ABSTRACT

ASSESSING SPATIAL AND TEMPORAL PATTERNS OF HUMAN-CAUSED ELEPHANT MORTALITY IN TSAVO EAST NATIONAL PARK, KENYA

by Daniel Muteti Kyale

Elephant mortality data available for the Tsavo East National Park for the 1990 – 2005 period were used to describe patterns of poaching-induced elephant mortality. Relationships between poaching-induced elephant mortality and human and biophysical factors were also examined. Elephant poaching occurred in clustered patterns and was significantly correlated with land cover, proximity to surface water, ranger patrol bases, park gates, roads, and park boundaries, and elevation. However, none of these factors individually explained more than 40% of the observed variation in poaching induced elephant mortality. These factors were used model risk to elephant poaching based on elephant poaching mortality data for the wet and dry seasons, and on all elephant poaching mortality regardless of season. Risk maps generated can be improved by careful selection of additional risk factors not considered here. These risk maps serve as a useful tool to guide and inform park managers involved in elephant conservation in Kenya. ASSESSING SPATIAL AND TEMPORAL PATTERNS OF HUMAN-CAUSED ELEPHANT MORTALITY IN TSAVO EAST NATIONAL PARK, KENYA

A Thesis

Submitted to the Faculty of Miami University in partial fulfillment of the requirements for the degree of Master of Arts Department of Geography

By Daniel Muteti Kyale Miami University Oxford, Ohio 2006

Approved: Advisor______(Dr. John K. Maingi)

Reader______(Dr. Kimberly E. Medley)

Reader______(Dr. Mary C. Henry) Table of Contents Page List of Figures……………………………………………………………………iv List of Tables……………………………………………………………………..v Acknowledgements……………………………………………………………...vi CHAPTER ONE…………………………………………………………….……1 1.0 Introduction…………………………………………………………………..1 1.1 Statement of the Problem…………………………………………………..2 1.2 Justification for the Study……………………………………………….…..3 1.3 Research Goal and Objectives…………………………………………….3 1.4 Thesis Organization………………………………………………………....4 CHAPTER TWO…………………………………………………………...... ….5 2.0 Wildlife Conservation in Kenya……………………………………………..5 2.1 Pre-colonial Period (Before 1895)………………………………………….5 2.2 Colonial Period (1895 - 1963)……………………………………………….6 2.3 Post-Colonial Period (1963 - Present)……………………………………...9 2.4 Kenya Wildlife Service………………………………………………………10 CHAPTER THREE………………………………………………………………14 3.0 Literature Review…………………………………………………………..14 3.1 Spatial Point pattern analysis……………………………………………15 3.1.1 First Order Statistics………………………………………………………16 3.1.1.1 Quadrat Analysis………………………………………………………..16 3.1.1.2 Kernel Density Analysis………………………………………………...18 3.1.1.3 Nearest Neighbor Analysis………………………………………….....19 3.1.1.4 Standard Deviation Ellipse……………………………………………..20 3.1.2 Second Order Statistics…………………………………………………...20 3.2 Human-Caused Wildlife Mortality………………………………………..22 3.3 Land Cover Mapping from Satellite Imagery……… ………………….25 3.4 Research Hypothesis……………………………………………………...27 CHAPTER FOUR……………………………………………………………...…28 4.0 Data and Methods…………………………………………………………..28

ii

4.1 Study Area…………………………………………………………………....28 4.1.1 The Tsavo Elephant Problem…………………………………………....30 4.2 Data…………………………………………………………………………...31 4.2.1 Elephant Mortality Data…………………………………………………...32 4.2.2 Satellite Data…………………………………………………………….....33 4.2.3 GIS Data………………………………………………………………..…..35 4.3 Methods………………………………………………………………………37 4.3.1 Describing Patterns of Human-Induced Elephant Mortality…………...39 4.3.1.1 Quadrat analyses………………………………………………………..39 4.3.1.2 Nearest Neighbor Analyses…………………………………………….40 4.3.1.3 Standard Deviation Ellipse Analyses……………………………….....40 4.3.1.4 Kernel Density Analyses………………………………………………..40 4.3.2 Land Cover Mapping……………………………………………………...41 4.3.2.1 Classification Scheme…………………………………………..……...41 4.3.2.2 Image Classification…………………………………………………….43 4.3.3 Exploring relationships between elephant mortality patterns and biophysical and human variables………………………….……………..44 4.3.4 Generating Risk to Poaching Maps……………………………………...45 CHAPTER FIVE………………………………………………………………….46 5.0 Results and Discussions …………………………………………………46 5.1 Elephant Mortality Analysis Results………………………………….…….46 5.2 Land Cover Mapping………………………………………………….……..54 5.2.1 Accuracy Assessment……………………………………………….…….56 5.3 Relationships between elephant mortality patterns and Biophysical and human factors……………………………….……….……57 5.4 Risk to Poaching Maps………………………………………………………61 CHAPTER SIX……………………………………………………………………69 6.0 Conclusions ………………………………………………………………….69 References……………………………………………..…………………………71

iii

List of Tables Table 1: Landsat ETM+ images used in the study……………………………35 Table 2: Categories of mortality datasets used to describe elephant mortality patterns ……………………………………….….39 Table 3: The land cover classification scheme used for the study…………42 Table 4: Quadrat Analysis Results for Poaching-based Mortality…………..46 Table 5: Nearest Neighbor Analysis Results for Poaching-based Mortality……………………………………………..47 Table 6: Sizes (hectares) of Land Cover classes mapped for TENP………56 Table 7: Error matrix for TENP land cover map created from Landsat ETM+ images………………………………………………...57 Table 8: Relationship between elephant mortality and land cover types...... 58 Table 9: Spearman’s Rank correlates for elephant mortality in TENP...... 59

iv

List of Figures Figure 1: Map of Kenya showing selected parks and reserves……………..11 Figure 2: Spatial point patterns…………………………………………………15 Figure 3: Location of Tsavo East National Park in Southern Kenya………..29 Figure 4: 2004 Rainfall distribution based on data obtained from TENP records………………………………………………….. 30 Figure 5: Elephant mortality distribution and TENP 10 km buffer…………..33 Figure 6: Landsat WRS-2 Path and Row scene intersections in TENP…...34 Figure 7: GIS layers generated…………………………………………………36 Figure 8: A flowchart of data and methods used in the study……………….38 Figure 9: (a) Images acquired on 3/4/2001 (b) Images acquired on 1/22/2000………………………………………………….….…..41 Figure 10: Kernel Density and Standard Deviation Ellipse Results for Overall Poaching Pattern ……………………………….……...48 Figure 11: Kernel Density and Standard Deviation Ellipse Results for Dry Season Elephant Poaching ………………………………49 Figure 12: Kernel Density and Standard Deviation Ellipse Results for Wet Season Elephant Poaching………………………………50 Figure 13: Kernel Density and Standard Deviation Ellipse Results for 1990 – 1997 Elephant Poaching Period……………………...52 Figure 14: Kernel Density and Standard Deviation Ellipse Results for 1998 – 2005 Elephant Poaching Period...…………………....53 Figure 15: Land Cover Map for TENP………………………………………….55 Figure 16: Relationship between dry season poaching and distance to main rivers……………………………………………...61 Figure 17: Risk to elephant poaching surface for Wet Season……………...63 Figure 18: Risk to elephant poaching surface for Dry Season………………65 Figure 19: Annual Risk to Elephant Poaching………………………………...67

v

Acknowledgements

I feel indebted to the department of Geography for granting me the opportunity for training experience in Miami University and Kenya Wildlife Service for allowing me study leave to further my studies. I sincerely thank Dr. John Maingi for his academic advising, inspiring ideas and for always finding an extra minute to address issues arising during the thesis process. I also thank Dr. Kimberly Medley and Dr. Mary Henry for their constructive inputs into the thesis and the various staff in Kenya Wildlife Service who helped in one way or another during my fieldwork in Kenya. Finally I register my special thanks to the departments of Geography and Zoology for their financial support that enabled me to conduct my fieldwork in Kenya.

vi

CHAPTER ONE

1.0 Introduction

Kenya’s tourism is founded on the country’s rich wildlife resources and contributes to approximately 20% of the Gross Domestic Product (Government of Kenya, 2005). In 2005, Kenya earned Kshs. 49 billion (US$ 677 million) up from Kshs. 42 billion earned in 2004 and it is projected that the tourism sector will earn the country Kshs. 56 billion by the end of 2006 (Odhiambo and Moraa, 2006). The contribution of the African elephant to nature-based tourism cannot be overlooked being a member of the “Big Five” family (elephant, rhino, lion, buffalo and leopard) that forms a star tourist attraction in sub-Saharan Africa. Additionally, African elephants are the largest terrestrial animals and keystone species whose feeding behaviors have contributed to maintaining African savannah ecosystems (Wijgaarden, 1985; Leuthold, 1996; Western and Maitumo, 2004). Elephant populations especially in eastern Africa had declined over the years due to human activities. Although loss of habitat to agriculture and human settlements (Turner et al., 2001; Makonjio and Warui, 2004) has been argued as a major long-term threat to the survival of elephants in situ, the ultimate direct threat to elephant survival has been identified as poaching (Western, 1987; Leader-Williams et al., 1990; Armbruster and Lande, 1993; Burton, 1999; Heltberg, 2001). Organized poaching had been responsible for the downfall of elephant populations in their natural ranges in Africa (Futurist, 1987), which necessitated the Convention on International Trade in Endangered Species of Flora and Fauna (CITES) to ban ivory trade in 1987. CITES believed that a ban on ivory trade would afford the elephant population a chance to recover from years of massive slaughter. Currently, however, the question of how best to conserve and manage elephants remains a thorny issue among elephant range states as well as among stakeholders in conservation (Economist 2000a&b). Southern Africa states led by South Africa, Namibia, Zimbabwe and Botswana have been advocating for a lift in the ban on trade in ivory, arguing that they experience

1 local overpopulation of elephants following successful conservation and protection strategies (Economist, 2000a). Against this idea are eastern African states led by Kenya and Asian countries led by India that strongly believe lifting the ban on ivory trade is will lead to an escalation in organized elephant poaching that has continued to haunt these states (Economist, 2000b). Against this backdrop, African Elephant Specialist Group (AFESG, 1997) identified five important issues that need to be addressed concerning conservation of African elephants: (a) law enforcement, poaching and ivory trade, (b) loss of elephant habitats, (c) local overpopulation, (d) improvement in elephant surveys, and (e) human-elephant conflict.

1.1 Statement of the Problem

The International Union for Conservation of Nature (IUCN, 2000) classifies the African elephant (Loxodonta africana) as an endangered species. Elephants have been poached to provide ivory for the multimillion-dollar illegal trade that is said to rival drug trafficking (Hakansson 2004; Jyoti and Harford, 1996). Kenya lost over 80% of its elephant populations between 1970 and 1989 (Smith and Kasiki, 2000). In 1989, CITES banned trade in ivory to curb elephant poaching at the international level (Burton, 1999). Despite the ban, Kenya still experienced high levels of elephant poaching. Worse still, CITES bowed to pressure from southern African countries, partially lifting the ivory ban in 1997 (Hertberg, 2001; Jachmann and Billiouw, 1997) and down-listing southern Africa elephants from Appendix I to II of IUCN red list of endangered species in 2004, thus allowing some limited trade in ivory (CITES, 2004). It is feared that this will most likely result in an upsurge of elephant poaching because of loopholes in CITES policing system that make it easy for traders to launder illegal ivory (Pearce, 2004). There is need therefore, for preventive strategies to avert elephant poaching at the national level and careful regulation of the ivory trade at the international level (Milner-Gulland and Leader-Williams, 1992). Available statistics from Born Free Foundation (2002) suggest elephant poaching had increased following the partial lift on the ban in trade in ivory in 1997. Such elephant poaching need to be stemmed through improved patrols by game rangers. The rising

2 incidences of elephant poaching make it necessary for the Kenya Wildlife Service (KWS) to increase and improve the quality of its anti-poaching patrols.

1.2 Justification for the Study

Poaching, law enforcement and ivory trade have been identified as among the top five priority issues affecting the African elephant in the elephant range states (AFESG, 1997; WWF, 1997). Protection of elephants in their natural ranges has been estimated to cost approximately US$ 200 per km2 annually (Cumming et al. 1990). Kenya, like many other poor Sub-Saharan African countries struggling to feed their people and keep their economies afloat is unable to meet this financial obligation and is thus unlikely to allocate more funds towards wildlife conservation. Jachmann and Billiouw (1997) estimated that there should be a minimum of one park ranger for every 24 km2 of wildlife reserve if effective patrolling and policing is to be realized. KWS, like most wildlife departments in other African countries, is understaffed with approximately one ranger per 100 km2 of wildlife reserve. It is therefore important to explore strategies that involve more efficient use of the limited available resources. By assessing spatial and temporal patterns of elephant mortality, important insights on characteristics of Tsavo East National Park (TENP) areas elephants are more vulnerable to human-induced death can be generated, which in turn can help guide effective deployment of policing resources. TENP was chosen as the preferred study area because it is the largest park in the country (KWS, 2003), and has both the highest concentration of elephants and highest incidences of elephant poaching reported (Economist, 2002; Hammer, 1993; Robinson, 2000). TENP is a predominantly semi-arid bushland with only a small area of the park developed and open for tourism. Unfavorably hot climate, poor accessibility, and the large size of the park, make patrolling difficult and more challenging with currently available resources (Kioko, 2002).

1.3 Research Goal and Objectives

In the present study, I used GIS to describe patterns of elephant mortality attributed to poaching and identified areas of the TENP that are considered to be at

3 greater risk to elephant poaching. Such information would be useful in guiding the deployment of policing resources in the park and its immediate vicinity. This goal was achieved by describing spatial and temporal patterns of human-induced elephant mortality for the period 1992 - 2005, and investigating the relationship between elephant mortality and biophysical and human factors that influence the distribution of elephants in TENP. A recent and detailed land cover map was required in these analyses but existing maps such as the Africover (FAO, 2003) was much generalized and did not reflect the variety of land cover classes in the TENP. The specific objectives for the study were therefore, to: (1) Describe how poaching-induced elephant mortality is distributed through space and time in TENP; (2) Create a land cover map for TENP in order to quantify elephant poaching mortality by land cover types; (3) Examine the relationships between observed patterns of elephant mortality and various biophysical and human factors that characterize the TENP; (4) Construct models describing elephant risk to poaching.

1.4 Thesis Organization

In chapter two of this thesis, wildlife conservation in Kenya is examined with particular reference to elephants. In chapter three, literature on point patterns analysis techniques, human-induced wildlife mortality, and land cover mapping from satellite imagery is reviewed. A description of the study area is provided in chapter four along with data, methods and analytical procedures used in the present study. In chapter five, I present then discuss results obtained and finally conclusions are provided in chapter six.

4 CHAPTER TWO

2.0 Wildlife Conservation in Kenya

Wildlife conservation is a term often used interchangeably with biodiversity or nature conservation. In its broadest sense, wildlife conservation refers to preservation, protection and management of wildlife and wildlife resources, be it flora, fauna or the soil or habitat on which these life forms occur. It is also an academic discipline or the study of biodiversity conservation (Soule, 1985; Miller, 1989). In this chapter I examine the history of wildlife conservation in Kenya with an emphasis on elephants. The chapter has been structured into four sections representing significant temporal periods in the country’s conservation history.

