Biogeographic Patterns of Lemur Species Richness and ... · Travis Scott Steffens Doctor of Philosophy Department of Anthropology University of Toronto 2017 Abstract Determining the

Biogeographic Patterns of Lemur Species Richness and Occurrence in a Fragmented Landscape

by

Travis Scott Steffens

A thesis submitted in conformity with the requirements for the degree of PhD

Department of Anthropology University of Toronto

© Copyright by Travis Scott Steffens 2017

ii

Biogeographic Patterns of Lemur Species Richness and

Occurrence in a Fragmented Landscape

Travis Scott Steffens

Doctor of Philosophy

Department of Anthropology University of Toronto

2017

Abstract

Determining the factors that affect species richness and occurrence is vital to the study of

primate biogeography. In this study, I investigate the biogeographic patterns of a lemur

community within a fragmented landscape in Ankarafantsika National Park, Madagascar. The

landscape consists of 42 deciduous dry forest fragments ranging in size from 0.23 to 117.7 ha. I

conducted a total of 1218 surveys for lemurs between June and November 2011 in the 42

fragments. I recorded a total of 1023 individual or group sightings of six species. I conducted

vegetation sampling in 38 of 42 forest fragments. I measured human disturbance within each

fragment and determined fragment isolation and proximity to human settlements. I explored

biogeographic patterns of lemur species richness and occurrence using the species-area

relationship, metapopulation dynamics, and landscape ecology. I found that lemurs in a

fragmented landscape show a species-area relationship in the form of a convex power model. I

did not find a sigmoidal pattern for the species-area relationship and I found no evidence of a

“small island effect.” Human disturbance and tree height also influence species richness, but it is

unclear how. Lemur species form different metapopulations within the same landscape.

Metapopulation dynamics suggest that area was a stronger factor determining individual lemur

iii

species occurrence than fragment isolation. However, for Microcebus species, area seems to

have less influence than for other species (Cheirogaleus medius and Eulemur fulvus). I

investigated the landscape ecology of lemur species occurrence and found species-specific scale

responses to habitat amount. Area predicted species occurrence for C. medius and M. murinus,

but not for M. ravelobensis. M. ravelobensis occurrence may be mediated by factors other than

area, such as dispersal ability and edge tolerance. My study shows the importance of a multi-

scale approach to lemur biogeography and how it is critical for understanding how lemur species

respond to high amounts of forest loss and fragmentation.

iv

Acknowledgments This dissertation would not have been possible without the incredible support of many people,

institutions, and funding agencies. I would like to thank my advisor Shawn Lehman. His

feedback and support helped turn a silly idea into a dissertation. Shawn has given me every

opportunity to seek my own path. He provided me with the necessary encouragement and

advice, allowing me to complete an expensive and logistically challenging project. Shawn has

continuously supported my academic career and helped me develop as a researcher.

I would like to thank the Department of Palaeontology at the University of Antananarivo. I am

very grateful for the support and assistance from the staff of the Madagascar Institut pour la

Conservation des Ecosystèmes Tropicaux (MICET/ICTE), including the director Benjamin

Andriamihaja, Benji Randrianambina, Jean Rakotoarison, Haza Rasoanaivo, and Tina.

I would like to thank Mamy Razafitsalama and Rindra Rakotoarvony. Both their efforts were

indispensable for this research project. Mamy especially provided unwavering support for my

project. He became a leader and took responsibility of many elements of my research. He was

not only a great field assistant but also an amazing project manager and friend. Of all the people

I have met in Madagascar, Mamy is the single most intelligent, personable, and hard working. I

look forward to a long and illustrious collaboration with Mamy on current and future research

and conservation efforts.

