and D. Cumming. 1999. Functional significance of ungulate diversity in African savannas and the ecological implications of the spread of pastoralism. Biodiversity and Conservation 8 :1643 1661. ESRI. 2009 ArcMap 9.3.1. Environmental Systems Resource Institute, Redlands, California. Evans, K. L., P. H. Warren, and K. J. Gaston. 2005. Species energy relationships at the macroecological scale: a review of the mechanisms. Biological reviews of the Cambridge Philosoph ical Society 80 :1 25. Fox, J. I am grateful to Bates, D., Firth, D., Friendly, M., Gorjanc, G., Graves, S., Heiberger, R., Monette, G., Nilsson, H., Ogle, D., Ripley, B., Weisberg, S., and Zeileis, A. for various suggestions and contributions. 2009. car: c ompanion to a pplied r egression. R package version 1.2 16. http://CRAN.R project.org/package=car Fritz, H., and P. Duncan. 1994. On the carrying capacity for large ungulates of African savanna ecosystems. Proceedings: Biological Sciences 256 :77 82. Fryxell, J. M., P. A. Abrams, R. D. Holt, J. F. Wilmshurst, A. R. E. Sinclair, and R. Hilborn. 2008. Spatial dynamics and coexistence of the Serengeti grazer community. Pages 277 300 in A. R. E. Sinclair, C. Pack er, S. A. R. Mduma, and J. M. Fryxell, editors. Serengeti III: Human Impacts on Ecosystem Dynamics. The University of Chicago Press, Chicago, Illinois, USA. Fryxell, J. M., J. F. Wilmshurst, and A. R. E. Sinclair. 2004. Predictive m odels of movement By Ser engeti grazers. Ecology 85 :2429 2435. Gaston, K. J., T. M. Blackburn, J. J. D. Greenwood, R. D. Gregory, R. M. Quinn, and J. H. Lawton. 2000. Abundance occupancy relationships. Journal of Applied Ecology 37 :39 59. Gause, G.F. 1934. The struggle for existen ce. Williams and Wilkins, Baltimore, Maryland, USA. Grange, S., P. Duncan, J. M. Gaillard, A. R. E. Sinclair, P. J. P. Gogan, C. Packer, H. Hofer, and M. East. 2004. What limits the Serengeti zebra population? Oecologia 140 :523 32.

PAGE 160

160 Hansen, M., DeFries, R., Townshend, J. R., Carroll, M., Dimiceli, C., and Sohlberg, R. 2003. Vegetation continuous f ields MOD44 B, 2001 p ercent tree c over, Collection 3, University of Maryland, College Park, Maryland, 2001. Hardin, G. 1960. The c ompetitive e xclusion p rinciple. Sci ence 131 :1292 1297. Hassan, S. N., G. M. Rusch, H. Hytteborn, C. Skarpe, and I. Kikula. 2008. Effects of fire on sward structure and grazing in western Serengeti, Tanzania. African Journal of Ecology 46 :174 185. Hilborn, R., P. Arcese, M. Borner, J. Hando, G. Hopcraft, M. Loibooki, S. Mduma, and A. R. E. Sinclair. 2006. Effective enforcement in a conservation area. Science 314 :1266. Holdo, R. M., J. M. Fryxell, A. R. E. Sinclair, A. Dobson, and R. D. Holt. 2011. Predicted impact of barriers to migration on the Serengeti wildebeest population. PloS One 6 :e16370. Holdo, R. M., R. D. Holt, and J. M. Fryxell. 2009. Opposing rainfall and plant nutritional gradients best explain the wildebeest migration in the Serengeti. American Naturalist 173 :431 445. Holt, R.D. 1977. Predation, apparent competition, and the structure of prey communities. Theoretical Population Biology 12 :197 229. Holt, R.D. 2001. Species coexistence. Pages 413 426 in S. Levin, ed. The Encyclopedia of Biodiversity, Vol. 5. Academic Press. Homewood, K., E. F. Lambin, E. Coast, A. Kariuki, I. Kikula, J. Kivelia, M. Said, S. Serneels, and M. Thompson. 2001. Long term changes in Serengeti Mara wildebeest and land cover: pastor alism, population, or policies? Proceedings of the National Academy of Sciences of the United States of America 98 :12544 9. Hopcraft, J. G. C., T. M. Anderson, S. Prez Vila, E. Mayemba, and H. Olff. 2012. Body size and the division of niche space: food an d predation differentially shape the distribution of Serengeti grazers. The Journal of Animal Ecology 81 :201 13. Hopcraft, J. G. C., A. Sinclair, and C. Packer. 2005. Planning for success: Serengeti lions seek prey accessibility rather than abundance. Jour nal of Animal Ecology 74 :559 566. Hubbell, S. P. 2001. The Unified Neutral Theory of Biodiversity and Biogeography. Princeton University Press, Princeton, N ew Jersey, USA. Hutchinson, G. E. 1957. Animal ecology and demography concluding remarks. Cold Spr ing Harbor Symposia on Quantitative Biology 22 :415 427.