2.1 Pre-colonial Period (Before 1895)

Many scholars who have studied African communities argue that these indigenous communities lived in balance with nature having coevolved with wildlife since the beginning of time (Fitter, 1986). Along the realm of coexistence, native communities developed useful knowledge of working with their natural environment without jeopardizing its ability to recover. It is widely believed that the attachment the indigenous African communities had to their natural environment made them learn to respect and nurture wildlife. Many communities in Kenya had community norms that governed use of wildlife resources. It was taboo to kill certain species of animals while others were totems among many Kenyan communities. Such conservation knowledge enshrined in the indigenous people’s cultural practices was passed from one generation to the other via folklore and songs. The Samburu, for instance, had a popular legend that they shared similar origin with elephants and thus killing or eating elephant meat was a taboo (Kuriyan, 2002). The Maasai did not hunt wildlife for food unless during times of environmental adversities, but only used them for traditional rituals and ceremonies (Collett, 1987). There were, however, other communities whose livelihoods depended entirely on hunting of wildlife and gathering of wild fruits and vegetables. Among these

5 communities were the Waliangulu and the Waata (Kassam and Bashuna, 2004). The Kamba, the Somali, the Galla and the Giriama supplemented their livestock with bush meat. Most importantly, however, was the participation of the hunting communities in trade with wildlife parts, most notably being ivory and rhino horn (Stone, 1972; Dalleo, 1979). The literature does not precisely indicate when trade in wildlife parts began, but some scholars seem to suggest that trade in ivory and rhino horn has taken place for a period of over ten millennia (Sharp, 1997). Inland hunting communities acted as producers and/or middlemen supplying Arab and Indian merchants with ivory and rhino horns at the coast (Stone, 1972; Dalleo, 1979). It has often been argued that hunting by the indigenous communities did not pose significant threat to wildlife as at this time because human population was low, the methods of hunting were not quite effective and wildlife herds were large (Harper, 1960). It is not clear when hunting became a threat to wildlife survival; however, scholars believe the proliferation of firearms made available by European hunters and settlers was a turning point on wildlife conservation in Kenya and Africa as a whole (Stone, 1972).

2.2 Colonial Period (1895 - 1963)

The arrival of Europeans marked the integration of Kenya into global trade and a shift in the way the rural Africans interacted with wildlife (Akama, 1998). Kenya came under British rule in 1895 when it became part of the British East African Protectorate. At this time, Europeans were settling and establishing themselves as large scale farmers in Kenya. They also hunted African game for trophies, horns, skins and other body parts (Akama, 1998; Adams, 2004). Large scale agriculture deprived wildlife of their natural habitats and hunting become highly effective with firearms. Meanwhile, firearms were also finding their ways to traditionally hunting indigenous communities. Game hunting by both the white settlers and the native hunting communities took a toll on wildlife. The situation was aggravated by the white settlers’ perceptions that wildlife acted as vermin, competing directly with their livestock for pasture and acting as hosts to livestock pests like tsetse flies and ticks (Anderson and Grove, 1987). It became

6 increasingly clear to the colonial administrators that something needed to be done to avert imminent African wildlife exterminations (Stone, 1972; Adams, 2004). By the turn of 19th century, pioneer game laws were put in place to regulate hunting and preserve African wildlife, beginning with the establishment of the Southern and Northern game reserves in areas of Kajiado and Laikipia districts respectively (Hulme and Murphree, 2001). In 1903, the Society for the Preservation of the Wild Fauna of the Empire, renamed in 1919 as the Society for the Preservation of the Fauna of the Empire (SPFE) was founded to rally support for conservation of wildlife in British colonies (Adams, 2004). In 1906, the establishment of Southern and Northern game reserves was legalized (gazettement) and discussions to regulate hunting began by 1908. It is at this time that the Game Department was established to enforce game laws (Kassam and Bashuna, 2004). However, some scholars believe that establishment of game reserves and introduction of game laws were not meant to conserve wildlife per se. They argue that game reserves were set aside to hold wildlife for future generations of white hunters and as sources of revenue for the colonial government via licensed hunting (Akama, 1998; Dalleo, 1979; Stone, 1972). Unlicensed hunting thus became poaching, which was a crime punishable by law (Steinhart, 1994; Kassam and Bashuna, 2004; Were, 2005). Game reserves were considered hunting blocks where licensed hunting and other human activities like collecting firewood and herding livestock took place. The biggest problem, however, was that boundaries of the game reserves were not clearly defined and at times were shifted by colonial governors at will. SPFE felt that there was a need for more permanent wildlife protected areas. A series of meetings between colonial authorities and SPFE took place culminating in the establishment of an international office for nature conservation in Brussels, Belgium in 1928 (Akama, 1998). The SPFE began to study the status of wildlife in Kenya in 1930 with the objective of finding ways of easing hunting pressure on African game. The results of the study were reviewed in a meeting held in London in 1933 to pave way for establishment of national parks. In 1939, a British game committee was set up to work on modalities of creating National Parks (Akama, 1998). SPFE strongly felt that establishment of National Parks was the only way that would ensure feasible in situ conservation of wild flora and

7 fauna, ecosystems and natural habitats and recovery of populations from intensive hunting. Unlike game reserves, national parks were to be considered pristine areas whose boundaries could not be amended by colonial governors except through comprehensive legislative supplements (Adam, 2004). The colonial government passed the Royal National Parks Order in 1945 thus giving direction on how national parks were to be established, managed and governed (Were, 2005). A government body independent of the game department was established to manage the national parks. It was referred to as the Royal National Parks of Kenya. Nairobi National Park was the first to be created in 1946 followed by TENP and Tsavo West National Park (TWNP) in 1948 (KWS, 1996). When establishing the National Parks, the colonial government borrowed heavily from the Yellowstone National Park (USA) model of 1872 in which national parks were treated as pristine and marketed to earn revenue through tourism (Anderson and Grove, 1987). Any form of unauthorized wildlife utilization was prohibited within the parks through Wild Animal Protection Ordinance issued in 1951. By early 1950s, the initially traditional hunter communities had transformed themselves into well organized poaching groups linked through networks of trophy producers, receivers, middlemen and merchants (Stone, 1972). Ivory and rhino horn prices had hit $5.5 and $800 a kilo respectively (Western, 1987). In response to the rising prices of ivory and rhino horns in the international market, the Waata were joined by armed Abyssinian bandits from Ethiopia in search of the trophies in northern Kenya, while the Waliangulu, Wakamba and Giriama hunted for the same trophies in and around the Tsavo plains (Kasiki, 1998; Kassam and Bashuna, 2004). Although the literature does not quantify the number of elephants and rhinos killed during this period, there was overwhelming evidence that heavy elephant and rhino poaching took place (Stone, 1972; Steinhart, 1994). It took the colonial government concerted efforts to contain the poaching menace by 1957, through countrywide anti-poaching operations (Sheldrick, 1973). This was at a time when many African states were struggling to free themselves from the yoke of colonialism.

8 2.3 Post-Colonial Period (1963 - Present)

When Kenya gained independence in 1963, tourism was thriving. Wildlife populations increased until 1969 when a severe drought hit the country. Besides the drought, poaching returned in full force in mid 1970s. By this time, prices of ivory and rhino horn in the international market had increased to $100 and $8000 per kilo respectively (Western, 1987). Coupled with renewed poaching, the drought is said to have reduced elephants from an estimated 35000 in 1974 to about 14000 in 1980 in the Tsavo ecosystem alone (Corfield, 1973; Cobb, 1976). It became necessary for the government to concentrate available resources towards protection of wildlife, particularly elephants and rhinos and to stamp out poaching. In 1976, Wildlife Conservation and Management Department (WCMD) was established through Wildlife Conservation and Management Act CAP 376 Laws of Kenya by integrating Royal National Parks of Kenya and the Game Department (Akama, 1998). The creation of WCMD did little to reduce poaching because law enforcement officials lacked modern equipment, rangers were poorly paid, and further compounding this problem, the prices of ivory and other wildlife merchandise were at-all time high (KWS, 1994; Akama, 1998; Were, 2004). Rangers were unable to contain rampant poaching because besides being poorly equipped, many were also complacent due to poor supervision and few incentives. Increased poaching compelled the government to issue a presidential degree banning wildlife hunting (Legal Notice number 120, Legislative Supplement number 26 of 1977), and in 1978 banning trade in wildlife products through Legal order number 181 (Kenya government, 1978; KWS 1994). All government efforts to stamp out elephant and rhino poaching failed. By late 1980s, Kenya elephant populations had declined from an estimated 170,000 in 1963 to only 16,000 countrywide (KWS, 2005). In 1989, President Daniel Arap Moi led the Kenyan government in burning about twelve tons of raw ivory in Nairobi National Park recovered from poachers to demonstrate to the world the country’s commitment to eradicating poaching. In the same year, CITES (to which Kenya is a signatory) banned international ivory trade in an attempt to curb poaching at the international level (Burton, 1999) and placed elephants in Appendix I of the IUCN red list of endangered species (IUCN, 2000). At this time the government felt an urgent need for a vibrant and well

9 motivated conservation body to protect and manage Kenya’s wildlife heritage. KWS was thus established in 1989 through Wildlife Conservation and Management (Amendment) Act CAP 376 Laws of Kenya to address the shortfalls of WCMD (Were, 2005).

2.4 Kenya Wildlife Service

Section 3 of the Wildlife Conservation and Management (Amendment) Act CAP 376 of 1989 gives KWS the mandate to protect, conserve and manage the country’s biodiversity as represented by its flora and fauna on behalf of the people of Kenya (KWS, 2005). Currently, Kenya has over fifty National Parks and Reserves covering about 8% of the country (KWS, 1996). The Parks and Reserves represent the diverse ecosystems found in the country ranging from mountains through semi-arid savannas and wetlands to marine (Figure 1). With Dr. Richard Leakey as its first director, KWS received moral and financial support from government and donors alike. It was well funded and equipped to face the challenge of poaching head on. Given the deplorable state of wildlife conservation at the inception of KWS, it became necessary for Leakey to adopt radical measures to counter poaching activities. It was during Leakey’s tenure that the government introduced a “shoot to kill” order in support of the director’s philosophy of “hard-edged” boundaries (Economist, 1999). Leakey believed that park boundaries characterized by electric fences and guarded by armed rangers were necessary to control human encroachment into parks, keep out unauthorized people who destroyed wildlife and their habitats, and contain the wildlife within parks, thus minimizing human-wildlife conflicts. This militaristic approach instilled discipline and restored order to conservation evidenced by an increase in elephant populations in the following years.

10

Figure 1: Map of Kenya showing selected parks and reserves. Parks and Reserves data obtained from Global Land Cover Facility.

As poaching subsided, an emerging challenge was to harmonize land use conflicts between parks and adjacent rural communities. Land use conflict, which still continues to be a difficult issue for KWS is largely attributed to the way wildlife policy in Kenya (and

11 generally in Africa) was implemented. Conservation in Africa was a purely Euro- American idea that did not consider the needs of rural African communities who had coexisted with wildlife over the ages (Anderson and Grove, 1987; Hulme and Murphree, 2001; Adams, 2004). In 1992, KWS introduced the Community Wildlife Service (CWS) department to partner with stakeholders in wildlife in order to create conservation awareness and pass tangible benefits to local communities (KWS, 2005). This was at a time when community conservation paradigm was taking firm hold on wildlife conservation in Africa. CWS however, was not as active as the militaristic Anti-poaching Unit, which made a section of stakeholders in conservation dissatisfied. Dr. Leakey resigned in 1994 citing political interference that made discharging his duties difficult (Morell, 1994). Dr. David Western, an ecologist by training took over the leadership of KWS after Leakey’s resignation. Unlike his predecessor, Western conceded that Kenyan protected areas were too small to shoulder the ecological impacts of confining wildlife within them (Baskin, 1994). Further more, an estimated 70% of wildlife in Kenya was believed to reside outside the protected areas (KWS, 1994; Were, 2005). To ease the ecological pressure on Kenya’s parks and reserves and ensure benefits accruing from conservation tickled down to the rural communities, Western embarked on promoting community conservation through his philosophy of “Parks beyond Parks” (Western, 1989). It was hoped that by enabling local people to realize tangible benefits from conservation, they would become more tolerant of wildlife and allow it to utilize their lands. Western’s policy, though well intended lacked a mechanism to ensure conservation benefits reached these local people who bear the cost of living with wildlife (Akama, 1998). European and American donor countries viewed Western’s policy with suspicion (Earth Watch Journal, 1998); meanwhile, refusing to avail financial aid due to political differences with the Kenyan government. Dr. Western’s tenure in KWS was plagued by financial, political and social problems that disrupted operations of the organization (Butler, 1998; McRae, 1998). Law enforcement was no longer a priority and elephant poaching returned in full force. The situation was further aggravated when CITES partially lifted the ban on ivory trade to

12 allow southern African countries to sell their stock piles. Western was fired in 1998 allowing Dr. Leakey to return to conservation in Kenya (Malakoff, 1998), which he quit six months later. Since then, KWS has been headed by more than six directors (Ng’ang’a, 2004), but the goals of wildlife conservation still remain the same as stipulated in the Wildlife Conservation and Management (Amendment) Act CAP 376 Laws of Kenya. It is worth mentioning that the question on how best to conserve wildlife and wildlife resources in Kenya and all of Africa has remained a controversial issue among conservationists, ecologists, government agencies and other wildlife conservation stakeholders (Turner, 1999; Attwell, 2000). Besides, conservation decisions are subject to political manipulations. There is always a need to consider needs of wildlife on one hand and the needs of the people on the other. Achieving a balance between the two makes conservation policy hard to develop and implement.

13 CHAPTER THREE

3.0 Literature Review

The first objective of this study was to establish how human-induced elephant mortality was distributed in TENP. Elephant mortality locations are point events that have occurred in a given location at a certain point in time. Point events have zero dimensions thus a valid measure of their distribution is the number of times they occur in a given space with respect to their respective geographic locations (Burden, 2003). Distribution of point events is analyzed using point pattern analysis techniques. The techniques can either be first order statistics or second order statistics. The first section (3.1) of this chapter therefore, discusses commonly used point pattern analysis techniques. Subsection 3.1.1 covers first order statistics including quadrat analysis, nearest neighbor analysis, kernel density and standard deviation ellipse and subsection 3.1.2 discusses Ripley’s K-function under second order statistics. Any point pattern is created by a certain process taking place in a given space over a period of time. The process that produces the point pattern may be influenced or related to external factors so that the resultant point pattern is in response to those external factors. The second objective in the present study was to establish what biophysical and human factors may have influenced the observed human-caused elephant mortality in TENP. Section 3.2 therefore, reviews the literature on human-caused wildlife mortality so that possible biophysical and human factors that might have influenced the observed human-induced elephant mortality can be understood. One important biophysical factor that influences elephant distribution in their natural habitats, and hence elephant mortality patterns is land cover (Verlinden and Gavor, 1998; De Boer et al. 2000). To analyze the relationship between the observed elephant mortality patterns and land cover in TENP, it was necessary to generate a land cover map of the study area from satellite imagery; an endeavor expressed in the third objective of the present study. The final section of this chapter (3.3) thus discusses methods commonly used to generate land cover maps from satellite imagery.

14 3.1 Spatial Point pattern analysis

Point pattern analysis is a statistical technique that is used to describe patterns of point events that have occurred in a specified geographic area (Gatrell, et al., 1996). In point pattern analysis, the researcher endeavors to ascertain whether mapped point events exhibit random, clustered or regular distribution (Figure 2). The researcher can further test the degree of significance the pattern exhibits and attempt to understand the process that might have produced the observed pattern (Wiegand and Moloney, 2004). For a researcher to perform pattern analysis on point data, Burden (2003) provides a set of five qualities the data should exhibit: (a) the pattern must be mapped on a plane, (b) a study area must be selected and determined before the analysis, (c) the point data should not be a selected sample, but the entire set of data to be analyzed, (d) there should be a one-to- one correspondence between objects in the study area and events in the pattern and (e) the points must be true incidents with real spatial coordinates.