I would like to thank the tireless efforts of the residents of Andranohobaka and Maevatanimbary

whose help was essential to the completion of this project. The community members provided

our food, water, and personnel, without which this project could not have survived past the first

week. Specifically, I would like to thank Jean Paul, Velontsara, Nada, Rollin, and also Alpha

(who is from Ambodimanga village). I would like to thank Madagascar National Park and their

staff who helped me to conduct my research in a remote portion of Ankarafantsika National

Park. Specifically, I would like to thank the head of research, Jacqueline Razaiarimanana, the

head of conservation and research, Justin Rakotoarimanana, as well as the Park Director at the

time, Rene Razafindrajery.

v

I send my warmest thanks to Fanja and her family at La Maison du Pyla, whose comforting

home and amazing food was a welcome respite to the challenges faced in the field.

I would like to thank my friends and colleagues who helped support me and provide invaluable

assistance for this dissertation. Specifically, I am grateful for the help and support of Jarred

Heinrich, who basically taught me how to use R and was a sounding board for any and all ideas.

I thank Yuri Fraser for his immense efforts in Madagascar. I would also like to thank Vincent

Dorie, Kim Valenta, Abigail Ross, Ryan Burke, Steve Miller, Jamie Sharpe, and Greg Bridgett

for the varied and important contributions to this dissertation. I would like to thank my

committee members for their support and feedback for my dissertation.

I would like to thank Ken and Sheena McGoogan for their immeasurable support during my

dissertation.

I would like to thank my mother, Cecile Patterson. She gave me life, she provided me with

support through the most trying times, she gave me every opportunity to succeed, and she

encouraged me to live my dream.

Finally, I would like to thank my partner Keriann McGoogan who not only helped me collect

data, edit manuscripts, negotiate contracts, navigate through the wilderness, and help answer all

my questions, but also was there for me in a way no other person could be. Her care and support

helped me go from idea to dissertation and she is the biggest reason for any success I receive.

Financial support for this work was provided by the following intuitions and organizations:

Sigma Xi Grants in Aid of Research, American Society of Primatology (Conservation Small

Grant), Calgary and Edmonton Valley Zoos, Primate Conservation, Inc., The Explorers Club

(Exploration Fund), University of Toronto School of Graduate Studies Travel Grant, the Ontario

Government (Ontario Graduate Scholarship), and the Natural Sciences and Engineering

Research Council of Canada (Discovery Grant).

vi

Table of Contents

Acknowledgments .......................................................................................................................... iv

Table of Contents ........................................................................................................................... vi

List of Tables ................................................................................................................................. xi

List of Figures ............................................................................................................................... xii

List of Appendices ....................................................................................................................... xiii

Chapter 1: Area, Patches, Landscapes and the Determinants of Lemur Species Occurrence .......................................................................................................................................1

1.1 Introduction ..........................................................................................................................1

1.2 Background on Madagascar and Lemurs .............................................................................1

1.3 Habitat Loss and Fragmentation ..........................................................................................3

1.4 Species Richness and Occurrence ........................................................................................3

1.5 Patch and Landscape Effects and Their Relation to Habitat Loss and Fragmentation ........4

1.6 Patch-Level Effects on Primate Communities: The Species-Area Relationship .................5

1.7 Patch-Level Effects on Primate Occurrence: Metapopulation Dynamics ...........................6

1.8 Landscape-Level Effects on Primate Occurrence ................................................................6

1.9 A Note on the Format of the Thesis .....................................................................................7

1.10Dissertation Goals ................................................................................................................7

Chapter 2: Species-Area Relationships of Lemurs in a Fragmented Landscape in Madagascar ......................................................................................................................................9

2.1 Introduction ..........................................................................................................................9

2.2 Choosing a Species-area Model ...........................................................................................9

2.3 Types of Species-Area Curves ...........................................................................................13

2.3.1 Convex Models without an Asymptote ..................................................................13

2.3.2 Convex Models with an Asymptote .......................................................................14

2.3.3 Sigmoidal Models without an Asymptote ..............................................................14

vii

2.3.4 Sigmoidal Models with an Asymptote ...................................................................15

2.4 Processes Determining the Species-Area Pattern ..............................................................15

2.5 Other Non-Area Factors Affecting Primate Species Richness ..........................................16

2.6 Species-Area Relationships in Primates ............................................................................18

