Mammalian Diversity and Distribution in Human-Altered Tropical Landscapes

Joseph Alexander Smith

A thesis submitted for the degree of Doctor of Philosophy at Imperial College London, May 2009

Declaration

I, Joseph Alexander Smith, confirm that the work presented in this thesis is my own with the following acknowledgement:

Chapter 5: Rodolphe Bernard collated and processed the satellite imagery used in this chapter. Arpat Ozgul provided advice on the estimation of connectivity measures. The concept for this chapter, the analyses and writing are my own work.

Throughout the remainder of the thesis, where information has been derived from other sources, I confirm that this has been indicated. The material contained in this thesis has not previously been submitted for a degree at Imperial College London or any other university.

Abstract

Habitat loss at the hands of human enterprise continues to drive the global decline in biodiversity. While much attention has been placed on the use of protected areas as a means of conservation, there is an increasing need to understand the capacity of unprotected, human-altered landscapes to provide refugia and connectivity at larger spatial scales. This study evaluates the mammalian diversity that persists under alternative land management regimes and degrees of landscape change in south-central Sumatra, Indonesia. Species occurrence data compiled from extensive field surveys across 1600km2 form the basis for analyses of community composition and species- specific responses to the current landscape. Results indicate that species richness declined with increased landscape alteration. The lowest observed species numbers were in areas of industrial scale oil palm production rather than scrub habitats or degraded forest. Endangered mammals that persisted in the wider matrix were extirpated from the oil palm dominated areas. Comparisons between the ecological traits shared by persistent versus locally extirpated species revealed that in the initial stages of landscape change there is the capacity to support large specialist species with slow life histories. As landscape degradation continues to an agricultural matrix only habitat and diet generalists persisted.

Tests of species-specific responses to landscape alteration focussed on the occurrence patterns of Sumatran tigers (Panthera tigris sumatrae) and four principal prey species. Measures of human prevalence derived from survey data and a novel application of occupancy estimation techniques, identified significant negative responses to higher levels of landscape development. Satellite derived measures of habitat connectivity and localised landcover degradation found that connectivity to areas of least disturbed forest was more important for reclusive species such as tapir (Tapirus indicus) and red muntjac (Muntiacus muntjak), while the occurrence of the wide-ranging tiger was more strongly influenced by local landcover degradation. The capacity of human altered landscapes to contribute to the conservation of mammalian communities is closely allied to the availability of degraded forests rather than alternative human altered landcovers. Given that these areas of forest are increasingly subject to degradation and conversion, spatial planning and proactive management are required to safeguard these resources.

Acknowledgements

I am indebted to my supervisors Chris Carbone and Tim Coulson for all of the advice and support that they have provided throughout my research. While Chris and Tim have steered my academic development, it was the ZSL Indonesia Programme that kept me on course during many months in Indonesia and provided the means to collect the data on which this thesis has been built. I am especially grateful to Tom Maddox, Dolly Priatna, Elva Gemita and Adnun Salampessy for their friendship and support. Huge credit must also go to the survey teams that had to endure many months of fieldwork in difficult conditions in order to help me collect the ones and zeros with which I was so obsessed. Emily Fitzherbert has been a great friend and provided much support throughout our time together in Indonesia and also here in the UK. Ian Belcher managed to not lose his patience with me despite endless questioning and many weeks spent fighting with Access databases. I am in your debt once again!

I have been lucky enough to know two generations of the Coulson lab: Luca Borger, Tom Ezard and Kelly Moyes initially and then Fanie Pelletier, Arpat Ozgul, Isabelle Smallegange and Aurelio Malo more recently. You have all been fantastic and played a large part in the development of my thoughts on conservation biology. Special thanks must go to Arpat, Isabelle and Aurelio for their encouragement during my write-up period. Thanks also to Jose Lahoz-Monfort for advice on satellite imagery analyses and to Rodolphe Bernard for arriving at Silwood just in time to rescue me from the nightmare of stripy Landsat imagery. At the IOZ, Amy Dickman has been a great ally throughout the past three and a half years, as have Maurus Msuha, Esteban Payan, Nicky Jenner, Ben Collen, Rob Pickles and Patricia Brekke.

The Natural Environment Research Council (NERC) funded my university studies and the Panthera Corporation funded my fieldwork through two Kaplan Scholarships. The Indonesian Institute of Science (LIPI) provided research permits for all fieldwork. Thanks and apologies in equal measure to my friends and family that have endured the stresses and strains associated with the past few years. Hopefully, it’s all over now! Finally, very special thanks to Tola Oni for always believing in me.

Table of Contents

Declaration 2

Abstract 3

Acknowledgements 4

Table of Contents 5

List of Tables 8

List of Figures 9

1 Research Background 13

1.1 Introduction 13 1.1.1 Industrial agents of landscape change 13 1.1.2 Human agents of landscape change 14 1.1.3 The effects of landscape change on biodiversity 15 1.1.4 Species responses to landscape change 16

1.2 Objectives 18

1.3 Study landscape 19

1.4 Tables & Figures 21

2 The Effects of Anthropogenic Landscape Change on Tropical Mammalian Diversity 25

2.1 Abstract 25

2.2 Introduction 25

2.3 Methods 27 2.3.1 Study Sites 27 2.3.2 Field Methods 28 2.3.3 Analyses 29

2.4 Results 31

2.5 Discussion 34

2.6 Tables and Figures 39

3 Ecological Traits and Mammalian Persistence in Human Altered Landscapes 48

3.1 Abstract 48

3.2 Introduction 48

3.3 Methods 50 3.3.1 Study sites 50 3.3.2 Field methods 50 3.3.3 Analyses 52

3.4 Results 55

3.5 Discussion 56

3.6 Tables and Figures 59

4 Human Agents of Landscape Change 64

4.1 Abstract 64

4.2 Introduction 64

4.3 Methods 66 4.3.1 Study Sites 66 4.3.2 Field methods 66 4.3.3 Analyses 67

4.4 Results 70

4.5 Discussion 72

4.6 Tables and Figures 76 6

5 Prospects for Tiger Conservation in Human-Altered Tropical Landscapes 85

5.1 Abstract 85

5.2 Introduction 85

5.3 Method 87 5.3.1 Study Sites 87 5.3.2 Field Methods 87 5.3.3 Analyses 89

5.4 Results 92

5.5 Discussion 94

5.6 Tables and Figures 97

6 General Discussion 104

7 Appendix 110

7.1 An example of the human activity datasheets 110

7.2 An example of the detection/non-detection datasheets 111

7.3 An example of the sampling cell maps used by the field teams during active search periods 112

8 Literature cited 113

List of Tables

Table 2.1 Regional species pool of nonvolant mammals expected to occur in undisturbed, central Sumatran lowland forests. 39

Table 3.1 Description of explanatory variables used to describe species resilience to landscape alteration. 59

Table 4.1 Human activity categories with details of specific contributory indicators and general descriptions. 76

Table 4.2 Principal component loadings and the directions of effect from six human activity categories compiled from landscape surveys. 77

Table 4.3 Summary of model selection and parameter estimates for tigers and four principal prey species. Data derived from intensive field surveys. 78

Table 5.1 Estimated range kernels for tiger and three principal prey species. 97

Table 5.2 Summary of model selection and parameter estimates for tigers and three principal prey species. Data derived from satellite imagery. 98

List of Figures

Figure 1.1 Landsat ETM+ imagery mosaic (band TM5), south-central Sumatra. 21

Figure 1.2 Species distributions with respect to human land use intensity. 22

Figure 2.1 Study site locations with respect to protected areas, agri-industrial land uses and Sumatran provincial borders. 40

Figure 2.2 Sample-based rarefaction curves including 95% confidence intervals. Data derived from species detection/non-detection using active search periods and camera traps in 131 survey cells. 41

Figure 2.3 Observed ecological group representation across landscape alteration classes, presented as proportions of the regional species pool. 42

Figure 2.4 Representation of observed species IUCN threat status in the regional species pool and the intermediate and high landscape alteration classes. 43

Figure 2.5 Individual point estimates of species richness with associated 95% confidence intervals from each of four land management areas surveyed in south-central Sumatra. 44

Figure 3.1 Study site locations with respect to principal protected areas, agri-industrial land uses and Sumatran provincial borders. 60

Figure 3.2 Classification tree analysis for the 51 species of mammal assigned to the regional species pool. 61

Figure 3.3 Classification tree analysis for 27 mammal species detected in the human altered landscape overall (i.e. all four and management areas). 62

Figure 4.1 Study site locations with respect to principal protected areas, agri-industrial land uses and Sumatran provincial borders. 79

Figure 4.2 Proportion Area Occupied (PAO) by each of six human activity categories throughout the sampled area (n = 131 survey cells). 80

Figure 4.3 Prevalence of human activities in individual land management areas. 81

Figure 5.1 Landsat ETM+ imagery mosaic (band TM5), south-central Sumatra. 99

Figure 5.2 Probabilities of species occurrence with respect to human land use intensity. 100

Plate 5.1 Photograph of the soil roads that are widespread throughout the landscape. 102

Chapter 1

Research Background

1 Research Background

1.1 Introduction

Throughout the humid tropics the extraction of forest commodities and the expansion of plantation agriculture are principal drivers of deforestation and subsequent declines in biodiversity (Geist & Lambin, 2002; Pimm & Raven, 2000). This exploitation of tropical landscapes is set to continue, with the appropriation of land for agriculture expected to continue rising for the next 50 years. In support of a growing human population, forecasted to reach 9 billion people by 2050 (UN, 1999), an area of 109 ha of natural ecosystems will be converted to agriculture (Tilman et al., 2001). Within the resulting matrix of agriculture, industry and rural settlement a growing proportion of tropical biodiversity must persist if it is to survive (Daily, 2001). While protected areas are invaluable tools in our efforts to stem the decline of biodiversity they are limited by their geographic coverage (Rodrigues et al., 2004) and susceptibility to fragmentation at the hands of increasingly pervasive human populations (e.g. Brooks et al., 2004; DeFries et al., 2005; Wittemyer et al., 2008) As such, the success of efforts to conserve tropical species will ultimately depend on our ability to utilise degraded lands to provide refuge in their own right and connectivity in support of protected areas (Daily, 2001; Daily et al., 2003).

1.1.1 Industrial agents of landscape change

In Indonesia the principle industrial practices that drive landscape change are the logging of natural forest concessions, agro-forestry and oil palm production (Elaeis guineensis). The legal extraction of timber has supplied international export markets while concurrent illegal logging meets much of Indonesia’s domestic demand (Klassen, 2006). Illegal logging has been estimated to account for as much of 50% of total timber production from Indonesian forests (ITFMP, 1999), and is therefore a major contributor to the unsustainable use of forest resources. Although there has been a decline in natural forest concessions in recent years, this has been countered by the proliferation of agro-forestry. Fast-growing tropical wood plantations for the pulp and paper industry (Cossalter & Pye- Smith, 2003) have expanded through the conversion of natural forests (Barr, 2001) and are predicted to continue expanding throughout Indonesia in years to come (FWI/GFW, 2002). The implications of these landuses for Asian biodiversity are the subject of ongoing research (Nyphus & Tilson, 2004; Nasi et al., 2008) but in the New World tropics research has demonstrated that these landuses can support more forest adapted species than alternative intensive agricultural systems (e.g. Gardner et al., 2008).

Oil palm is the most valuable plantation economy of the tropical world (Henderson & Osborne, 2000) and is responsible for much of Indonesia’s agricultural expansion. Currently, Indonesia meets ±43% (FAOSTAT 2007) of global demand for this, the worlds most traded oil seed crop (Carter et al., 2007). Oil palm has risen to prominence as a global commodity because of numerous applications in the food and cosmetics industries (Casson, 2003). In addition, the European biofuel markets are providing further stimulus for the expansion of oil palm production (Clay, 2004; Danielsen et al., 2008). Although the proliferation of oil palm has raised concerns over the impacts on biodiversity, perhaps the greater threat lies in the continuation of this trend, with an additional 43% increase in production forecast before 2023 (Casson, 2003) with production largely concentrated on Sumatra.

Recent research indicates that oil palm plantations specifically are a poor substitute for native tropical forests; supporting impoverished floral and faunal communities typically dominated by a few non-forest species (Danielsen et al., 2008; Fitzherbert et al., 2008). In addition, an increase in oil palm concessions is likely to confer an increase in the marginal, degraded lands that routinely follow industrial scale land clearance. These areas are highly dynamic, transitional landscapes dominated by rural people and one example of the secondary consequences of agricultural expansion that broaden its implications.

1.1.2 Human agents of landscape change

The Indonesian government’s transmigration programme has lead to the ongoing development of Sumatra through the movement of people from Java and Bali. This initiative amounted to the largest organised resettlement ever recorded (Leinbach, 1989). At its peak between 1969 and 1993 the initiative moved in excess of eight million people and stimulated the clearance of 1.7 million hectares of agricultural land (GOI 1993 as cited in Barber & Schweithelm 2000). In the latter stages of the programme emphasis was moved away from the establishment of subsistence agriculture in favour of wage labour allied to industrial scale, concessionary agriculture and resource extraction (e.g. oil palm estates and logging concessions (Potter & Lee, 1998)). Agricultural expansion leads to the direct transformation of natural landcover and also a more insidious displacement of people into frontier lands. The transmigration programme had more profound environmental consequences than anticipated largely due to poor relocation sites and highly destructive land use practices (World Bank, 1994). Such extensive movement of people combined with uncertain land tenure laws have produced social issues that now underpin the degradation of rural landscapes on Sumatra. For example, it is not

uncommon for fire to be used as a weapon where land tenure disputes lead to conflict ( Tomich et al., 1998; Applegate et al., 2001; Suyanto, 2007).

As such, we can identify two key processes in the development of rural landscapes in this region: industrial transformation and chronic degradation by rural people. Profound changes at the hands of commercial enterprise are readily monitored by satellites but these sensors rarely capture the more subtle effects of landcover degradation (Nepstad et al., 1999; Achard et al., 2002; Butler & Laurance, 2008). Degradation can have profound effects on the integrity of native landcovers and associated biodiversity, and as such is an important consideration for conservation strategy (Phillips, 1997; Sodhi et al., 2009).

1.1.3 The effects of landscape change on biodiversity

While landcover degradation has implications for conservation strategy, there is also the complex issue of variation between taxa in their responses to these changes (Peh et al., 2004; Barlow et al., 2007). In the New World tropics recent research has demonstrated that degraded and production forests support more forest adapted species than alternative or intensive agricultural land uses (Lindenmayer & Franklin, 2002; Barlow et al., 2007). Meta-analyses by Danielsen et al. (2008) indicate that oil palm plantations support impoverished native floral and faunal communities, typically dominated by a few non-forest species. Total vertebrate species richness on oil palm plantations was less than half (38%) that of natural forest areas.

A key objective within conservation science is to identify why certain species and species-groups are more susceptible to extinction than others (McKinney, 1997). While the overall result of anthropogenic landscape change appears to be a decline in biodiversity (e.g. Fitzherbert et al., 2008), there are stark differences between individual species’ responses (Davies et al., 2000; Owens & Bennett, 2000; Daily, 2001). The groups of species that occur in modified landscapes are typically subsets of an ancestral species pool, altered in composition and structure by the loss of species that were unable to persist in a changed landscape (Duchamp & Swihart, 2008). Intrinsic biological traits influence species’ responses to changing environments and ultimately their risk of extinction (Bennett & Owens, 1997; Purvis et al., 2000a; Jones et al., 2003; Cardillo et al., 2004). Collecting trait data from within communities that persist under different degrees of landscape alteration would allow us to identify the traits shared by species that persist and those that become locally extinct. Overall, the main intrinsic traits linked to extinction are body size, habitat and/or dietary specialisation (Owens & Bennett, 2000; Purvis et al., 2000b). Larger bodied species are consistently associated with higher probabilities of

extinction (Fisher & Owens, 2004) and within mammals specifically, threatened species are on average an order of magnitude heavier than non-threatened species (Cardillo et al., 2005). In a changing landscape, broader habitat and/or dietary requirements would be expected to confer a greater advantage to species survival. This idea is supported in the literature where ecological specialisation is associated with higher risk of extinction in both mammals (Haracourt et al., 2002; Boyles & Storm, 2007) and birds (Julliard et al., 2003; Shultz et al., 2005).

1.1.4 Species responses to landscape change

There is considerable interest in understanding patterns of species occurrence as a function of landscape factors or habitat characteristics (Scott et al., 2002). These techniques discriminate between locations that are and are not occupied by the species of interest (e.g. Hirzel et al., 2002). In recent years this approach has been used to model habitat relationships of several mammal species (Carroll et al., 1999; Reunanen et al., 2002; Linkie et al., 2006; Buij et al., 2007). Throughout this body of research, false absences in which sites are occupied by the species of interest but the species remains undetected are a source of bias that must be explicitly accounted for (MacKenzie et al., 2002). The occupancy estimation and modelling techniques developed by MacKenzie et al. (2002) represent generalised linear regression analyses that accommodate these false absences. For rare species that are often the focus of conservation science, traditional methods of monitoring, e.g. estimates of abundance, are often unattainable. However, presence/absence data collected and analysed within an occupancy framework could be used to understand the landscape traits that determine where species occur. The tiger (Panthera tigris) is one such rare species.

The tiger represents one of the greatest challenges to conservation because of the space it requires, its value as a saleable commodity and its tendency toward conflict with people (Sunquist & Sunquist, 2002). Research has shown that availability of prey is the key factor driving tiger habitat selection; there is a positive relationship between tiger and prey densities (Schaller, 1967; Seidensticker & McDougal, 1993; Sunquist, 1981). Thus, there is a need to conserve habitat at a local level for prey as much as for tigers. Where prey data is not available, researchers use geographic proxies to differentiate the areas where tigers do and do not occur e.g. Linkie et al. (2006).

The early stages of habitat modification by humans may in certain circumstances be beneficial to tigers. The creation of forest mosaic habitats and edge environments provide cover for concealment and stalking during hunting (Sunquist, 1981; Prins & Iason, 1989). Edge habitats also support herbivorous prey species such as wild boar (Sus scrofa) and

sambar (Rusa unicolor), and productivity may be greater than in closed forest (Santiapillai & Ramono, 1987; Nowell & Jackson, 1996). However, this only holds true if there are relatively low levels of human persecution and controlled habitat modification, otherwise severe habitat degradation can result (McNeely, 1994). In contrast, complete removal of native vegetation often results in insufficient ground level cover and a lack of food for the ungulate prey base (Sawyer, 1993).

Recent research to identify where wild tigers can persist in the long term has focussed on mapping suitable forest cover at large spatial scales (Wikramanayake et al., 1998; Dinerstein et al., 2007). The extensive Tiger Conservation Landscapes (TCLs) that result are designed to harbour self-sustaining populations of tigers and prey indefinitely. However, there is increasing recognition of the importance of the surrounding matrix on the viability of tiger populations in reserves (Ranganathan et al., 2008), but little is known about the determinants of tiger occurrence outside of protected areas (see Maddox et al., 2007; Linkie et al., 2008). Excluding other risk factors (e.g. poaching, disease etc), the longterm viability of remnant tiger populations, and the populations of prey on which they rely will ultimately be dependent upon the connections within and between TCLs (Wikramanayake et al., 2004). This will inevitably require that species are able to persist, or at very least pass-through, the intervening human dominated matrix. Connecting habitat patches through commercial land use may ultimately improve the prospects for fragile, small populations (Noss, 1987, 1991).

In summary, current thinking suggests that the future of tropical biodiversity conservation lies in the management of human–altered landscapes (Daily, 2001; Lindenmayer & Franklin, 2002; Gardner et al., 2009). This philosophy requires that human dominated landscapes provide requisite wildlife habitats amidst a largely tolerant human population. Although research suggests that many components of biodiversity can persist in these landscapes (Daily et al., 2003; Barlow et al., 2007), this does require that they are viable in the long-term and not subject to significant degradation. Given that human-dominated landscapes are generally complex matrices of different land uses it is likely that they are highly susceptible to degradation by human activities (Daily, 2001). These pervasive human activities can blur the distinction between legally designated protected areas and the wider matrix. This realisation drives current interest in the conservation potential of wider landscapes and poses the question: to what extent could a latent conservation value in human altered landscapes be used to mitigate the decline of tropical biodiversity?

1.2 Objectives

In this research I have combined extensive field surveys and quantitative analyses to identify patterns in mammalian occurrence throughout a representative, human altered matrix in south-central Sumatra. Although mammals are a well-studied taxonomic group that draw considerable research interest, the capacity to conserve these species amidst developing human dominated landscapes remains largely unknown. My specific objectives were (1) to describe changes in mammalian species and community composition that occur across different land management regimes and the associated gradient of landcover alteration, (2) to investigate the intrinsic ecological traits associated with patterns of mammalian persistence and local extinction, (3) to identify the human drivers of current landscape degradation and subsequent effects on the local occurrence of tigers and prey species, and (4) to test a conceptual model of the landscape traits that influence the occurrence of tigers and prey species at wider landscape scales. This thesis contains six chapters: a general introduction to the research field and study landscape (Chapter 1), four research papers (Chapters 2 – 4) and a general discussion (Chapter 6). The research papers reflect two main themes:

Mammalian diversity and species persistence In Chapter 2, I combined species inventories from field surveys and a regional species pool compiled from the literature to describe the decline in species richness, changes in community composition and representation of threatened species with increasing landscape alteration. In Chapter 3, I investigated patterns of mammalian persistence and local extinction as functions of ecological traits. Specifically, I investigated the importance of body size and a number of reproductive, dietary and behavioural traits in patterns of species persistence.

Patterns of species occurrence in anthropogenic landscapes: In Chapter 4, I used detection/non-detection data and occupancy analyses (MacKenzie et al., 2002) to identify the prevalence of human activities and subsequent effects on probabilities of species occurrence. In Chapter 5, I developed a conceptual model to determine if an aversion to sparse landcover or connectivity to least disturbed forest best explained patterns of tiger and prey occurrence in this landscape.

1.3 Study landscape

This study was conducted on Sumatra, the second largest island in the Indonesian archipelago. The island’s size (476,000km2) and past connection to mainland Asia, coupled with its diverse natural vegetation support exceptional biodiversity that includes 201 species of mammal (Whitten et al., 2000). The specific study sites were located in the adjoining provinces of Jambi and South Sumatra situated in the south of the island (latitude 1o 53´ to 2o 35´S, longitude 103o 2´ to 104o 9´E). Based on forest cover maps reproduced by Whitten et al. (2000), these sites lay in contiguous forest that spanned from east coast to west until at least 1982 (FAO/MacKinnon, 1982). However, subsequent maps produced in the mid 1990s indicate that in the intervening 14 years the study sites became isolated forming a discrete patch of forest.

Field data were collected from two commercial land management areas (an oil palm plantation and a logging concession) and two legally designated wildlife reserves (Dangku and Bentayan). Individual study sites were surveyed in quick succession between March and September 2006 in the following order: (i) Bentayan wildlife reserve (BWR, 300-km2), a low-level protected area (IUCN category IV) subject to extensive human settlement and associated land cover conversion; (ii) a selective logging concession (SLC, 800-km2), subject to industrial legal logging and small-scale uncontrolled logging; (iii) an oil palm concession (OPC, 270-km2), where an oil palm monoculture was managed amidst an adjacent matrix of scrub habitats that have been extensively colonised by pioneer land-settlers, and (iv) Dangku wildlife reserve (DWR, 250-km2), a second low-level protected area (IUCN category IV) extensively populated by rural settlers with associated uncontrolled logging and land cover conversion. Maps summarizing the locations where footprints of the focal species (tiger, sambar, Malayan tapir and red muntjac) were recorded are presented in Figures 1.2a and 1.2b. Similar maps are not presented for wild pig as this species was extremely abundant and consequently was only recorded on the first detection in each sampling occasion/cell.