Figure 2: Spatial point patterns.

Point pattern analysis traces its origin from botanists and ecologists in 1930s (Burden, 2003) and has since spread to many fields. Developments in GIS have led to increased use of point pattern analysis in diverse fields such as epidemiology (Gatrell, et al., 1996; O’Brien, et al., 1999), ecology (Wiegand and Moloney, 2004; Bonnicksen and Stone, 1981), and criminology (Bowers and Hirschfield, 1999; Ackerman and Murray, 2004) among others. GIS facilitates hypotheses generation because of its capacity to

15 overlay and integrate spatial information, substantiate quantitative analyses through its ability to handle large amounts of data efficiently and with little effort, and its visualization ability in the traditional exploratory analysis in statistics (Bolstad, 2002; Kliskey, 1995). With its advanced geospatial analytical tools GIS enable their users to explore point pattern analyses well beyond establishing whether distribution of geographic features deviates from complete randomness, to establishing the nature and significance of the spatial arrangement. Point pattern analysis techniques can be classified as either first order or second order statistics.

3.1.1 First Order Statistics

First order statistical point pattern analysis techniques describe first order properties of a spatial point process. They are area-based and describe large scale variation in concentrations of mapped events across a study area (Gatrell et al., 1996; Wiegand and Moloney, 2004). The point intensity, λ is the number of point events occurring per unit area, thus all density-based point pattern analysis techniques describe first order properties of a point process. The commonly used techniques are: (a) Quadrat method (b) Kernel density estimate, (c) Nearest neighbor distance and (d) Standard deviation ellipse.

3.1.1.1 Quadrat Analysis In quadrat analysis, the analyst overlays a grid of cells of identical shape and size across the study area where the point events have been recorded. The analyst then counts the frequency distribution of events within each cell (Lee and Wong, 2001). The frequency distribution of quadrat counts is compared to the distribution expected for a spatial process model hypothesized to be responsible for the observed pattern. The commonly hypothesized process is random Poisson process (Wall et al., 1985) and uses parametric statistics like Kolmogorov-Smirnoff (K-S) to test for goodness of fit. The observed frequency distribution is converted to cumulative frequency and then compared to cumulative frequency generated using a Poisson process (Lee and Wong, 2001). A D statistic is computed as follows:

16 D = Max│Oi-Ei│ Where: Max│ │ denotes maximum difference

Oi is observed frequency distribution at quadrat i

Ei is the expected frequency at quadrat i

The D statistic is compared to Critical Dα 0.05 calculated as follows:

Critical Dα 0.05 = 1.36/√M Where: M is the number of quadrats in the study area.

If D is greater than Dα 0.05, then the observed distribution is significantly different from a random one generated using a Poisson process and vice versa. By further exploring the properties of a Poisson process, the analyst can proceed to establish whether the observed distribution is more clustered or more dispersed than a random one using Variance-Mean Ratio (VMR) (Lee and Wong, 2001). In a Poisson distribution, the variance of the distribution equals its mean, hence a random distribution has a VMR of one (VMR = 1). An observed distribution exhibiting clustering thus will have a VMR greater than one, while a dispersed distribution has a VMR close to zero. Plante et al. (2004) used quadrat analysis to study the distribution of deer in Anticosti Island of Québec province in Canada. The researchers established that deer distribution in winter was not random, but instead the deer preferred areas where balsam fir occurred and where forest regeneration was abundant. Bonannicksen and Stone (1981) analyzed spatial patterns of a giant Sequoia mixed conifer forest community in California and managed to demonstrate presence of aggregations of even-aged trees using quadrat analysis method. The technique has also been used to demonstrate diffusion pattern of fowl pest disease in Wales, Britain (Gilg, 1973) and the spread of chronic bronchitis in urban areas of Leeds, UK (Girt, 1972). Although quadrat analysis is useful for comparing an observed point pattern with a theoretical random pattern, it has several shortcomings. The size of the quadrat influences the observed pattern in that a large quadrat would most likely have many events recorded while a small quadrat may have little or no events et all. Large quadrats are thus likely to produce clustered patterns of the point process under examination compared to patterns likely to be produced by small-sized quadrats. It is important

17 therefore, to compute an optimal quadrat size. Lee and Wong (2001) suggest that the optimal quadrat size should be twice the study area divided by the total number of point events recorded. In addition, quadrat method considers events within the quadrats, but not between quadrats. This means that the method does not consider relative location of events. Lastly, the choice of the origin and orientation of quadrat affects observed frequency distribution.

3.1.1.2 Kernel Density Analysis Kernel density estimates involve counting events within a moving window. The Kernel is a moving three-dimensional function used to weight point events in accordance to their distances from where the estimate is made (Gatrell et al., 1996). This method results in smooth estimates of variation in point concentration. The degree of smoothness is determined by the bandwidth used in the analysis and consequently the resultant estimated density surface. The bandwidth is the length of the search radius within which the kernel exerts its influence. Kernel density is a very good way of visualizing a point pattern to detect occurrence hotspots. Kernel density analysis produces a map of estimates of the local intensity of any spatial process. The bandwidth used in a kernel density analysis should have some meaning in the context of the study. For instance, the distribution patterns of a given species of wildlife may be estimated in a kernel density analysis using a bandwidth relative to its home range. More complex variations of basic kernel density estimation function are used that weight nearby events more heavily than distant ones in estimating local density (Gatrell et al., 1996). It is however, important to note that kernel density estimation suffers edge effects that can be rectified by leaving a suitable buffer around the study area or normalized by modifying the kernel estimate function such that points near the edges are weighted heavier than those further away. Kernel density estimation has been used to model hotspot areas of wildlife mortality along a major road in Australia (Ramp et al., 2005) and to isolate problem neighborhoods characterized as crime hotspots in Lima, Ohio (Ackerman and Murray 2004), Trisalyn and Boots (2005) identified hotspot areas infested with pine beetles in

18 British Columbia forests, and Johnson et al. (2006) estimated the risk of exposure to West Nile virus in New York State using kernel density estimation.

3.1.1.3 Nearest Neighbor Analysis Nearest neighbor distance analysis measures distances between sample points and their nearest neighboring points (Lancaster and Downes, 2004). The concept behind nearest neighbor analysis just like quadrat analysis is randomness (Wall et al., 1985). However, unlike quadrat analysis which uses a grid of cells, this method is based on comparing the observed average distances between nearest neighboring point events and those expected in a random point process (Lee and Wong, 2001). The nearest neighbor statistic (R) is the ratio of the average distance from the closest neighbor to each point event and the distance expected based on chance. This means that R = 1 if the observed average distance between nearest neighboring points is equal to the mean distance expected in a random point process. When R < 1, the point process is clustered, but dispersed if R > 1. However, analysts like to find out whether a certain distribution may have occurred purely by chance. In this case, the standard error of the distribution is used to compute Z score, in which significant distribution is exhibited when -1.96 ≥ Z ≥ 1.96. Nearest neighbor analysis is a more popular point pattern analysis technique in many fields because it lacks many problems associated with quadrat analysis. This is because it takes into account spatial arrangement of point events. However, the method is prone to errors that may be introduced in the analysis by irregular study area boundaries. In addition, nearest neighbor analysis results are dependent on the geographic scale used in the analysis. For example, analyzing the distribution of cities in Ohio at the State scale would produce different results from the patterns that would result from the same analysis at country, or continental scales. Wall et al. (1985) used nearest neighbor analysis to explore spatial distribution of accommodation units in Toronto Canada. They found that the method captured dispersed patterns adequately; however, it depicted some seemingly clumped patterns as random. Heupel and Simpfendorfer (2005) used nearest neighbor analysis to study the distribution

19 of juvenile black sharks in Terra Ceia bay, Florida. The technique captured adequately temporal aggregation of juvenile sharks over a period of three years.

3.1.1.4 Standard Deviation Ellipse Some point processes are characterized by directional bias. For instance, many residential neighborhoods follow road network. In such a case, standard deviation ellipse explores best how distribution of point events deviates from the mean center of the distribution. It shows the directional bias of the point process and the degree of eccentricity (Gesler, 1986). Standard deviation ellipse is a very good way of visualizing a point pattern to detect temporal changes in directional distribution of the point process. However, it is affected by boundaries and does not establish aggregation or dispersion in a point process. Wall et al. (1985) used standard deviation ellipse to analyze spatial distribution and temporal trends in accommodation units in Toronto, Canada. Unlike quadrat and nearest neighbor analyses, he found out that standard deviation ellipse performed well in estimating pattern orientation, shape and spatial shift in the distribution. Standard deviation ellipse has also been used to analyze daily activities of health care seekers among black residents of Washington, D.C (Shannon et al,. 1978).

3.1.2 Second Order Statistics

A point pattern may be produced by a process that exhibits varying distribution patterns if examined over a range of distance scales. For instance, a seemingly clumped distribution may exhibit dispersed or random pattern when examined at multiple spatial scales (Lancaster and Downes, 2004). Second order statistics are thus used to describe correlation or covariance between values of the point process at different spatial scales (Gatrell et al., 1996; Wiegand and Moloney, 2004) and are based on distribution of pairs of point-to-point distances in the study area (Barff and Hewitt, 1989). Second order statistics can detect spatial heterogeneity in a point process, and thus show varying scales at which a particular pattern exhibits strongest clustering.

20 The commonly used second order analysis of a point pattern is the Ripley’s K- function (Ripley, 1981). Ripley’s K-function assumes that the point process is stationary and isotropic. Gatrell et al. (1996) describe a stationary and isotropic point process as one in which the intensity of the point process is constant across the study area and is only influenced by the point-to-point distance, but not its orientation. Assuming isotropy and stationarity in the point process, Ripley’s K-function is defined as:

K(d) = λ-1E(Nd) Where: λ is the number of point events per unit area E( ) denotes expectation operator Nd describes the number of point events within a distance d of an arbitrary event selected randomly from the study area (Diggle, 1983).

The K-function within a specified distance d relates the observed events within distance d of each other to the number of events expected to occur within distance d of each other under assumption of a null distribution. The commonly used null distribution is Complete Spatial Randomness (CSR) generated using a homogeneous Poisson process. Under CSR, K(d) = πd2. If K(d) > πd2 the observed distribution exhibits clustering. Likewise, an observed distribution in which K(d) < πd2 exhibits dispersion at d. By plotting K(d) - πd2 against d, we can visualize the scale of d at which the point process exhibits varying degrees of clustering or dispersion. Monte Carlo simulations are used to compute minimum and maximum K(d) under CSR to test the significance of the distribution. This creates an upper and lower envelope above which K(d) exhibits significant clustering and below which it exhibits significant dispersion. Khaemba (2001) explored and quantified clustering of wildlife populations from aerial survey in Laikipia District in Kenya using Ripley K-function. He observed that Ripley K-function is suitable for identifying clustering in observed spatial distribution and for testing hypotheses related to animal distribution. O’Brien et al. (1999) used Ripley K-function to assess spatial and temporal aggregation of cancers in dogs in Michigan between 1964 and 1994. The researchers concluded that the Ripley K-function has a good potential for assessing and generating hypotheses about disease aggregation

21 especially where it is difficult to ascertain the population at risk of being infected. However, like all distance-based functions, K-function suffers edge effects when the number of neighboring points is underestimated as a result of some events lying outside the study area boundary (Wiegand and Moloney, 2004). Such effects become pronounced especially when the number of point events in the pattern is small. Edge correction is done using Ripley’s local weighting factor (Lancaster and Downes, 2004), or alternatively by extending a buffer zone around the study area (Gatrell et al., 1996; Wiegand and Moloney, 2004). The Ripley’s local weighting factor attaches more weight to the points close to the boundary thus offsetting the edge effects. Edge correction in itself can, however, introduce errors in the analysis if the assumption that the point density in the outside of the study area is similar to that within it does not hold (Lancaster and Downes, 2004).

3.2 Human-Caused Wildlife Mortality

Overexploitation, whether legal (hunting) or illegal (poaching) has been identified as an important factor driving wildlife to the verge of extinction (Diamond, 1984; Harper, 1960). Interestingly, there has been little research on spatial patterns of human-caused wildlife mortality because many researchers concentrate on examining spatial aspects of live animals. What has always struck me as a field wildlife officer has always been what can be learned from assessing patterns of human-induced wildlife mortality and the usefulness of such information in the management and conservation of the living wildlife. Spatial aspects of human-induced wildlife mortality are not well documented and the available literature is little. Nielsen et al. (2004) used GIS to model the spatial distribution of human-caused grizzly bear mortality in the Central Rockies ecosystem of Canada. A bear’s relative risk to death was found to be related to those landscape biophysical and human factors that described human accessible habitats in those locations bears were likely to frequent. These biophysical and human factors included access to water, roads, slope, elevation, terrain ruggedness and vegetation. The relationship between bear mortality and the biophysical and human factors explains why bear deaths

22 were found to be localized around Lake Louise, the city of Banff and the Red Deer River northwest of Calgary, rather than being random or dispersed within the study area. Biophysical and human factors influence distribution and habitat preferences for different animal species in their natural habitats (Musiega and Kazadi, 2004). As such, an animal will select a habitat that enhances its chances of survival and reproduction. Unfortunately, not all the preferred habitat actually enhance an animal’s survival and reproduction success, but can instead become mortality zones if they are located where wildlife and human interests are bound to clash (Delibes et al., 2001), or at worse are targeted by poachers. Various studies have been conducted on spatial aspects of elephants including population size and distribution (Barnes et al. 1995; Omullo et al. 1998), home range estimation and habitat preferences (Douglas-Hamilton, 1998), and human-elephant conflicts (Hoare, 2000; Smith and Kasiki, 2000; Sitati et al. 2003). However, little has been done to study human-caused elephant mortality. Nevertheless, available literature provides invaluable information about biophysical and human factors that tend to influence elephant distribution and by extension may influence patterns of elephant mortality. De Boer et al. (2000) studied diet and distribution of elephants in Maputo Elephant Reserve in Mozambique and established that elephants preferred denser forested patches and proximity to fresh water during the dry season. They also found out that elephants preferred to feed on grass during the rainy season, while in the dry season they concentrated on the few available browse plants. Babaasa (2000) studied habitat preferences by elephants in Bwindi Impenetrable National Park in Uganda and also concluded that elephant occurrence coincided with seasonal changes in rainfall and food availability. In another study carried out in northern Botswana, Verlindern and Gavor (1998) established that distribution of elephants was influenced by habitat type and proximity to permanent surface water during the dry season and food nutrition during the wet season. Khaemba and Stein (2000) found elephant occurrence in the Maasai Mara National Reserve was positively correlated with tall grass and proximity to water, but was negatively correlated with proximity to roads. Smith and Kasiki (2000) while studying human-elephant conflicts in the Tsavo ecosystem found that elephants tended to avoid

23 areas of steep slopes and higher elevation, and remained in close proximity to permanent water. Ben-Shahar (1999) studied elephants in woodland habitats of northern Botswana and established that the elephants’ diet consisted of 60% grass and 40% woody plants. He also found proximity to surface water during the dry season determined the distribution of elephants. Demeke and Bekele (2000) examined elephants of Mago National Park in Ethiopia using interviews and dung counts along transects. They found that the elephants remained in the patchy forest bush habitats and opted to move along foothills dominated by bush vegetation, a maneuver suggesting the elephants were avoiding human contact. The authors also established that elephant carcasses occurred mainly along riverine and forest areas during the dry season. Mpanduji et al. (2002) studied movement of elephants along Selous-Niassa corridor in southern Tanzania and found the key factors responsible for elephant movements were water, food, and to some extent, human disturbance. A study on population and distribution of elephants in the central sector of Virunga National Park in the Democratic Republic of Congo revealed that elephants preferred thick-bushed grassland and forest with easy access to water, but avoided steep rocky slopes (Mubalama, 2000). The study also established that elephants exhibited compressed distribution preferring to aggregate in localized places, a characteristic reminiscent of elephant populations that had experienced heavy of poaching (Ruggiero, 1990). Pilgram and Western (1986) and Leader-Williams et al. (1990) hypothesize that poaching occurs in remote areas with abundant elephants logically because remote areas are difficult for patrol rangers to access due to inadequate routes. Inaccessibility to remote areas can thus afford poachers a chance to poach elephants without their activities being detected by patrol rangers. From the literature reviewed, it is clear that proximity to surface water (Ben- Shahar, 1999) and food availability (Babaasa, 2000) influence elephant distribution. In addition, elephant distribution varies with land cover (Verlinden and Gavor, 1998; De Boer et al. 2000), slope and elevation (Smith and Kasiki, 2000; Mubalama, 2000) and is also influenced by human factors. The human factors include distance to roads (Khaemba and Stein, 2000) and proximity to human habitation (Mpanduji et al. 2002; Douglas- Hamilton, 1998).