2.7 Testing for a Species-Area Relationship ............................................................................19

2.8 Justification ........................................................................................................................19

2.9 Goals ..................................................................................................................................20

2.10Methods..............................................................................................................................21

2.10.1 Study Site and Study Species .................................................................................21

2.10.2 Question 1: What is the SAR Pattern? ...................................................................24

2.10.3 Question 2: What Other Non-Area Factors Affect Lemur Species Richness? ......28

2.10.4 Statistical Analysis .................................................................................................29

2.11Results ................................................................................................................................32

2.11.1 Fragment Area and Survey Results ........................................................................32



2.12Discussion ..........................................................................................................................37



2.12.3 Suggestions for Conservation ................................................................................41

2.13Conclusion .........................................................................................................................42

Chapter 3: Population Dynamics of Lemurs in a Fragmented Landscape in Madagascar .......43

3.1 Introduction ........................................................................................................................43

3.2 Types of Metapopulations ..................................................................................................44

3.3 Metapopulation Models .....................................................................................................46

viii

3.4 Dispersal ............................................................................................................................48

3.5 Effect of Additional Variables within Incidence Function Models ...................................49

3.6 Simulating Metapopulation Dynamics Over Time ............................................................50

3.7 Metapopulation Dynamics: Forest Loss and Fragmentation Effects on Primate Occurrence .........................................................................................................................50

3.8 Justification ........................................................................................................................52

3.9 Goal ....................................................................................................................................53

3.10Methods..............................................................................................................................55


3.10.2 Question 1: Do Lemur Species Form Metapopulations? .......................................57

3.10.3 Question 2: What are the Separate Effects of Area (Extinction Risk) and Connectivity/Isolation (Colonization Potential) within a Lemur Metapopulation? ....................................................................................................62

3.10.4 Question 3: Within Simulated Metapopulations Over Time, How Do Area and Connectivity/Isolation Affect Occurrence? What are the Conservation Implications? ..........................................................................................................63

3.11Results ................................................................................................................................64


3.11.2 Question 2: What are the Separate Effects of Area (Extinction Risk) and Connectivity/Isolation (Colonization Potential) on a Lemur Metapopulation? .....67

3.11.3 Question 3: Within Simulated Metapopulations Over Time, How Do Area and Connectivity/Isolation Affect Occurrence? .....................................................68

3.12Discussion ..........................................................................................................................71


3.12.2 Question 2: What are the Separate Effects of Area (Extinction Risk) and Connectivity/Isolation (Colonization Potential) on a Lemur Metapopulation? .....76

3.12.3 Question 3: Within Simulated Metapopulations Over Time, How Do Area and Connectivity/Isolation Affect Occurrence? What are the Conservation Implications? ..........................................................................................................78

3.13Suggestions for Conservation ............................................................................................79

ix

3.14Conclusion .........................................................................................................................79

Chapter 4: Lemur Species-Specific Scale Responses to Habitat Loss in Fragmented Landscapes in NW Madagascar .....................................................................................................81

4.1 Introduction ........................................................................................................................81

4.2 Landscape-Level Effect of Habitat Amount on Species Occurrence. ...............................82

4.3 Species-Specific Scale Responses to Habitat Amount. .....................................................84

4.3.1 Scale and Landscapes. ...........................................................................................84

4.3.2 “Scale of Effect.” ...................................................................................................84

4.3.3 Species-Specific “Scale of Effect” .........................................................................85

4.4 Justification ........................................................................................................................86

4.5 Goal ....................................................................................................................................86

4.6 Methods..............................................................................................................................87


4.6.2 Question 1: What is the Scale of Species Response to Habitat Amount? .............87

4.6.3 Question 2: What is the Landscape-Level Effect of Habitat Amount on Species Occurrence? ..............................................................................................92

4.7 Results ................................................................................................................................93

4.7.1 Description of Spatial Autocorrelation Among Landscapes. ................................93


4.7.3 Question 2: What is the Landscape-Level Effect of Habitat Amount on Species Occurrence? ..............................................................................................96