To record evidence of mammal species and concurrent human activities I conducted repeated detection/non-detection surveys (MacKenzie et al., 2002) in 4km2 sampling cells drawn from each management area. The data derived from these field surveys were treated differently in Chapters 2 and 3 versus Chapters 4 and 5. Specifically, for Chapters 2 and 3 the cells are grouped by management area to compile species inventories. Chapters 4 and 5 use detection histories generated from the repeated samples of each cell.

It is important to note that the study design described here introduces pseudoreplication to the data - repeated samples were taken from groups of survey cells within individual land management areas. In light of this, care should be taken in considering the communities recorded in this work as representative of the communities in similar land uses in the wider landscape. However, the land uses sampled in this study represent a gradient of land use intensity similar to that found elsewhere in the region and as such the data and analyses presented should be interpreted with respect to this gradient.

Please note that because each of the data chapters in this thesis is written as a standalone paper there is some repetition in the sections titled Field Methods and Study Sites.

1.4 Tables & Figures

Figure 1.1 Landsat ETM+ imagery mosaic (band TM5), south-central Sumatra. (a) Sumatran protected areas are indicated in black; they include IUCN categories Ia (Nature reserve), II (National Park), IV (Wildlife reserve) and VI (Protection forest); (data from the World Protected Area Database, available at www.unep-wcmc.org). Survey locations lay within the white polygon. (b) Individual management areas in detail, black points indicate survey cell centres (n = 131). Areas in dark grey indicate area of least disturbed forest of >1km2 that persist amidst the human-dominated matrix of degraded natural landcover and agri-industrial land uses (indicated in light grey). Bare soil and the areas of most sparse vegetation are indicated in white. 21

(a)

(b)

Figure 1.2a Species distributions with respect to human land use intensity. Panels (a) and (b) indicate the locations of tiger and sambar footprints respectively. Black polygons indicate the boundaries of individual land management areas with the locations of 4km2 survey cells (n = 131) also shown. Dark grey patches indicate areas of least disturbed forest. The human dominated matrix of degraded natural vegetation, settlements and agricultural land uses are shown in grey with areas of bare soil and sparse vegetation indicated in white.

(c)

(d)

Figure 1.2b Species distributions with respect to human land use intensity. Panels (c) and (d) indicate the locations of tapir and red muntjac footprints respectively. Black polygons indicate the boundaries of individual land management areas with the locations of 4km2 survey cells (n = 131) also shown. Dark grey patches indicate areas of least disturbed forest. The human dominated matrix of degraded natural vegetation, settlements and agricultural land uses are shown in grey with areas of bare soil and sparse vegetation indicated in white.

Chapter 2

The Effects of Anthropogenic Landscape Change On Tropical Mammalian Diversity

2 The Effects of Anthropogenic Landscape Change on Tropical Mammalian Diversity

2.1 Abstract

Throughout the tropics, human-altered landscapes are expanding behind agri-industrial frontiers. The success of efforts to conserve tropical diversity is increasingly dependent on our ability to secure the conservation potential of these human altered landscapes; but in many settings insufficient data exist to capitalise on this perceived potential. Here I present an inventory of the nonvolant mammals detected during extensive field surveys throughout the agri-industrial matrix of south-central Sumatra. Results indicate that highest mammalian richness was associated with management areas dominated by degraded forests and that these areas supported the larger and most highly threatened (IUCN categories endangered and critically endangered) members of the regional species pool. In contrast, the mammalian community associated with oil palm dominated areas was significantly impoverished compared to other management areas sampled and no longer supported highly threatened species. Areas of remnant, degraded forest offer the best hope for the conservation of threatened mammals in the human altered landscapes of Sumatra. However, the proliferation of the oil palm industry and continued degradation of native vegetation threatens to undermine the conservation potential of these areas.

2.2 Introduction

Throughout the humid tropics the extraction of forest commodities and the expansion of plantation agriculture are principal drivers of deforestation and subsequent declines in biodiversity (Pimm & Raven, 2000; Geist & Lambin, 2002; Laurance & Peres, 2006). The exploitation of tropical landscapes is set to continue, with the appropriation of land for agriculture expected to continue rising for the next 50 years (Tilman et al., 2001). In the wake of this prolific landscape change, human-altered landscapes now dominate. It is within these matrices of agriculture, industry and rural settlement that a growing proportion of tropical biodiversity must persist if it is to survive (Daily, 2001). Although strictly protected areas (e.g. IUCN categories I-III) can provide refuge for many species, these areas are not infallible and are increasingly undermined by the spread of human activities (e.g. Woodroffe & Ginsberg, 1998; Brooks et al., 2004; DeFries et al., 2005). As such, the success of efforts to conserve tropical species will ultimately depend on our ability to utilise degraded lands to provide refuge in their own right and connectivity in support of protected areas (Daily, 2001; Daily et al., 2003).

South East Asian forests are the most species rich but also the most threatened globally (Laurance, 2007). Indonesia holds much of South East Asia’s remaining primary forests (Koh, 2007) but these forests are subject to the same direct and indirect pressures that affect tropical forests at large. In recent years, although there has been a decline in natural forest concessions this has been countered by the proliferation of agricultural land uses, particularly agro-forestry and oil palm (Elaeis guineensis) plantations.

In the preceding 40 years, Indonesia has seen deforestation rates of ~1.3 million hectares per year (FAO, 2005). The legal extraction of timber from Indonesian natural forest concessions has supplied international export markets while concurrent illegal logging has addressed much of the domestic demand (Klassen, 2006). Illegal logging has been estimated to account for as much of 50% of total timber production from Indonesian forests (ITFMP, 1999) and is therefore a major contributor to the unsustainable use of forest resources. Fast-growing tropical wood plantations for the pulp and paper industry (Cossalter & Pye-Smith, 2003) have expanded through the conversion of natural forests (Barr, 2001) and are predicted to continue expanding throughout Indonesia in years to come (FWI/GFW, 2002). The implications for Asian biodiversity of an increase in this specific landuse are the subject of ongoing research (Nyphus & Tilson, 2004; Nasi et al., 2008) but in the New World tropics research has demonstrated that degraded and production forests can support more forest adapted species than more intensive agricultural systems (Lindenmayer & Franklin, 2002; Barlow et al., 2007).

Much of Indonesia’s agricultural expansion is attributable to the oil palm industry. Currently, Indonesia meets ±43% (FAOSTAT 2007) of global demand for this, the worlds most traded oil seed crop (Carter et al., 2007). Oil palm is the most valuable plantation economy of the tropical world (Henderson & Osborne, 2000) and production is increasing by 9% per year, largely in response to the European biofuel market and demand from the Asian food industry (Clay, 2004). Recent research indicates that oil palm plantations specifically are a poor substitute for native tropical forests; supporting impoverished floral and faunal communities typically dominated by a few non-forest species (Danielsen et al., 2008; Fitzherbert et al., 2008). In addition, an increase in oil palm concessions is likely to confer an increase in the marginal, degraded lands that routinely follow industrial scale land clearance. These areas are highly dynamic, transitional landscapes dominated by rural people, rarely considered in the scientific literature. They are one example of the secondary consequences of agricultural expansion that broaden its implications.

In this study I investigate the implications of anthropogenic landscape change on the mammalian diversity of south-central Sumatra. Using a combination of extensive field surveys and a regional species pool compiled from the literature I describe the decline in species richness and changes in community composition that occur with increasing landscape-alteration. Simple ecological traits were used to summarize different groups in the species pool and which of these persisted or were locally extirpated under different degrees of relative landscape change. Finally, I focus on the impoverished mammalian communities identified in oil palm dominated areas and discuss the implications of continued landscape change for the future conservation of mammals in this region.

2.3 Methods

2.3.1 Study Sites

Between March and September 2006 I sampled four land management areas in south- central Sumatra (latitude 1o 53´ to 2o 35´S, longitude 103o 2´ to 104o 9´E). The individual study sites encompassed a gradient of landscape alteration. Management areas dominated by degraded forest and extensive natural vegetation were structurally more complex and subject to less human disturbance than areas of oil palm crop or the adjacent scrub habitats that were colonised by illegal land settlers. This coarse description of landcover complexity was used to group study sites as either intermediate or highly altered landscapes. The individual land management areas assigned to the intermediate landscape alteration class were (i) a selective logging concession (SLC, 800-km2, 45 cells), subject to industrial legal logging and small-scale illegal logging; (ii) Dangku wildlife reserve (DWR, 250-km2, 28 cells), a low-level protected area (IUCN category IV) extensively populated by rural settlers with associated illegal logging and land cover conversion and, (iii) Bentayan wildlife reserve (BWR, 300-km2, 30 cells), a second low-level protected area (IUCN category IV) also subject to extensive human settlement and associated land cover conversion. The alternative high landscape alteration class contained two components of an oil palm concession (OPC, 270-km2), (i) an oil palm monoculture (OPC_M, 14 cells) and (ii) an adjacent matrix of scrub habitats (OPC_S, 14 cells) that had been extensively colonised by pioneer land-settlers.

2.3.2 Field Methods

I sampled 131 2 x 2-km cells drawn from the aforementioned land management areas. A combination of active search periods (trained field personnel) and camera traps were used to detect nonvolant mammals. Data collection methods were consistent across sites, with survey cells drawn from simple random samples of the overall management area (OPC, DWR), sub-sections thereof (SLC) or a uniform grid initiated from a randomised start location (BWR). This approach to cell selection balanced the logistical constraints of cell accessibility, geographic coverage of the overall management area and the desire to generalise findings to the wider landscape. Overall, survey cells covered 29% of the total area (1620-km2) encompassed by the management boundaries.

Five survey teams, each of two people, were established in March 2006. Teams were led by individuals of equivalent training and field experience. Each team operated independently within the survey cells and visited each cell only once. Four teams were active on a given day with teams rotating between proximate cells once three hours of active search effort had been completed. Teams travelled to and between cells by motorbike. For logistical reasons cells were surveyed in groups, with clusters of four neighbouring cells surveyed on 3-4 sampling occasions (85% of cells = 4 sampling occasions) over a two-day period. This approach provided repeated independent samples of each sampling cell.

During each three-hour sampling occasion teams aimed to travel widely throughout the cell (mean average two-dimensional distance was 4.8km, derived from GPS odometers on sampling occasions with ~3hrs GPS coverage, n = 202) searching a representative sample of the available habitats and in turn maximising the probability of encountering mammal signs. Direct mammal signs were sightings or audio cues and indirect signs were predominantly footprints but included other indicators such as faeces where these could be assigned to a target species with confidence. Team leaders geo-referenced all direct and indirect signs using Garmin 60c global positioning systems (GPS) (Garmin International Inc., Olathe, KS) in universal transverse mercator (UTM) coordinates. All teams were equipped with footprint identification guides and followed rigorous species identification protocols to minimise the risk of false positive species detection through misidentification of signs.

height of 30 – 45cm depending on the terrain and vegetation structure in the immediate vicinity of the trap site. The camera placement protocol was designed to maximise the probability of detecting mammals without introducing systematic bias toward a particular species.

2.3.3 Analyses

Observed species richness Sample-based rarefaction was used to test the completeness of the species inventory from the human altered landscape overall (i.e. all management areas combined). The rarefaction curve was compiled using the Mao Tau estimator (Colwell et al., 2004) implemented in EstimateS software (Colwell, 2008). Input data were binary species incidences (1/0) derived from species’ detections/non-detections during active search periods and camera trap photographs recorded in each survey cell (Figure 2.2).

Regional species pool I defined a regional species pool (sensu Cam et al., 2000; Daily et al., 2003) of mammal species that would be expected to occur in a primary, central Sumatran, lowland forest ecosystem based on species distributions published by Payne & Francis (1985), Nowak (1999) and Macdonald (2006). This reference community was restricted to species that could have been detected by our sampling methods; defined as species ≥ 0.65kg adult mass, ≥ 40cm head-body length and either partially terrestrial or if strictly arboreal primates, group living and vocal. Species meeting these criteria were considered sufficiently detectable by our combination of active search periods and camera trap sampling. The final regional species pool contained 51 species drawn from eight taxonomic orders and 18 families (Table 2.1).

Observed ecological group representation Each of the 51 species in the regional pool was assigned to a simple ecological group based on body mass and trophic group; requisite data were compiled from Nowak (1999), Macdonald (2006) and Jones et al. (2009). The frequency distribution of species’ body mass estimates across trophic groups was positively skewed; mammals of ≤17.6kgs represented 84% of the overall regional pool (specifically, 94% of carnivores, 76% of herbivores, 83% omnivores). The largest mammals were >162kg (N = 5) with intermediate body sizes (17.6 - 162kgs) represented by three omnivores (Sus barbatus; 135.8kgs, Sus scrofa; 84.5kgs and Helarctos malayanus; 57.1kgs). I applied a 21kg cut- off to distinguish small and large bodied mammals across all three trophic groups. This body mass cut-off had a clear biological interpretation for carnivores as it represented the distinction between small prey and large prey feeding strategies (Carbone et al., 1999;

Carbone et al., 2007). I applied the same cut-off to herbivores and omnivores in the interests of consistency and because this value served as an effective distinction between modal body mass estimates and the tail of the distributions from these trophic groups.

This list of species per ecological group was subset by management area (DWR, BWR, SLC, OPC_S, OPC_M) and ultimately collapsed into a two level factor that described broad levels of landscape alteration: Intermediate (composed of DWR, BWR and the SLC) and High (composed of OPC_M and OPC_S). This approach provided pair-wise comparisons between the regional species pool and each of the landscape alteration classes (Figure 2.5).

Analysis was by generalised linear model (glm) with a binomial error structure. The response variable was the proportion of species from each ecological group in the regional pool that was detected in each landscape alteration class. Explanatory variables were categorical: trophic group (three levels, herbivore, omnivore, carnivore), body mass (two levels, small ≤21kgs, large >21kgs) and the relative level of landscape alteration (two levels, low = degraded forest matrices, high = agriculture/scrub habitat matrices). The minimum adequate model was derived by backward stepwise removal of non- significant terms (p >0.05) from the saturated model following analysis of variance (anova, test = Chi) (Crawley 2008). Analyses were conducted using the statistical software package R, version 2.7.2 (R Development Core Team, 2008).

Observed IUCN threat status To assess the conservation implications of reduced species richness under different levels of landscape alteration I used chi-squared goodness of fit tests to identify significant differences in the representation of different threat categories between landscape alteration classes with respect to the regional species pool. The IUCN threat categories Near Threatened (NT), Vulnerable (V), Endangered (E) and Critically Endangered (CE) were reduced to a two level factor: Low threat (NT and V combined) and High Threat (E and CE combined). In pair-wise comparisons between high and low levels of relative landscape alteration, the observed species richness in the lesser-altered landscape represented the expected species richness of the higher alteration landscape under the null hypothesis.

Estimated species richness Point estimates of species richness and associated 95% confidence intervals were calculated for each land management area using three estimators that accommodate heterogeneity in detection probabilities between both species and sampling cells. I used model(h), also known as the Incidence-based Coverage Estimator (ICE), model(th) (Lee & Chao, 1994; Shen, 2003), and the first-order jackknife estimator (Burnham & Overton, 1978). These estimators are implemented in the free software package Species Prediction and Diversity Estimation (SPADE) (Chao & Shen, 2003). Using a range of estimators in this way provides a conservative approach to species richness estimation. This analysis tests for a coarse gradient of change in species numbers across alternative land management areas and the relative landscape alteration classes that they represent. I calculated mean average species richness from the individual point estimates in each landscape alteration class. This measure of community integrity sensu Cam et al. (2000) allows for species detection probabilities of <1 and ensures that comparisons between landscape alteration classes are relative by accounting for the diminished regional species pool in progressively more altered landscapes.

Ordination of species richness Non-metric multidimensional scaling ordination (MDS) (Clark & Warwick, 2001) was used to evaluate community composition between land management areas and landscape alteration classes. Input data were species detection/non-detection (1/0) per survey cell (n = 131). I used MDS (PRIMER software, version 5) rather than an alternative ordination technique because this approach makes fewer assumptions about the distribution of the underlying data and provides a visually intuitive summary of community similarity between sampling locations (Clark & Warwick, 2001).

2.4 Results

Observed species richness Sample-based rarefaction curves indicated the completeness of the species inventory from the human altered landscape overall (Figure 2.2). In total, 27 mammal species were detected in the human altered landscape overall. This observed species richness represents 53% of the species assigned to the regional species pool (n = 51 species) of south-central Sumatran lowland forest mammals that were susceptible to detection by the sampling methods used. This species inventory contained representatives from 15 taxonomic families, three fewer than occurred in the regional pool since both representatives of the Herpestidae (Herpestes brachyurus, Herpestes semitorquatus) and the individual representatives of the Mephitidae (Mydaus javanensis) and Rhinocerotidae

(Dicerorhinus sumatrensis) were absent from our sample. Negligible differences existed between observed species numbers in qualitatively similar management areas. Within the wildlife reserves, 26 mammal species were detected versus 24 in the selective logging concession; with 89% of species shared between these land management areas. Within OPC_S areas, 17 species were detected compared to 14 in the OPC_M; 78% of species were shared between these areas.

Ecological group representation Carnivores, herbivores and omnivores were approximately equally represented in the overall regional species pool (31.4%, 33.3% and 35.3% respectively). In analyses of ecological group representation (Figure 2.3), the minimum adequate model retained body mass and landscape alteration class as significant variables (p < 0.001). Back- transformation of the coefficients from the linear predictor indicate that on average, 26% fewer small mammals (≤ 21kg body mass) were detected in the altered landscapes compared to the regional species pool and that areas of intermediate landscape alteration contained 72% more mammal species than areas of higher relative alteration. Large carnivores (Panthera tigris sumatrae) were not detected in the high alteration landscape class.

Observed IUCN threat status Representation of Low and High threat status species (see Analyses) declined significantly in pair-wise comparisons between the regional species pool and the High landscape alteration class (X2 = 12.58, df =1, p<0.05). There was evidence of a declining trend in species numbers between the regional species pool and the Intermediate landscape alteration class (X2 = 3.38, df =1, p = 0.07). Numbers of threatened species declined significantly between the Intermediate and High landscape alteration classes (X2 = 5.08, df =1, p <0.05). Specifically, low threat species declined and high threat species were not detected at all in the high landscape alteration class (Figure 2.4).

Estimated species richness Estimated levels of species richness varied little among the wildlife reserves (DWR, BWR) and selective logging concession (SLC) that represent areas dominated by degraded forests and disparate rural settlements. In contrast, there was a notable decline in species richness associated with the oil palm concession (OPC_S and OPC_M) areas (Figure 2.5). The model(h) estimator indicated a significant (p <0.05) decline in estimated species richness in the oil palm dominated matrix (OPC_M) compared to the management areas that represented the intermediate landscape alteration class (BWR, DWR and SLC). Observed and estimated species richness in the oil palm dominated matrix (OPC_M) were the lowest values recorded in this study and as such are consistent

with other research indicating that this landcover hosts impoverished vertebrate communities and low overall biodiversity (Danielsen et al., 2008; Fitzherbert et al., 2008).

Mean average point estimates of species richness from management areas of intermediate landscape alteration (BWR, DWR and SLC) indicated that 0.51 (0.47 - 0.70 95% CI) of the regional species pool was likely to occur in those areas compared to 0.36 (0.32 - 0.53 95% CI) in the high landscape alteration areas (OPC_S and OPC_M). The lowest point estimate in this study, 0.32 (0.28 - 0.49 95% CI), was derived from the oil palm dominated matrix (OPC_M) indicating that ~0.70 of the regional species pool was extirpated from this highly altered landscape.

Ordination: community composition Differences in community composition were reflected in MDS ordination plots (Figure 2.6) in which the distances between points (i.e. survey cells, n = 131) is directly proportional to the observed dissimilarity in species detection/non-detection. Survey cells drawn from the OPC_S and OPC_M areas form a discrete group in multi-dimensional space, indicating the relative similarity of these areas in terms of mammalian community composition, and therefore, their dissimilarity from the majority of other survey cells. Tests of analysis of similarity (ANOSIM) between intermediate and high landscape alteration classes identified significant differences (R = 0.35, p = 0.001) in community composition between these areas. Similar results were produced when the data were restricted to only those collected from active search periods (Figure 2.7).

2.5 Discussion

This research makes five principal contributions to our understanding of how landscape change has impacted tropical mammalian diversity in south-central Sumatra. First, I present a representative and largely complete inventory of the nonvolant, forest adapted mammals that occur in this context. Second, I estimate the species richness allied to alternative land management regimes and I highlight the significantly impoverished mammalian communities associated with oil palm dominated areas. Third, I identify that the species that are lost following increasing landscape change, are predominantly the smaller bodied (<21kg) members of the regional species pool. Fourth, I demonstrate that species of highest conservation concern (listed in the IUCN threat categories, Endangered and Critically Endangered) are extirpated from the oil palm dominated areas. Finally, I discuss the continued degradation of remnant forest and the implications this has for the conservation of threatened mammals.

Since I was unable to sample a pristine lowland rainforest system I used a species list compiled from the literature to reflect regional mammalian diversity (Cam et al., 2000; Daily et al., 2003). Results indicate that in south-central Sumatra, the conversion of contiguous lowland forests to a human-altered industrial matrix has resulted in a significant decline in mammalian species richness. Specifically, areas dominated by degraded forest (the selective logging concession, Bentayan wildlife reserve and Dangku wildlife reserve) have lost ~0.5 of the regional species pool, compared to a loss of ~0.7 in areas of dominated by oil palm. Comparisons between the wildlife reserves and the logging concession showed non-significant differences in estimated species richness. These areas were of qualitatively similar vegetation: dominated by degraded forest that provided relatively contiguous natural vegetative cover and so it was unsurprising that these areas contained similar mammalian diversity.

Estimates of mammalian species richness indicated that highest richness was allied to management areas dominated by degraded forest and lowest in areas dominated by oil palm crop. I found evidence of a step-like decline in species richness between these areas. However, a statistically significant difference (p <0.05) in species richness was only detected between the oil palm dominated matrix and all other areas sampled. The disparity between oil palm dominated areas and other components of the sampled landscape was supported by analyses of community composition in which these areas formed a discrete cluster in multidimensional space indicative of the significantly impoverished communities in these areas. These results for species numbers and community composition are in keeping with research on other taxa that have demonstrated that oil palm dominated areas support only low levels of overall biodiversity and severely impoverished remnant communities (Danielsen et al., 2008; Fitzherbert et

al., 2008). The proliferation of oil palm crop will create increasingly large areas of land that are inhospitable to 70% of the Sumatran forest-mammal community described here. In other regions, significant differences in mammalian species richness have also been found between forest dominated areas and agricultural areas; forest areas again supporting the highest mammalian richness (Daily et al., 2003).

I used simple ecological groups to categorise the mammals in the regional species pool, based on trophic group (carnivore, omnivore, herbivore) and body size (small <21kg, large >21kg). The majority of species in the regional pool were smaller bodied with essentially equal representation between trophic groups. The factors influencing the proportion of regional pool species occurring in the altered landscape were body size and relative landscape alteration class. On average, a higher proportion of the small mammals were lost in the transition from the regional pool to the altered landscapes. This is in contrast to the pattern observed in other systems where larger species were removed (Daily et al., 2003). Although I detected all of the larger mammals in the human altered landscape overall it is important to note that these species were allied to areas dominated by degraded forest and that these larger species were not detected in oil palm dominated areas. Overall, areas of intermediate landscape alteration contained 72% more mammal species, on average, than areas of higher relative alteration.