PAGE 161

161 Jarman, P. J. 197 4 The s ocial o rganisation of a ntelope in r elation to t heir e cology. Behaviour 48 :215 267. Jarvis, A., Reuter, H. I., Nelson, A., and Guevara, E. 2008. Hole filled seamless SRTM data V4, International Centre for Tropical Agriculture (CIAT), available from http://srtm.csi.cgiar.org Kingdon, J. 1997. The Kingdon field guide to African mammals. Academic Press, London England, UK Kleynhans, E. J., A. E. Jolles, M. R. E. Bos, and H. Olff. 2011. Resource partitioning along multiple niche dimensions in differently sized African savanna grazers. Oikos 120 :591 600. Klop, E., and H. H. T. Prins. 2008. Diversity and species composition of West African ung ulate assemblages: effects of fire, climate and soil. Global Ecology and Biogeography 17 :778 787. Klop, E., and J. van Goethem. 2008. Savanna fires govern community structure of ungulates in Bnou National Park, Cameroon. Journal of Tropical Ecology 24 :39 47. Legendre, P. 1993. Spatial autocorrelation: trouble or new paradigm? Ecology 74 :1659 1673. Legendre, P. and L. Legendre. 1998. Numerical ecology 2nd edition. Elsevier, Amsterdam, Netherlands. Lichstein, J. W., T. R. Simons, S. A. Shriner, and K. E. F ranzreb. 2002. Spatial a utocorrelation and autoregressive m odels in e cology. Ecological Monographs 72 :445 463. Lindenmayer, D., and M. Hunter. 2010. Some guiding concepts for conservation biology. Conservation Biology 24 :1459 1468. MacArthur, R. H. 1972. G eographical ecology: patterns in the distribution of species. Princeton University Press, Princeton, New Jersey, USA. MacArthur, R., H. Recher, and M. Cody. 1966. On the relation between habitat selection and species diversity. The American Naturalist 100 : 319 332. Magurran, A. E. 2004 Measuring b iological d iversity. Blackwell Publishing Oxford, England, UK Martinko, E. a., R. H. Hagen, and J. a. Griffith. 2006. Successional change in the insect community of a fragmented landscape. Landscape Ecology 21 :711 721.

PAGE 162

162 Matsuura, K., and Willmott, C. J. 2009. Terrestrial p recipitation: 1900 2008 g ridded monthly t ime s eries v ersion 2.01. May, R.M. 1975. Patterns of species abundance and diversity. In Ecology and evolution of communities (ed. M.L. Cody and J.M. Di amond), pp. 81 120. Harvard University Press, Cambridge, Massachusetts, USA. McCune, G. and J. B. Grace. 2002. Analysis of ecological communities. MjM software Design, Gleneden Beach, Oregon, USA. McGill, B. J., B. J. Enquist, E. Weiher, and M. Westoby. 20 06. Rebuilding community ecology from functional traits. Trends in Ecology and Evolution 21 :178 185. McNaughton, S. 1985. Ecology of a grazing ecosystem: the Serengeti. Ecological Monographs 55 :259 294. Mduma, S. A. R., and J. G. C. Hopcraft. 2008. Appendi x: The main herbivorous mammals and crocodiles in the greater Serengeti ecosystem. Pages 497 505 in A. R. E. Sinclair, C. Packer, S. A. R. Mduma, and J. M. Fryxell, editors. Serengeti III: h uman impacts on ecosystem dynamics. University of Chicago Press, C hicago Illinois, USA Metzger, K., A. Sinclair, K. Campbell, R. Hilborn, J. Hopcraft, S. Mduma, and R. Reich. 2007. Using historical data to establish baselines for conservation: The black rhinoceros (Diceros bicornis) of the Serengeti as a case study. Bi ological Conservation 139 :358 374. Metzger, K. L., A. R. E. Sinclair, R. Hilborn, J. G. C. Hopcraft, and S. A. R. Mduma. 2010. Evaluating the protection of wildlife in parks: the case of African buffalo in Serengeti. Biodiversity and Conservation 19 :3431 3 444. Mittelbach, G., C. Steiner, S. Scheiner, K. Gross, H. L. Reynolds, R. B. Waide, M. R. Willig, S. I. Dodson, and L. Gough. 2001. What is the observed relationship between species richness and productivity? Ecology 82 :2381 2396. Motomura, I. 1932. On th e statistical treatment of communities. Zool. Mag. Tokyo 44 :379 383 (in Japanese). Murray, M. G., and D. Brown. 1993. Niche s eparation of grazing u ngulates in the Serengeti: a n e xperimental t est. Journal of Animal Ecology 62 :380 389. Murray, M. G., and A. W. Illius. 2000. Vegetation modification and resource competition in grazing ungulates. Oikos 89 :501 508. Musiega, D. E., and S. N. Kazadi. 2004. Simulating the East African wildebeest migration patterns using GIS and remote sensing. African Journal of Ec ology 42 :355 362.