24 3.3 Land Cover Mapping from Satellite Imagery

Wildlife managers need reliable land cover information on which to base decisions on conservation of wildlife and wildlife habitats and for policy formulation, development and implementation. Such information is often insufficient in developing countries owing to high cost of production, inadequacy of technical know-how and inaccessibility due to poor infrastructure among others hurdles (Haack and English, 1996). Consequently, conservation decisions are largely opportunistic and ad hoc principle (Leader-Williams et al., 1990). Remote sensing is an indispensable tool that occupies an important information niche for natural resource managers and decision makers as a source of information for land cover mapping (Wyatt, 2000; Han et al., 2004). Land cover maps are increasingly being incorporated with other spatial data such as digital elevation model (DEM), drainage, communication networks, and human population distribution to address many wildlife problems of spatial concern. For instance, land cover maps are being incorporated in studies involving human-wildlife conflict (Sitati et al., 2003), home range studies (Douglas-Hamilton, 1998) and population distribution (Omullo et al., 1998). Traditionally, land cover mapping has been based on aerial photo interpretation and in situ data collection (King, 2002; Millington and Alexander, 2000). However, such methods are laborious and quite expensive especially if large areas are to be mapped (Mertens and Lambin, 2000). The launch of the first Landsat MSS in 1972 ushered in the era of multspectral satellite imagery. Satellite imagery provides a synoptic view of large areas that can be replicated through repeat coverage and stored for permanent record (Haack and English, 1996). Satellite imagery provides information in digital format, which allows for automated statistical analysis of data for land cover mapping and easy integration in a GIS, thus saving time and processing resources (Lillesand and Kiefer, 2000). The transformation of satellite imagery into valid vegetation maps with physiographic descriptors has been proposed as an essential goal for wildlife managers (Franklin et al., 2002). Generating a land cover map from satellite imagery involves automated digital processing and classification of multispectral satellite images. Image classification relies mainly or solely on spectral reflectance values of each pixel in the image scene (King,

25 2002). Spectral characteristics exhibited by the various land covers present in a study area form distinct spectral signatures that analysts use to discriminate them (Lillesand and Kiefer, 2000). Decision rules based on statistical pattern recognition techniques are then applied to extract land cover information. Supervised and unsupervised image classification algorithms are the commonly used thematic information extraction techniques (Jansen, 2005). In unsupervised classification, the analyst applies algorithms that examine the unknown pixels aggregating them according to similarities in their spectral reflectance values (Martinez-Casasnovas, 1998; Lillesand and Kiefer, 2000). Spectrally separable classes are defined, and then information class labels assigned by comparing the classes with ground referenced data. Mixed pixels are usually masked out then reclassified until satisfactory separation is reached. Several clustering algorithms are used in unsupervised classification, including Iterative Self-Organizing Data Analysis (ISODATA) method, chain method, and Cluster Busting (Jansen, 2005). Unsupervised classification is valuable especially when one is working in an area he/she is not quite conversant with. It does not require one to have a foreknowledge of the land cover classes (Liu, 2005). In the process of land cover classification, many errors occur (Jansen, 2005; Congalton and Green, 1999). These errors lead to some pixels being given wrong labels (error of commission) while others are not included in their correct information classes (error of omission). It is thus important to assess the degree of accuracy the produced map exhibits. A simple qualitative accuracy assessment involves scanning the map to detect presence of mistakes. These are obvious errors, for example, an agricultural field mapped in a lake or a city on top of a mountain. However, the commonly used accuracy assessment method is the Error Matrix. An error matrix constitutes a square grid of numbers in columns representing reference data and rows representing the thematic information generated by the classification (Congalton and Green, 1999). From the error matrix, overall accuracy, producer’s accuracy and user’s accuracy can be computed. In addition, statistical error assessment can be done using the various forms of Kappa statistics present (Jansen, 2005; Lillesand and Kiefer, 2000; Congalton and Green, 1999).

26 3.4 Research Hypothesis The working hypothesis for this study was that human-induced elephant mortality is not random, and is described by an association of biophysical and human factors that influence distribution of elephants in a habitat area and access by poachers to these habitats elephants are likely to frequent. Based on available literature, I expect elephant mortality to occur close to permanent sources of water, and in denser vegetation. I also anticipate higher levels of poaching further from patrol bases and human settlements. This is because proximity to surface water (Ben-Shahar, 1999) and food availability (Babaasa, 2000) influence elephant distribution. In addition, elephant distribution varies with land cover (Verlinden and Gavor, 1998; De Boer et al., 2000) and is influenced by human disturbance as indicated by distance to roads (Khaemba and Stein, 2000) and proximity to human habitation (Mpanduji et al., 2002; Douglas-Hamilton, 1998). Slope and elevation have also been found to influence elephant distribution (Smith and Kasiki, 2000; Mubalama, 2000).

27 CHAPTER FOUR

4.0 Data and Methods

4.1 Study Area

TENP and Tsavo West National Park (TWNP) form the core of the Tsavo ecosystem, with Chyulu Hills National Park in Kenya and the Mkomazi Game Reserve in Tanzania forming important peripheral components (Wijgaarden, 1985). Gazetted in 1948, TENP is among the oldest parks in Kenya (Smith and Kasiki, 2000). It covers approximately 12000 km2, accounting for about 40% of the total area covered by parks in Kenya (KWS, 2003). Like all other parks in Kenya, TENP is held in trust by the KWS, which is charged with the duties of administering and managing national parks on behalf of the people of Kenya. The park and its wildlife resources are thus protected through Wildlife Conservation and Management (Amendment) Act CAP 376 Laws of Kenya. As such, only non-consumptive tourism is legally permissible within the confines of the park. TENP is located in southern Kenya on the eastern part of Yatta plateau and north of Nairobi-Mombasa highway approximately 330 km from Nairobi (Figure 3). Elevation within the park increases westwards from 150 m in the eastern park boundary to 1200 m in the western boundary (Tolvanen, 2004). Rainfall is bimodal with the long rains occurring between March and May, and the short rains occurring between October and December. March – May and October – December therefore, represent the wet season, while January – February and June – September for the dry season in TENP. Like the elevation, the rainfall increases from about 250 mm in the eastern part of the park to about 450 mm in the western parts (KWS, 2003). However, the rainfall is variable and inconsistent in space and time (McKnight, 2000), sometimes starting or ending before or after the anticipated time as was observed in 2004 (Figure 4). The even topography characterizing much of TENP is occasionally punctuated by isolated hills, although steeper slopes occur in the western parts (Kasiki, 1998).

28

Figure 3: Location of Tsavo East National Park in Southern Kenya.

The main source of permanent water in the park is the Galana River, which is formed by the union of Athi and Tsavo rivers. Seasonal sources of water include: the Tiva and Voi rivers, the Aruba dam and a few scattered ponds and swamps (Figure 3). Vegetation within TENP is mainly bushland/grassland savannah and semi-arid Acacia and Commiphora woodlands with Premna, Bauhinia and Sericocomorpsis scrub scattered with Delonix eleta and Melia volkensii trees and interspersed with open plains (McKnight, 2000). Riverine vegetation dominated by Acacia elatior, Hyphaene compressa, and Suaeda monoica occurs along the rivers. The vegetation is generally denser in the western part of the park and lighter in the eastern part, corresponding to a decreasing rainfall gradient (KWS, 2003).

29 200 180 160 140

m) 120 m (

ll 100 a f

in 80 a

R 60 40 20 0 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Month

Figure 4: 2004 Rainfall distribution based on data obtained from TENP records

Owing to its large size, TENP has the highest abundance of wildlife in Kenya and is thus perceived as an important world biodiversity area (Smith and Kasiki, 2000). Available records show over 32 species of large mammals and over 324 species of birds occur in Tsavo ecosystem (Cobb, 1976), but it is elephants that are considered the keystone species because their feeding behaviors have profound ecological effect on the ecosystem (Leuthold, 1977; Wijgaarden, 1985).

4.1.1 The Tsavo Elephant Problem

It is argued that the portion of land covered by TENP and TWNP was declared protected largely because the colonial government perceived the area as being of marginal agricultural potential and largely unoccupied (Kasiki, 1998). However, hunter- gatherer and pastoral communities utilized this area. Notable among the communities that hunted elephants were Waliangulu, Wakamba and Giriama, who supplied ivory to Arabs, European, and Indian ivory merchants in the Kenyan coast (Parker and Amin, 1983; Ville, 1995; Kassam and Bashuna, 2004). Their traditional hunting techniques using bows and arrows had little impact on the Tsavo elephant population, which still remains the largest in Kenya. When TENP and TWNP were gazetted in 1948, some Waliangulu and Wakamba men turned to fulltime poaching of elephants and rhinos. This prompted

30 the colonial government to mount aggressive anti-poaching campaign between 1956 and 1957 (Sheldrick, 1973; Akama, 1998), which according to Steinhart (1994) was the most successful in Africa. The earliest elephant census shows an estimated population of 9413 in 1962 (Laws, 1969). Subsequent elephant censuses denote an increase in population to the tune of 13068 in early 1970s (Leuthold, 1977). However, a major drought followed that claimed over 50% of the TENP elephant population (Corfield, 1973). The die-off provided ready ivory for the poachers and ivory trade thrived to an extent that it attracted men of Somali origin (Parker and Amin, 1983). With the Somalis came automatic weapons that hastened the decline of the already dwindling TENP elephant population. Poaching by Kamba and Somalis armed with firearms returned in full force between mid 1970s and late 1980s (Kassam and Bashuna, 2004). This decimated TENP elephant population to less than 20% of the population recorded in early 1970s (Poole et al., 1992). Poachers of Somali origin thought to come from the neighboring Tana River District in Kenya and across the international border in Somalia still continue to wreak havoc in TENP (Kioko, 2002). Although elephants in the general Tsavo ecosystem have registered between 3-4% annual population growth since 1990 (Society, 2005), persistent poaching in TENP requires KWS to rethink its strategies to effectively protect Kenya’s elephants. According to Born Free Foundation (2000) the level of elephant poaching increased fivefold between 1998 and 1999.

4.2 Data

Data obtained for the present study included elephant mortality, topographic maps, Landsat ETM+ and MODIS satellite imagery, digital elevation model (DEM) and field data. All subsequent analyses were performed using ArcGIS 9.1, ArcView 3.2, and ERDAS IMAGINE 8.7 software.

31 4.2.1 Elephant Mortality Data

Over the years, KWS has recorded elephant mortality reported countrywide as a routine practice. These data are archived at KWS headquarters in Nairobi and were obtained for this study during the 2005 summer fieldwork. 210 elephant mortality records were available as from October 10th 1989 to July 2nd 2005. These data included locality where death occurred, date of death, geographic location, and cause of death. Unfortunately, some of the earlier records were missing spatial reference and were therefore omitted. There were 203 complete elephant mortality records obtained for the period December 18th 1990 to July 2nd 2005 for the area within and in the vicinity of TENP. Causes of elephant mortality were categorized as poaching, control, conflict, accident, sickness, natural, euthanasia, and in some cases it was unknown. Elephant mortality data were first entered in an Excel spreadsheet with the following fields: date of mortality, UTM X-coordinate, UTM Y-coordinate, cause of mortality, and place mortality occurred. The data were then converted into DBF 4 (dBASE IV) format and imported into ArcGIS to create a point shapefile of elephant mortality and associated attribute data (Figure 5). A 10-km buffer of TENP was generated (Figure 5) and used to clip elephant mortality data to the buffer area. It was felt that the 10-km buffer would be sufficient to reduce edge effects during point pattern analyses analyses. The spatial extent of all subsequent data sets generated for this study were based on the 10-km buffer.

32

Figure 5: Elephant mortality distribution and TENP 10 km buffer.

4.2.2 Satellite Data

Four recent Landsat ETM+ images were acquired to cover the location of the study area (Figure 6). Landsat ETM+ images were preferred because of the reasonably good spatial resolution and widespread success in land cover mapping for projects, especially those involving natural resources (McCarthy et al., 2005; Muller et al. 1999; Cherrill et al., 1994). The four images selected for the study (Table 1) were selected because they were the most recent, were obtained during same season, and had minimal

33 cloud cover. The data were obtained at no cost from the Global Land Cover Facility (GLCF) server at the University of Maryland.

Figure 6: Landsat WRS-2 Path and Row scene intersections in TENP

Downloaded Landsat ETM+ images that excluded the thermal band (ETM+ 6) were mosaicked and reprojected from the default WGS 84 UTM ZONE 37N to WGS 84 UTM ZONE 37S. Next, a mask of the study area which included a 10-km buffer was used to extract only the area corresponding to it from the mosaic of the four Landsat ETM+ images.

34 Table 1: Landsat ETM+ images used in the study Image Path Row Date taken 1 167 61 March 4th 2001 2 167 62 March 4th 2001 3 166 62 January 22nd 2000 4 166 63 January 22nd 2000

MODIS Normalized Difference Vegetation Index (NDVI) images were generated from 32-day 500m composites downloaded from GLCF, University of Maryland. The images were available from November 16th 2000. The MODIS data came in Interrupted Goode’s Homolosine projection and covered the whole of Africa. The data were subset to Kenya then reprojected to UTM zone 37S. Next the, study area was extracted from the MODIS images using a mask derived from applying a 10-km buffer to the TENP boundary. The resulting 2-band MODIS images were used to generate NDVI images. The NDVI composites would be a good indicator of the spatial and temporal variability in available food resources for the elephants because NDVI is a good indicator of rainfall variability (Eklundh, 1998) and vegetation biomass (Tappan et al., 1992; Moreau et al., 2003). The NDVI composites were also useful for broadly identifying different vegetation types through interpretation of seasonal changes in vegetation phenology (Moleele et al., 2001).

4.2.3 GIS Data

Roads, rivers and waterholes for TENP were digitized in ArcGIS using the mosaic of Landsat ETM+ images (Figure 7). However, some features like established animal trails were difficult to distinguish from park roads and many of the waterholes were indistinguishable in the Landsat ETM+ images because they were smaller in size than the image pixels. 1:50000 topographic maps based on aerial photography were thus used to supplement information generated from the Landsat ETM+ images.