To identify the effects of landscape alteration on species of conservation concern I reduced the IUCN threat categories (IUCN, 2008) to a two level factor: low threat (near threatened and vulnerable) and high threat (endangered and critically endangered). There was evidence of a declining trend in species numbers between the regional species pool and the areas dominated by degraded forest. In addition, numbers of high and low threat species declined significantly in the transition to areas of high landscape alteration. Specifically, low threat species declined and high threat species were not detected at all in the high landscape alteration class. The loss of high threat species in the oil palm areas equates to the only large carnivore in this system, the Sumatran tiger (Panthera tigris sumatrae).

There are several points that the critical reader should consider when interpreting these results. Observed species richness does not account for detection probabilities of <1 and as such will typically represent underestimates of true richness. Consequently, I restrict the use of observed species numbers to summary analyses that are dependent on species identity (i.e. allocation to IUCN threat categories and simple ecological groups). Provided the species inventory is relatively complete (see Figure 2.2) use of these naïve species numbers in these contexts are justifiable as indicators of general trends. Although sampling methods were consistent between land management areas, the area of each differed. Potentially, fewer species might have occurred in certain land

management areas by virtue of their smaller size rather than the adverse effects of the landcover itself.

It is important to note that the study design described here introduces pseudoreplication to the data because repeated samples were taken from groups of survey cells within individual land management areas. In light of this, care should be taken in considering the communities recorded in this work as representative of the communities in similar land uses in the wider landscape. However, the land uses sampled in this study represent a gradient of land use intensity similar to that found elsewhere in the region and so the data and analyses presented here should be interpreted with respect to this gradient.

In addition, individual land management areas were relatively heterogeneous and included a gradient of relative human alteration. This was apparent in Figures 2.6 and 2.7 in which several survey cells from Bentayan wildlife reserve lay in the multidimensional space occupied by the most highly altered survey cells from the plantation concession. This was not unexpected as areas of Bentayan wildlife reserve were subject to qualitatively higher levels of landscape change.

The proximity of individual land management areas and our interest in relatively large bodied, wide-ranging mammals allows species “spill over” from patches of native vegetation in the wider matrix into highly altered areas of land. This is a particular issue when local-regional species richness plots are employed to test for species saturation (Srivastava, 1999), and although I do not employ this analytical technique specifically, I acknowledge the implications of species spill over between sampling areas. However, in the landscape described here the direction of this species spill over was from less altered to more altered landscapes, since there were no species unique to the more altered areas. The net effect of this process is that my estimates of species richness for individual management areas, and the alteration gradient they represent, can be interpreted as the upper bounds of species richness. Spatially explicit analyses that specifically address these issues are the subject of current collaborative analyses. Preliminary results from these analyses indicate that mammalian diversity within the interior of an oil palm monoculture is in fact reduced to four species of mammal (Fitzherbert et al. in prep); representing just 8% of the regional species pool described here.

In light of forecasted agricultural and population expansion, the spread of the human altered landscapes described here are to a large extent inevitable (Daily, 2001). Results from this study suggest that for forest adapted mammals in south-central Sumatra this could lead to a net loss of between 50-70% of the regional species pool, depending on the level of alteration.

The projected expansion of the oil palm industry (Clay, 2004) could impact our capacity to conserve mammalian diversity by directly replacing valuable forest habitats and by indirectly encouraging the colonisation of these areas by rural people (Sandker et al., 2007). Results presented here indicate that oil palm dominated areas support low levels of mammalian diversity, a similar trend to that reported in other taxa (Danielsen et al., 2008; Fitzherbert et al., 2008). No evidence has yet been found of highly threatened species such as the Sumatran tiger using oil palm dominated areas and this is likely to be a consequence of inadequate ground level cover between the mature palms. A key step to mitigate the impact of oil palm would be to increase ground level cover to provide connectivity for wide-ranging mammals (Maddox et al., 2003, 2007).

An increase in oil palm concessions also confers an increase in scrub habitats and rural settlement in areas bordering the concession (Sandker et al., 2007). These areas are highly dynamic, transitional landscapes dominated by rural people. Tigers were previously widespread throughout marginal scrub habitats that bordered the oil palm concession, but it is believed that the colonisation and further degradation of these areas by illegal land settlers was responsible for the extirpation of these animals. It is likely that this was a product of both landcover degradation and poaching (Maddox et al., 2003, 2007).

As oil palm expands, so the habitats that support endangered species in human altered landscapes will diminish. Since the species of greatest concern to conservation are extirpated from this land management regime they will be confined to degraded forest habitats contained within selective logging concessions and small wildlife reserves. These areas could be appropriated under provincial spatial planning to provide refugia and landscape connectivity for wide-ranging mammals, such as the Sumatran tiger, for which isolated protected areas are insufficient (Woodroffe & Ginsberg, 1998).

Many species threatened by the loss of native habitats actually use human altered landscapes and could in principle be conserved through appropriate management of these landscapes (McNeely & Scherr, 2002). Results from this study indicate that the conservation value of human altered landscapes lies principally in areas of degraded forest. However, my observations suggest that this landcover is a transient system progressing towards rural settlement and smallholder agriculture and will therefore be very difficult to secure. On a wider scale, some researchers have suggested that biodiversity loss associated with deforestation can be offset by regenerating degraded lands and plantation forests (Wright & Muller-Landau, 2006). However, others have argued that we still do not know enough about the conservation potential of these areas to be so confident in their utility (Brook et al., 2006; Gardner et al., 2007). To fully

understand the conservation potential of human altered landscapes we need data from all of the landcover types (Craig et al., 1999; Daily et al., 2003; Barlow et al., 2007). Where research has been done, there have been issues with study design, bias toward certain taxa (Gardner et al., 2007) and reliance on species richness as a response variable (Su et al., 2004). However, recognition of these issues has encouraged more rigorous studies intent on resolving these specific issues for multiple taxa (e.g. Barlow et al., 2007).

Wider recognition that landscapes dominated by degraded forests have a particular conservation value, could ultimately provide support for protected areas that otherwise would be left to preserve tropical forest biodiversity in relative isolation. In light of the speed and magnitude with which anthropogenic landscape change is advancing, it is imperative that conservation scientists identify the true capacity for conservation in the wider human-altered matrix.

2.6 Tables and Figures

Table 2.1 Regional species pool of nonvolant mammals expected to occur in undisturbed, central Sumatran lowland forests. Taxonomy follows Wilson and Reeder (2005). IUCN Red List status is denoted as follows: Not Listed (NL), Data Deficient (DD), Least Concern (LC), Near Threatened (NT), Vulnerable (V), Endangered (E) and Critically Endangered (CE). Size classes refer to adult body mass: small <21kg, large >21kg. Detection method indicates whether a species was detected during active search periods (AS), by camera trap (CT), or by both (AS+CT). 1 1 0 1 0 0 0 0 0 0 1 0 0 0 0 0 0 0 1 0 0 0 1 0 1 0 1 1 0 0 0 0 0 0 1 0 0 0 1 1 0 0 0 0 1 0 0 0 1 0 0 14 OPC_M Detected in 1 1 1 1 1 0 0 0 0 0 1 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 1 1 0 0 0 0 0 0 1 1 0 0 1 1 0 0 0 0 1 0 0 0 1 0 1 17 OPC_S High landscape alteration class Detected in 1 1 1 1 1 1 1 0 1 1 1 0 0 0 0 0 0 0 1 0 0 0 1 0 1 0 1 1 0 0 0 0 0 0 1 1 0 1 1 1 0 0 0 0 1 1 0 0 1 0 1 24 SLC Detected in 1 1 0 1 1 1 1 0 1 1 1 0 1 0 0 0 0 0 1 0 0 0 1 0 1 0 1 1 0 0 0 0 0 0 0 1 0 0 1 1 1 0 0 0 1 1 0 0 1 0 1 23 DWR Detected in 1 1 0 1 1 1 0 0 1 0 1 0 0 0 0 0 0 0 0 0 0 0 1 0 1 0 1 1 0 0 0 0 0 0 1 1 0 1 1 1 0 0 0 0 1 1 0 1 1 0 1 21 BWR Intermediate landscape alteration class Detected in CT CT AS AS AS AS AS AS AS AS CT ND ND ND ND ND ND ND ND ND ND ND ND ND ND ND ND ND ND ND ND ND ND ND ND AS+CT AS+CT AS+CT AS+CT AS+CT AS+CT AS+CT AS+CT AS+CT AS+CT AS+CT AS+CT AS+CT AS+CT AS+CT AS+CT method Detection Size Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Small Large Large Large Large Large Large Large Large class Herbivore Herbivore Omnivore Omnivore Herbivore Herbivore Carnivore Carnivore Carnivore Carnivore Carnivore Carnivore Carnivore Carnivore Omnivore Omnivore Omnivore Omnivore Carnivore Carnivore Carnivore Carnivore Omnivore Carnivore Omnivore Omnivore Omnivore Omnivore Omnivore Omnivore Omnivore Carnivore Omnivore Omnivore Carnivore Herbivore Herbivore Carnivore Omnivore Omnivore Herbivore Herbivore Herbivore Herbivore Herbivore Herbivore Herbivore Herbivore Herbivore Herbivore Herbivore Trophic group E V V V V V V V E V V E V LC LC LC LC LC LC LC LC LC LC LC LC LC LC LC LC LC LC LC LC LC LC LC NL LC LC NT NT NT NT NT NT NT NT NT CE CE DD IUCN Red list status English name sambar red muntjac bearded pig wild boar greater mouse-deer lesser mouse-deer dhole Asian golden cat clouded leopard tiger leopard cat fishing cat marbled cat flat-headed cat short-tailed mongoose collared mongoose sunda stink badger hog badger oriental small-clawed otter hairy-nosed otter smooth-coated otter European otter yellow-throated marten Malayan weasel sun bear binturong Asian palm civet Malayan civet small-toothed palm civet masked palm civet banded palm civet otter civet small Indian civet banded linsang moonrat Malayan tapir Sumatran rhinoceros pangolin crab-eating macaque southern pig-tailed macaque banded sureli white-thighed surili silvery lutung Sumatran sureli agile gibbon siamang lar gibbon Asian elephant Malayan porcupine Sumatran porcupine long-tailed porcupine Binomial Rusa unicolor Muntiacus muntjak Sus barbatus Sus scrofa Tragulus napu Tragulus kanchil Cuon alpinus Catopuma temminckii Neofelis nebulosa Panthera tigris Prionailurus bengalensis Prionailurus viverrinus Pardofelis marmorata Prionailurus planiceps Herpestes brachyurus Herpestes semitorquatus Mydaus javanensis Arctonyx collaris Aonyx cinerea Lutra sumatrana Lutrogale perspicillata Lutra lutra Martes flavigula Mustela nudipes Helarctos malayanus Arctictis binturong Paradoxurus hermaphroditus Viverra tangalunga Arctogalidia trivirgata Paguma larvata Hemigalus derbyanus Cynogale bennettii Viverricula indica Prionodon linsang Echinosorex gymnura Tapirus indicus Dicerorhinus sumatrensis Manis (Paramanis) javanica Macaca fascicularis Macaca nemestrina Presbytis femoralis Presbytis siamensis Trachypithecus cristatus Presbytis melalophos Hylobates agilis Symphalangus syndactylus Hylobates lar Elephas maximus Hystrix (Acanthion) brachyura Hystrix (Thecurus) sumatrae Trichys fasciculata Family Cervidae Cervidae Suidae Suidae Tragulidae Tragulidae Canidae Felidae Felidae Felidae Felidae Felidae Felidae Felidae Herpestidae Herpestidae Mephitidae Mustelidae Mustelidae Mustelidae Mustelidae Mustelidae Mustelidae Mustelidae Ursidae Viverridae Viverridae Viverridae Viverridae Viverridae Viverridae Viverridae Viverridae Viverridae Erinaceidae Tapiridae Rhinocerotidae Manidae Cercopithecidae Cercopithecidae Cercopithecidae Cercopithecidae Cercopithecidae Cercopithecidae Hylobatidae Hylobatidae Hylobatidae Elephantidae Hystricidae Hystricidae Hystricidae Order Total counts: Artiodactyla Artiodactyla Artiodactyla Artiodactyla Artiodactyla Artiodactyla Carnivora Carnivora Carnivora Carnivora Carnivora Carnivora Carnivora Carnivora Carnivora Carnivora Carnivora Carnivora Carnivora Carnivora Carnivora Carnivora Carnivora Carnivora Carnivora Carnivora Carnivora Carnivora Carnivora Carnivora Carnivora Carnivora Carnivora Carnivora Erinaceomorpha Perissodactyla Perissodactyla Pholidota Primates Primates Primates Primates Primates Primates Primates Primates Primates Proboscidea Rodentia Rodentia Rodentia

Figure 2.1 Study site locations with respect to protected areas, agri-industrial land uses and Sumatran provincial borders. (a) Protected areas (dark grey) on Sumatra include IUCN categories Ia (Nature reserve), II (National Park), IV (Wildlife reserve) and VI (Protection forest). Data were from the World Protected Area Database, available at www.unep-wcmc.org. Human altered landscapes (light grey) dominated by estate crops, plantations and logging concessions. Data were from the World Forest Watch, available at www.globalforestwatch.org. The management areas surveyed are indicated in black. Note that the oil palm concession and selective logging concession share boundaries and as such appear as one area. (b) The individual management areas in detail. Black points indicate survey cell centres (n = 131), black triangles correspond to oil palm dominated cells in the oil palm concession (OPC_M). Public roads (i.e. asphalt) and logging roads are also shown.

Figure 2.2 Sample-based rarefaction curves including 95% confidence intervals. Data derived from species detection/non-detection using active search periods and/or camera traps in 131 survey cells. The uppermost grey curves indicate the intermediate landscape alteration class. The lower black curves indicate the highly altered landscape class.

Figure 2.3 Observed ecological group representation across landscape alteration classes, presented as proportions of the regional species pool. Significantly fewer small- bodied mammals (<21kg) were detected in the human altered landscapes overall (p <0.05) and a significantly higher proportion of the ecological groups were detected in the intermediate landscape alteration class than the high alteration class (p <0.05).

Figure 2.4 Representation of observed species IUCN threat status in the regional species pool and the intermediate and high landscape alteration classes. Lower relative IUCN threat status is indicated in grey and is composed of IUCN categories Near Threatened (NT) and Vulnerable (V). Higher relative IUCN threat status is indicated in black and is composed of IUCN categories Endangered (E) and Critically Endangered (CE). Mammal species of greatest conservation concern - Endangered (E) and Critically Endangered (CE) - were not detected in the highly altered landscapes represented by the oil palm concession (OPC_S and OPC_M areas combined).

Figure 2.5 Individual point estimates of species richness with associated 95% confidence intervals from each of four land management areas surveyed in south-central Sumatra. Species richness was consistent across the management areas dominated by degraded forest (wildlife reserves and selective logging concession) but there is evidence of a step- like decline in the plantation concession, particularly the oil palm matrix, in which the lowest observed and estimated species richness was recorded. The model(h) estimate indicates that significantly (p <0.05) fewer species occur in the plantation’s oil palm dominated areas than in the areas of intermediate landscape alteration.

2 Stress 0.22

-2 -1 1 2

-1

-2

Figure 2.6 Pattern of community composition (2D multidimensional scaling ordination plot) across 131 survey cells drawn from five land management areas. Data are a combination of active search periods and camera traps. Survey cells drawn from the oil palm concession are black points (OPC_M cells) and hollow grey circles (OPC_S), with all other survey cells coloured light grey.

2 Stress 0.22

-2 -1 1 2

-2

-1

-2

Figure 2.7 Patterns of community composition (2D multidimensional scaling ordination plot) across 131 survey cells drawn from five land management areas. Data are derived from only active search periods. Survey cells drawn from the oil palm concession are black points (OPC_M cells) and hollow grey circles (OPC_S), with all other survey cells indicated by light great points.

Chapter 3

Ecological Traits and Mammalian Persistence in Human Altered Landscapes

3 Ecological Traits and Mammalian Persistence in Human Altered Landscapes

3.1 Abstract

Intrinsic ecological traits affect species persistence in the face of environmental change. Identifying the mechanics of this process will help us understand species responses to the human-altered landscapes that are an inevitable feature of the 21st century. Results suggest that in the initial stages of landscape change there is the capacity to support large specialist species, with slow life histories. However, as landscape degradation continues to an agricultural matrix, only habitat and diet generalists persist. Thus, there exists a small window of opportunity for the conservation of the larger, typically vulnerable species, as further habitat degradation favours only a few generalists.

3.2 Introduction

With anthropogenic landscape change recognised as the greatest threat to biodiversity (Pimm et al., 1995; Chapin III et al., 2000) there is a great need to understand how biological traits influence species’ responses to changing environments and ultimately their risk of extinction (Bennett & Owens, 1997; Purvis et al., 2000a; Jones et al., 2003; Cardillo et al., 2004). Collectively, intrinsic ecological traits contribute to a species’ niche, which essentially defines the conditions under which a species exists and how it interacts with its environment (Grinnell, 1917; Hutchinson, 1957; Chase & Leibold, 2003). In a changing landscape, a broader niche would be expected to confer a greater advantage to species survival and this idea is supported in the literature where ecological specialisation has been found to influence extinction risk in both mammals (Harcourt et al., 2002; Boyles & Storm, 2007) and birds (Julliard et al., 2003; Shultz et al., 2005).

While the overall result of anthropogenic landscape change appears to be a decline in biodiversity (e.g. Fitzherbert et al., 2008), there are stark differences between individual species’ responses (Daily, 2001; Davies et al., 2000; Owens & Bennett, 2000). The groups of species that occur in modified landscapes are typically subsets of an ancestral species pool, altered in composition and structure by the loss of species that were unable to persist in a changed landscape (Duchamp & Swihart, 2008). In the most heavily altered landscapes, communities become dominated by a few extremely adaptable species (Wood & Chung, 2003).

Key results from recent work indicate that species extinction risk is driven by a combination of both intrinsic (species biology) and extrinsic (environmental) factors (Cardillo et al., 2005). Overall, the main intrinsic traits linked to extinction are body size, habitat and/or dietary specialisation (Owens & Bennett, 2000; Purvis et al., 2000b). In general, larger body size is associated with a slow life history strategy; a mechanism that is thought to have evolved under more stable environments to produce longer lived animals with reduced reproductive output. In contrast, smaller species typically utilise a fast strategy that persists in less stable, changing environments and is associated with higher reproductive outputs (Stearns, 1983). For larger species with large space requirements, small habitat patches may not be sufficient. However, body size also affects the ability to move across the landscape (Hanski, 1998; Etienne & Heesterbeek, 2001) and this increased mobility may enable larger species to more effectively utilize remnant habitat patches. By comparing species’ traits within communities that occupy different components of a landscape, it would be possible to identify the relative influence of these different factors on patterns of species persistence and local extinction.

Profound landscape change has occurred across much of south-central Sumatra in recent years. In this study I investigate patterns of mammalian persistence in this region to explore how species’ traits influence their response to different levels of habitat degradation. This approach informs our understanding of how mammalian biodiversity is affected by the continued intensification of landscape change. Here, I compare the importance of body size and a number of reproductive, dietary and behavioural traits on observed patterns of species persistence. First, I consider the traits associated with persistence in a less severe form of landscape change by comparing a theoretical species pool to an observed species inventory from degraded forests. Second, I identify the traits associated with species that are known to occur within the most severely altered areas of the study landscape – an oil palm (Elaeis guineensis) matrix.

3.3 Methods

3.3.1 Study sites

Between March and September 2006 I sampled four land management areas in south- central Sumatra (latitude 1o 53´ to 2o 35´S, longitude 103o 2´ to 104o 9´E). The individual study sites encompassed a gradient of landscape alteration. Management areas dominated by degraded forest and extensive natural vegetation were structurally more complex and subject to less human disturbance than areas of oil palm crop or the adjacent scrub habitats that were colonised by illegal land settlers. This coarse description of landcover complexity was used to group study sites as either intermediate landscape alteration or high landscape alteration. The individual land management areas assigned to the intermediate landscape alteration class were (i) a selective logging concession (SLC, 800-km2, 45 cells), subject to industrial legal logging and small-scale illegal logging; (ii) Dangku wildlife reserve (DWR, 250-km2, 28 cells), a low-level protected area (IUCN category IV) extensively populated by rural settlers with associated illegal logging and land cover conversion and (iii) Bentayan wildlife reserve (BWR, 300-km2, 30 cells), a second low-level protected area (IUCN category IV) also subject to extensive human settlement and associated land cover conversion. The alternative high landscape alteration class contained two components of an oil palm concession (OPC, 270-km2), (i) an oil palm monoculture (OPC_M, 14 cells) and (ii) an adjacent matrix of scrub habitats (OPC_S, 14 cells) that have been extensively colonised by pioneer land-settlers.

3.3.2 Field methods

I sampled 131 2 x 2-km cells drawn from the aforementioned land management areas. A combination of active search periods (trained field personnel) and camera traps were used to detect nonvolant mammals. Data collection methods were consistent across sites, with survey cells drawn from simple random samples of the overall management area (OPC, DWR), sub-sections thereof (SLC) or a uniform grid initiated from a randomised start location (BWR). This approach to cell selection balanced the logistical constraints of cell accessibility, geographic coverage of the overall management area and the desire to generalise findings to the wider landscape. Overall, survey cells covered 29% of the total management area (1620-km2) encompassed by the management boundaries.

Five survey teams, each of two people, were established in March 2006. Teams were led by individuals of equivalent training and field experience. Each team operated independently within the survey cells and visited each cell only once. Four teams were active on a given day with teams rotating between proximate cells once three hours of active search effort had been completed. Teams travelled to cells by motorbike and for logistical reasons cells were surveyed in groups, with clusters of four neighbouring cells surveyed on 3-4 sampling occasions (85% of cells = 4 sampling occasions) over a two- day period. This approach provided repeated independent samples of each sampling cell.

During each three-hour sampling occasion teams aimed to travel widely throughout the cell (mean average two-dimensional distance was 4.8km, derived from GPS odometers on sampling occasions with ~3hrs GPS coverage, n = 202) searching a representative sample of the available habitats and in turn maximising the probability of encountering mammal signs. Mammal signs were either obtained directly through sightings or audio cues, or indirectly predominantly through recording footprints but also other indicators such as faeces where these could be assigned to a target species with confidence. Team leaders geo-referenced all direct and indirect signs using Garmin 60c global positioning systems (GPS) (Garmin International Inc., Olathe, KS) in universal transverse mercator (UTM) coordinates. Each team was equipped with footprint identification guides and followed rigorous species-sign identification protocols to minimise the risk of wrongly identifying a species, leading to a false positive record of species presence.

In addition to active search periods, Deer cam DC300 (Non-typical inc., USA) 35mm film camera traps were placed at the intersection of wildlife trails within survey cells to provide an additional form of sampling with which to establish species detection/non-detection. Cameras were placed in 85% (n = 111) of survey cells for a mean average of 18 trap nights per cell, providing 1991 camera-trap nights in total. Cameras were placed at a height of 30 – 45cm depending on the terrain and vegetation structure in the immediate vicinity of the trap site. The camera placement protocol was designed to maximise the probability of detecting mammals without introducing systematic bias toward a particular species.

3.3.3 Analyses

This study sought to identify the ecological traits associated with species persistence in the forest-adapted mammals of south-central Sumatra. First, I define which of the species in the ancestral regional pool persist into the present landscape. Second, I define where these persistent species occur in the present day landscape. Throughout this study I have defined persistent species as those detected and non-persistent species as those that were not found. This approach assumes that species non-detection is equivalent to true absence, see below for further details.