PAGE 163

163 NASA. n.d. Land Processes Distributed Active Archive Center (LP DAAC). MODIS. Nelson, F., and A. Agrawal. 2008. Patronage or p articipation? Community based n atural r esource m anagement r eform in Sub Saharan Africa. Development and Change 3 9 :557 585. Newmark, W. D. 2008. Isolation of African protected areas. Frontiers in Ecology and the Environment 6 :321 328. Norton Griffiths, M. 1973. Counting the Serengeti migratory wildebeest using two stage sampling. African Journal of Ecology 11 :135 149 Norton Griffiths, M., D. Herlocker, and L. Pennycuick. 1975. The patterns of rainfall in the Serengeti ecosystem, Tanzania. East Africa Wildlife Journal 13 :347 374. Ogutu, J. O., and H. T. Dublin. 2004. Spatial dynamics of lions and their prey along an e nvironmental gradient. African Journal of Ecology 42 :8 22. Ogutu, J. O., H. P. Piepho, H. T. Dublin, N. Bhola, and R. S. Reid. 2008. Rainfall influences on ungulate population abundance in the Mara Serengeti ecosystem. Journal of Animal Ecology 77 :814 29. Olff, H., and J. G. C. Hopcraft. 2008. The resource basis of human wildlife interaction. Pages 95 133 in A. R. E. Sinclair, C. Packer, S. A. R. Mduma, and J. M. Fr yxell, editors. Serengeti III: h uman i mpacts on ecosystem d ynamics. University of Chicago P ress, Chicago Illinois, USA Olff, H., M. E. Ritchie, and H. H. T. Prins. 2002. Global environmental controls of diversity in large herbivores. Nature 415 :901 4. Paradis E., Claude J. and Strimmer K. 2004. APE: analyses of phylogenetics and evolution in R language. Bioinformatics 20 : 289 290. Pennycuick, L. 1975. Movements of the migratory wildebeest population in the Serengeti area between 1960 and 1973. East Africa Wildlife Journal 13 :65 87. a, F., I. Gordon, and C. Nores. 2001. Evolutionary transitions among feeding styles and habitats in ungulates. Evolutionary Ecology Research 3 :221 230. Pettorelli, N., J. Bro Jrgensen, S. M. Durant, T. Blackburn, and C. Carbone. 2009. Energy availability and density estimates in African ungulates. The American Naturalist 173 :698 704. Pickett, S. T. A., and M. L. Cadenasso. 1995. Landscape ecology: spatial heterogeneity in ecological systems. Science 269 :331.

PAGE 164

164 Pinheiro, J., Bates, D., DebRoy, S., Sarkar, D. and the R Core team. 2009. nlme: l inear and n onlinear m ixed e ffects m odels. R package version 3.1 96. Prins, H. H. T., and H. Olff. 1998. Species richness of African grazer assemblages: towards a functional explanation. Pages 449 490 in D. M. Newbery, H. H. T. Prins, and N. Brown, editors. Dynamics of tropical communities: 37th symposium of the British Ecological Society. Blackwell Science, Oxford England, UK Pulliam, H. R. 2000. On the relationship between niche and distribution. Ecology Letters 3 :349 361. R De velopment Core Team. 2009. R: a l anguage and environment for s tatistical c omputing. R Foundation for Statistical Computing, Vienna, Austria. ISBN 3 900051 07 0, URL http://www.R project.org Rabinowitz, D. 19 81. Seven forms of rarity. Pages 205 217 in H. Synge, editor. The biological aspects of rare plants conservation. John Wiley and Sons, Chichester, UK Rowe, R. J., R. C. Terry, and E. A. Rickart. 2011. Environmental change and declining resource availabili ty for small mammal communities in the Great Basin. Ecology 92 :1366 1375. Savage, V. M., J. F. Gillooly, W. H. Woodruff, G. B. West, A. P. Allen, B. J. Enquist, and J. H. Brown. 2004. The predominance of quarter power scaling in biology. F unctional Ecology 18 :257 282. Seagle, S. W., and S. J. McNaughton. 1992. Spatial variation in forage nutrient concentrations and the distribution of Serengeti grazing ungulates. Landscape Ecology 7 :229 241. Scott, J.M. P.J. Heglund, M.L. Morrison. 2002. Pr edicting species occurrences: issues of accuracy and scale. Island Press, Washington DC, USA. Setsaas, T., T. Holmern, G. Mwakalebe, S. Stokke, and E. Roskaft. 2007. How does human exploitation affect impala populations in protected and partially protected areas? A case study from the Serengeti Ecosystem, Tanzania. Biological Conservation 136 :563 570. Shinoda, M. 1995. Seasonal phase lag between rainfall and vegetation activity in tropical Africa as revealed by NOAA satellite data. International Journal o f Climatology 15 :639 656. Shorrocks, B. 2007. The biology of African savannahs. Oxford University Press, Oxford England, UK