35

Figure 7: GIS layers generated

Thirty topographic maps based on aerial photography of between 1954 and 1989 were obtained from Kenya Department of Resource Survey and Remote Sensing (DRSRS) during fieldwork in summer 2005. The topographic maps were scanned at 400 dpi and saved in TIFF format. The TIFF files were imported to ERDAS IMAGINE and geometrically corrected. The maps were subsequently reprojected to WGS 84 UTM Zone 37S, the common projection for all data sets. A 90m DEM compiled from the Shuttle Radar Topographic Mission (SRTM) was used to obtain elevation grid and slope of the study area. Lastly, all the six ranger patrol

36 bases (Figure 7) were visited during 2005 summer fieldwork and their locations determined using a hand-held global positioning system (GPS). The information collected with the GPS was downloaded and imported into ArcGIS’ ArcMap to create a point shapefile.

4.3 Methods

The data and analytical methods used in the present study are presented in a simple schematic diagram below (Figure 8)

37 Figure 8: A flowchart of data and methods used in the study

Obtained from Obtained from Obtained from GLCF data server at KW S DRSRS University of Maryland

Elephant Landsat Topographic MODIS mortality ETM+ maps images data images

Roads, rivers, GIS database for waterholes and TENP NDVI images elephan t mortality boundary digitized generated

Acquired in the field (TENP) Point pattern Patrol base Ground Image analyses location referenced classification data data

Mortality Land distribution and density cover map maps

Obtained from SRTM Correlation Slope and elevation & regression generated DEM analyses

Risk to poaching models

38 4.3.1 Describing patterns of Human-Induced Elephant Mortality

The spatial patterns of elephant mortality attributed to poaching were examined based on wet and dry seasons in TENP, and on two periods in the history of KWS. The periods of interest in history of KWS were 1990 – 1997 when CITES’ complete ban on ivory trade was in force and 1998 – 2005 after CITES partially lifted the ban. In addition, elephant mortality patterns were also examined for non-poaching mortality irrespective of cause or season. For each elephant mortality category analyzed, the mortality dataset was selected by attribute then exported to create point shapefiles in ArcGIS’ ArcMap. Six elephant mortality point shapefiles were generated for mortality pattern analyses (Table 2). The elephant mortality datasets were then subjected to four exploratory first order descriptive statistical analyses including quadrat analysis, nearest neighbor analysis, standard deviation ellipse and kernel density estimation. It was felt that exploring elephant mortality using these point pattern techniques would show both graphically and statistically spatial patterns the elephant mortality in each category assumed. The use of multiple point pattern analysis techniques would further show among them, the stronger indicator of elephant mortality patterns for the present study.

Table 2: Categories of mortality datasets used to describe elephant mortality patterns Point shapefile Number of records Overall poaching 75 Wet season poaching 40 Dry season poaching 35 1990 – 1997 poaching 24 1998 – 2005 poaching 51

4.3.1.1 Quadrat analyses Quadrat analysis was performed for each mortality dataset from point pattern analysis implemented as a menu item in ArcView 3.2. A complete grid of 77 squares each measuring 21315.5m in length was laid across the study area and analyses performed at 0.05 level of confidence. The grid of 77 squares was calculated using the

39 formula provided by Lee and Wong (2001). For each mortality dataset, the following variables were recorded: (a) Number of mortality records per quadrat [λ] (b) Variance (c)

Variance-Mean Ratio [VMR] (d) K-S D statistic (e) Critical D statistic [Dα0.05] and (f) Remarks.

4.3.1.2 Nearest Neighbor Analyses First order nearest neighbor analysis was performed for each elephant mortality dataset using ArcView 3.2. The following variables were recorded for each mortality dataset: (a) Observed average distance (m) between nearest neighboring mortality records [Observed neighbor distance] (b) Expected average distance (m) between nearest neighboring mortality records [Expected neighbor distance] (c) Nearest neighbor statistic [R statistic] (d) Z score and (e) Remarks.

4.3.1.3 Standard Deviation Ellipse Analyses ArcGIS was used to generate standard deviation ellipses associated with each poaching-induced elephant mortality in order to ascertain its spatial spread and directionality. In each case, the mean center of spatial distribution and standard deviation ellipse for each mortality dataset was recorded.

4.3.1.4 Kernel Density Analyses Kernel density analyses were performed to identify areas within the study area that were hotspots for elephant poaching. Kernel density analyses for different elephant mortalities due to poaching were performed using ArcGIS. A band width of 23.9km was selected in the analysis because it corresponds to mean home range size for TENP elephants. Female elephants in TENP have an average home range of 2400 km2 while that for males averages at 1200 km2 (Leuthold and Sale, 1973). During the dry season, however, mean elephant home range expands (Leuthold and Sale, 1973), and as such, a wider band width (31.5km) was used for analysis of dry season poaching.

40 4.3.2 Land Cover Mapping

Although acquisition of the Landsat ETM+ images used in the present study was near anniversary, there were still some phenological differences evident in the images; a problem frequently encountered in semi-arid regions because of erratic and highly variable rainfall pattern. The classifications were therefore, done in two separate mosaics of the Landsat ETM+ images (Figure 9) in order to minimize classification errors resulting from phonological differences.

(a) (b)

Figure 9: (a) Images acquired on 3/4/2001 (b) Images acquired on 1/22/2000

4.3.2.1 Classification Scheme The land cover classification scheme used in this study consisted of nine classes (Table 3). These classes were selected based on descriptions of the vegetation types of the Tsavo ecosystem (Wijgaarden, 1985; KWS, 1996; KWS, 2003).

41 Table 3: The land cover classification scheme used for the study Land cover type Description Forest Occur along permanent rivers and semi-permanent streams. It is also found in high elevation areas including the Yatta plateau and on hills. Dominant species include Acacia elatior, Hyphaene compressa and Suaeda monoica. Woodland Areas covered by trees with largely open canopies. The trees are more spaced than forests and shed their leaves during the prolonged dry spell between June and October. Dominant plants are Acacia and Commiphora species Bushland Areas dominated by short closely spaced multi-stemmed plants. The plants lack definite crown and they can be erect or spread. Common species are Premna, Bauhinia and Sericocomorpsis with occasional Delonix elata and Melia volkensii trees. Open bushland Areas characterized by vegetation similar to that found in bushland, but the plants are relatively smaller and highly scattered. Herbs sprout in this vegetation during rain seasons, but die immediately there after. Herbaceous cover Occurs on temporarily flooded soils characteristic of black cotton soils. Vegetation in this category consists of mostly coarse grasses, herbs and forbs Grassland A vegetation community in which grasses are the most conspicuous plants. It is also dotted with Acacia spp., Commiphora spp., Delonix elata and Melia volkensii Active Agriculture Areas under cultivated crops that are growing. They occur outside the park boundary in areas inhabited by people. Barren Land Soil surface without plant material growing on it. Areas covered by rocks, dry stream beds, built up areas, fallow farms and roads fall in this category. Water Areas covered by open surface water. Permanent rivers, dams and waterholes.

42 4.3.2.2 Image Classification Unsupervised classification was used in this classification because it is a more suitable technique in classifying heterogeneous land covers such as those present in TENP. Moreover, unsupervised classification was more suitable because obtaining adequate representative samples for the different land cover types was constrained by the large size of the study area, its remoteness, and inadequate timeline for the fieldwork. Before performing an ISODATA classification, an NDVI image was created and thresholded to create a vegetated and non-vegetated images which were then classified separately. The number of spectral clusters specified in each classification was set to twice the number of land cover classes targeted for classification and the number of iterations set to twenty six in order to ensure that the classification was not terminated before the specified number of clusters was fully separated. In the initial classifications, clouds and shadows were masked out. Pixels that were not successfully separated in each classification were masked from the original Landsat ETM+ images and from spectral enhancements created using optimal band combinations for separating the mixed pixels then reclassified until satisfactory pixel separation were achieved. Pixels that were satisfactorily separated in each classification attempt were assigned a number representing their informational class labels. The information class labels were assigned based on interpretation of the Landsat ETM+ images and photographs taken during fieldwork. Often, however, single-date satellite imagery were insufficient for identifying land cover types because of phenological differences between the images. For example, senescent grass in a Landsat ETM+ image could be incorrectly assigned to the bare soil class. Using multitemporal NDVI images derived from MODIS allowed different spectral clusters to be assigned to their appropriate informational classes as these data sets capture phenology of different land cover types. A final map was created by overlaying all the satisfactorily separated pixels in each classification and/or reclassification. During the overlay, all spectral clusters were recoded to their appropriate information classes. In the final map, informational class labels were replaced with the actual land cover names.

43 Classification accuracy assessment was performed on the final map based on reference points collected during fieldwork and others obtained from interpretation of 1:50000 topographic maps based on 1954 – 1989 aerial photography. An error matrix was constructed and producer’s, user’s accuracy, overall accuracy, and Kappa statistics calculated.

4.3.3 Exploring relationships between elephant mortality patterns and biophysical and human variables

Elephant mortality caused by wet season poaching, mortality caused by dry season poaching, and mortality caused by annual elephant poaching irrespective of season were correlated with each of the following biophysical and human factors: (a) distance to patrol bases (b) distance to park gates (c) distance to park roads (d) distance to park boundary (e) distance to main rivers (f) distance to seasonal rivers (g) distance to waterholes (h) elevation (i) slope. In addition, the elephant poaching datasets were explored in relation to TENP land cover types. ArcGIS’ Spatial Analyst was used to create distance surfaces to ranger patrol bases, park gates, park roads, park boundary, permanent rivers, seasonal rivers, and waterholes. Next, the distance surfaces were used to extract distances of all elephant poaching mortality locations in every poaching category to the biophysical and human factors. Additionally, elevation and slope, and land cover types at every elephant mortality location were extracted from DEM and TENP land cover map respectively for each poaching category. Kernel density surfaces for each elephant poaching category were generated using a band width of 5640m, which is the park area a ranger is expected to patrol in TENP Ottichilo (1987) also used the same scale to study elephant mortality in Tsavo ecosystem. Next, densities of elephant mortality in all poaching categories were extracted at every mortality location from the kernel density surfaces. All extracted elephant mortality density values in each poaching category were correlated with corresponding distance (or elevation and slope) values extracted for each biophysical and human factor under examination. Before correlation for each set of

44 variables was done however, the datasets were tested for normality in distribution using Kolmogorov-Smirnov method. Normality was assumed when P > 0.05. Spearman’s rank correlates were thus used because almost all the datasets were not normally distributed. Land cover types were category rather than numerical data and therefore, their relationships with the various poaching categories needed to be assessed differently. The land cover types were extracted at each mortality location for all elephant poaching categories examined in this analysis. The number of times (frequency) elephant mortality occurred in each land cover type was tabulated for every elephant poaching category and its respective percentage calculated for comparison with percentage size of corresponding land cover.

4.3.4 Generating Risk to Poaching Maps

Biophysical and human factors that exhibited significant correlation with the various categories of elephant poaching mortality were used to generate risk to poaching surfaces for TENP using the Model Builder in ArcGIS. Each biophysical or human factor used to create risk to poaching models was reclassified into ten classes using equal interval method so that elephant risk to poaching would be assessed in a scale of 1 – 10. For each biophysical or human factor classified, new values were assigned according to the relationship the biophysical or human factor exhibited when correlated with the various elephant poaching categories analyzed. For instance, distance to main rivers, which exhibited negative correlation with dry season poaching was recoded into 10 classes. A new value of 10 was assigned to the shortest mortality location distance to the nearest main river, while a new value of 1 was assigned to the longest mortality location distance to the nearest main river. For a positive correlation, a new value of 10 was assigned to the longest mortality location distance, and a new value of 1 assigned to the shortest distance to the biophysical or human factor being reclassified. During reclassification of land covers, new values were assigned to each land cover type in accordance to the elephant mortality density occurring in it for every elephant poaching category examined. No weighting was applied to the different reclassified datasets during overlaying because their relative importance to elephant mortality was unknown.

45 CHAPTER FIVE

5.0 Results and Discussion

5.1 Elephant Mortality Analysis Results

Results from quadrat analyses showed that elephant mortality in TENP was not random (D > Dα0.05) but exhibited clustered patterns (VMR > 1) irrespective of season or period of KWS history examined (Table 4). In addition to showing that elephant mortality in TENP was clustered (R < 1), nearest neighbor analyses confirmed that certain factors other than chance (Z > 1.96) influenced elephant mortality patterns (Table 5).

Table 4: Quadrat Analysis Results for Poaching-based Mortality

Poaching λ Variance VMR K-S D Dα 0.05 Remarks Category Statistic Overall 0.974 4.879 24.388 0.369 0.157 Clustered poaching Wet season 0.519 4.031 29.846 0.955 0.215 Clustered poaching Dry season 0.455 4.883 36.590 1.200 0.275 Clustered poaching 1990 - 1997 0.312 5.105 52.151 2.212 0.301 Clustered poaching 1998 – 2005 0.662 3.867 24.193 0.641 0.190 Clustered poaching

Studies on social organization of elephants have shown that elephants of TENP exhibit aggregation all year round (McKnight, 2000). The aggregations constitute large groups that are not family units but those formed in response to stress, harassment and

46 lack of matriarchs to lead family units as a result of previous heavy poaching (Lewis, 1986; Ruggiero, 1990). The aggregating behavior of previously heavily poached elephant populations may therefore, explain why elephant mortality exhibited clustered patterns in TENP irrespective of season or KWS historical period analyzed. In addition, poachers often kill more than one large individual elephant in herd in an effort of maximize harvest, thus resulting to clusters of poached elephant carcasses.

Table 5: Nearest Neighbor Analysis Results for Poaching-based Mortality Mortality Observed Expected R statistic Z score Remarks category distance (m) distance (m) Overall 2,736.1 12,005.9 0.228 12.793 Clustered poaching Wet season 4,201.9 16,439.8 0.256 9.007 Clustered poaching Dry season 3,294.7 17,574.9 0.187 9.197 Clustered poaching 1990 - 1997 4,470.1 19,458.3 0.230 7.220 Clustered poaching 1998 – 2005 3,408.2 13,348.3 0.255 10.174 Clustered poaching

Standard deviation ellipse results showed a spatial distribution for overall poaching centered slightly in the northern part of the park (Figure 10). The results indicate a higher concentration overall poaching mainly in the central and northern parts of TENP. Kernel density results depicted overall poaching hotspots as occurring mainly along the main river courses in the park (Figure 10). Concentration of poaching hotspots along the main rivers indicates the influence of proximity to surface water on the distribution of elephants in the park because elephant drink and mud-bathe daily (Estes, 1999). The results are in agreement with Ottichilo’s (1987) findings that elephant

47 poaching was concentrated along the central part of Galana River and in the north and northwestern parts of the park.

Figure 10: Kernel Density and Standard Deviation Ellipse Results for Overall Poaching Pattern

In both quadrat and nearest neighbor analyses, dry season poaching exhibited higher levels of clustering compared to wet season poaching (Tables 4 & 5). The results imply that elephant poaching was more localized in TENP during the dry season than in wet season (Figures 11 & 12). Analysis using the standard deviation ellipse indicated a much narrower poaching pattern in the dry season compared to the wet season. Both dry and wet season elephant poaching ellipses, however, indicate that seasonal elephant

48 poaching exhibited similar spatial orientation and their centers of distribution lay approximately in the central part of the park. Kernel density analysis depicted dry season poaching hotspots on the Tiva and Galana rivers in the northwest and central-western parts of TENP, respectively (Figure 11). Sites with moderate elephant poaching included the southeastern tip of the park on Voi River and the central-eastern part. There were two main wet season poaching hotspots occurring along the lower part of Galana River across the park (Figure 12). Moderate wet season poaching sites were spread along Tiva River to the north of TENP through the central parts of the park, to the southeastern tip on Voi River.