These analyses rely on three species inventories. First, a regional pool of species expected to occur in a pristine, lowland Sumatran forest landscape (the dominant landcover in this region prior to extensive human alteration) was compiled from the literature (n = 51 species). Second, the total observed species richness from field surveys across all four of the land management areas (n = 27 species). Third, the total observed species diversity from the survey cells within the oil palm dominated matrix (n = 14 species). Using these species lists I identified the ecological traits associated with species persistence under two scenarios: (i) persistence from the ancestral species pool to the present human altered landscape overall, and (ii) persistence within the most highly altered components of the present landscape – the oil palm dominated matrix.

Species detected at least once by either active search period or camera trap were considered persistent in a given context; the remainder were classified as non-persistent. It is likely, however, that a number of species were present in the human altered landscape overall but remained undetected by the sampling methods described here (see Chapter 2). Consequently, my interpretation of detection/non-detection equating directly to species persistence/non-persistence must be interpreted in the knowledge that the record may contain false absences. This issue is more relevant in analyses based on the regional species pool as several of these species would have been difficult to detect, requiring camera trap pictures or direct sightings (see Discussion).

The Mammalian Community I defined a regional pool of mammal species that were (i) detectable using the sampling methods described here and (ii) likely to occur in the study region based on species distributions published in the literature (e.g. Payne & Francis, 1985; Nowak, 1999; Macdonald, 2006). Of this regional pool, 49% of species were listed as near threatened or threatened according to IUCN Red List criteria (specific categories: near threatened, vulnerable, endangered and critically endangered).

To minimise the risk of excluding species which actually persist in the region the reference community was restricted to nonvolant species that could have been detected by the sampling methods employed, defined as species ≥ 0.65kg adult mass, ≥ 40cm head-body length and either partially terrestrial or if strictly arboreal primates, group living and vocal. The lower body mass limit was based on the smallest species that could be Persistent (11 sp.) reliably identified from its footprints, the Malayan weasel (Mustela nudipes). The smallest Class probability: 0.91 species detected by camera trap during field surveys was the common tree shrew sambar red muntjac (Tupaia glis), weighing 90 – 190 grams, head-body length 135-205mm. These criteria bearded pig wild boar dhole exclude members of the Scandentia and Rodentia as detections of such small species clouded leopard tiger are very rare events, essentially occurring by chance. Species meeting the sun bear Malayan tapir aforementioned criteria were considered sufficiently detectable to be detected at least Asian elephant Sumatran rhino once during active search periods and/or camera trap sampling, although I acknowledge that for many of these species, detection probabilities would be very low. The final regional species pool contained 51 species drawn from eight taxonomic orders and 18 families (Chapter 2, Table 2.1). In total, 27 species from the regional pool were detected during field surveys across the overall sampled area (i.e. both wildlife reserves the logging concession and the oil palm concession).

Explanatory Variables For 51 species of nonvolant mammal assigned to the regional species pool I collated data on nine candidate traits; each chosen because of clear potential to influence species’ capacity to persist in human altered landscapes (Table 5.1). Continuous explanatory variables were collated from the Pantheria dataset (Jones et al., 2009), specifically these were, adult body mass, age at sexual maturity, inter-birth interval, litter size and diet breadth. Behavioural traits were categorical and were sourced from the literature e.g. Nowak (1999), Macdonald (2006) and Francis (2008), and included, activity period (three levels; diurnal, nocturnal, day-and-night active), terrestriality (three levels; terrestrial, arboreal or tree-climbing) and sociality (two levels, solitary or forms transient pairs, group living).

Classification Tree Models I used classification tree models (Breiman et al., 1984; Clark & Pregibon, 1992; Ripley, 1996; De'ath & Fabricius, 2000) to assign mammal species to a binary response variable: ‘persistent’ or ‘non-persistent’ according to whether a species was detected at least once during active search periods or by the camera traps during the field surveys. Classifications were based on a suite of ecological traits expected a priori to influence species’ ability to persist in human-altered landscapes. All data analyses were performed using the statistical software package R (R Development Core Team, 2008) and the library of tree model routines implemented in the package, rpart (Recursive PARTitioning) (Therneau & Atkinson, 2008).

Models were fitted using binary recursive partitioning, whereby the data are repeatedly split along coordinate axes of the explanatory variables. Splits are allocated where the distinction between the response variable is maximised at each node. Each explanatory variable is assessed in turn (De'ath & Fabricius, 2000). From all possible splits of all the explanatory variables, the one that maximised the homogeneity of the two resulting groups was selected. I used the Gini impurity criterion to determine the optimal splitting criterion, with splits attempted wherever three or more species were present and allowing terminal nodes to contain as few as one species. Optimal tree sizes were obtained from 50 10-fold cross-validations. The final trees (Figures 3.2 and 3.3) correspond to the modal tree sizes using the 1-SE rule, as advocated by Breiman et al. (1984).

The species-trait dataset contained missing values because many of the species in the regional pool are poorly studied. This issue was resolved by using surrogate variables in the classification tree models (Therneau & Atkinson, 2008). In this approach, once a splitting variable and specific split point were identified from the dataset overall, e.g. inter- birth intervals of ‘< 12 months’ and ‘≥ 12 months’, the partitioning algorithm was re- applied to predict these categories using the other explanatory variables. Numerous surrogate variables are identified and then ranked against one another based on misclassification error rates. Finally, where missing values exist, the surrogate variables are used in rank-order to split these species. In the analyses described here surrogate variables were used on several occasions and have been added to Figures 5.4 and 5.5 accordingly. In essence, the use of surrogate variables simply means that a specific split is the result of two explanatory variables working in combination rather than just one. An alternative solution to missing trait values would have been to estimate the missing trait values based on similar species. However, in many cases alternative but similar species were also poorly studied and so this approach would have been problematic to execute and ultimately difficult to defend.

3.4 Results

Ancestral regional pool vs. the human altered landscape inventory In a comparison between the species in the ancestral pool and those detected in the present landscape, the persistent species were generally larger (adult body mass > 14kg). Of the smaller species, those that produced smaller litters (< 2 offspring per litter) were more likely to persist. Eight species lacked trait values for the variable litter size and so a surrogate variable, trophic group, split these species. The persistent species contained a higher proportion of omnivores and the non-persistent species tended to be herbivores or carnivores. Therefore, in the transition from the ancestral regional pool to the present landscape, there was a tendency for larger bodied species and those that produced smaller litters or had broader diets to persist.

Human altered landscape inventory vs. highly altered areas The distinction between the species that persisted in the wider human altered landscape and those that also occurred within the highly altered, oil palm areas, was largely attributed to body mass, reproductive and dietary traits. Species that were able to use the oil palm dominated areas were smaller with broad diets and higher reproductive outputs. The only larger mammals using these areas were omnivorous e.g. wild boar (Sus scrofa).

Although, inter-birth interval was the primary splitting variable, body mass was a strong surrogate variable and was used to split 12 species that lacked trait values for the primary variable. In combination these variables split the larger bodied species with the longest inter-birth intervals from the remainder and assigned them to a non-persistent group. Trophic group provided the second highest splitting criterion, with all omnivores classed as persistent. The remaining species (the herbivores and carnivores) were again split by a dietary trait, diet breadth, in which the majority of persistent species (n = 9) had relatively broader diets (>1.5 dietary groups). These nine species were split by two reproductive traits: Age at Sexual Maturity and Litter Size. The penultimate split, Age at Sexual maturity, classified one species as ‘persistent’. The final eight species were subsequently classified based on litter size; five species with small litters of <1.5 offspring were classified as non-persistent and three species with relatively larger litters were classified as persistent. Overall, in these latter stages of the tree, the persistent species were characterised by broader diets and produced slightly larger litters.

3.5 Discussion

This study sought to identify the ecological traits associated with species persistence in the face of anthropogenic landscape change. The traits chosen could confer an advantage within a single generation (e.g. broader diets allowing species to find food in changing landscapes) or between generations (e.g. high reproductive output and a generally fast life history strategy). First, I defined which of the species from the ancestral regional pool persisted into the current landscape. Second, I identified the circumstances under which these persistent species occur in the modern landscape. In this study I defined persistent species as those detected and non-persistent species as those that remained undetected. This approach requires that I treat non-detection as a true absence.

From a suite of ecological traits expected to influence species’ capacity to persist in the face of landscape change, I identified that a combination of body size, reproductive rate and dietary specialisation were best able to distinguish between species that persisted and those that did not. I found a pattern of species persistence that was indicative of the ideas embodied by the fast/slow life history strategies (sensu Stearns, 1983). Broadly speaking, in the initial stages of landscape alteration the larger bodied slow strategists, with associated low reproductive output, were found to persist in the less disturbed areas in the landscape matrix. In areas with more severe levels of landscape change, the balance of persistence shifted in favour of fast strategists. In addition, I also found evidence of diet specialisation being an important factor; omnivores and species with broad diet breadth were more likely to persist. These analyses also indirectly measured habitat specialisation. The species that persist in the most altered landcover (oil palm) also occur in all other habitats and therefore represent habitat generalists.

Research has shown that the main ecological traits linked to extinction risk are body size, habitat and/or dietary specialisation (Owens & Bennett, 2000; Purvis et al., 2000b). This body of research is directly relevant to the work described here as it also tries to understand how ecological traits influence persistence. Much of this research has focussed on extinction because ultimately conservation science seeks to understand drivers of decline in order to move from a reactive to a predictive capacity able to mitigate these declines (Mattila et al., 2008; Terborgh et al., 2001).

Since body mass is correlated with many other ecological characteristics including mobility and reproductive strategy (Laurance, 1991) it is not surprising that this trait was an important explanatory variable. Larger, slow strategists are more likely to survive in

mature ecosystems (pristine and degraded forest) in which they put more resources into rearing fewer young. Also, animal ranging patterns increase with increasing body size (Jetz et al., 2004) and this may allow larger species to move between remnant natural habitats distributed throughout the wider matrix (Hanski, 1998; Etienne & Heesterbeek, 2001). However, the lower reproductive outputs associated with larger species have in turn been linked to an increased risk of extinction due to a reduced capacity to respond to population declines (Owens & Bennett, 2000; Purvis et al., 2000b). In contrast, smaller ‘fast’ species are better adapted to surviving in habitats subject to disturbance and tend to produce more offspring to compensate for their lower survivorship under these conditions.

The influence of diet on extinction risk has been studied at a trophic level (Purvis et al., 2000b) and in terms of niche breadth (Safi & Kerth, 2004). In the system described here, persistent species tended to be omnivores or species with broader diets. It is likely that species with narrow dietary requirements would be more sensitive to loss of food or the destruction of habitats from which their food originates (Laurance, 1991).

Habitat specialisation has been shown to influence species’ extinction probability (Owens & Bennett, 2000; Purvis et al., 2000b). Although my analyses did not include a direct measure of habitat specialisation, I still address this issue through patterns of species presence/absence in different habitats. The species that persist in the most altered areas can be considered the least specialised as they occur in all habitats from pristine forest (regional species pool) down to oil palm dominated areas.

With the exception of the suids (Sus scrofa, Sus barbatus) and cervids (Rusa unicolor, Muntiacus muntjak), larger mammals were allied to the areas of degraded forest found in the logging concession and wildlife reserves. Management areas dominated by this landcover are subject to considerable pressure from rural human populations (see Chapter 4) and are typically in a state of progressive degradation. As such, these larger species, e.g. Sumatran tiger (Panthera tigris sumatrae) and Malayan tapir (Tapirus indicus), are of high conservation concern as they persist in this landscape only by virtue of these extremely valuable degraded forest habitats.

Tree models have provided a useful approach in the analysis of a relatively small dataset with missing trait values for rare and poorly known species e.g. (Peh et al., 2004; Olden et al., 2008). The critical reader may be concerned that my approach does not sufficiently address the implications of species missing from the inventories. In my approach, a species that occurred in the landscape but evaded detection during sampling would be misclassified as non-persistent. If these misclassifications were sufficient to alter the representation of ecological traits, this could potentially influence my conclusions

regarding mammalian persistence in this context. The species most likely to have been missed are reclusive, e.g. golden cat (Catopuma temminckii), or have highly specialised habitat requirements e.g. otter species. A small proportion of these species could be reclassified as persistent to test if this changes the conclusions from the tree models. Adding the short-tailed mongoose, a species that I did not detect but which occurs in the landscape (Maddox et al., 2007), did not affect model tree size or the splitting criterion. Also, removing two species, marbled cat (Pardofelis marmorata) and the banded surili (Presbytis femoralis) that were detected anecdotally did not affect the results from the tree models. Provided the species inventories are largely complete and representative of the community I sought to describe, the important traits will have been detected and my conclusions as to their relative importance will be robust.

Results from this study suggest that in the initial stages of landscape change enough suitable habitats remained to support large specialist species. However, it appears that the transition to an agricultural matrix could only support the more robust species best described as small bodied generalists, e.g. Malayan civet (Viverra tangalunga). The few large bodied species that persisted were omnivores (e.g. wild boar). It is interesting to note that extrinsic processes, such as the off-take of larger mammals by hunting (Escamilla et al., 2000; Cardillo & Bromham, 2001; Daily et al., 2003) could also affect perceptions of species persistence. In general, hunting has played an important role in population declines in Asian forests (Corlett, 2007). In this landscape, tigers are thought to have been persecuted during the colonisation of marginal lands on the plantation concession, ultimately leading to their extirpation from this area. (Maddox et al., 2003; Maddox et al., 2007).

In summary, this study suggests that the larger mammals of conservation concern (slow strategy species) persist only in the early stages of landscape change. There exists a small window of opportunity for the conservation of these species as further degradation favours the fast strategy species. In addition, even if conservation interventions were swift and effective, there is a high probability that these populations carry an extinction debt (Tilman et al., 2004). This time lag between habitat loss and extinction could undermine the perceived potential for conservation in human-dominated landscapes.

3.6 Tables and Figures

Table 3.1 Description of explanatory variables used to describe species resilience to landscape alteration.

Regional Pool : Human Altered Landscape : Human Altered Landscape Oil Palm Matrix

Explanatory Class Type Range (mean) Type Range (mean) variable (No. data (No. data points) points)

Adult body Mass Continuous 0.6-3269.8 (111.9) Continuous 0.8 - 3269.8 (160.3) mass (kg) (49) (27)

Age at sexual Reproductive Continuous 5.1-132.0 (40.7) Continuous 5.1 - 132.0 (35.8) maturity (24) (18) (months)

Inter-birth Reproductive Continuous 5.0-42.0 (15.0) Continuous 5.0 - 39.0 (14.3) interval (22) (15) (months)

Litter size Reproductive Continuous 1.0-6.6 (2.1) Continuous 1.0 - 6.6 (2.0) (43) (25)

Diet breadth Dietary Continuous 1.0-6.0 (3.3) Continuous 1.0 - 6.0 (3.0) (45) (25)

Trophic group Dietary Categorical carnivore, Categorical carnivore, (51) herbivore, omnivore (51) herbivore, omnivore

Activity cycle Behavioural Categorical diurnal, nocturnal, day Categorical diurnal, nocturnal, day (51) & night active (51) and night active

Group size Behavioural Categorical ≤ 2 individuals, >2 Categorical ≤ 2 individuals, >2 (51) individuals (51) individuals

Terrestriality Behavioural Categorical terrestrial, arboreal, Categorical terrestrial, arboreal, (51) terrestrial & arboreal (51) terrestrial and arboreal

Figure 3.1 Study site locations with respect to principal protected areas, agri-industrial land uses and Sumatran provincial borders. (a) Protected areas on Sumatra include IUCN categories Ia (Nature reserve), II (National Park), IV (Wildlife reserve) and VI (Protection forest); indicated in dark grey (data from the World Protected Area Database, available at www.unep-wcmc.org). Human altered landscapes dominated by estate crops, plantations and logging concessions are coloured light grey (data from World Forest Watch, available a www.globalforestwatch.org). The management areas surveyed are indicated in black. Note that the oil palm concession and selective logging concession share boundaries and as such appear as one area. (b) The individual management areas in detail, black points indicate survey cell centres (n = 131). Public roads (i.e. asphalt) and logging roads are also shown.

(11sp.)

ambar

Persistent 0.91 Class probability: s redmuntjac pig bearded wild boar dhole clouded leopard tiger sunbear tapir Malayan elephant Asian Sumatranrhino

14 (kg) 14 >

hedsurili

thig - tailedporcupine - long Sumatranporcupine Malayan porcupine gibbon lar siamang gibbon agile Sumatransurili silverylutung white surili banded

(kg) Mass

clawedotter

- tailed macaque tailed - (20sp.)

eating macaque eating - sser(Javan) mouse deer Persistent 0.65 Class probability: greater(Malay) mouse deer le golden Asian cat small Oriental civet Malayan banded palmcivet moonrat pangolin crab southernpig offspring omnivore <2

<14 (kg) <14

et Trophic group Littersize

(surrogate variable)

throatedmarten - toothed palmcivet -

bandedlinsang ottercivet Asian palmciv yellow

small Indiancivet masked palm civet small binturong Malayan weasel European otter

offspring

carnivore herbivore >2

(20sp.)

coatedotter - tailed mongoosetailed nosedotter - - headedcat - otpersistent

N 0.8 Class probability: cat leopard cat fishing cat marbled flat badger hog sundastink badger short collared mongoose hairy smooth

Figure 3.2 Classification tree analysis for the 51 species of mammal assigned to the regional species pool. Light grey species names indicate classification errors. The response variable was binary, with species detected at least once during field surveys of the human altered landscape overall (four management areas) assigned to the response class ‘persistent’, otherwise ‘non-persistent’.

sp.) tailed macaque tailed - (7

throatedmarten - eating macaque eating - ersistent

P 1.00 Class probability: wild boar sunbear yellow civet Malayan palm civet Asian crab southernpig

Omnivore sp.) (1

>8.2

Persistent Class probability: 1.00 Malayan porcupine

clawedotter - (3sp.) (months)

7.5

<1 Trophic group SexualMaturity >1.2 >1.5 Persistent 1.00 Class probability: cat leopard small Oriental moonrat

< 8.2 Herbivore

Littersize 0 sp.) (5 Carnivore,

Diet Diet breadth <1.2 er(Malay) mouse deer 0

s(kg) Notpersistent 0.8 Class probability: redmuntjac great lesser(Javan) mouse deer surili banded gibbon agile (months) sp.) birthinterval - (1 Mas

Inter tailedporcupine (surrogate variable) - <2.8

ersistent P 0.8 Class probability: long

<1.5 (kg)

Mass

0 sp.) .8

(4 ≥2

3 sp.) (6

Notpersistent 0.8 Class probability: pangolin dhole clouded leopard cat marbled

17.5

≥

iger Notpersistent 0.8 Class probability: tapir Malayan siamang elephant Asian t pig bearded sambar

Figure 3.3 Classification tree analysis for 27 mammal species detected in the overall human altered landscape (i.e. all four and management areas). Light grey species names indicate classification errors. These 27 species represent a local species pool from which the 14 species detected in the most profoundly altered landscapes were drawn. Species that were detected at least once in areas dominated by scrub/oil palm matrix were listed as ‘persistent’ while the remaining species were listed as ‘not persistent’ in the binary response variable.

Chapter 4

Human Agents of Landscape Change

4 Human Agents of Landscape Change

4.1 Abstract

Pervasive human populations can have significant effects on the integrity of rural landscapes. With increasing emphasis placed on the role of landscape-scale conservation strategies there is an urgent need to understand the extent of human influence in remnant wildlife habitats and the effects on the species they support. In this study, I have identified the types and relative intensities of human activity that occur across the south-central Sumatran landscape, particularly within the commercial forestry and small protected areas that support larger, threatened mammal species such as the Sumatran tiger (Panthera tigris sumatrae) and Malayan tapir (Tapirus indicus). Using a novel application of the ecological state variable proportion area occupied, I estimate that ~80% of these individual management areas are subject to illegal logging and land tenure activities. In tests between species occurrence and measures of human disturbance, significant correlations were identified but species’ responses differed in keeping with their ecology. The continued degradation of forest habitats in this region threatens to undermine the conservation potential of both the protected areas and the human-altered landscape at large.

4.2 Introduction

Human impacts on tropical landscapes are widely summarized as rates of deforestation and landcover conversion. These profound changes, now largely at the hands of commercial enterprise, are readily monitored by satellites but these sensors rarely capture the more subtle effects of landcover degradation (Nepstad et al., 1999; Achard et al., 2002; Butler & Laurance, 2008). With respect to forests specifically, degradation has been defined as a process of ‘temporary or permanent deterioration in the density or structure of vegetation cover or its species composition’ (Grainger, 1993) and in many cases is directly associated with human disturbance (Lambin, 1999). Degradation can have profound effects on the integrity of native landcovers and associated biodiversity, and as such is an important consideration for conservation strategy (Phillips, 1997; Sodhi et al., 2009).

Current thinking suggests that the future of tropical biodiversity conservation lies in the management of human–altered areas that provide support for otherwise isolated protected areas (Daily, 2001; Lindenmayer & Franklin, 2002; Gardner et al., 2009). This philosophy requires that human dominated landscapes provide requisite wildlife habitats amidst a largely tolerant human population. Research suggests that many components of biodiversity in fact persist in these landscapes and could be preserved under this 64

approach (Daily et al., 2003; Barlow et al., 2007). However, this does require that these resources are viable in the long-term and not therefore subject to significant deterioration. Given that human-dominated landscapes are generally complex matrices of different land uses and subject to ongoing development, these areas are highly susceptible to continuing degradation (Daily, 2001). In addition, since the efficacy of protected areas is inexorably linked to the matrix in which they are embedded (Ranganathan et al., 2008; Wittemyer et al., 2008), degradation at park margins and human colonization of the interior could seriously undermine these areas.

While landcover degradation has clear implications for conservation strategy, there is also the complex issue of variation among taxa in response to these changes (Peh et al., 2004; Barlow et al., 2007). Animals respond differently to gradients of habitat degradation and in some instances low levels of degradation may in fact benefit certain species, for example increased availability of browse in secondary forests can support higher densities of some forest ungulates (Fragoso, 1991). However, understanding the range of these responses necessitates concurrent data on species occurrence and human activities that drive degradation.

In this study, I used detection/non-detection surveys and occupancy modeling techniques (MacKenzie et al., 2002; Tyre et al., 2003) to identify human drivers of landcover degradation and subsequent effects on species occurrence. The study system included large areas of plantation agriculture, industrial forestry and legally designated wildlife reserves (IUCN category IV protected areas). Having identified the types and prevalence of human activities across the landscape, I tested for species-specific responses to a gradient of human disturbance using occurrence data for Sumatran tigers (Panthera tigris sumatrae) and their principal ungulate prey base: wild boar (Sus scrofa), sambar (Cervus unicolor), red muntjac (Muntiacus muntjac) and tapir (Tapirus indicus).

4.3 Methods

4.3.1 Study Sites

Between March and September 2006 I conducted repeated detection/non-detection surveys across a network of 131 2 x 2-km sampling cells in south-central Sumatra (Figure 4.1). Clusters of cells were drawn from four land management areas (latitude 1o 53´ to 2o 35´S, longitude 103o 2´ to 104o 9´E), each subject to a varying amount of anthropogenic influence. Individual land management areas were: (i) Bentayan wildlife reserve (BWR, 300-km2, 30 cells), a low-level protected area (IUCN category IV) subject to extensive human settlement and associated land cover conversion; (ii) a selective logging concession (SLC, 800-km2, 45 cells), subject to industrial legal logging and small-scale illegal logging; (iii) an oil palm concession (OPC, 270-km2, 28 cells), where an oil palm monoculture was managed amidst an adjacent matrix of scrub habitats that have been extensively colonised by pioneer land-settlers, and (iv) Dangku wildlife reserve (DWR, 250-km2, 28 cells), a second low-level protected area (IUCN category IV) extensively populated by rural settlers with associated illegal logging and land cover conversion.