PAGE 165

165 Sinclair, A. R. E. 2003. Mammal population regulation, keystone processes and ecosystem dynamics. Philosophical transactions of th e Royal Society of London. Series B, Biological sciences 358 :1729 40. Sinclair, A. R. E., J. G. C. Hopcraft, H. Olff, S. A. R. Mduma, K. A. Galvin, and G. J. Sharam. 2008. Historical and future changes to the Serengeti ecosystem. Pages 7 46 in A. R. E. Sin clair, C. Packer, S. A. R. Mduma, and J. M. Fryxell, editors. Serengeti III: human i mpacts on e cosystem d ynamics. University of Chicago Press, Chicago Illinois, USA Sinclair, A., S. Mduma, and J. S. Brashares. 2003. Patterns of predation in a diverse p redator prey system. Nature 425 :288 290. Sinclair, A., S. A. R. Mduma, J. G. C. Hopcraft, J. M. Fryxell, R. Hilborn, and S. Thirgood. 2007. Long t erm e cosystem d ynamics in the Serengeti: l essons for c onservation. Conservation Biology 21 :580 590. Sinclair, A., and M. Norton Griffiths. 1982. Does competition or facilitation regulate migrant ungulate populations in the Serengeti? A test of hypotheses. Oecologia 53 :364 369. Tanzania Wildlife Conservation Monitoring. 1994. Status and trends of wildebeest on th e Serengeti plains. Frankfurt Zoological Society, P.O. Box 3134, Arusha, Tanzania. Tilman, D. 2004. Niche tradeoffs, neutrality, and community structure: a stochastic theory of resource competition, invasion, and community assembly. Proceedings of the Nat ional Academy of Sciences of the United States of America 101 :10854 61. Tucker, C. J., Pinzon, J. E., and Brown M. E. 2004. Global inventory m odeling and m apping s tudies, Global Land Cover Facility, University of Maryland, College Park, Maryland. Turner, M.G. 2005. Landscape ecology: what is the state of the science? Annual Review of Ecology Evolution and Systematics 36 :319 344. Whittaker, R. H. 1952. A study of summer foliage insect communities in the Great Smoky Mountains. Ecological Monographs 22 :2 44. Whittaker, R.H. 1965. Dominance and diversity in land plant communities. Science 147 :250 260. Wilmshurst, J. F., J. M. Fryxell, and C. M. Bergman. 2000. The allometry of patch selection in ruminants. Proceedings of the Royal Society of London. Series B, Bi ological Sciences 267 :345 349.

PAGE 166

166 Wilsey, B. J. 1996. Variation in use of green flushes following burns among African ungulate species: the importance of body size. African Journal of Ecology 34 :32 38. Wittemyer, G., P. Elsen, W. T. Bean, A. C. O. Burton, and J. S. Brashares. 2008. Accelerated human population growth at protected area edges. Science 321 :123. Zuur, A. F., E. N. Ieno, and C. S. Elphick. 2010. A protocol for data exploration to avoid common statistical problems. Methods in Ecology and Evolution 1 :3 14. Zuur, A. F., E. N. Ieno, N. J. Walker, A. A. Saveliev, and G. M. Smith. 2009. Mixed e ffects models and extensions in ec ology with R. Springer, New York New York, USA

PAGE 167

167 BIOGRAPHICAL SKETCH Smriti Bhotika was born in Bombay, India in 1981. Three y ears later, her family moved to the United States of America and settled in Durham, North Carolina where she grew up. She graduated from the North Carolina School of Science and Mathematics in 1999. She continued her education at the University of North Carolina at Chapel Hill where she earned her Bachelor of Science in computer science with a minor in biology in 2003. Upon graduating, she worked in the Genome Sequencing and Computational Biology Division of Los Alamos National Laboratory in New Mexico f or two years. During this time she explored her interests further and discovered ecology, a subject which combined her background in biology, mathematics, and computer science with her innate passion for nature. This realization led her to pursue graduat e studies at the University of Florida where she earned her Doctor in Philosophy in interdisciplinary ecology with a concentration in geographic information systems in 2012.