Figure 11: Kernel Density and Standard Deviation Ellipse Results for Dry Season Elephant Poaching

49 The difference in seasonal poaching mortality patterns may be attributed to changes in elephant behavior as influenced by seasonal changes in weather conditions. Although TENP elephants have been found to aggregate all year round, mixed herds have been found aggregate in larger numbers during wet seasons as bulls from bachelor herds join females for mating (Leuthold, 1976). Besides, resources are plenty of during wet season. The larger aggregates of elephant herds during wet season are distributed relative to available food resources (McKnight, 2000). Clusters of wet season elephant poaching will therefore be widely dispersed within TENP compared to dry season poaching.

Figure 12: Kernel Density and Standard Deviation Ellipse Results for Wet Season Elephant Poaching

50 During dry season, resource scarcity (food and water) limit distributions of elephants. Bachelor herds are the poachers’ main target because they have the largest tusks (Poole and Thomsen, 1989). These herds have been found to increase from an average of 22 to 50 individuals in TENP during dry season (Leuthold, 1976). Because of limited food and water resources in this season, elephant home ranges increase considerably (Leuthold and Sale 1973). The expanding home ranges largely overlap in areas with food and water resources. Elephant herds thus congregate near permanent surface water sources in the dry season. Leuthold and Sale (1973) established that Tsavo elephants actually returned to the same grazing areas every year during dry season unlike during wet season when their distribution closely followed the rainfall patterns. The concentration of elephants near permanent water sources thus, may explain why dry season poaching was concentrated near permanent rivers. Results from quadrat and nearest neighbor analyses depict higher levels of clustering of elephant poaching in the 1990 – 1997 period compared to the 1998 – 2005 period (Tables 4 & 5). Elephant poaching was more localized in a small area of the park in the 1990 – 1997 period when CITES’ total ban on ivory trade was in place. In the 1998 – 2005 period, however, elephant poaching expanded to previously unaffected areas of the park (Figures 13 & 14). Kernel density and standard deviation ellipse results both show a dramatic increase in the area of TENP under elephant poaching pressure from the 1990 – 1997 period to the 1998 – 2005 period. Kernel density analyses of elephant poaching in 1990 – 1997 period identified three hotspots along the Tiva River, along Athi River in the central-west, and along the Galana River near Sangayaya and Rhino patrol bases (Figure 13). In the 1998 – 2005 period, elephant poaching spread drastically all over the park with hotspots along all the main rivers (Figure 14).

51

Figure 13: Kernel Density and Standard Deviation Ellipse Results for the 1990 – 1997 Elephant Poaching Period

Before CITES banned ivory trade in 1989, Kenyan elephant populations had plummeted to dangerously low levels mainly due to poaching. The effect of the ivory trade ban coupled with the establishment of KWS with highly motivated, well trained and well-equipped anti-poaching units explains the low poaching that occurred in the 1990 – 1997 period. After CITES partially lifted the ban on ivory trade in 1997, Kenya experienced an upsurge in poaching. This is because poachers could smuggle ivory across international borders so that it was sold together with that harvested legally from southern Africa countries. The CITES policing system could not effectively control ivory laundering (Pearce, 2004).

52

Figure 14: Kernel Density and Standard Deviation Ellipse Results for 1998 – 2005 Elephant Poaching Period

The increased elephant poaching is illustrated clearly by all mortality analyses results in this study. In the 1998 – 2005 period, elephant poaching occurred at a rate of 9 elephants a month compared to 4 in the 1990 -1997 period. The results further validate Kenya’s argument during CITES annual general meeting of 1997 that a partial lifting of the ban on ivory trade would lead to an escalation in elephant poaching (Economist, 2000a & b). In summary, all the descriptive statistical methods to analyze elephant mortality within and in the immediate vicinity of TENP appear to complement one another. Quadrat and nearest neighbor methods provided statistics that showed the nature and significance of elephant mortality patterns. Standard deviation ellipse and kernel density

53 analyses on the other hand provided good visualizations of the mortality patterns that agreed with the corresponding quadrat and nearest neighbor analyses results. Standard deviation ellipses displayed the mean center and spatial orientation of elephant mortality distribution in each category examined and kernel density analyses depicted clearly where mortality hotspots in every category occurred in the study area. However, I found the nearest neighbor and kernel density analysis a good combination for exploring the elephant mortality patterns. This is especially so considering that unlike nearest neighbor analysis, quadrat analysis does not indicate the significance of the observed mortality patterns. Similarly, kernel density analysis is superior to the standard deviation ellipse when analyzing elephant mortality patterns because of its ability to present visually the nature and location of mortality hotspots.

5.2 Land Cover Mapping

Initial unsupervised classification resulted in mixed pixels within riverine forests, woodlands, and forests on hills and higher elevation areas along the Yatta plateau. Landsat ETM+ bands 1, 2 and 6 were found ideal for separating forests from woodlands. Riverine forests and forests on hilly areas were spectrally inseparable and were consequently lumped together to form the forest class. Senescent grassland, open bushland and barrenland were initially mixed. ETM2 and ETM4 successfully separated open bushland from dry grassland and barrenland. Water and cloud shadows were initially mixed and were successfully discriminated using ETM2 and ETM3. Among the land cover classes in TENP (Figure 15), bushland was the most dominant, accounting for 41.0% of the study area (Table 6). Bushland was mainly concentrated in the northeast and central parts of TENP. Other major land cover types included grassland (18.2%), woodland (15.9%) and open bushland (13.8%). Forest and woodland generally were found in the western parts of the park, the occurrence coinciding with an increasing rainfall gradient westward (KWS, 2003). Similarly, grassland and open bushland dominance in the eastern region of TENP is attributed to a decreasing rainfall gradient eastward.

54

Figure 15: Land Cover Map for TENP

55 Table 6: Sizes (hectares) of Land Cover classes mapped for TENP # Land Cover Area (Ha) Percentage Cover (%) 1 Forest 84,011.9 4.4 2 Woodland 306,315.0 15.9 3 Grassland 351,109.0 18.2 4 Bushland 792,563.0 41.0 5 Open Bushland 266,359.0 13.8 6 Herbaceous Cover 48,937.8 2.5 7 Active Agriculture 10,623.0 0.6 8 Water 3,182.0 0.2 9 Barrenland 68,272.9 3.5

5.2.1 Accuracy Assessment

The land cover map produced for TENP had an overall accuracy of 85.4% and a Kappa statistic value of 0.811 (Table 7). Barrenland and forest were easily discriminated from the other land cover types and therefore, exhibited high producer’s accuracy. However, some senescent grassland pixels were confused with barrenland and some woodland pixels misclassified as forest. Some woodland and open bushland pixels were confused with bushland. Active agriculture, herbaceous vegetation and water were excluded from the accuracy assessment because besides being very small land cover types, reference information on these classes was unavailable.

56 Table 7: Error matrix for TENP land cover map created from Landsat ETM+ images

Reference Data

Classified Producer's User's

t Open

Fores Accuracy Accuracy Bushland Bushland Woodland Grassland Classified Data Barrenland Totals (%) (%)

1. Forest 24 3 0 2 0 0 29 92.31 82.76

2. Woodland 2 35 0 4 0 0 41 79.55 85.37

3. Grassland 0 0 39 6 2 0 47 88.64 82.98

4. Bushland 0 5 2 93 5 0 105 85.19 87.62

5. Open Bushland 0 1 0 3 31 0 35 79.49 88.57

6. Barrenland 0 0 3 1 1 18 23 100 78.26

Reference totals 26 44 44 109 39 18 280

Overall Classification Accuracy = 85.4%

Overall Kappa Statistic = 0.811

5.3 Relationships between elephant mortality patterns and biophysical and human variables

Poaching-caused elephant mortality occurred at the highest concentrations in bushland, and moderately in grassland, open bushland and woodland. No elephant poaching mortality was recorded in active agriculture or in water (Table 8). However, there was a bias introduced due to the fact that the land cover types were not of equal sizes. This meant that the larger the land cover type, the more likely that poaching would occur in it, and the small the land cover, the lower the likelihood that elephant poaching would occur. To eliminate this bias, density of poaching-caused mortality was computed and explored in relation to each land cover type. High mortality densities were observed within open bushland, grassland, bushland and herbaceous vegetation for overall elephant poaching. Grassland, bushland and open bushland experienced high poaching densities during wet season, while during dry season, elephant poaching density was high in woodland, open bushland, bushland and

57 herbaceous vegetation. The observed high densities of overall elephant poaching in grassland, herbaceous vegetation and open bushland reflect elephant distribution in TENP. Leuthold (1976) indicated that TENP elephants prefer open rather than densely vegetated areas. In addition, during wet season, grasslands and open bushlands green-up thus providing elephants with ample food. During dry season, however, grasslands become depleted forcing elephants to shift to woodlands. The high elephant poaching mortality density in the dry season in herbaceous vegetation may be attributed to the occurrence of this land cover type on frequently flooded areas characterized by black cotton soil. The soil retains moisture for a long period after the end of the rainy season, hence vegetation growing on it remains green and palatable longer thus attracting elephants.

Table 8: Relationship between elephant mortality and land cover types

Elephant mortality recorded

Land cover Overall poaching Wet season poaching Dry season poaching Name Area(Ha) Frequency Density Frequency Density Frequency Density Forest 84011.9 3 0.00003571 1 0.00001190 0 0.00000000 Woodland 306315 9 0.00002938 2 0.00000653 9 0.00002938 Grassland 351109 18 0.00005127 14 0.00003987 3 0.00000854 Bushland 792563 27 0.00003407 17 0.00002145 14 0.00001766 Open bushland 266359 14 0.00005256 5 0.00001877 8 0.00003003 Herbaceous vegetation 48937.8 3 0.00006130 0 0.00000000 1 0.00002043 Agriculture 10623 0 0.00000000 0 0.00000000 0 0.00000000 Water 3181.99 0 0.00000000 0 0.00000000 0 0.00000000 Barrenland 68272.9 1 0.00001465 1 0.00001465 0 0.00000000

∑ 1931373.6 75 0.00003883 40 0.00002071 35 0.00001812

Elephant poaching during wet season was positively correlated with proximity to ranger patrol bases (P< 0.01), seasonal rivers (P<0.05) and park roads (P<0.01), but negatively correlated with proximity to waterholes (P<0.05) and elevation (P<0.01) (Table 9). The results suggest that elephant distribution is not constrained by resources during the wet season as there is plenty of food and water around. This may explain why main rivers, which are sources of surface water, did not show significant correlation with

58 elephant poaching patterns observed during wet season. It is during wet season that TENP elephants aggregate in large numbers and the herds move further apart in response to their expanding home ranges (Mcknight, 2000). However, elephants need to drink and mud-bathe daily (Estes, 1999), hence the significant negative correlation between wet season poaching and proximity to waterholes. On the other hand, poachers target areas with plenty of elephants, but that are remote and infrequently patrolled by park rangers (Pilgram and Western, 1986; Leader-Williams et al, 1990). Areas close to ranger patrol bases and roads are thus avoided because they are areas frequently patrolled by rangers and any poaching activities can easily be detected. Elevation exhibits significant negative correlations with wet season poaching suggesting that poachers target elephants at low elevations during wet season; a phenomenon reflecting preference of low elevations by TENP elephants (Smith and Kasiki, 2000).

Table 9: Spearman’s Rank correlates for elephant mortality in TENP

Dist. to Dist. to Dist. to Dist. to Dist. to Dist. to Dist. to Elevation Slope

Patrol Park Main Seasonal Waterholes Park Park (m) (Degrees)

Bases Gates Rivers Rivers (m) Roads Boundary

(m) (r, p) (m) (r, p) (m) (r, p) (m) (r, p) (r, p) (m) (r, p) (r, p) (r, p) (r, p)

Annual 0.247* 0.065 -0.238* 0.073 -0.216 0.408** 0.229* -0.258* 0.015 poaching 0.033 0.580 0.040 0.534 0.062 0.000 0.048 0.014 0.888

Wet season 0.422** 0.022 0.139 0.338* -0.317* 0.466** 0.200 -0.424** 0.039 poaching 0.007 0.895 0.392 0.033 0.047 0.002 0.216 0.006 0.810

Dry season -0.148 0.444** -0.686** -0.409* -0.022 0.147 0.378* -0.169 -0.046 poaching 0.398 0.007 0.000 0.015 0.892 0.400 0.025 0.333 0.792

Levels of significance: *p<0.05, **p<0.01

Dry season poaching was positively correlated with distance to park gates (P<0.01) and park boundary (P<0.05), but negatively correlated with main rivers (P<0.01) and seasonal rivers (P<0.05). Proximity to patrol bases, slope and elevation, however, showed no significant correlation with dry season poaching. The results indicate that elephants are distributed close to sources of permanent surface water during dry season. TENP elephant home ranges shrink considerably during dry season as food

59 and water resources become scarce (Leuthold and Sale, 1973). The elephants then retreat to areas along Tiva, Galana and Voi rivers (Kasiki, 1998) because these areas have the resources necessary for the survival of the elephants during this time of the year. Unfortunately, the same areas provide good elephant harvesting areas for poachers hence the significant negative correlations between dry season poaching and proximity to both main and seasonal rivers. Influence of surface water on elephant distribution has also been observed in Maputo elephant reserve in Mozambique (De Boer et al, 2000) and in northern Botswana (Verlindern and Gavor, 1998). However, poachers target elephants in remote areas where they are unlikely to be detected by park rangers on patrol, hence the reason poachers keep away from areas close to park gates and park boundary. Ehrlich (1973) and Milner-Gulland and Leader-Williams (1992) argue that the fear of being detected by law enforcement authorities is a more effective deterrent to a commission of a crime than the actual punishment a criminal would receive if caught. The fear of being detected by park rangers therefore, explains why no poaching occurred near Voi gate/patrol base although Kasiki (1998) observed that elephants tended to move westwards to aggregate around this area during dry season. Annual elephant poaching irrespective of climatic season was positively correlated with proximity to ranger patrol bases (P<0.05), park roads (P<0.01) and park boundary (P<0.05), but negatively correlated with proximity to main rivers (P<0.05) and elevation (P<0.05). Poaching is therefore, likely to occur close to sources of permanent surface water irrespective of changes in weather conditions. Similar observations were made by Ottichilo (1987) in TENP and Demeke and Bekele (2000) in Mango National Park in Ethiopia. Moreover, poachers maximize their hunting success by targeting areas where elephants are concentrated (close to main rivers and at low elevations), while minimizing risk of being detected by keeping in areas farthest from patrol bases, park roads and park boundary. Of all the biophysical and human factors examined in relation to poaching- induced elephant mortality, proximity to main rivers exhibited the highest correlation (- 0.686) with dry season poaching. This factor, however, explained only about 40.7% of the observed variability in dry season poaching (Figure 16). This finding indicates that there were other important factors influencing poaching-induced elephant mortality that

60 were not measured in the current study. Some of these factors may include rainfall distribution, locations of lodges and campsites and park ranger observation posts, and political, and socio-economic factors that may influence elephant poaching.