4.3.2 Field methods

Detection/non-detection field surveys were conducted to record signs of recent mammal presence (e.g. footprints) and concurrent human activities. Base camps were maintained within each management area for the ~25 days required to survey each site. From these camps, field teams systematically surveyed clusters of adjacent cells throughout the local cell network. Within each land management area, survey cells were drawn from a random sample from the overall area (OPC, DWR), sub-sections thereof (SLC) or a uniform grid initiated from a randomised start location (BWR). Minimum inter-cell distances varied among management areas (mean average distance between cell centres = 2.8-km, range 2 - 4-km). This approach to cell selection balanced the logistical constraints of cell accessibility, geographic coverage of the overall management area and the desire to generalise findings to the wider landscape. Overall, survey cells covered 29% of the total area (1620-km2) encompassed by the management areas.

Five survey teams, each of two people, were established in March 2006. Teams were led by individuals of equal field experience and training. Each team operated independently within the survey cells and visited each cell only once. Four teams were active on any one day with teams rotating between proximate cells once three hours of active search effort had been completed. Teams travelled to and between cells by motorbike. Groups of

four neighbouring cells were surveyed on 3-4 sampling occasions (85% of cells = 4 sampling occasions) over a two-day period; providing repeated independent samples.

During each three-hour sampling occasion teams aimed to travel widely throughout the cell (mean average two-dimensional distance was 4.8km, derived from GPS odometers on sampling occasions with ~3hrs GPS coverage, n = 202) searching a representative sample of the available habitats and in turn maximising the probability of encountering mammal signs. Direct mammal signs were sightings or audio cues and indirect signs were predominantly footprints but included other indicators such as faeces where these could be assigned to a target species with confidence. Indicators of human activity, such as motorcycle tracks, fresh-sawn timber etc were recorded and allocated to one of six predefined human activity categories. Team leaders geo-referenced all direct and indirect signs using Garmin 60c global positioning systems (GPS) (Garmin International Inc., Olathe, KS) in universal transverse mercator (UTM) coordinates. Each team was equipped with footprint identification guides and followed rigorous species identification protocols to minimise the risk of false positive species’ detections through misidentification of direct or in-direct signs.

4.3.3 Analyses

In this study, I sought to identify the prevailing human activities across different land management areas and in turn test for correlations between measures of relative human disturbance and the occurrence of tigers and their prey species.

Detection/non-detection data Field data were collected under an occupancy framework (MacKenzie et al., 2002) to estimate (i) the proportion of the sampled areas in which different human activities occurred (PAO) and (ii) the probabilities of species occurrence. Under this approach detection probabilities are derived from repeated sampling of survey cells. These are then incorporated into logistic regression analyses to provide unbiased estimates of probabilities of species occurrence.

The parameter of interest measured under occupancy theory is the probability of occurrence in a sampling unit, ψ. Estimates of this probability are typically used in two ways. Firstly, this probability describes the fraction of the sampled area in which a species occurs. This is the ecological state variable, proportion area occupied (PAO). I used this technique to estimate the prevalence of different human activities within the sampled area. Secondly, the parameter ψ can be used to investigate patterns of species occurrence.

Detection/non-detection methods were intended to distinguish between two states of the sampling units – species present (occupied) and species absent (unoccupied) while accounting for detection errors (e.g., when the sampling units are smaller than the home range of a species). Where true and pseudo absences cannot be differentiated, estimates of occupancy (ψ) are more aptly interpreted as intensity of use. Since space use varies considerably with animal body size (Jetz et al., 2004) and life history, but our sampling scale remained constant, I interpret all results that relate to species occurrence in terms of the intensity of use.

Detection histories To estimate probabilities of occurrence, detection histories were compiled for each cell and each sampling occasion in which a ‘1’ denoted detection and a ‘0’ denoted non- detection of a given species or human activity. Cell specific detection histories were compiled for each species of mammal and for each of the six human activities. For simplicity, the following example refers only to the compilation and analysis of detection histories for individual human activities, but the process used for the five mammal species was identical.

I assigned each record of human activity to one category, allowing 2 - 9 specific indicators per category (Table 4.1). Records of at least one direct or indirect sign in a given sampling occasion conferred detection of a given activity category. This is analogous to allowing multiple signs of species presence to confer detection in more traditional applications of this approach (e.g. Linkie et al., 2006). If this approach were adopted as part of a landscape monitoring programme, reducing the number of indicators assigned to a given category could make this technique more sensitive to changes in the prevalence of specific activities.

Detection histories (h) were used to estimate detection probabilities specific to each human activity. For example, a detection history for site i (hi) of 1100 would represent detections of a particular activity on the first and second sampling occasions only and the probability of acquiring history hi would be calculated as,

Pr(hi = 1100) = ψp1p2(1-p3)(1-p4) where pj is the probability of detecting the activity during sampling period j (=1,…,4) assuming that evidence of that activity was present. Detection histories for each category were analysed using logistic regression implemented in program MARK version 5.1 (White & Burnham, 1999) to obtain unbiased estimates of the PAO by each human activity category. I used the constant model ψ(.)p(.) (MacKenzie et al., 2002) to provide a

measure of the relative prevalence of each human activity across the landscape. The constant model assumes that detection probabilities (denoted p) and occurrence probabilities (denoted ψ) are constant across the sampling units i.e. survey cells.

Species occurrence and covariates of human influence Detection histories for tigers and the prey species were compiled in the same way as described for the human activity categories. To test for the effects of human activities on the occurrence probabilities of these species, covariates were produced to summarise the relative intensity of human disturbances.

Counts in each activity category were converted to a mean average per cell per sampling occasion. This standardised count provided the covariate “total”. This provided a broad description of overall human activity prevalence in each cell. I acknowledge that all of the variation in these counts cannot be attributed to underlying abundances without testing for a monotonic relationship between the two (e.g. Williams et al., 2002). However, in this context these counts should be sufficient to indicate relative abundance of human activity within mammal habitats.

Next, these count data were reduced to a set of orthogonal standardized linear combinations using principal component analysis, specifically the function prcomp implemented in the statistical software package R (R Development Core Team, 2008). Input variable loadings were used to assess the relative influence (i.e. direction and magnitude) of each activity category on the principal components; providing biologically meaningful gradient of relative human influence across the study landscape.

Finally, I used the single-season, single-species, occupancy model developed by MacKenzie et al. (2002) to estimate occupancy probabilities and test the additive and multiplicative effects of cell-specific covariates (i.e., ‘total’ and two principal components). The effect of these covariates on species occurrence probability was tested by setting the detection probability constant p(.) and modelling the occurrence probability as a linear function of these covariates e.g., ψ(total) or ψ(PC1*PC2). Comparisons between candidate models were based on the difference in Akaike information criterion values adjusted for small sample sizes (∆AICc), and their Akaike weights (wi) (Burnham & Anderson, 2002). In an effort to be conservative, the final model was chosen as the model with the least number of parameters within two AIC units (∆AICc) of the top ranked model.

4.4 Results

In this study, I sought to identify the types and relative intensities of human activity that occur in a modern tropical landscape, and in turn examine the effects of these disturbances on probabilities of species occurrence.

Prevalence of human activities Evidence of human activity was widespread (see Figure 4.2). Across the surveyed area (n = 131 cells), I recorded 3796 individual signs of human activity. Indicators of permanent landcover change - agricultural plots, land tenure - were least abundant, occurring in 59 and 56 of the 131 survey cells, ~0.45 of the sampled area (naïve PAO estimate). Semi-permanent structures and associated satellite activities – use of vehicles, people roaming and illegal logging – were significantly (p < 0.05) more widespread, occurring in 103, 96, 117 and 105 of the 131 survey cells respectively. Consistently, this equated to >70% of the sampled area subject to these human activities; naïve PAO estimates of 0.79, 0.73, 0.89 and 0.80 respectively.

Within the individual study sites evidence of human traffic (people roaming on foot and travelling by motor vehicle) was estimated to occur in 90 – 100% of sampled areas (see Figure 4.3a and 4.3b). Illegal logging was also widespread and most prevalent in the commercial logging concession (95% of the sampled area) and the IUCN category IV protected areas (60% – 80% of Dangku and Bentayan wildlife reserves respectively). The high prevalence of structures within protected areas (80%) indicated that these sites were subject to extensive colonisation. Land tenure and agriculture were consistently the least prevalent of the six human activity categories considered (40-60% of sampled areas), with the highest levels recorded in the oil palm concession (Figure 4.3b).

Relative Human Influence Individual human activity categories were highly correlated (p < 0.001), but principal component analysis produced two orthogonal measures of the relative human influence detected across the cell network. Combined, these principal components explained 59.5% of total variance in the original counts of human activity (PC1 = 38.9%, PC2 = 20.6% of total variance explained). Input variable loadings indicated that PC1 described a gradient of relative landscape development with many structures and high levels of human traffic, both in terms of people roaming on foot and using vehicles to travel. From here on I refer to this principle component as ‘Development’. In contrast, PC2 describes a transition from areas subject to illegal logging and land tenure towards the early stages of development (Table 4.2). This principle component I refer to as ‘Frontier’. It is noteworthy, that this principle component has strong negative loadings from the input variables illegal logging and land tenure. Consequently, when a species is negatively influenced by the

covariate ‘frontier’ (see Table 4.3), I conclude that the probability of occurrence is highest in areas subject to illegal logging and land tenure.

Tiger occurrence probabilities and covariate effects Tigers were detected in 24 of 131 sampling cells (naïve occurrence probability, 0.18). Tiger occurrence probability was negatively influenced by the covariate total (β = -1.86; 95% CI: -2.89, -0.84) and also development (i.e., PC1) (β= -0.94; 95% CI: -1.47, -0.40). However, these models (numbered 1.1 and 1.2 in Table 4.3) did not differ significantly (∆AICc <2), and I chose to focus further discussions on model ψ(Development)p(.) (model 1.2 in Table 4.3). Higher values of Development were associated with lower probabilities of tiger occurrence (Figure 4.3).

Tigers were negatively associated with covariate Development and the simple count of human activities per sampling cell (covariate ‘total’). For this species, relatively few data were available to test for fine-scale effects of different human activities on occurrence. As such, I suggest that the most appropriate conclusion for this species in this context is that tigers occurred in the least developed areas subject to only low levels of human activity.

Prey species occurrence probabilities and covariate effects Wild boar were detected in every survey cell (naïve occurrence probability, 1.00) making covariate effects on occurrence probabilities inestimable. However, this is compelling evidence in itself of this species insensitivity to human activities, at the very least in terms of presence/absence at a landscape scale.

The occurrence probabilities of the three remaining species were negatively influenced by human activities; higher values of the significant covariates being associated with lower probabilities of species occurrence. The two cervids, muntjac and sambar, were detected in 88 and 107 of the cells (naïve occupancy estimates of 0.67 and 0.82 respectively). Among the candidate models, the most parsimonious model for each species contained the individual covariate Frontier (models 2.1 and 3.2 respectively in Table 4.3); occurrence probabilities of these species were negatively influenced by this covariate (β= -1.47; 95% CI: -2.32, -0.61 and β = -2.91; 95% CI: -4.20, -1.61 respectively) (Figure 4.3). This indicates that the highest probabilities of occurrence were in areas subject to illegal logging and land tenure.

Tapir were detected in 68 of the sampling cells, providing a naïve occupancy estimate of 0.52. Occurrence probabilities of this species were negatively influenced by Development and Frontier covariates (β = -0.65; 95% CI: -0.96, -0.35 and β = -1.15; 95% CI: -1.74, - 0.55 respectively) (Figure 4.3); the most parsimonious model included the additive effects

of these covariates (model 4.1 in Table 4.3). This indicates that tapir were more likely to occur in areas subject to logging and land tenure, but with the lowest levels of Development.

4.5 Discussion

In this study, I sought to identify the types and prevalence of human activities across different landuse types and their effects on species occurrence. Using a novel application of the ecological state variable proportion area occupied (MacKenzie et al., 2002) I found that destructive illegal activities such as logging and land tenure were extremely widespread in both commercial landuses and wildlife reserves alike. Secondly, significant correlations were found between species occurrence and measures of human disturbance. Tiger and tapir were found in areas subject to least human development, whereas sambar and muntjac were found in frontier areas. The high levels of destructive human activity raise concerns about the longterm viability of remnant forests and protected areas for conservation (Sodhi & Brook, 2008).

Drivers of Landcover Degradation Across the landscape overall, evidence of human activities was widespread. Of the six categories measured, illegal logging, people roaming, built structures and motor vehicles were extremely common (~80% of the area surveyed). Land tenure and smallholder agriculture were less prevalent, but still occurred in approximately 50% of the area surveyed.

With respect to individual management areas, evidence of human traffic (people roaming on foot and travelling by motor vehicle) was extremely common. Although not directly associated with the degradation of wildlife habitats this result indicates the accessibility of these areas and the extent to which humans moved throughout the landscape. In contrast, illegal logging has clear potential to directly impact wildlife habitats and was also widespread. This activity was most prevalent in the commercial logging concession and both wildlife reserves. It represents degradation additional to legal, mechanised logging in the concession and encroachment into these protected areas. The consistently high 72

prevalence of structures indicates that these sites are also subject to extensive and permanent colonisation. The highest levels of land tenure and smallholder agriculture were recorded in the oil palm concession. This result reflects the colonisation of marginal scrub vegetation at the boundaries of a commercial land holding. The high levels of logging in this site were also predominantly associated with these marginal areas not under active management or producing oil palm crop.

Although landcover degradation doesn’t always result in an outright loss of native vegetation (Phillips, 1997), in south-central Sumatra the trend is for pioneer activities in frontier areas to move progressively towards rural development. Specifically, individual huts built at the side of roads quickly coalesce into small communities with associated smallholder agriculture. This trend has arisen and persists because of weak land tenure laws (Sandker et al., 2007). This type of rural settlement is particularly common at the edges of commercial land uses e.g. industrial forestry and oil palm plantations (Maddox et al., 2007) and protected areas (Wittemyer et al., 2008). These rural communities are largely responsible for the human activities described here and subsequent degradation of wildlife habitats.

Although protected areas are an invaluable resource in our efforts to conserve tropical biodiversity, these areas are not the inviolate refugia that their name suggests (Liu et al., 2001; Curran et al., 2004; Wittemyer et al., 2008). In fact, the pattern of human activities detected in the protected areas (IUCN category IV) was consistent with that of the wider landscape. These particular reserves support high mammalian richness and the largest, most threatened mammals in the regional species pool - Asian elephant and Sumatran tiger (see Chapter 2). However, more than 80% of these two sites were subject to active illegal logging and pioneer land settlement. These activities degrade habitats and expose wildlife to human development and persecution (Phillips, 1997).

These results have important implications for the application of conservation strategy. Current thinking suggests that the future of tropical biodiversity conservation lies in the management of human–altered areas in support of protected areas i.e. landscape-scale conservation (Daily, 2001; Lindenmayer & Franklin, 2002; Gardner et al., 2009). However, our results draw attention to the fact that at local scales the ongoing degradation of wildlife habitats threatens to undermine their capacity to function as meaningful refugia. Ultimately, the capacity to physically conserve taxa at landscape scales will be a function of the available habitats, the prevailing human activities and the ecology of the specific taxa concerned.

Mammalian Occurrence in Human-Dominated Landscapes The results presented here suggest that species-specific occurrence probabilities lie along a gradient of relative human activity. The gradient from early frontier activities - in which tigers and tapir exist - progresses to one of land degradation driven by illegal logging and tenure in which sambar and muntjac were detected. Finally, the most developed landscapes dominated by people are used by wild boar – a notoriously resilient species (Francis, 2008).

Research suggests that the early stages of forest transformation (e.g. the frontier areas) could create favourable habitats for browsing herbivores (Eisenberg, 1980; Fragoso, 1991). Sambar, red muntjac and tapir are known to occur in a variety of forest types, including logged forests (Heydon, 1994; Brooks et al., 1997; Davies et al., 2001; Holden et al., 2003) and to utilise areas disturbed by shifting agriculture (Geist, 1998). However, the pattern of co-occurrence between species and areas of frontier activity may in reality be driven more by people colonising these areas than these animals deriving benefits from the early stages of landcover change. In this context, the negative response of tapir to ‘development’ is not unexpected in light of this species reclusive nature (Nowak, 1999). Tigers are known to have very general habitat requirements but have previously been shown to occur in more remote areas, farther from public roads and associated human development (Linkie et al., 2006). In some respects, this conclusion is in contrast to the local pattern observed by Maddox et al. (2003) in which tigers were known to range throughout areas of scrub habitat directly adjacent to a developed and expanding oil palm plantation. However, these marginal scrub habitats were progressively colonised by illegal land settlers and the tigers were quickly extirpated due to loss of habitat, direct persecution, or more likely a combination of these factors (Maddox et al., 2007). Analyses conducted here also suggest that despite the presence of a large proportion of the tigers prey base in relatively disturbed habitats, tigers were not detected in all of these areas. This indicates that additional factors other than prey distribution influence the occurrence of tigers in these landscapes. Therefore, conservation planning should incorporate a range of potential determinants when trying to secure landscapes for large carnivores.

In this paper I have restricted analyses to a single-season, single-species models developed by MacKenzie et al (2002), and specifically the constant model – ψ(.)p(.). However, it is noteworthy that this modelling framework was designed to accommodate covariates that are able to explain both probabilities of occurrence and of detection. In this context, covariates could be used to identify which geographic factors are associated with high probabilities of specific activities such as illegal land tenure. These models could ultimately be used to predict the areas most susceptible to particular anthropogenic threats. In addition, multi-season models (MacKenzie et al., 2003) could be used to monitor key sites over a number of years. Collectively, these analytical techniques could

prove very useful in our efforts to understand the role of humans in landcover degradation and species declines.

These analyses demonstrate the pervasive nature of human activities in rural tropical landscapes. Destructive activities such as illegal logging and land tenure threaten to undermine the capacity for conservation in both individual protected areas and the wider landscape. The relocation of rural communities may be an option with which to secure some protected areas (e.g. Ghate, 2003; McLean & Straede, 2003) but this is irrelevant at larger scales and so alternative approaches are desperately required.

In south-central Sumatra the trend is for early degradation (frontier) to shift towards development. Tests for species-specific responses indicate that the transition from frontier landscapes through to development change the capacity of the landscape to support different species. As lands become more developed, fewer opportunities exist for reclusive, forest-adapted species such as the tiger and tapir. Collecting concurrent human activity data from ground surveys allows us to identify the beginnings of landcover degradation. Intervention at this stage could prevent the progression of frontier landscapes into those of development. The window of opportunity for conservation is closing in south-central Sumatra; attention must be focussed on the management of rural human populations if wildlife habitats are to remain.

4.6 Tables and Figures

Table 4.1 Human activity categories with details of specific contributory indicators (signs) and a general description.

Human Activity Specific Indicators Description/Example Categories

Structures Sightings of encampments, Point sources of disturbance with radiating influence huts, bivouacs from associated people. Often pre-cursors to land settlement

Motor vehicles Sightings, sounds and Regular linear disturbance. Typically motorbikes tracks from motorcycles, commuting between local towns and 4x4s and trucks representatives of the Structures category (above)

Uncontrolled logging Sawn timber, felled trees, Aggregated, shifting, point disturbance with timber extraction rails, radiating effect from associated people. Camps of temporary sawmills, 2-4 structures occupied by 5-10 people for ~1 chainsaws seen/heard month

Land tenure Land-claim demarcation, Permanent landcover conversion creating 'gaps' in active “Slash and burn” the landscape, typically 1 - 3 ha. Often associated with pioneer land settlement

Small-holder agriculture Rubber tapping, rice fields, Conversion of forest habitats to non-native oil palm, vegetable plots vegetation and small-scale monocultures. Typically subsistence agricultural plots such as rice fields or oil palm of (~1-2ha)

People roaming Sightings, sounds and Transient, linear disturbance, highly pervasive, footprints of people or additive effect with increasing numbers of people. recently cut trails Often represented by rural settlers searching for forest resources

Table 4.2 Principal component loadings and the directions of influence (positive/negative) from six human activity categories compiled from landscape surveys. The principal components explained 59.5% of the total variance in the underlying count data.

Human Activity Category PC1 PC2

‘Development’ ‘Frontier’

Semi-Permanent Structures 0.395 0.140

Motor Vehicles 0.494 0.311

Illegal Logging 0.179 -0.746

Land Tenure 0.349 -0.530

Agricultural Plots 0.359 0.206

People Roaming 0.564 0.063

Table 4.3 Summary of model selection and parameter estimates (with standard errors in parentheses) for tigers and four principal prey species. Symbol ψ, is the probability that a cell is occupied by the species of interest and p is the probability of detecting the species in the jth survey where ψ(.)p(.) assumes that species presence and detection probability are constant across time and sampling cells. Total denotes the covariate derived from a standardised total count of human activity signs per sampling occasion, Development and Frontier denote the covariates derived from the first and second principal components of the human activity dataframe respectively. A plus sign (+) denotes additive effects, whereas an asterisk (*) denotes multiplicative effects between the covariates. No. Par. is number of parameters in the model, ∆AICc is the difference in AIC values between each model with the low-AIC value correction, wi is the AIC model weight. Note that occurrence models fit to the wild boar detection history do not extend beyond the constant model because this species was detected in every cell on virtually every sampling occasion, so covariate effects are inestimable. The most parsimonious models for each species are indicated in bold text. Where relevant, the slope parameters for the covariate effects on ψ are also given.

Figure 4.1 Study site locations with respect to principal protected areas, agri-industrial land uses and Sumatran provincial borders. (a) Protected areas on Sumatra include IUCN categories Ia (Nature reserve), II (National Park), IV (Wildlife reserve) and VI (Protection forest); indicated in dark grey (data from the World Protected Area Database, available at www.unep-wcmc.org). Human altered landscapes dominated by estate crops, plantations and logging concessions are coloured light grey (data from World Forest Watch, available a www.globalforestwatch.org). The management areas surveyed are indicated in black. Note that the oil palm concession and selective logging concession share boundaries and as such appear as one area. (b) The individual management areas in detail, black points indicate survey cell centres (n = 131). Public roads (i.e. asphalt) and logging roads are also shown.

Figure 4.2 Proportion Area Occupied (PAO) by each of six human activity categories throughout the sampled area (N = 131 survey cells). Grey bars indicate naïve occupancy estimates (the number of cells in which each activity category was detected at least once), horizontal black bars indicate estimates of PAO of specific human activity categories derived from the constant model - ψ(.)p(.) - (MacKenzie et al. 2002) and vertical black bars represent 95% confidence intervals surrounding the point estimates.

Figure 4.3a Prevalence of human activities in Bentayan wildlife reserve (upper panel) and Dangku wildlife reserve (lower panel). Solid bars indicate naïve proportions of the sampled area, points represent estimates from the constant model ψ(.)p(.) and vertical bars indicate 95%confidence intervals. Confidence intervals are absent where estimates equal 1.00.

Figure 4.3b Prevalence of human activities in the selective logging concession (upper panel) and the oil palm concession (lower panel). The estimates from the oil palm concession include the areas of marginal scrub habitats and rural settlement (i.e. OPC_S). Solid bars indicate naïve proportions of the sampled area, points represent estimates from the constant model ψ(.)p(.) and vertical bars indicate 95%confidence intervals. Confidence intervals are absent where estimates equal 1.00.