Figure 16: Relationship between dry season poaching and distance to main rivers

5.4 Risk to Poaching Maps

Three risk to poaching maps were generated for TENP, each for wet season, dry season, and for annual elephant poaching irrespective of season. The risk to poaching maps were generated using biophysical and human factors that had been found to be significantly correlated with elephant poaching mortality. Input data into the wet season risk to elephant poaching map included distances to patrol bases, park roads, waterholes, seasonal rivers, the elevation layer, and the land cover map. The risk to elephant poaching map for wet season indicated areas at highest risk as occurring mainly in the central-east, south of Sala gate toward the southeastern edge of park, and in the

61 northeastern parts of TENP (Figure 17). Ottichilo (1987) observed similar poaching concentrations in the central-eastern border of TENP near the Galana ranch. Other areas at higher risk to elephant poaching during wet season were in the central parts of Voi River, and in the immediate vicinity of the areas with the highest risk to poaching. The areas of TENP with high risk to poaching were concentrated mainly within bushland, open bushland and grassland reflecting elephant preference for more open land covers in TENP during wet season. In addition, areas of high risk to poaching were not concentrated along the main rivers because distribution of elephant during the wet season is not limited by availability of surface water. Kasiki (1998) found that elephants were widely dispersed in areas without permanent surface water during the wet season.

62

Figure 17: Risk to elephant poaching surface for Wet Season

Risk to elephant poaching in the dry season was generated using the distance surfaces to park gates, main rivers, seasonal rivers, park boundary, and the land cover

63 map. The area of highest risk to elephant poaching occurred along the Tiva River (Figure 18). Other areas of high risk to elephant poaching in the dry season were along the central parts of the Galana and Voi rivers. In contrast to the elephant risk to poaching map for wet season, that generated for the dry season indicated areas of highest risk as concentrated along the main rivers, where water is available throughout the year. According to Kasiki (1998), elephants occur in large numbers along Galana, Voi and Tiva rivers during the dry season. However, the western portion of the Voi River had a low poaching risk because of its proximity to the Voi patrol base and park gate. Ottichilo (1987) found similar patterns with elephant poaching concentrated along the central part of Galana River and on the northern and northwestern parts of TENP.

64

Figure 18: Risk to elephant poaching surface for Dry Season

65 An overall risk to poaching map that did not factor the seasonality of poaching was created using distance surfaces to patrol bases, main rivers, park roads and park boundary, elevation, and the land cover map. Areas with the highest risk to elephant poaching occurred in the central part of the Tiva River and the central-eastern parts of the Galana River (Figure 19). This risk to elephant poaching map had similarities to the dry season risk to elephant poaching map by showing the highest risk to poaching near sources of surface water. It also indicated that TENP elephants were at a high risk to poaching in the more open central and eastern parts than the dense west dominated by forests and woodlands. Areas that could be considered relatively safe for elephants were found near park gates and patrol bases, except for the Sala Gate. The location of Sala Gate close to Galana River coupled with the fact that the gate is located at the border closest to Tana River district where poachers of Somali origin often access the park may explain why elephants in this region are not safe. In a nutshell, elephants are limited in their distribution by food and water. Seasonal variations in availability of food and water resources explain observed distribution of elephants in TENP. Poachers on the other hand do not just kill elephants in any area where elephants occur in large numbers, but instead target those elephants in remote locations where poaching activities are likely to remain undetected.

66

Figure 19: Annual Risk to Elephant Poaching

A major limitation encountered when developing risk to elephant poaching models was the inability to quantifying the degree of influence each layer (biophysical or

67 human) input in the models had on the risk to poaching. As a result, no weights were applied to the different input surfaces in the risk model. Weighting the biophysical and human factor datasets based on expert information would most likely improve the quality of the risk to poaching maps because not all the biophysical and human factors used had equal influence on observed elephant mortality patterns.

68 CHAPTER SIX

6.0 Conclusions

This study set out to establish whether there were areas of TENP in which elephants were at higher risk of being poached based on available elephant mortality data (1990 – 2005). This research goal was achieved by first examining the spatial and temporal patterns of poaching-induced elephant mortality, and then establishing what biophysical and/or human factors were correlated to the observed patterns of elephant mortality. The biophysical and human factors found significantly correlated with poaching-caused elephant mortality patterns were combined in GIS models to generate corresponding risk to poaching maps. Results obtained from quadrat and nearest neighbor analyses indicated that elephant poaching was not a random event in TENP, and instead exhibited clustered patterns irrespective of season for which poaching-induced elephant mortality was examined or in the pre- and post-CITES ban on ivory trade.. In addition, nearest neighbor analysis indicated that the clustering of poaching-induced elephant mortality did not occur by chance. Standard deviation ellipse and kernel density results agreed with corresponding quadrat and nearest neighbor results, and in addition displayed graphically the spatial orientation and center of distribution, and the location of poaching hotspots, respectively. However, nearest neighbor analysis and kernel density analysis were the best combination for analyzing elephant mortality patterns because of the former’s ability to statistically test the significance of the elephant mortality patterns and the latter’s ability to visualize hotspots. Overall poaching was more clustered than non-poaching mortality because poachers often maximize harvest by killing more than one large individual elephant in a herd. Dry season poaching was more clustered compared to wet season poaching, this reflecting the limited food and water resources available for elephants during the dry season. There was a dramatic increase in the degree of poaching from the 1990 – 1997 to the 1998 – 2005 period. This increase is a reflection of the effects of CITES’ partial lift on the ban on ivory trade in 1997.

69 Different biophysical and human factors were correlated with observed patterns of poaching-induced elephant. Land cover type, availability of surface water and elevation were the most important biophysical factors limiting elephant distribution, while proximity to park roads, gates, park boundary and patrol bases were significant deterrents to poaching. The biophysical and human factors influencing poaching-caused elephant mortality, however, varied depending on season of elephant mortality. Slope was not significantly correlated with poaching-induced elephant mortality. Risk to poaching maps for both the dry season and overall poaching depicted highest risk to poaching in areas close to sources of permanent water. Risk to poaching map for the wet season indicated highest risk to poaching in the more open land cover types located furthest from main rivers. However, none of the biophysical and human factors individually explained more than 40% of the observed variation in poaching induced elephant mortality. Future research should seek to uncover other factors responsible for the observed poaching-induced elephant mortality patterns. In addition, there is need to quantify the influence of each biophysical or human factor in the elephant poaching risk map so that they can be weighted accordingly. The present study was unable to meet that condition. Weighting of elephant poaching risk factors can be achieved by seeking expert opinions from experienced field wildlife officers. Nevertheless, the information generated in this study can be useful in guiding KWS in elephant conservation and management decisions for TENP. Deployment of policing resources (equipment and personnel), establishment of more strategically located patrol bases, opening up of more park roads and digging waterholes are some of the decisions that can be informed by the results of this study.

70 References

Ackerman, W. V. and Murray, A. T. (2004). Assessing spatial patterns of crime In Lima, Ohio. Cities, 21(5): 423-437.

Adams, W. M. (2004). Against Extinction: The Story of Conservation (Chapter 5 & 7). Earthscan, London.

AFESG, (1997). Review of African Elephant Conservation Priorities (Thouless, C. R, eds). Working document of the IUCN/SSC African Specialist Group, Nairobi, Kenya.

Akama, J. S. (1998). The evolution of wildlife conservation policies in Kenya. Journal of Third world Studies, Vol XV(2): 103 – 117.

Anderson, D. and Grove, R. (eds). (1987). Conservation in Africa: People, policies and practice. Cambridge University press, New York.

Armbruster, P. and Lande, R. (1993). A Population Viability Analysis for African Elephant (Loxodonta Africana): How big should a reserve be? Conservation Biology 7(3): 602-610.

Attwell, C. M. (2000). Postmodernism and African conservation science. Biodiversity and Conservation, 9(5):559-577.

Babaasa, D. (2000). Habitat Selection by Elephants in Bwindi Impenetrable National park, South Western Uganda. African Journal of Ecology, 38: 116-122.

Barff, R. A. and Hewitt, D. E. (1989). Second order analysis of bivariate point patterns. Professional Geographer, 41(2): 183 – 189.

71 Barnes, R.F.W., Blom, A., Alers, M. P. T. and Barnes, K. L. (1995). An estimate of the numbers of forest elephants in Gabon. Journal of Tropical Ecology 11: 27-37.

Baskin, Y. (1994). There is a new wildlife policy in Kenya: use it or lose it. Science, Vol. 265(5173): 733 – 734.

Ben-Shahar, R. (1999). Elephants and their Woodland Habitats in Northern Botswana. Pachyderm, 27: 101-104.

Bolstad, P. (2002). GIS Fundamentals. A First Text on Geographic Information Systems. Eider Press, White Bear Lake.

Bonnicksen, T. M. and Stone, E. C. (1981). The Giant Squoia-mixed conifer forest community characterized through pattern analysis as a mosaic of aggregations. Forest Ecology and Management, 3: 307-328.

Born Free Foundation (2000). “Stop the Clock” Report 4th April 2000. Executive Summary. Born Free Foundation. West Sussex.

Born Free Foundation (2002). 10 elephants butchered by ivory poachers in Tsavo East, Kenya. http://www.bornfree.org.uk/elefriends/news008.htm

Bowers, K. and Hirschfield, A. (1999). Exploiting links between crime and disadvantage in northwest England: an analysis using geographical information systems. Geographical Information Science, 13(2): 159 – 184.

Burden, F. F. (2003). Point pattern analysis. GIS Resource Document 03-41 (GIS_RD_03-41. http://64.233.167.104/

Burton, M (1999). An Assessment of Alternative Methods of Estimating the Effects of Ivory Trade Ban on Poaching Effort. Ecological Economics 30:93-106.

72

Butler, V. (1998). Controversy in Kenya. International Wildlife, 00209112, Nov/Dec98, Vol. 28, Issue 6.

Cherrill, A. J., Lane, A. and Fuller, R. M. (1994). The use of Landsat 5 Thematic Mapper imagery in the characterization of landscape composition: a case study in northern England. Journal of Environmental Management, 40: 357 – 377.

CITES (2004). Thirteenth meeting of the Conference of the Parties Bangkok (Thailand), 2-14 October 2004. Interpretation and implementation of the Convention Species trade and conservation issues. Elephants. http://www.cites.org/eng/cop/13.

Cobb, S. (1976). The abundance and distribution of large herbivores in Tsavo national park, Kenya. Unpublished PhD dissertation, Oxford university.

Collett, D. (1987). Pastoralists and wildlife: Image and reality in Kenya Maasailand. In Conservation in Africa: People, policies and practice. Anderson, D. and Grove, R. (eds). Cambridge University Press, New York.

Congalton, R. G. and Green, K. (1999). Assessing the accuracy of remotely sensed data: Principles and practices. Lewis Publishers, Boca Raton.

Corfield, T. F. (1973). Elephant mortality in Tsavo National Park. East African Wildlife Journal, 11: 339 – 368.

Cumming, D. H. M, du Toit, R. F. and Stuart, S. N. (eds) (1990). African Elephants and Rhinos: Status Survey and Conservation Action Plan. IUCN. Gland.

Dalleo, P. T. (1979). The Somali role on organized poaching in northeastern Kenya, 1909 – 1939. The International Journal of African Historical Studies, 12(3): 472 – 482.

73 De Boer, W. F., Ntumi, C. P., Correia, A. U. and Mafuca, J. M. (2000). Diet and Distribution in the Maputo Elephant Reserve, Mozambique. African Journal of Ecology, 38: 188-201.

Delibes, M, Goano, P. and Ferreras, P. (2001). Effects of an Attractive Sink Leading to Maladaptive habitat Selection. American Naturalist 158: 277-285.

Demeke, Y. and Bekele, A. (2000). Study on the Elephants of Mago National Park, Ethiopia. Pachyderm, 28: 32-43.

Diamond, J. (1984). “Normal “extinctions of isolated populations. In M. H. Nitecki (ed). Extinctions. University of Chicago Press, Chicago.

Diggle, P. J. (1983). Statistical analysis of spatial point patterns. Academic Press, New York.

Douglas-Hamilton, I. (1998). Tracking African elephants with a global positioning system (GPS) radio collar. Pachyderm, 25: 81-92.

Earth Watch Journal, (1998). Community conservation collapse. Earth Watch Journal, Vol. 13(4): 15 – 15.

Economist (1999). A hard-edged attitude to wildlife conservation. Economist, Vol. 351, Issue 8123.

Economist (2000a). An elephant problem. Economist, Vol. 355(8166): 17 – 17.

Economist (2000b). Excitement at CITES. Economist, Vol. 355(8166): 87- 88.

Economist (2002). Closing in for the Kill? After a decade of quiet, ivory poaching has started again in Kenya. Economist, 363(8273): 76-77.

74

Ehrlich, I. (1973). Participation in Illegitimate Activities: A Theoretical and Empirical Investigation. Journal of Political Economy, 81 (3): 521-566.

Eklundh, L. (1998). Estimating relations between AVHRR, NDVI and rainfall in East Africa at 10-day monthly time scales. International Journal of Remote Sensing, 19: 563- 570.

Estes, R. (1999). The safari companion: Guide to watching African mammals including hoofed mammals, carnivores, and primates. Chelsea Green Publishing Company, White River Junction.

FAO (2003). Multipurpose Africover databases on environmental resources (MADE) available for 10 countries. http://www.africover.org

Fitter, R. (1986). Wildlife for man. Collins Harper Publishers, London.

Franklin, S. E., Peddle, D. R., Dechka, J. A. and Stenhouse, G. B. (2002). Evidential reasoning with Landsat TM, DEM and GIS data for landcover classification in support of grizzly bear habitat mapping. International Journal of Remote sensing, 32(21): 4633- 4652.

Futurist (1987). Africa’s threatened wildlife. Futurist, Vol. 21(2): 45 – 45.

Gatrell, C. A., Bailley, T. C. Diggle, P. J. and Rowlingson, S. B. (1996). Spatial point pattern analysis and its application in geographic epidemiology. Trans. Inst. Geogr., 21: 256-274.

Gesler, W. (1986). The uses of spatial analysis in medical geography: a review. Soc. Sci. Med., 23(10). 963-973.

75 Gilg, A. W. (1973). A study in agricultural disease diffusion: the case of the 1970-71 fowl-pest disease. Trans. Inst. Br. Geogr. 59: 77-79.

Girt, J. L. (1972). Simple chronic bronchitis and urban ecological structure. In Medical Geography: Techniques and Field Studies (McGlashan, N. D. ed). Methuen, London.

Government of Kenya (2005). http://www.kenya.go.ke/scategory.php?c_id=3&category=Tourism

Haack, B. and English, R. (1996). National land cover mapping by remote sensing. World development 24(25): 845-855.

Hakansson, N. T. (2004). The human ecology of world systems in East Africa: The impact of ivory trade. Human Ecology, 32(5): 569-591.

Hammer, J. (1993). The war to Save the Tsavo Reserve. Newsweek Vol 121(15): 3-4.

Han, K., Champeaux, J. and Roujean, J. (2004). A land cover classification product over France at 1km resolution using SPOT4/VEGETATION data. Remote Sensing of Environment, 92: 52-66.

Harper, C. L. (1960). Environment and Society: Human Perspectives on Environmental issues. Prentice-Hall. New Jersey.

Hertberg, R. (2001). Impacts of the Ivory Trade Ban on Poaching Incentives: A Numerical example. Ecological Economics 36: 189-195.

Heupel, M. R. and Simpfendorfer, C. A. (2005). Quantitative analysis of aggregation behavior in juvenile blacktip sharks. Marine Biology, 147(5): 1239 – 1249.