Figure 4.4 Slope parameter estimates for the effect of individual covariates on occurrence probabilities (with associated 95% confidence intervals) for tigers and three principal prey species: red muntjac, sambar and Malayan tapir. Statistically significant (p < 0.05), negative responses to a given covariate occur where 95% confidence intervals lie entirely below zero.

Chapter 5

Prospects for Tiger Conservation in Human-Altered Tropical Landscapes

5 Prospects for Tiger Conservation in Human-Altered Tropical Landscapes

(BWR) 5.1 Abstract

The conservation of tigers requires the management of human-altered landscapes. To better understand the distribution of tigers and their prey in these areas I developed an a priori concept of how landscape traits might influence the distribution of these species in south-central Sumatra. Specifically, I tested whether an aversion to profound land cover change or connectivity to less disturbed forest best explained patterns of species occurrence. Tiger distribution was significantly influenced by profound landcover change. Sambar were widespread and essentially unaffected by the covariates considered. However, connectivity to patches of least disturbed forest was important for reclusive ungulates such as tapir and muntjac. Severe landcover change has the potential to influence the distribution of tigers within the matrix and the degradation of remnant forests could impact prey species. The continuation of this process could undermine the capacity of these landscapes to provide refugia and connectivity for this assemblage and other large mammals.

5.2 Introduction

Habitat loss and fragmentation are considered the greatest threats to terrestrial biodiversity (Sala et al., 2000). The collective effects of these processes are often described using quantitative summaries of community level change such as species richness, community structure or composition (e.g. Barlow et al., 2007). However, since different components of biodiversity exhibit different responses to these processes, there is also a need to understand species-specific responses to landscape change (sensu Gardner et al., 2009) and the implications this has for the long-term viability of particular populations. Tigers (Panthera tigris) represent one of the greatest challenges to conservation because of the space they require, their value as a saleable commodity and their tendency toward conflict with humans (Sunquist & Sunquist, 2002). Furthermore, as a forest adapted, wide-ranging species, tigers are particularly susceptible to the deleterious effects of habitat loss and fragmentation. Nonetheless, as arguably the ultimate flagship species, tigers are the subject of intensive conservation measures throughout their remaining range (Sanderson et al., 2006).

Recent research to identify where wild tigers can persist in the long term has focussed on mapping suitable forest cover at large landscape scales (Wikramanayake et al., 1998; Dinerstein et al., 2007). The extensive Tiger Conservation Landscapes (TCLs) that result are designed to harbour self-sustaining populations of tigers and prey indefinitely. Many TCLs include sizeable protected areas, but evidence suggests that even when these are high profile national parks (IUCN category II) they are not the inviolate refugia their names suggest (Liu et al., 2001; Wittemyer et al., 2008; Gaveau et al., 2009). As such, TCLs will inevitably be subject to some anthropogenic landscape change and degradation. Recent research has highlighted the importance of the surrounding matrix on the viability of tiger populations in reserves (Ranganathan et al., 2008), but little is known about the determinants of tiger occurrence outside of protected areas (see Maddox et al., 2007; Linkie et al., 2008).

On the Indonesian island of Sumatra, the majority of tigers are likely to persist in large blocks of forest (Wikramanayake et al., 1998), but the intervening landscape is generally far less suitable as it is increasingly human-dominated and at lower elevations is subject to prolific agricultural expansion – notably at the hands of the burgeoning oil palm industry (Kinnaird et al., 2003). Excluding other risk factors (e.g. poaching, disease), the longterm viability of remnant tiger populations, and the populations of prey on which they rely will ultimately be dependent upon the connections within and between TCLs (Wikramanayake et al., 2004). This will inevitably require that species are able to persist in, or at very least pass-through, the intervening human dominated matrix.

Previous research on Sumatran tiger occurrence has largely been restricted to the west of the island, in Kerinci Seblat National Park (Linkie et al., 2006), its immediate surroundings (Linkie et al., 2008) and also Bukit Barisan Selatan National Park (O'Brien et al., 2003). These sites are essentially composed of contiguous forest habitats, often in rugged and mountainous terrain with human disturbances largely confined to adjacent lowland areas. Analyses by Linkie et al. (2006) modelled occurrence probabilities as functions of landscape covariates and described prevailing environmental characteristics and geographic proxies of human disturbance (e.g. distance to public roads, settlements etc). Key findings from this work include the negative relationship between tiger occurrence and the distribution of public roads.

In the study system described here, contiguous tropical forest has been fragmented, converted and degraded by human activities. To better understand the implications for large mammals I used extensive detection/non-detection surveys (MacKenzie et al., 2002) to identify patterns of habitat use by Sumatran tigers (Panthera tigris sumatrae) and their principal ungulate prey species: sambar (Rusa unicolor), red muntjac (Muntiacus muntjac) and Malayan tapir (Tapirus indicus). I hypothesised that the degree

of relative landscape alteration would distinguish between areas where these species do and do not occur. I used satellite imagery to derive measures of landscape quality at two spatial scales. Specifically these were (i) local levels of land clearance and sparse native vegetation, and (ii) landscape-connectivity to areas of least disturbed forest. Essentially, this study seeks to understand how the extremes of anthropogenic landscape change affect the distribution of tigers and prey at wider-landscape scales.

5.3 Method

5.3.1 Study Sites

5.3.2 Field Methods

During each three-hour sampling occasion teams aimed to travel widely throughout the cell (mean average two-dimensional distance was 4.8km, derived from GPS odometers on sampling occasions with ~3hrs GPS coverage, n = 202) searching a representative sample of the available habitats and in turn maximising the probability of encountering mammal signs. Direct mammal signs were sightings or audio cues and indirect signs were predominantly footprints but included other indicators such as faeces where these could be assigned to a target species with confidence. Indicators of human activity, such as motorcycle tracks, fresh-sawn timber etc were recorded and allocated to one of six predefined human activity categories. Since search effort was directed primarily toward detecting mammals, the prevalence of human activity signs recorded per cell reflects the level of human activity that occurred within areas used by mammals. Team leaders geo- referenced all direct and indirect signs using Garmin 60c global positioning systems (GPS) (Garmin International Inc., Olathe, KS) in universal transverse mercator (UTM) coordinates. Each team was equipped with footprint identification guides and followed rigorous species identification protocols to minimise the risk of false positive species’ detections through misidentification of direct or in-direct signs.

Vegetation assessments were conducted adjacent to wildlife detections in each cell. Survey teams recorded broad habitat descriptions and structural vegetation assessments based on a simple version of the Landcover Classification System (LCCS) (Di Gregario & Jansen, 2000) using an ordinal scale in 10m radius plots. These landcover reference points were used to identify areas of least degraded forest and areas corresponding to most sparse native vegetation in subsequent satellite imagery analyses.

5.3.3 Analyses

To understand the distribution of tigers and their prey I developed a conceptual model that describes the probability of species occurrence with respect to habitat connectivity and local levels of degradation. I derived estimates of connectivity based on a dispersal kernel tailored for the tiger (Moilanen, 2004). I used satellite imagery to derive two measures of landscape quality; (i) local levels of land clearance and sparse native vegetation, and (ii) relative connectivity to areas of least disturbed forest. Essentially, these tests identify whether an aversion to profound local degradation or connectivity to least disturbed areas of forest is the best fit to empirical occurrence data.

Local habitat quality For large bodied, secretive species adapted to dense cover, areas of sparse vegetation are likely to reduce the probability of species occurrence. To test this hypothesis, field data on the location and condition of vegetation were collected during extensive detection/non-detection surveys. These geo-referenced vegetation assessments and accompanying photographs were used to identify the extremes of relative landscape alteration throughout the sampled area. Highly altered areas can be summarized as: degraded natural vegetation (typically sparse trees or shrubs), newly planted crops (e.g. oil palm), and areas of active slash-and-burn.

Areas of closed canopy forest remain in discrete patches throughout the landscape. The largest of these patches remain in the selective logging concession and Dangku wildlife reserve (see Figure 5.1). These areas are the closest approximation to intact forest that remain in this landscape. They are subject to the lowest levels of human alteration by virtue of their relative inaccessibility. I hypothesized that these areas might represent refugia for many of the forest-adapted species in the landscape. As such, the occurrence of tiger and prey species may be a function of the connectivity to these areas. These areas of forest were clearly identifiable from false-colour, Landsat satellite images (see Spatial Data). Point locations of closed and open canopy forest recorded during detection/non-detection surveys were used to refine the colour pixel threshold values for this landcover class.

Spatial data Four satellite images, acquired by the Landsat Enhanced Thematic Mapper Plus Sensor (ETM+) between September 2006 and October 2007 (paths: 124,125; rows: 61,62) were sourced from the Global Land Cover Facility (accessed via: https://wwwlandcover.org). The images were affected by clouds and also regular lines of missing data caused by the failure of this instrument’s scan line detector (SLC) in 2003. To mitigate the effects of these missing data, 13 additional ETM+ images, acquired between March 2005 and

October 2007, were used to produce a series of binary masks that ‘gap-filled’ areas of missing data in the main images. Specifically, binary masks were produced from bands TM1 (wavelength: 0.45-0.52 µm, spectral band: blue), TM4 (0.76-0.90 µm, spectral band: near infrared), and TM5 (1.55-1.75 µm, spectral band: mid infrared) to identify clouds, cloud shadows and lines of missing data respectively. These masks were combined and used to filter band TM5 from each of four main images. This produced a series of images composed of TM5 pixel values and ‘No Data’. Variation in reflectance between the images was corrected using mean average differences in reflectance from regular point layers of ≥500 sampling locations. Final images were produced from combinations of TM5 masks, weighted to indicate their relative contributions to each of the main images. These were combined in one mosaic covering ~160,000 km2.

Connectivity Within the context of this study, connectivity (or its inverse, isolation) refers to the relative accessibility of a given patch to the least disturbed forest patches embedded in the wider landscape. To model the effect of connectivity on species occurrence probabilities, I adapted a method from the patch-occupancy literature that weights measures of connectivity with the species’ dispersal ability (Moilanen, 2004). To reflect differences in species space use, I developed an equivalent ‘range kernel’ based on the radius of a home range:

(1.1)

where, dij is the distance between survey cells i and j, and parameter α defines the distribution of local ranging distances for a given species (1/α is the average ranging distance).

In the absence of robust home range estimates for our four focal species, I employed an established scaling rule that describes the relationship between body size and animal space use. Specifically, I used the scaling exponents and slopes from log-log plots of area as a linear function of mass produced by Jetz et al. (2004) to estimate home range sizes for tigers and the three principal prey species considered here:

Y = (H.M)m

Where Y is the estimated home range area (km2), H is the observed scaling relationship per individual area use, M is the mean average adult body mass (kg) and m the calculated scaling exponent (slope). The radius of these home range estimates was used

to approximate the distances over which each species would be expected to range (Table 5.1).

I used this measure to estimate the relative connectivity between each survey cell and the areas of least disturbed forest in the surrounding landscape (Moilanen & Nieminen, 2002):

Si = exp(!"dij)Aj (1.2) #

Where Si denotes relative connectivity, exp(–αdij) the range kernel and Aj the area of patch j.

Covariate modelling Occupancy methods were designed to distinguish between two states of the sampling units – species present (occupied) and species absent (unoccupied). However, pseudo- absences can arise where sampling units are smaller than the areas occupied by a target species e.g. the home range. Signs of species presence will not be available for detection where a sampling unit lies within a home range but this unit has not been visited by the target species. Where true and pseudo absences cannot be differentiated, estimates of occupancy (ψ) are more aptly interpreted as usage. Since space use varies considerably with animal body size (Jetz et al., 2004) but our sampling scale remained constant, I interpret all results that relate to species occurrence in terms of the intensity of habitat use.

I used the single-species, single-season occupancy model developed by MacKenzie et al. (2002). Essentially this model represents generalised linear regression analyses that accommodate detection probabilities <1. In the null model there are two parameters, ψ (the probability of species occurrence) and p (the probability of detecting a species when it is present). To identify factors that influence these probabilities, measured covariates can be fitted using an appropriate link function e.g. the logit link. In this study I fitted two covariates to the probability of species occurrence: (i) connectivity to areas of least disturbed forest >1km2, and (ii) the area of sparse landcover within sampling cells. In addition, I tested two potential sources of variation in detection probability: (i) the local prevalence of soil roads and (ii) differences in survey team performance between sampling occasions. In many instances the soft substrate associated with soil roads provided the best medium with which to detect species footprints (Plate 5.1). I used the Zonal statistics tool in ArcGIS 9.2 (ESRI Inc., Redlands, CA) to estimate the distance

from 400 points within each survey cell to the adjacent soil roads. I used the mean average of these distances as a covariate for each survey cell. I compared the detection probabilities from different sampling occasions to test for differences in team leader ability.

Maximum likelihood estimates for the parameters ψ and p and associated covariates effects were derived using Program MARK version 5.1 (White & Burnham, 1999). Comparisons between candidate models were based on the differences between Akaike information criterion values, adjusted for small sample sizes (∆AICc), and their Akaike weights (wi) (Burnham & Anderson, 2002). The model that was within two AIC units (∆AICc) of the top ranked model and that had the least number of parameters was considered the most parsimonious and final model.

5.4 Results

Detection probability The probability of detecting secondary signs of recent tiger activity (e.g. footprints) were high in the landscape overall, with a mean average estimate from the constant model, ψ(.)p(.), of 0.72 (95% CI: 0.62, 0.81). The probability of detecting these signs was significantly (p <0.05), negatively influenced by the covariate soil roads, (β = -4.349; 95% CI: -7.852, -0.846), indicating that p declines in areas with fewer soil roads and trails.

Mean average detection probabilities for the three prey species were 0.60 (95% CI: 0.53, 0.66) for tapir, 0.62 (95% CI: 0.57, 0.67) for sambar and 0.49 (95% CI: 0.43, 0.55) for muntjac based on estimates derived from the constant model. Similarly to tiger, the probability of detecting signs of tapir was significantly negatively influenced by this covariate (β = -1.680; 95% CI: -2.948, -0.413). The probabilities of detecting sambar and muntjac were not affected significantly by the prevalence of soil roads within sampling cells (β = 0.518; 95% CI: -0.247, 1.283 and β = -0.607; 95% CI: -1.564, 0.349 respectively).

When considered as an individual effect, the covariate soil roads had a significant effect on both tiger and tapir detection probabilities. Models testing for significant differences in detection probability between sampling occasions – ψ(.)p(time) – were not significant for any of the species; ranked >2 ∆AICc below the constant model and containing three additional parameters. This indicates that detection probabilities remained consistent between successive sampling occasions and the different team leaders that these involved. 92

Occurrence probability Tigers were detected in 24 of 131 sampling cells (naïve occurrence probability, 0.18). Of the two covariates considered, the extent of sparse land cover within each sampling cell received the most support (within 37.56 AICc of the constant model, see Table 5.2) and had a significant, negative effect on tiger occurrence probability (β = -6.969; 95% CI: - 10.858, -3.081). Probabilities of tiger occurrence declined steeply in cells that contained 5 - 30% of sparse vegetation and were negligible (i.e. <0.1) where cells contained >30% of this profound landcover change (Figure 5.2a). Models containing covariates for connectivity to areas of least disturbed forest (connectivity to forest >1km2) received little support and were within 2 AICc of the constant model, ψ(.)p(.).

Sambar deer were the most widely distributed prey species, detected in 107 of the 131 survey cells (naïve occupancy of 0.82). Muntjac and tapir were detected in 88 and 68 of the 131 survey cells, or naïve occupancy of 0.67 and 0.52 respectively. Of the principal prey species, tapir and muntjac were significantly, positively influenced by the covariate connectivity to forest >1km2, see Table 5.2 and Figures 5.2a and 5.2b. For both species, the most parsimonious model contained only the connectivity covariate (tapir: β = 0.005;

95% CI: 0.001, 0.009; muntjac: β = 0.004; 95% CI: 0.0004, 0.0084), and were ranked (respectively) 42.69 and 15.51 ∆AICc above the alternative constant models. For sambar, the covariate connectivity to forest >1km2 was also the most parsimonious (∆AICc 30.192), but the positive effect of this covariate was barely significant (β = 0.161; 95% CI: -0.056, 0.377), see Figure 5.2b.

5.5 Discussion

Evidence suggests that the future of tropical biodiversity conservation lies in the management of human–altered landscapes (Daily, 2001; Lindenmayer & Franklin, 2002; Gardner et al., 2009). Reconciling this philosophy amidst an expanding human population poses a number of formidable challenges, not least the need to understand how human activities influence specific components of biodiversity (e.g. Andren, 1997; With & Crist, 1995). In this study I developed an a priori concept of how the distribution of large mammals might be influenced by landscape traits in this highly altered matrix. This research sought to resolve the importance of species mobility (described by a range kernel) and habitat selection (local prevalence of sparse vegetation) on patterns of species occurrence. I found that connectivity to patches of undisturbed forest was significant for Malayan tapir and red muntjac occurrence, but that tiger were more strongly influenced by the level of sparse vegetation. Sambar were widespread and not significantly affected by the covariates considered.

Tigers were detected in discrete regions of the logging concession and Dangku wildlife reserve with a limited overall distribution (naïve occupancy 0.18) and were less likely to occur in areas dominated by sparse vegetation. As a reclusive, cryptic species an important requirement of tiger habitat is available cover from vegetation or terrain (Sunquist & Sunquist, 2002). This structural requirement of their immediate surroundings makes areas of sparse landcover a potential barrier to movement. Sparse ground level cover is a key feature of oil palm crop and is believed to be one of the principal reasons why tigers have not been detected within the crop itself despite making use of adjacent scrub and degraded forest areas (Maddox et al., 2007). Where tigers were found to occur, the prevalence of soil roads had a significant positive effect on the probability of detection. This effect was expected a priori as field observations suggest that these roads provided useful highways for many species and were of a consistent soft substrate ideal for leaving footprints. This observation is akin the use of pre-existing animal trails and mountain ridges observed in the mountainous habitat of south-west Sumatra (O'Brien et al., 2003).

Connectivity to areas of least disturbed forest was not important for tigers and there are four possible explanations for this. First, given their large home ranges they are sufficiently mobile that they are relatively insensitive to the distribution of least disturbed forest and able to move between these areas as part of their normal ranging patterns. Therefore, they are inherently more connected to undisturbed forest by virtue of their mobility. Second, tigers’ habitat requirements may be too broad for their occurrence to be solely dependent on patches of undisturbed forest (e.g. Sunquist & Sunquist, 2002). 94

Third, the scale of sampling was relatively small compared to the distances over which tigers can travel, and as such it could be difficult to resolve the true importance of connectivity for tigers in our sampled area (see Moilanen, 2004). Sampling over a larger spatial scale coupled with alternative classifications of forest quality might allow us to better resolve the importance of connectivity and landcover type. Finally, although species were widespread in the sampled landscape, the surveys did not estimate their abundance. Prey abundance is recognised as one of the principle determinants of tiger distribution and without a measure of prey abundance it is not possible to understand the capacity of this landscape to support a tiger population long-term (Karanth et al., 2004).

For the prey species the amount of sparse vegetation does not appear to affect occurrence probability. Despite similarities in ranging patterns, tapir and sambar differed in their distribution throughout the landscape, tapir occurring in a smaller proportion of the sampled area (60%). Tapir were also found to walk along soil roads (Holden et al., 2003), and the prevalence of these roads had a significant positive effect on their detection probability. For the other prey species, soil roads were unimportant. For tapir and muntjac, the higher the level of connectivity, the higher the probability of their occurrence. In contrast, sambar occurrence was not significantly influenced by either connectivity or local levels of degradation. They were widely distributed throughout the cell network (naïve ψ 0.82), which suggests that they have wide habitat requirements (Schaller, 1967).

Whilst levels of sparse vegetation are an important factor for tigers, it was not an effective explanation for patterns of ungulate occurrence. Forest ungulates are known to occur in a variety of landcovers (Brooks et al., 1997; Davies et al., 2001; Heydon, 1994; Holden et al., 2003). Sambar are known to have very broad dietary requirements and utilise agricultural areas close to human habitation (Francis, 2008; Geist, 1998; Schaller, 1967). This may in part explain their wide distribution in the landscape described here. Although sambar and tapir have qualitatively similar habitat and dietary requirements, tapir may be less tolerant of these areas and are thus allied to areas of less disturbed forest. Results suggest that red muntjac are more allied to areas of undisturbed forest in this landscape than the other cervid, sambar. As such, this species shares the significant positive association with undisturbed forest detected for tapir.

Where can tigers persist in current human altered landscapes? Conservation biologists approach the conservation of large wide-ranging species from the perspective of metapopulation management (Linkie et al., 2006). The implementation of these approaches for tigers require that we work extensively within the human-altered matrix. Results presented here suggest that within this matrix it is important to minimise the areas of sparse vegetation as these have been shown to negatively impact their probability of occurrence. Since these areas are often a by-product of both industrial

scale development and the activities of rural communities, it is difficult to foresee how this can be prevented.

Furthermore, for certain species it has been suggested that patch quality may be a more important predictor of occupancy than connectivity (Fleishman et al., 2002). The satellite imagery suggested three broad classes of landcover in this landscape: closed canopy forests, sparse vegetation and an intervening matrix of highly heterogeneous vegetation. The covariates used here essentially modeled the two extremes, but for a habitat generalist such as the tiger, the intervening matrix would likely also be suitable habitat under many circumstances (Ricketts, 2001). An understanding of patch quality is not limited to habitat type (Prugh et al., 2008) but incorporates the availability of other resources such as prey, and to threats such as persecution by people or landscape development. As such the connectivity measure described here may underestimate the true area of suitable habitat patches available in the landscape. However, in this landscape the remnant forest and other areas of native vegetation are under considerable pressure from the rural population and are therefore likely to decline (see Chapter 4). This would undermine the potential of these areas to support both tigers and many other species.

Whilst connectivity to areas of least disturbed forest was not a good predictor of tiger occurrence at the local scales studied here, at larger spatial scales habitat connectivity is known to be an important factor affecting metapopulation dynamics ( Wiens, 1996; Linkie et al., 2006). On a landscape scale, the Sumatran TCLs are likely to hold the larger source populations, and the human-altered matrix the smaller sink populations (Wikramanayake et al., 2004). A wider appreciation of the ecological value of human altered landscapes will be key to their inclusion in wider-scale, practical conservation measures for tigers and many other species.

5.6 Tables and Figures

Table 5.1 Estimated range kernels for tiger and three principal prey species.

Trophic Species Mean Scaling Scaling Home Home range Mean group average relationship per exponent range size radius (km): α average body mass individual area (m)** (km2) movement (M)* (kg) use (H)** (1/α)

Herbivore tapir 311 2.06 1.02 7.20 1.5 0.67 Herbivore sambar 177 2.06 1.02 4.00 1.1 0.91 Herbivore muntjac 18 2.06 1.02 0.40 0.3 3.33 Carnivore tiger 162 52.07 1.2 233.30 8.6 0.12

*Jones et al. (2009) ** Jetz et al. (2004)

Table 5.2 Summary of model selection and parameter estimates (with standard errors in parentheses) for tigers and three principal prey species. Symbol ψ, is the probability that a cell is occupied by the species of interest and p is the probability of detecting the species in the jth survey where ψ(.)p(.) assumes that species presence and detection probability are constant across time and sampling cells, ψ(km2 converted land) denotes the covariate representing the extent of each survey cell subject to severe landcover degradation, ψ(connectivity to forest >1km2) denotes the covariate for relative connectivity to undisturbed forest patches >1km2, #Par. is number of parameters in the model, ∆AICc is the difference in AIC values (with the low-AIC value correction) between each model and the highest ranked model in the candidate set, wi is the AIC model weight. The most parsimonious models for each species are indicated in bold text.

Figure 5.1 Landsat ETM+ imagery mosaic (band TM5), south-central Sumatra. (a)Sumatran protected areas are indicated in black; they include IUCN categories Ia (Nature reserve), II (National Park), IV (Wildlife reserve) and VI (Protection forest); (data from the World Protected Area Database, available at www.unep-wcmc.org). Survey locations lay within the white polygon. (b) Individual management areas in detail, black points indicate survey cell centres (n = 131). Areas in dark grey indicate area of least disturbed forest of >1km2 that persist amidst the human-dominated matrix of degraded natural landcover and agri-industrial land uses (indicated in light grey). Bare soil and the most sparse vegetation are indicated in white.

Figure 5.2a Probabilities of species occurrence with respect to human land use intensity. The upper panel shows the probability of tiger occurrence with respect to the percentage of converted land (sparse vegetation) within sampling cells. The lower panel shows the probability of tapir occurrence with respect to relative connectivity to areas of least disturbed forest. These covariates were retained in the most parsimonious models of these species occurrence. Estimates are shown across a range of observed variation (5th to 95th percentiles) in each environmental covariate.

100

Figure 5.2b Probabilities of species occurrence with respect to human land use intensity. The upper and lower panels show the probabilities of sambar and red muntjac occurrence respectively and each with respect to connectivity to areas of least disturbed forest. Estimates are shown across a range of observed variation (5th to 95th percentiles) in this environmental covariate.

101

Plate 5.1 Soil roads are widespread throughout this landscape. The substrate is generally soft at the road edge and can hold the impressions of mammal footprints for ~10 days provided rain is infrequent or light.

102

Chapter 6

General Discussion

103

6 General Discussion

Habitat loss at the hands of human enterprise continues to drive the global decline in biodiversity. Throughout the tropics the extraction of forest commodities and the expansion of plantation agriculture are the principal drivers of this decline (Pimm & Raven, 2000; Geist & Lambin, 2002). This exploitation of tropical landscapes is set to continue, and in the wake of this prolific landscape change, human-altered landscapes now dominate. It is within these matrices of agriculture, industry and rural settlement that a growing proportion of tropical biodiversity must persist if it is to survive (Daily, 2001). Although protected areas (e.g. IUCN categories I-VI) can provide refuge for many species, these areas are not infallible and are increasingly undermined by the spread of human activities (e.g. Woodroffe & Ginsberg, 1998; Brooks et al., 2004; DeFries et al., 2005). As such, the success of efforts to conserve tropical species will ultimately depend on our ability to reconcile the potential for conservation in degraded lands amidst growing pressure from burgeoning human populations.

Comparisons of mammalian species richness between land management areas indicated that highest richness was associated with management areas dominated by degraded forests and that these areas supported the larger and most highly threatened members of the regional species pool. Similar differences between forest dominated and agricultural areas have also been reported from equivalent Neotropical landscapes (Daily et al., 2003). In contrast the mammalian community associated with oil palm dominated areas was significantly impoverished compared to other management areas sampled and no longer supported highly threatened species. These results are in keeping with research on other taxa (Danielsen et al., 2008; Fitzherbert et al., 2008).

In order to try to understand how these patterns of species persistence and local extinction had arisen, I tested a suite of ecological traits expected to influence species’ capacity to persist in the face of landscape change. Collectively these traits contribute to a species’ niche, which essentially defines the conditions under which a species exists and how it interacts with its environment (Grinnell, 1917; Hutchinson, 1957; Chase & Leibold, 2003). In a changing landscape, a broader niche would be expected to confer a greater advantage to species survival and this idea is supported in the literature where ecological specialisation has been found to increase extinction risk in both mammals (Haracourt et al., 2002; Boyles & Storm, 2007) and birds (Julliard et al., 2003; Shultz et al., 2005). Results from large-scale comparative analyses indicated that species extinction risk is ultimately driven by a combination of both intrinsic (species biology) and extrinsic (environmental) factors (Cardillo et al., 2005), but that overall, the main intrinsic

104

traits linked to extinction were body size, habitat and/or dietary specialisation (Owens & Bennett, 2000; Purvis et al., 2000b).

I identified that a combination of body size, reproductive rate and dietary specialisation were best able to distinguish between species that persisted and those that did not. I found a pattern of species persistence that was indicative of the ideas embodied by the fast/slow life history strategies (sensu Stearns, 1983). Broadly speaking, in the initial stages of landscape alteration the larger bodied slow strategists, with associated low reproductive output, were predicted to persist. With more severe levels of landscape change, the balance of persistence shifted in favour of the fast strategists. In addition, I also found evidence of diet specialisation being an important factor; omnivores and species with broad diet breadth were more likely to persist. These analyses also indirectly measured habitat specialisation. The species that persist in the most altered landcover (oil palm) also occur in all other habitats and therefore represent habitat generalists. Collectively, the distribution of mammalian adversity in human altered landscapes suggests that there exists a small window of opportunity for the conservation of the larger, typically threatened species in these areas as further development favours only a handful of species equipped with fast life histories.

Having identified the distribution of mammalian diversity and considered how this specific pattern may have arisen with respect to species’ relative resilience, I wanted to investigate the principle human activities that shape these landscapes.

The negative impacts of oil palm production on mammalian richness and community composition are of grave concern because much of Indonesia’s agricultural expansion is directly attributable to this industry. Currently, Indonesia meets ±43% (FAOSTAT 2007) of global demand for this, the worlds most traded oil seed crop (Carter et al., 2007). Results from this study indicate that the proliferation of oil palm will create increasingly large areas of land that are inhospitable to approximately 70% of the Sumatran forest- mammal community described here. In addition, an increase in oil palm concessions is likely to confer an increase in the marginal, degraded lands that routinely follow industrial scale land clearance. These areas are highly dynamic, transitional landscapes dominated by rural people, rarely considered in the scientific literature. They are one example of the secondary consequences of agricultural expansion that broaden its implications. Many species were found to use these marginal habitats but the prevalence of people and destructive activities mean that these areas are often in a state of chronic deterioration.

Realizing the potential to conserve wildlife in human altered landscapes requires that the resources on which species depend are available in the long-term and not subject to significant deterioration. However, given that human-dominated landscapes are generally

105

complex matrices of different land uses and subject to ongoing development this suggests that these areas are in fact highly susceptible to degradation by human activities (Daily, 2001). In addition, since the efficacy of protected areas is inexorably linked to the matrix in which they are embedded (Ranganathan et al., 2008; Wittemyer et al., 2008), degradation at park margins and human colonization of the interior could seriously undermine these areas.

Although landcover degradation doesn’t always result in an outright loss of native vegetation (Phillips, 1997), in south-central Sumatra the trend is for pioneer activities in frontier areas to move progressively towards rural development. Specifically, individual huts built at the side of roads quickly coalesce into small communities with associated smallholder agriculture. This trend has arisen and persists because of weak land tenure laws (Sandker et al., 2007). This type of rural settlement is particularly common at the edges of commercial land uses e.g. industrial forestry and oil palm plantations (Maddox et al., 2007) and protected areas (Wittemyer et al., 2008). In this landscape the highest levels of land tenure and smallholder agriculture were recorded in the marginal scrub vegetation of the oil palm concession.

Evidence of human activities was not restricted to the edges of land management areas, destructive activities such as illegal logging and land tenure were found in the interior of the commercial logging concession and the wildlife reserves. In addition, the high levels of human traffic recorded (people roaming on foot and travelling by motor vehicle) indicate the accessibility of these areas and the extent to which humans moved throughout the landscape. The prevalence of structures indicates that these sites are also subject to extensive and permanent colonisation.

These analyses demonstrate that as oil palm production increases across Sumatra, so the habitats that support endangered species in human altered landscapes will diminish. Since the species of greatest concern to conservation are extirpated from oil palm dominated areas they will be confined to degraded forest habitats contained within

106

selective logging concessions and small wildlife reserves. However, these areas are subject to intensive human activity and as such are in state of decline. The relocation of rural communities may be an option with which to secure some protected areas (e.g. McLean & Straede, 2003) but this is irrelevant at larger scales and so alternative approaches are desperately required.

Having identified patterns of diversity I then focused on species-specific responses to the current landscape. I focused on a specific assemblage: Sumatran tigers (Panthera tigris sumatrae) and their principal ungulate prey species. Tigers represent one of the greatest challenges to conservation because of the space they require, their value as a saleable commodity and their tendency toward conflict with humans (Sunquist & Sunquist, 2002). Furthermore, as a forest adapted, wide-ranging species, tigers are particularly susceptible to the deleterious effects of habitat loss and fragmentation. Nonetheless, as arguably the ultimate flagship species, tigers are the subject of intensive conservation measures throughout their remaining range (Sanderson et al., 2006).

In south-central Sumatra the trend is for early degradation (frontier areas) to shift towards development. Tests for species-specific responses indicate that the transition from frontier landscapes through to development change the capacity of the landscape to support different species. As lands become more developed, fewer opportunities exist for reclusive, forest-adapted species such as the tiger and tapir. Analyses conducted in Chapter 4 suggest species-specific occurrence probabilities lie along a gradient of relative human activity. The gradient from early frontier activities - in which tigers and tapir exist - progresses to one of land degradation driven by illegal logging and tenure in which sambar and muntjac persisted. Finally, the most developed landscapes dominated by people are used by wild boar – a notoriously resilient species (Francis, 2008).

Conservation biologists often approach the conservation of large wide-ranging species from the perspective of metapopulation management (Linkie et al., 2006). The implementation of these approaches for tigers requires that we work extensively within the human-altered matrix. (Wikramanayake et al., 2004). A wider appreciation of the

107

ecological value of these altered landscapes is required if they are to contribute to larger scale conservation efforts.

Chapter 5 considers the landscape traits that discriminate between the areas where tigers occur and where they do not. To better understand the distribution of tigers and their prey in this landscape I developed an a priori concept of how landscape traits might influence the distribution of these species in south-central Sumatra. Specifically, I tested whether an aversion to profound land cover change or connectivity to less disturbed forest best explained patterns of species occurrence. I found that connectivity to patches of undisturbed forest was significant for tapir and muntjac occurrence, but that tiger were negatively influenced by the level of sparse vegetation within sampling cells. Sambar were widespread and not significantly affected by connectivity or the degree of landcover.

For certain species it has been suggested that patch quality may be a more important predictor of occupancy than connectivity (Fleishman et al., 2002). Results presented here suggest that within this matrix it is important to minimise the areas of sparse vegetation as these have been shown to negatively impact the probability of tiger occurrence. The covariates tested here essentially modelled two extremes in the landscape, but for a habitat generalist such as the tiger, the intervening matrix of heterogeneous vegetation may also be suitable habitat under many circumstances (Ricketts, 2001). As such human altered landscapes have the capacity to provide both refugia in their own right and a medium through which individuals can pass (Ranganathan et al., 2008). An understanding of patch quality is not limited to habitat type (Prugh et al., 2008) but incorporates the availability of other resources such as prey, and threats such as persecution by people or landscape development. As such the connectivity measures used here may underestimate the true area of suitable habitat patches available in the landscape. However, in this landscape the remnant forest and other areas of native vegetation are under considerable pressure from the rural population and are therefore likely to decline (see Chapter 4). This would undermine the potential of these areas to support both tigers and many other species. Since these areas are often a by-product of both industrial scale development and the activities of rural communities, it is difficult to foresee how this can be prevented.

If a combination of the ground surveys and imagery analyses used here could be replicated on a larger scale, early intervention could prevent the progression of frontier landscapes into those of development. In addition, a key step to mitigate the impact of oil palm on reclusive and wide-ranging mammals is to test whether increased ground level cover in some areas of the crop could provide contiguous vegetative cover and therefore a degree of habitat connectivity for these mammals, particularly tigers and more forest dependent species such as the Malayan tapir and red muntjac (Maddox et al., 2007).

108

The effects of anthropogenic landscape change on the mammal communities considered in this study are indicative of trends observed in other regions. Recent broad-scale analyses across South East Asia, report that the effects of human disturbances on forest biotas have widespread impacts irrespective of taxonomic group, type of human activity or the measure of biodiversity considered (Sodhi et al., 2009). In light of forecasted agricultural and population expansion, the spread of the human altered landscapes described here are to a large extent inevitable (Daily, 2001; Morton et al., 2006). However, many species threatened by the loss of native habitats actually use human altered landscapes and could in principle therefore be conserved through appropriate management of these areas (McNeely & Scherr, 2002). Results from this study indicate that the conservation value of human altered landscapes lies principally in areas of degraded forest. However, the proliferation of the oil palm industry and continued degradation of native vegetation at the hands of rural populations threatens to undermine the conservation potential of these landscapes.

This study has highlighted the pervasiveness of human activities across the landscape, the severity of ongoing habitat degradation in protected areas and the unmanaged nature of human development across the landscape as a whole. In light of the speed and magnitude with which anthropogenic landscape change is advancing, it is imperative that action is taken based on current knowledge. Government and stakeholder recognition of the conservation value of these areas will be key to securing their latent potential. The window of opportunity for conservation is closing in south-central Sumatra; attention must be focussed on the management of industrial expansion and rural populations if wildlife habitats are to remain.

109

7 Appendix

7.1 An example of the human activity datasheets

110

7.2 An example of the detection/non-detection datasheets

111

7.3 An example of the sampling cell maps used by the field teams during active search periods

112

8 Literature cited

Achard, F., Eva, H.D., Stibig, H., Mayaux, P., Gallego, J., Richards, T., & Malingreau, J. (2002) Determination of Deforestation Rates of the World's Humid Tropical Forests. Science, 297, 999-1002.

Andren, H. (1997) Population response to landscape chnages depends on specialization to different landcape elements. Oikos, 80, 193-196.

Applegate, G., Chokkalingam, U., & Suyanto, S. (2001). The Underlying Causes and Impacts of Fires in South-East Asia CIFOR, ICRAF, USAID and USFS, Bogor, Indonesia.

Barlow, J., Gardner, T.A., Araujo, I.S., Avila-Pires, T.C., Bonaldo, A.B., Costa, J.E., Esposito, M.C., Ferreira, L.V., Hawes, J., Hernandez, M.I.M., Hoogmoed, M.S., Leite, R.N., Lo-Man-Hung, N.F., Malcolm, J.R., Martins, M.B., Mestre, L.A.M., Miranda-Santos, R., Nunes-Gutjahr, A.L., Overal, W.L., Parry, L., Peters, S.L., Ribeiro-Junior, M.A., da Silva, M.N.F., da Silva Motta, C., & Peres, C.A. (2007) Quantifying the Biodiversity Value of Tropical Primary, Secondary, and Plantation Forests. Proceedings of the Natural Academy of Sciences, USA, 104, 18555– 18560.

Barr, C. (2001) Banking on sustainability : structural adjustment and forestry reform in post-suharto Indonesia. CIFOR and WWF's Macroeconomics Program, Bogor.

Bennett, P.M. & Owens, I.P.F. (1997) Variation in extinction risk among birds: chance or evolutionary predisposition? Proceedings of the Royal Society of London Series B, 264, 401-408.

Boyles, J.G. & Storm, J.J. (2007) The perils of picky eating: dietary breadth is related to extinction risk in insectivorous bats. PLoS One, 2, e672.

Breiman, L., Friedman, J.H., Olshen, R.A., & Stone, C.G. (1984) Classification and Regression Trees. Wadsworth International Group, Belmont, California, USA.

Brook, B.W., Bradshaw, C.J.A., Koh, L.P., & Sodhi, N.S. (2006) Momentum drives the crash: mass extinction in the tropics. Biotropica, 38, 302-305.

Brooks, D.M., Bodmer, R.E., & Matola, S. (1997). Tapirs - Status Survey and Conservation Action Plan. IUCN/SSC Tapir Specialist Group, Gland, Switzerland.

Brooks, T.M., Bakarr, M.I., Boucher, T., da Fonseca, G.A.B., Hilton-Taylor, C., Hoekstra, J.M., Moritz, T., Olivieri, S., Parrish, J., Pressey, R.L., Rodrigues, A.S.L., Sechrest, W., Stattersfield, A., Strahm, W., & Stuart, S.N. (2004) Coverage provided by the global protected-area system: Is it enough? . BioScience, 54, 1081-1091.

Buij, R., Mcshea, W.J., Campbell, P., Lee, M.E., Dallmeier, F., Guimondou, S., Mackaga, L., Guisseougou, N., Mboumba, S., Hines, J.E., Nichols, J.D., & Alonso, A. (2007) Patch-occupancy models indicate human activity as major determinant of forest elephant Loxodonta cyclotis seasonal distribution in an industrial corridor in Gabon. Biological Conservation, 135, 189-201. 113

Burnham, K.P. & Anderson, D.R. (2002) Model Selection and Multimodal Inference, 2nd edn. Springer-Verlag, New York.

Burnham, K.P. & Overton, W.S. (1978) Estimation of the size of a closed population when capture probabilities vary among animals. Biometrika, 65, 623-633.

Butler, R.A. & Laurance, W.F. (2008) New strategies for conserving tropical forests. TRENDS in Ecology and Evolution, 23, 469-472.

Cam, E., Nichols, J.D., Sauer, J.R., Hines, J.E., & Flather, C.H. (2000) Relative species richness and community completeness: avian communities and urbanization in the mid-Atlantic states. Ecological Applications, 10, 1196-1210.

Carbone, C., Mace, G.M., Roberts, S.C., & Macdonald, D.W. (1999) Energetic constraints on the diet of terrestrial carnivores. Nature, 402, 286-288.

Carbone, C., Teacher, A., & Rowcliffe, M. (2007) The costs of carnivory. PLOS Biology, 5, 363-368.

Cardillo, M. & Bromham, L. (2001) Body size and risk of extinction in Australian mammals. Conservation Biology, 15, 1435-1440.

Cardillo, M., Mace, G.M., Jones, K.E., Bielby, J., Bininda-Emonds, O.R.P., Sechrest, W., Orme, C.D.L., & Purvis, A. (2005) Multiple causes of high extinction risk in large mammal species. Science, 309, 1239-1241.

Cardillo, M., Purvis, A., Sechrest, W., Gittleman, J.L., Bielby, J., & Mace, G.M. (2004) Human population density and extinction risk in the world’s carnivores. PLOS Biology, 2, 909-914.

Carroll, C., Zielinski, W.J., & Noss, R.F. (1999) Using presence-absence data to build and test spatial habitat models for the fisher in Klamath region, USA. Conservation Biology, 13, 1344-1359.

Carter, C., Finley, W., Fry, J., Jackson, D., & Willis, L. (2007) Palm oil markets and future supply. European Journal of Lipid Science and Technology, 109, 307 - 314.

Casson, A. (2003). Oil Palm, Soybeans and Critical Habitat Loss. WWF Forest Conversion Initiative, Switzerland.

Chao, A. & Shen, T.J. (2003) Program SPADE (Species Prediction And Diversity Estimation). http://chao.stat.nthu.edu.tw.

Chapin III, F.S., Zavaleta, E.S., Eviner, V.T., Naylor, R.L., Vitousek, P.M., Sala, O.E., Reynolds, H.L., Hooper, D.U., Mack, M., Diaz, S.E., Hobbie, S.E., & Lavorel, S. (2000) Consequences of Changing Biodiversity. Nature, 405, 234-242.

Chase, J.M. & Leibold, M.A. (2003) Ecological niches: linking classical and contemporary approaches. University of Chicago Press, Chicago, Illinois, USA.

114

Clark, K.R. & Warwick, R.M. (2001) Change in marine communities: an approach to statistical analysis and interpretation. 2nd edn. Primer E, Plymout, United Kingdom.

Clark, L.A. & Pregibon, D. (1992). Tree-based models. In Statistical models in S (eds J.M. Chambers & T.J. Hastie), pp. 377-420. Wadsworth and Brooks, Pacific Grove, California, USA.

Clay, J. (2004) World Agriculture and the Environment: A Commodity-by-Commodity Guide to Impacts and Practices. Island Press.

Colwell, R.K. (2008) EstimateS: Statistical estimation of species richness and shared species from samples http://purl.oclc.org/estimates, University of Connecticut, Storrs, Connecticut.

Colwell, R.K., Mao, C.X., & Chang, J. (2004) Interpolating, extrapolating, and comparing incidence-based species accumulation curves. Ecology, 85, 2717-2727.

Corlett, R.T. (2007) The impact of hunting on the mammalin fauna of tropical asian forests. Biotropica, 39, 292-303.

Cossalter, C. & Pye-Smith, C. (2003). Fast-wood forestry, myths and realities. CIFOR, Bogor.

Craig, J., Mitchell, N., & Saunders, A.D. (1999) Nature Conservation in Production Environments: Managing the Matrix. Surrey Beatty and Sons, Sydney.

Curran, L.M., Trigg, S.N., McDonald, A.K., Astiani, D., Hardiono, Y.M., Siregar, P., Caniago, I., & EKasischke, E. (2004) Lowland forest loss in protected areas of Indonesian Borneo. Science, 303, 1000-1003.

Daily, G.C. (2001) Ecological forecasts. Nature, 245 - 245.

Daily, G.C., Ceballos, G., Pacheco, J., Suzan, G., & Sanchez-Azofeifa, A. (2003) Countryside biogeography of neotropical mammals: Conservation opportunities in agricultural landscapes of Costa Rica. Conservation Biology, 17, 1814 - 1826.

Danielsen, F., Beukema, H., Burgess, N.D., Parish, f., Bruhl, C.A., Donald, P.F., Murdiyarso, D., Phalan, B., Reijnders, L., Struebig, M., & Fitzherbert, E.B. (2008) Biofuel Plantations on Forested Lands: Double Jeopardy for Biodiversity and Climate. Conservation Biology, In Press.

Davies, G., Haydon, M., Leader-Williams, N., MacKinnon, J., & Newing, H. (2001). The effects of logging on tropical forest ungulates. In The Cutting Edge: Conserving Wildlife in Logged Tropical Forest (eds R.A. Fimbel, A. Grajal & J.G. Robinson). Columbia University Press, New York.

Davies, K.E., Margules, C.R., & Lawrence, J.E. (2000) Which traits of species predict population declines in experimental forest fragments? Ecology, 81, 1450-1461.

De'ath, G. & Fabricius, K. (2000) Classification and Regression Trees: A Powerful Yet Simple Technique for Ecological Data Analysis. Ecology, 81, 3178-3192. 115

DeFries, R., Hansen, A., Newton, A.C., & Hansen, M.C. (2005) Increasing isolation of protected areas in tropical forests over the past twenty years. Ecological Applications, 15, 19-26.

Di Gregario, A. & Jansen, L.J.M. (2000) Land Cover Classification System (LCCS): Classification Concepts and User Manual FAO. Available at:http://www.fao.org/docrep/003/x0596e/X0596e00.htm#P-1_0, Rome.

Dinerstein, E., Loucks, C., Wikramanayake, E., Ginsberg, J., Sanderson, E., Seidensticker, J., Forrest, J., Bryja, G., Heydlauff, A., Klenzendorf, S., Leimgruber, P., Mills, J., O'Brien, T., Schrestha, M., Simons, R., & Songer, M. (2007) The fate of wild tigers. BioScience, 57, 508-514.

Duchamp, J.E. & Swihart, R.K. (2008) Shifts in bat community structure related to evolved traits and features of human-altered landscapes. Landscape Ecology, 23, 849-860.

Eisenberg, J.F. (1980). The density and biomass of tropical mammals. In Conservation Biology: An Evolutionary-Ecological Perspective (eds M. Soule & B.A. Wilcox). Sinauer Associates, Sunderland, Massachusetts.

Escamilla, A., Sanvicente, M., Sosa, M., & Galindo-Leal, C. (2000) Habitat mosaic, wildlife availability, and hunting in the tropical forest of Calakmul, Mexico. Conservation Biology, 14, 1592-1601.