76 Hoare, R. E. (2000). African Elephants and Humans in Conflict: The outlook for coexistence. Oryx, 34: 34-38.

Hulme, D. and Murphree, M. (eds). (2001). African Wildlife and livelihoods: The promise and performance of community conservation. James Currey Ltd, Oxford.

IUCN (2000). Red list of Endangered Species. IUCN, Gland, Switzerland.

Jachmann, H and Billiouw, M. (1997). Elephant Poaching and Law Enforcement in the Central Luangwa Valley, Zambia. Journal of Applied Ecology 34(1): 233-244.

Jansen, J. R. (2005). Introductory Digital Image Processing: A remote Sensing Perspective 3nd ed. Prentice Hall, Upper saddle River.

Johnson, G. D., Eidson, M., Schmit, K., Ellis, A. and Kulldorff, M. (2006). Geographic Prediction of Human Onset of West Nile Virus Using Dead Crow Clusters: An Evaluation of Year 2002 Data in New York State. American Journal of Epidemiology, 163(2): 171 – 180.

Jyoti, K. and Harford, J. (1996). The ivory trade ban: Is it effective? Ecological Economics, 19(2): 147-155.

Kasiki, S. M. (1998). Human-elephant conflict in areas adjacent to the Tsavo National Parks, Kenya. Phd dissertation, University of Kent.

Kassam, A. and Bashuna, A. B. (2004). Marginalization of Waata Oromo hunter- gatherers of Kenya: insider and outsider perspectives. Africa, 74(2): 194 – 216.

Kenya government (1978). Policy on wildlife management. Sessional paper number 3. Government Printers, Nairobi.

77 Kenya Wildlife Service (1994). Wildlife-human conflicts in Kenya. Report of the five- person review group. Kenya Wildlife Service, Nairobi.

Kenya Wildlife Service (1996). National Parks of Kenya: 50 years of challenge and achievement. Nairobi, Kenya.

Kenya Wildlife Service (2003). Tsavo East National Park. http://www.kws.org/tsavoeast.htm

Kenya Wildlife Service (2005). http://www.kws.org

Khaemba, W. M. (2000). Spatial point pattern analysis of aerial Survey data to assess clustering in wildlife distributions. JAG, Vol.3(2): 139 – 145.

Khaemba, W. M. and Stein, A. (2000). Use of GIS for a Spatial and Temporal analysis of Kenyan Wildlife with Generalised Linear Modeling. International Journal of Geographical Information Science, 14 (8): 833-853.

King, R. B. (2002). Land cover mapping principles: a return to the interpretation fundamentals. International journal of Remote Sensing, 23(18): 3525 – 3545.

Kioko, J. M. (2002). KWS Press Release - Family of ten Elephants Gunned Down in Tsavo East, one Poacher killed, three still on the run. Kenya Wildlife Press Release. Nairobi.

Kliskey, A. D. (1995). The role and functionality of GIS as a planning tool in natural resource management. Comput., Environ. and Urban Systems, 19(1): 15 -22.

Kuriyan, R. (2002). Linking local perceptions of elephants and conservation: Samburu pastoralists in northern Kenya. Society and Natural Resources, 15: 949 – 957.

78 Lancaster, J. and Downes, B. J. (2004). Spatial point pattern analysis of available and exploited resources. Ecography, 27: 94-102.

Laws, R. M. (1969). The Tsavo research project. Journal of Reproductive Fertility, Supplement, 6: 495 – 531.

Leader-Williams, N, Albon, S. D. and Berry, P.S.M. (1990). Illegal Exploitation of Black Rhinoceros and Elephant Populations: Patterns of Decline, Law Enforcement and Patrol Efforts in Luangwa Valley, Zambia. Journal of Applied Ecology 27(3): 1055-1087.

Lee. J. and Wong, D. W. S. (2001). Statistical Analysis with ArcViwe GIS. John Wiley & Sons, New York.

Leica Geosystems (2003). ERDAS Field GuideTM 7th ed. GIS and Mapping, LLC. Atlanta, Georgia.

Leuthold, W. (1976). Group size in elephants of Tsavo National Park and possible factors influencing it. Journal of Animal Ecology, 45(2): 425 – 439.

Leuthold, W. (1977). Spatial organization and strategy of habitat utilization of elephants in Tsavo National Park, Kenya. Zeitschrift fur Saugetierkunde, 42: 358 – 379.

Leuthold, W. (1996). Recovery of woody vegetation in Tsavo National Park, Kenya, 1970 – 1994. African Journal of Ecology, 34: 101 – 112.

Leuthold, W. and Sale, J. B. (1973). Movements and patterns of habitat utilization of elephants in Tsavo National Park, Kenya. East African Wildlife Journal, 11: 369 – 384.

Lewis, D. M. (1986). Disturbance effects on elephant feeding: evidence for compression in Luangwa Valley, Zambia. African Journal of Ecology, 24: 227 – 241.

79 Lillesand, T. M. and Kiefer, R. W. (2000). Remote Sensing and Image Interpretation. 4th ed. John Wiley & Sons, New York.

Lillesand, T. M., Kiefer, R. W. and Chipman, J. W. (2004). Remote Sensing and Image Interpretation. 5th ed. John Wiley & Sons, New York.

Liu, X. (2005). Supervised Classification and Unsupervised Classification. http://cfa- www.harvard.edu/~xliu/presentations/SRS1_project_report.PDF

Makonjio, M. O. and Warui, J. K. (2004). Threats to biodiversity and their implications in protected areas of Kenya. Journal of Sustainable Tourism, 12 (1): 55 – 69.

Malakoff, D. (1998). Leakey back as head of wildlife service. Science, Vol. 282 Issue 5386, p19.

Martinez-Casasnovas, J. A. (2000). A cartographic and database approach for land cover/use mapping and generalization from remotely sensed data. International Journal of Remote sensing, 21(9): 1825-1842.

McCarthy, J., Gumbricht, T. and McCarthy, T. S. (2005). Ecoregion classification in the Okavango Delta, Botswana using multitemporal remote sensing. International Journal of Remote sensing, 26(19): 4339 – 4357.

McKnight, B. (2000). Changes in Elephant Demography, Reproduction and Group Structure in Tsavo East National Park (1966 - 1994). Pachyderm ,29: 15-24.

McRae, M. (1998). Survival test for Kenya’s wildlife. Science, Vol. 280, Issue 5363.

Mertens, B. and Lambin, E. (2000). Land cover change trajectories in southern Cameroon. Annals of association of American geographers, 90: 467 – 494.

80 Miller, G. T. (1989). Resource conservation and management. Wordsworth Publishing Company, Belmont.

Millington, A. C. and Alexander, R. (2000). Vegetation mapping in the last three decades of the twentieth century. In R. Alexander, and Millington A. C. (eds). Vegetation Mapping pp. 321-331. John Wiley and Sons Ltd, Chichester.

Milner-Gulland, E. J. and Leader-Williams, N. (1992). A Model of Incentives to Illegal Exploitation of Black Rhinos and Elephants: Poaching Pays in Luangwa Valley, Zambia. Journal of Applied Ecology, 29(2): 388-401.

Moleele, N., Ringrose, S., Arnberg, S., Lunden, B. and Vanderpost C. (2001). Assessment of vegetation indexes useful for browse (forage) prediction in semi-arid rangelands. International Journal of Remote Sensing, 22(5): 741-756.

Moreau, S., Bosseno, R., Gu, X. F. and Baret, F. (2003). Assessing the biomass dynamics of Andean bofedal and totora high-protein wetland grasses from NOAA/AVHRR. Remote Sensing of Environment, 85: 516-529.

Morell, V. (1994). Richard Leakey quits post, charges “smear campaign”. Science, 4/1/94, Vol. 264 Issue 5155, p23.

Mpanduji, D.G., Hofer, H., Hilderbrandt, T.B., Goeritz, F. and East, M.L. (2002). Movement of Elephants in the Selous-Niassa Wildlife Corridor, Southern Tanzania. Pachyderm, 33: 18-31.

Mubalama, L. (2000). Population and Distribution of Elephants (Loxodonta Africana africana) in the Central Sector of Virunga Natioanal park, Eastern DRC. Pachyderm, 28: 44-55.

81 Muller, S. V., Racoviteanu, A. E. and Walker, D. A. Landsat MSS-derived land-cover map of northern Alaska: Extrapolation methods and a comparison with photo-interpreted and AVHRR-derived maps. International Journal of Remote Sensing, 20(15& 16): 2921 – 2946.

Musiega, D. E. and Kazadi, S. (2004). Simulating the East African wildebeest migration patterns using GIS and remote sensing. African Journal of Ecology, 42: 355 – 362.

Ng’ang’a, N. (2004). Showdown looms at KWS meeting. East African Standard, Friday, August 20, 2004.

Nielsen, S. E, Herrero, S, Boyce, M. S, Mace, B. B, Gibeau, M. L. and Jevons, S. (2004). Modeling the Spatial Distribution of Human-Caused Grizzly Bear Mortalities in Central Rockies of Canada. Biological Conservation 120: 101-113.

Odhiambo, A. and Moraa, J. (2006). Toll stations driving tourists away from the Maasai Mara. Daily Nation, Wednesday, August 16, 2006. http://nationmedia.com

Omullo, L. B, Said, M. T and Warugte, P. W. (1998). Habitat Evaluation for Wildlife Conservation and Management Techinical Report 152. department of Resource Survey and Remote Sensing, Nairobi, Kenya.

Ottichilo, W. K. (1987). The cause of the recent heavy elephant mortality in the Tsavo ecosystem, Kenya, 1975 – 80. Biological Conservation 41: 279 – 289.

Parker, I. S. C. and Amin, M. (1983). Ivory Crisis. Chatto & Windus, London.

Pearce, F. (2004). Ivory sale fuels illegal trade fears. New scientist, 02624079, vol. 181 (2439).

82 Pilgram, T. and Western, D. (1986). Inferring Hunting Patterns on African Elephant from Tusks in the International Ivory Trade. Journal of Applied Ecology 23(2): 503-514.

Plante, M., Lowell, K., Potvin, F., Boots, B. and Fortin, M. (2004). Studying deer habitat on Anticosti Island, Québec: Relating animal occurrences and forest map information. Ecological Modelling, 174(4): 387 – 399.

Poole, J. H. and Thomsel, J. B. (1989). Elephants are not beetles: implications of the ivory trade for the survival of the African elephant. Oryx, 23: 188 – 198.

Poole, J. H., Aggarwal, N., Sinage, R. Nganga, S., Broten, M. and Douglas-Hamilton, I. (1992). The status of kenya’s elephants. Kenya Wildlife Service and Department of Resource Surveys and Remote Sensing, European Community.

Ramp, D.; Caldwell, J., Edwards, K. A., Warton, D. and Croft, D. B. (2005). Modeling of wildlife fatality hotspots along the Snowy Mountain Highway in New South Wales, Australia. Biological conservation 126(4): 474 – 490.

Ripley, B. D. (1981). Spatial Statistics. Wiley, London.

Robinnson, S. (2000). Dying for Ivory: Even African conservationists cannot agree on how best to protect elephant populations from the threat of poachers. Time Europe, 155(15): 44-45.

Ruggiero, R. (1990). The effects of poaching disturbance on elephant behavior. Pachyderm, 13: 42 – 44.

Shannon, G. W. et al. (1978). The search for medical care: an explosion of urban black behavior. International journal of health services, 8: 519-530.

83 Sharp, R. (1997). The African elephant: Conservation and CITES. Oryx, 31(2): 111 – 119.

Sheldrick, D. W. (1973). The Tsavo Story. Collins and Harvill, London.

Sitati, N. W, Walpole, M. J, Smith, J. and Leader-Williams (2003). Predicting Spatial Aspects of Human-Elephant Conflict. Journal of Applied Ecology 40:667-677.

Smith, R. J. and Kasiki, S. (2000). A Spatial Analysis of Human-Elephant Conflict in the Tsavo Ecosystem, Kenya. AfESG Report. IUCN/SSC, Gland.

Society (2005). Thrills of counting elephants. The Standard. Saturday February 12, 2005. http://www.eastandard.net

Soule, M. (1985) What is Conservation Biology? Bioscience, 35: 727- 734.

Steinhart, E. (1994). National parks and anti-poaching in Kenya, 1947 – 1957. The International Journal of African Historical Studies, Vol. 27(1): 59 – 76.

Stone, M. L. (1972). Organized poaching in Kitui district: A failure in district authority, 1900 to 1960. The International Journal of African Historical Studies, Vol. 3: 436 – 452.

Tappan, G. G., Tyler, D. J., Wehde, M. E. and More, D. G. (1992). Monitoring rangeland dynamics in Senegal with Advanced Very High Frequency Resolution Radiometer data. Geocarto International, 7: 87-98.

Tolvanen, R. (2004). Nature Conservation Areas in Kenya- Tsavo East and West National Parks. In Pellikka, P. J., Ylhaisi and Clark, B. (eds). Tait Hills and Kenya- Seminar, Reports and Journal of a Field Excursion to Kenya. Expedition Reports of the Department of Geography, University of Helsinki 40: 59-63.

84 Trisalyn, N. and Barry, B. (2005). Identifying insect infestation hot spots: an approach using conditional spatial randomization. Journal of Geographical Systems, 7(3-4): 291 – 311.

Turner, M. (1999). No space for participation: Pastoralist narratives and the etiology of park-herder conflict in southern Niger. Land Degradation and Development, 10(4):345- 363.

Turner, M. G, Garbner, R. H, and O’Neill, V. R. (2001). Landscape Ecology in Theory and Practice. Patterns and Processes. Springer-Veralg. New York.

Verlinden, A. and Gavor, I. K.N. (1998). Satellite Tracking of Elephants in Northern Botswana. African Journal of Ecology, 36: 105-116.

Ville, J. V. (1995). Man and Elephants in the Tsavo area of Kenya: an anthropological perspective. Pachyderm, 20: 69 – 72.

Wall, G., Dudycha, D. and Hutchinson, J. (1985). Point pattern analysis of accommodation in Toronto. Annals of tourism Research, 12: 603-618.

Were, E. M. (2005). The domain of authority and sphere of influence of wildlife conservation and management policy in Kenya. Journal of Third World Studies, Vol. XXII(2): 227 – 248.

Western, D. (1987). Africa’s Elephants and Rhinos: Flagships in crisis. Trees, Vol. 2(11): 343 – 346.

Western, D. (1989). Conservation without Parks: wildlife in rural landscape. In D. Western and M. C. Pearl. (eds). Conservation for the twenty-first Century (pp 158-165). Oxford University press, New York.

85 Western, D. and Maitumo, D. (2004). Woodland loss and restoration in a savannah park: a 20-year experiment. African Journal of Ecology. Vol. 42(2): 111 – 121.

Wiegand, T. and Moloney, K. A. (2004). Rings, circles, and null-models for point pattern analysis in ecology. Oikos, 104(2): 209-229.

Wijgaarden, W. (1985). Elephants – trees- grazers: Relationship between climate, soil, vegetation and large herbivores in a semi-arid ecosystem (Tsavo, Kenya). ITC Publication #4.

WWF, (1997). Conserving Africa’s Elephants: current issues and priorities for action (Dublin, H. T. McShane, T. O. and Newby, J. eds). WWF International, 1196 Gland, Switzerland.

Wyatt, B. K. (2000). Vegetation mapping from ground, air and space – Competitive or Complementary Techniques? In R. Alexander, and Millington A. C. (eds). Vegetation Mapping pp. 3-15. John Wiley and Sons Ltd, Chichester.

86