Etienne, R.S. & Heesterbeek, J.A.P. (2001) Rules of thumb for conservation of metapopulations based on a stochastic winking-patch model. The American Naturalist, 158, 389-407.

FAO (2005). State of the World's forests 2005. Food and Agriculture Organisation of the United Nations, Rome.

FAO/MacKinnon, J. (1982). National Conservation Plan for Indonesia. FAO, Bogor.

Fisher, D.O. & Owens, I.P.F. (2004) The comparative method in conservation biology. TRENDS in Ecology and Evolution, 19, 391-398.

Fitzherbert, E.B., Struebig, M.J., Morel, A., Danielsen, F., Brulh, C.A., Donald, P.F., & Phalan, B. (2008) How will oil palm expansion affect biodiversity? TRENDS in Ecology and Evolution, 23, 538 - 545.

Fleishman, E., Ray, C., Sjogren-Gulve, P., Boggs, C.L., & Murphy, D.D. (2002) Assessing the roles of patch quality, area, and isolation in predicting metapopulation dynamics. Conservation Biology, 16, 706-716.

Fragoso, J.M. (1991). The effect of selective logging on Baird's tapir. In Latin American Mammalogy: History, Biodiversity, and Conservation (eds M.A. Mares & D.J. Schmidley), pp. 295-304. Oklahoma Museum of Natural History, Norman, Oklahoma.

Francis, C.M. (2008) A field guide to the mammals of south-east Asia. New Holland, London, UK.

116

FWI/GFW (2002). The state of the forest: Indonesia. Forest Watch Indonesia, Bogor and Global Forest Watch, Washington DC.

Gardner, T.A., Barlow, J., Chazdon, R., Ewers, R.M., Harvey, C.A., Peres, C.A., & Sodhi, N.S. (2009) Prospects for tropical forest biodiversity in a human-modified world. Ecology Letters, 12, 1-21.

Gardner, T.A., Barlow, J., Parry, L., & Peres, C.A. (2007) Predicting the Uncertain Future of Tropical Forest Speciec in a Data Vacuum. Biotropica, 39, 25-30.

Gardner, T.A., Hernández, M.I.M., Barlow, J., & Peres, C.A. (2008) Understanding the biodiversity consequences of habitat change: the value of secondary and plantation forests for neotropical dung beetles. Journal of Applied Ecology, 45, 883-893.

Gaveau, D.L.A., Linkie, M., Suyadi, Levang, P., & Leader-Williams, N. (2009) Three decades of deforestation in southwest Sumatra: Effects of coffee prices, law enforcement and rural poverty. Biological Conservation, 142, 597-605.

Geist, H.J. & Lambin, E.F. (2002) Proximate Causes and Underlying Driving Forces of Tropical Deforestation. BioScience, 52, 143-150.

Geist, V. (1998). Three-pronged old world deer. In Deer of the World: Their Evolution, Behaviour and Ecology. Stackpole Books.

Ghate, R. (2003) Global gains at local costs: Imposing protected areas: Imposing restricted areas in India. Journal of Sustainable Development and World Ecology, 10, 377-395.

GOI (Government of Indonesia) (1993). Rencana Pembangunan Lima Tahun Keenam (Sixth Five-Year Development Plan), Jakarta.

Grainger, A. (1993) Controlling tropical deforestation. Earthscan Publications Ltd, London.

Grinnell, J. (1917) Field tests of theories concerning distributional control. American Naturalist, 51, 115-128.

Hanski, I. (1998) Metapopulation dynamics. Nature, 396, 41-49.

Harcourt, A.H., Coppeto, S.A., & Parks, S.A. (2002) Rarity, specialization and extinction in primates. Journal of Biogeography, 29, 445-456.

Henderson, J. & Osborne, D.J. (2000) The oil palm in all our lives: how this came about. Endeavour, 24, 63-68.

Heydon, M.J. (1994). The Ecology and Management of Rainforest Ungulates in Sabah Malaysia: Implications of Forest Disturbance. Institute of Tropical Biology, University of Aberdeen, Aberdeen, UK.

117

Hirzel, A.H., Hausser, J., Chessel, D., & Perrin, N. (2002) Ecological-niche factor analysis: how to compute habitat-suitability maps without absence data. Ecology, 83, 2027-2036.

Holden, J., Yanuar, A., & Martyr, D.J. (2003) The Asian Tapir in Kerinci Seblat National Park, Sumatra: evidence collected through photo-trapping. Oryx, 37, 34-40.

Hutchinson, G.E. (1957). Concluding remarks. In Population studies: animal ecology and demography, Vol. 22, pp. 415-427. Cold Spring Harbor Press, New York, USA.

ITFMP (1999). A draft position paper on threats to sustainable forest management in Indonesia: roundwood supply and demand and illegal logging. Indonesia-UK Tropical Forest Management Programme

IUCN (2008). IUCN Red List of Threatened Species. IUCN, Gland, Switzerland.

Jetz, W., Carbone, C., Fulford, J., & Brown, J.H. (2004) The scaling of animal space use. Science, 306, 266-268.

Jones, K.E., Bielby, J., Cardillo, M., Fritz, S.A., O'Dell, J., Orme, C.D.L., Safi, K., Sechrest, W., Boakes, E.H., Carbone, C., Connolly, C., Cutts, M.I.J., Foster, J.K., Grenyer, R., Habib, M., Plaster, C.A., Price, S.A., Rigby, E.A., Rist, J., Teacher, A., Bininda-Emonds, O.R.P., Gittleman, J., L., Mace, G.M., & Purvis, A. (2009) PanTHERIA: A species-level database of life-history, ecology and geography of extant and recently extinct mammals. Ecology.

Jones, K.E., Purvis, A., & Gittleman, J.L. (2003) Biological correlates of extinction risk in bats. American Naturalist, 161, 601-614.

Julliard, R., Jiguet, F., & Couvert, D. (2003) Common birds facing global changes: what makes a species at risk? Global Change Biology, 10, 148-154.

Karanth, K.U., Nichols, J.D., Kumar, N.S., Link, W.A., & Hines, J.E. (2004) Tigers and their prey: predicting carnivore densities from prey abundance. Proceedings of the Natural Academy of Sciences, 101, 4854-4858.

Kinnaird, M.F., Sanderson, E.W., O'Brien, T.G., Wibisono, H.T., & Woolmer, G. (2003) Deforestation trends in a tropical landscape and implications for endnagered large mammals. Conservation Biology, 17, 245-256.

Klassen, A.W. (2006). Forest concession policy in Indonesia: Incentive or disincentive for sustainable forest management. Tropical Forest Foundation.

Koh, L.P. (2007) Impending disaster or sliver of hope for Southeast Asian forests? The devil may lie in the details. Biodiversity and Conservation, 16, 3935-3938.

Lambin, E.F. (1999) Monitoring forest degradation in tropical regions by remote sensing: some methodological issues. Global Ecology and Biogeography, 8, 191-198.

Laurance, W.F. (1991) Ecological correlates of extinction proneness in Australian tropical rain-forest mammals. Conservation Biology, 5, 79-89.

118

Laurance, W.F. (2007) Forest destruction in tropical Asia. Current Science, 93, 1544 - 1550.

Laurance, W.F. & Peres, C.A. (2006) Emerging threats to tropical forests. University of Chicago Press, Chicago.

Lee, S.M. & Chao, A. (1994) Estimating population size via sample coverage for closed capture-recapture models. Biometrics, 50, 88-97.

Leinbach, T.R. (1989) The Transmigration Programme in Indonesian National Development Strategy: Current Status and Future Requirements. Habitat International, 13, 81-93.

Lindenmayer, D.B. & Franklin, J.F. (2002) Conserving Biodiversity: A Comprehensive Multiscaled Approach. Island Press, Washington, DC.

Linkie, M., Achmad Haidir, I., Nugroho, A., & Dinata, Y. (2008) Conserving tigers Panthera tigris in selectively logged Sumatran forests. Biological Conservation, 141, 2410-2415.

Linkie, M., Chapron, G., Martyr, D.J., Holden, J., & Leader-Williams, N. (2006) Assessing the viability of tiger subpopulations in a fragmented landscape. Journal of Applied Ecology, 1-11.

Liu, J.G., Linderman, M., Ouyang, Z.Y., An, L., Yang, J., & Zhang, H.M. (2001) Ecological degradation in protected areas: the case of Wolong Nature Reserve for giant pandas. Science, 292, 98-101.

Macdonald, D.W., ed. (2006) Encyclopedia of mammals. Oxford University Press, Oxford.

MacKenzie, D.I., Nichols, J.D., Hines, J.E., Knutson, M.G., & Franklin, A.B. (2003) Estimating site occupancy, colonization, and local extinction when a species is detected imperfectly. Ecology, 84, 2200-2207.

MacKenzie, D.I., Nichols, J.D., Lachman, G.B., Droege, S., Royle, A.J., & Langtimm, C.A. (2002) Estimating site occupancy rates when detection probabilities are less than one. Ecology, 83, 2248-2255.

Maddox, T.M., Gemita, E., Wijamukti, S., & Selampassy, A. (2003). Pigs, Palms, People and Tigers. Zoological Society of London.

Maddox, T.M., Priatna, D., Gemita, E., & Selampassy, A. (2007). The conservation of tigers and other wildlife in oil palm plantations. Jambi Province, Sumatra, Indonesia. Zoological Society of London, London (UK).

Mattila, N., Kotiaho, J.S., Kaitala, V., & Komonen, A. (2008) The use of ecological traits in extinction risk assessments: A case study on geometrid moths. Biological Conservation, 141, 2322-2328.

McKinney, M.L. (1997) Extinction vulnerability and selectivity: combining ecological and paleontological views. Annual Review in Ecology and Systematics, 28, 495-516.

119

McLean, J. & Straede, S. (2003) Conservation, relocation and the paradigms of park and people management - a case study of Padampur villages and the Royal Chitwan National Park, Nepal. Society and Natural Resources, 16, 509-526.

McNeely, J.A. (1994) Lessons from the past: forests and biodiversity. Biodiversity and Conservation, 3, 3-20.

McNeely, J.A. & Scherr, S.J. (2002) Ecoagriculture: strategies to feed the world and save wild biodiversity. Island Press, Washington, DC.

Moilanen, A. (2004) SPOMSIM: software for stochastic patch occupancy models of metapopulation dynamics. Ecological Modelling, 179, 533-550.

Moilanen, A. & Nieminen, M. (2002) Simple connectivity measures in spatial ecology. Ecology, 83, 1131-1145.

Morton, D.C., DeFries, R.S., Shimabukuro, Y.E., Anderson, L.O., Arai, E., del Bon Espirito-Santo, F., Freitas, R., & Morisette, J. (2006) Cropland expansion changes deforestation dynamics in the southern Brazilian Amazon. Proceedings of the National Academy of Sciences USA, 103, 14637-14641.

Nasi, R., Koponen, P., Poulsen, J.G., Buitenzorgy, M., & Rusmantoro, W. (2008) Impact of landscape and corridor design on primates in a large-scale industrial tropical plantation landscape. Biodiversity and Conservation, 17, 1105-1126.

Nepstad, D.C., Verissimo, A., Alencar, A., Nobre, C., Lima, E., Lefebvre, P., Schlesinger, P., Potter, C., Moutinho, P., Mendoza, E., Cochrane, M., & Brooks, M. (1999) Large-scale impoverishment of Amazonian forests by logging and fire. Nature, 398, 505-508.

Noss, R.F. (1987) Corridors in real landscapes: a reply to Simberloff and Cox. Conservation Biology, 1, 159-164.

Noss, R.F. (1991). Landscape connectivity: different functions at different scales. In Landscape, linkages and biodiversity (ed W.E. Hudson), pp. 27-39. Defenders of Wildlife, Washington, D.C.

Nowak, R.M., ed. (1999) Walker's mammals of the world, 6th edn. John Hopkins University Press, London.

Nowell, K. & Jackson, P. (1996) Wild Cats: Status Survey and Conservation Action Plan IUCN/SSC Cat Specialist Group, IUCN, Gland, Switzerland.

Nyphus, P. & Tilson, R. (2004) Agroforestry, elephants, and tigers: balancing conservation theory and practice in human-dominated landscapes of Southeast Asia. Agriculture ecosystems and Environment, 104, 87-97.

O'Brien, T., Wibisono, H.T., & Kinnaird, M.F. (2003) Crouching tigers, hidden prey: Status of Sumatran tigers in the Bukit Barisan Selatan National Park, Sumatra, Indonesia. Animal conservation, 6, 131-139.

120

Olden, J.D., Poff, N.L., & Bestgen, K.R. (2008) Trait synergisms and the rarity, extirpation, and extinction risk of desert fishes. Ecology, 89, 847-856.

Owens, I.P.F. & Bennett, P.M. (2000) Ecological basis of extinction risk in birds: habitat loss versus human persecution and introduced predators. Proceedings of the Natural Academy of Sciences (USA), 97, 12144-12148.

Payne, J. & Francis, C.M. (1985) A field guide to the mammals of Borneo The Sabah Society, World Wildlife Fund Malaysia, Kuala Lumpur, Malaysia.

Peh, K.S., de Jong, J., Sodhi, N.S., Lim, S.L., & Yap, C.A. (2004) Lowland rainforest avifauna and human disturbance: persistence of primary forest birds in selectively logged forests and mixed-rural habitats of southern Peninsular Malaysia. Biological Conservation, 123, 489-505.

Phillips, O.L. (1997) The changing ecology of tropical forests. Biodiversity and Conservation, 6, 291-311.

Pimm, S.L. & Raven, P. (2000) Biodiversity - Extinction by numbers. Nature, 403, 843 - 845.

Pimm, S.L., Russell, G., Gittleman, J., & Brooks, T. (1995) The Future of Biodiversity. Science, 269, 347-350.

Potter, L. & Lee, J. (1998). Tree Planting in Indonesia: Trends, Impacts and Directions. Centre for International Forestry Research (CIFOR), Adelaide, Australia.

Prins, H.H.T. & Iason, G.R. (1989) Dangerous lions and nonchalant buffalo. Behaviour, 108, 262-296.

Prugh, L.R., Hodges, K.E., Sinclair, A.R.E., & Brashares, J.S. (2008) Effect of habitat area and isolation on fragmented animal populations. Proceedings of the National Academy of Sciences USA, 105, 20770-20775.

Purvis, A., Agapow, P.M., Gittleman, J.L., & Mace, G.M. (2000a) Nonrandom extinction and the loss of evolutionary history. Science, 288, 328-330.

Purvis, A., Gittleman, J.L., Cowlishaw, G., & Mace, G.M. (2000b) Predicting extinction risk in declining species. Proceedings of the Royal Society B, 267, 1947-1952.

R Development Core Team (2008) R: a language and environment for statistical computing (Version 2.7.2). R Foundation for Statistical Computing, Vienna, Austria.

Ranganathan, J., Chan, K.M.A., Karanth, K.U., & Smith, J.L.D. (2008) Where can tigers persist in the future? A landscape-scale, density-based population model for the Indian subcontinent. Biodiversity and Conservation, 141, 67-77.

Reunanen, P., Nikula, A., Monkkonen, M., Hurme, E., & Nivala, V. (2002) Predicting occupancy for the Siberian flying squirrel in old-growth forest patches. Ecological Applications, 12, 1188-1198.

121

Ricketts, T.H. (2001) The Matrix Matters: Effective Isolation in Fragmented Landscapes. American Naturalist, 158, 87-99.

Ripley, B.D. (1996) Pattern recognition and neural networks. Cambridge University Press, Cambridge, UK.

Rodrigues, A.S.L., Andelman, S.J., Bakarr, M.I., Boitani, L., Brooks, T.M., Cowling, R.M., Fishpool, L.D.C., da Fonseca, G.A.B., Gaston, K.J., Hoffmann, M., Long, J.S., Marquet, P.A., Pilgrim, J.D., Pressey, R.L., Schipper, J., Sechrest, W., Stuart, S.N., Underhill, L.G., Waller, R.W., Watts, M.E.J., & Yan, X. (2004) Effectiveness of the global protected area network in representing species diversity. Nature, 428, 640-643.

Safi, K. & Kerth, G. (2004) A comparative analysis of specialization and extinction risk in temperate-zone bats. Conservation Biology, 18, 1293-1303.

Sala, O.E., Chapin, F.S., Armesto, J.J., Berlow, E., Bloomfield, J., Dirzo, R., Huber- Sanwald, E., Huenneke, L.F., Jackson, R.B., Kinzig, A., Leemans, R., Lodge, D.M., Mooney, H.A., Oesterheld, M., Poff, N.L., Sykes, M.T., Walker, B.H., Walker, M., & Wall, D.H. (2000) Glabal biodiversity scenarios for the year 2100. Science, 287, 1770-1774.

Sanderson, E., Forrest, J., Loucks, C., Ginsberg, J., Dinerstein, E., Seidensticker, J., Leimgruber, P., Songer, M., Heydlauff, A., O'Brien, T., Bryja, G., Wikramanayake, E., & Klenzendorf, S. (2006) Setting Priorities for the Conservation and Recovery of Wild Tigers: 2005-2015. The Technical Assessment. WCS, WWF, Smithsonian, NFWF-STF, Washington (DC).

Sandker, M., Suwarno, A., & Campbell, B.M. (2007) Will forests remain in the face of oil palm expansion? Simulating change in Malinau, Indonesia. Ecology and Society, 12.

Santiapillai, C. & Ramono, W.S. (1987). Tiger numbers and habitat evaluation in Indonesia. In Tigers of the world: the biology, biopolitics, management and conservation of an endangered species (eds R.L. Tilson & U.S. Seal). Noyes Publications, New Jersey.

Sawyer, J. (1993). Plantations in the tropics. IUCN, Gland, Switzerland.

Schaller, G.B. (1967) The Deer and the Tiger: A study of wildlife in India University of Chicago Press, Chicago.

Scott, J.M., Heglund, P.J., Morrisn, M.L., Haufler, J.B., Raphael, M.G., Wall, W.A., & Samson, F.B. (2002) Predicting Species Occurrences Island Press, Washington, D.C.

Seidensticker, J. & McDougal, C. (1993) Tiger predatory behaviour, ecology and conservation. Symposium of the Zoological Society of London, 65, 105-125.

Shen, T.J. (2003) Prediction of Biodiversity, National Tsing-Hua University, Hsin-Chu, Taiwan.

122

Shultz, S., Bradbury, R.B., Evans, K.L., Gregory, R.D., & Blackburn, T.M. (2005) Brain size and resource specialization predict long-term population trends in British birds. Proceedings of the Royal Society B, 272, 2305-2311.

Sodhi, N.S. & Brook, B.W. (2008) Fragile Southeast Asian biotas. Biological Conservation, 141, 883-884.

Sodhi, N.S., Lee, T.M., Koh, L.P., & Brook, B.W. (2009) A Meta-Analysis of the Impact of Anthropogenic Forest Disturbance on Southeast Asia's Biotas. Biotropica, 41, 103-109.

Srivastava, D.S. (1999) Using local-regional richness plots to test for species saturation: pitfalls and potentials. Journal of Animal Ecology, 68, 1-16.

Stearns, S.C. (1983) The impact of size and phylogeny on patterns of covariation in the life history traits of mammals. Oikos, 41, 173-187.

Su, J.C., Debinski, D.M., Jakubauskas, M.E., & Kindscher, K. (2004) Beyond Species Richness: Community Similarity as a Measure of Cross-Taxon Congruence for Coarse-Filter Conservation. Conservation Biology, 18, 167-173.

Sunquist, M. & Sunquist, F. (2002) Wild Cats of the World University of Chicago Press, London.

Sunquist, M.E. (1981) The social organization of tigers (Panthera tigris) in Royal Chitwan National Park, Nepal. Smithsonian Contributions to Zoology, 336, 1-98.

Suyanto, S. (2007) Underlying Cause of Fire: Different form of land tenure conflicts in Sumatra. Mitigation and Adaptation Strategies for Global Change, 12, 67-74.

Terborgh, J., Lopez, L., Nunez, P., Rao, M., Shahabuddin, G., Orihuela, G., Riveros, M., Ascanio, R., Adler, G.H., Lambert, T.D., & Balbas, L. (2001) Ecological meltdown in predator-free forest fragments. Science, 294, 1923-1926.

Therneau, T.M. & Atkinson, B. (2008) rpart: Recursive Partitioning. R port by Brian Ripley. http://mayoresearch.mayo.edu/mayo/research/biostat/splusfunctions.cfm

Tilman, D., Fargione, J., Wolff, B., D'Antonio, C., Dobson, A., Howarth, R., Schindler, D., Schlesinger, W.H., Simberloff, D., & Swackhamer, D. (2001) Forecasting agriculturally driven global environmental change. Science, 292, 281 - 284.

Tilman, D., May, R.M., Lehman, C.L., & Nowak, M.A. (2004) Habitat destruction and the extinction debt. Nature, 371, 65-66.

Tomich, T.P., Fagi, A.M., de Foresta, H., Michon, G., Murdiyarso, D., Stolle, F., & Van Noorwijk, M. (1998) Indonesia's Fires: Smoke As a Problem, Smoke As a Symptom. Agroforestry Today, 10, 4-7.

Tyre, A.J., Tenhumberg, B., Field, S.A., Niejalke, D., Parris, K., & Possingham, H.P. (2003) Improving precision and reducing bias in biological surveys: estimating false-negative error rates. Ecological Applications, 13, 1790-1801.

123

UN (1999). World Population Prospects: The 1998 Revision, vol. I, Comprehensive Tables. UN Department of Economic and Social Affairs, Population Division, New York.

White, G.C. & Burnham, K.P. (1999) Program MARK: Survival estimation from populations of marked animals. Bird Study, 46, 120-139.

Whitten, T., Damanik, S.J., Anwar, J., & Hisyam, N. (2000) The Ecology of Sumatra Periplus, Singapore.

Wiens, J.A. (1996). Wildlife in patchy environments: Metapopulations, mosaics, and management. In Metapopulations and wildlife conservation (ed D.R. McCullough). Island Press, Washington D.C.

Wikramanayake, E., McKnight, M., Dinerstein, E., Joshi, A., Gurung, B., & Smith, D. (2004) Designing a conservation landscape for tigers in human-dominated environments. Conservation Biology, 18, 839-844.

Wikramanayake, E.D., Dinerstein, E., Robinson, G., Karanth, K.U., Rabinowitz, A., Olson, D., Matthew, T., Hedao, P., Connor, M., Hemley, G., & Bolze, D. (1998) An ecology-based method of defining priorities for large mammal conservation: the tiger as a case study. Conservation Biology, 12, 865-878.

Williams, B.K., Nichols, J.D., & Conroy , M.J. (2002) Analysis and Management of Animal Populations. Academic Press, San Diego, CA.

With, K.A. & Crist, T.O. (1995) Critical Thresholds in Species Responses to Landscape Structure. Ecology, 76, 2446-2459.

Wittemyer, G., Elsen, P., Bean, W.T., Burton, C.O., & Brashares, J.S. (2008) Accelerated human population growth at protected area edges. Science, 321, 123-126.

Wood, B.J. & Chung, G.F. (2003) A critical review of the development of rat control in Malaysian agriculture since the 1960s. Crop Protection, 22, 445-461.

Woodroffe, R. & Ginsberg, J.R. (1998) Edge effects and the extinction of populations inside protected areas. Science, 280, 2126-2128.

World Bank (1994). Indonesia Transmigration Program: A Review of Five Bank Supported Projects, Washington, D.C.

Wright, S.J. & Muller-Landau, H.C. (2006) The future of tropical forest species. Biotropica, 38, 287-301.

124