4.8 Discussion ..........................................................................................................................96


4.8.2 Question 2: What is the Effect of Habitat Amount on Species-Specific Occurrence of Lemurs at the Landscape-Level? .................................................101

4.9 Suggestions for Conservation ..........................................................................................102

4.10Conclusion .......................................................................................................................102

x

Chapter 5: Patch and Landscape Determinants of Lemur Species Richness and Occurrence ...................................................................................................................................104

5.1 Conclusion .......................................................................................................................104

5.2 Summary of Results .........................................................................................................104

5.2.1 Species-Area Relationships in a Lemur Community ...........................................104

5.2.2 Metapopulation Dynamics of Lemur Species ......................................................106

5.2.3 Landscape Effects on Lemur Species ..................................................................107

5.3 Implications ......................................................................................................................108

5.3.1 Species-Area Relationship ...................................................................................108

5.3.2 Metapopulation Dynamics ...................................................................................108

5.3.3 Landscape Effects ................................................................................................109

5.4 Directions for Future Research ........................................................................................110

5.4.1 Species-Area Relationships .................................................................................110

5.4.2 Metapopulation Dynamics ...................................................................................111

5.4.3 Landscape Ecology ..............................................................................................112

5.5 Implications for Lemur Conservation ..............................................................................113

5.6 Significance ......................................................................................................................114

References ....................................................................................................................................116

Appendices ...................................................................................................................................131

xi

List of Tables Table 2.1: Primate Species in Ankarafantsika National Park. ..................................................... 23

Table 2.2: Survey Data of 42 Fragments in a 3,000 ha Fragmented Landscape. ........................ 26

Table 2.3: 10 Candidate Species-Area Models. ........................................................................... 28

Table 2.4: Potential Influence of 10 Predictor Variables on Species Richness. .......................... 31

Table 2.5: Fragment Characteristics and Species Richness. ........................................................ 33

Table 2.6: Fitted Parameters of 10 Candidate Species-Area Models Using Non-linear Least

Squares Regression for Primate Species in the 42 Fragments. .................................................... 34

Table 2.7: Non-linear Least Squares Regression Model Selection of Species-Area Models. ..... 35

Table 2.8: Comparison of Seven Generalized Additive Models Using Poisson Link Function

Predicting Lemur Species Richness. ............................................................................................ 37

Table 3.1: Primate Species in Ankarafantsika National Park Found within Study Site. ............. 55

Table 3.2: Lemur Patch Occupancy in a Fragmented Landscape. ............................................... 64

Table 3.3: Metapopulation Models of Six Lemur Species in 42 Fragments in a Fragmented

Landscape. ................................................................................................................................... 66

Table 3.4: T-test Results for the Mean Probability of Occurrence (Ji) for Four Lemur Species in

the Largest Versus Smallest Fragments. ...................................................................................... 67

Table 3.5: T-test Results for the Mean Probability of Occurrence (Ji) for Four Lemur Species in

the Most Versus Least Connected Fragments. ............................................................................. 68

Table 4.1: Study Species Characteristics and Landscape Scale Sizes ......................................... 88

xii

List of Figures Figure 2.1: Sigmoidal Species-Area Model. ................................................................................ 10

Figure 2.2: Convex Species-Area Models. .................................................................................. 12

Figure 2.3: Study Site and Distribution of Forest within Madagascar. ....................................... 22

Figure 2.4: Study Site. ................................................................................................................. 24

Figure 2.5: Five Competing Species-Area Models. ..................................................................... 35

Figure 2.6: Hierarchical Partitioning Model. ............................................................................... 36

Figure 3.1: Five Types of Metapopulations. ................................................................................ 45

Figure 3.2: Probability of Occurrence Among Patches for Four Lemur Species in a Fragmented

Landscape. ................................................................................................................................... 65

Figure 3.3: Simulations of Metapopulation Dynamics for Four Lemur Species Over 200 Time

Steps in a Fragmented Landscape When the Five Largest and Five Smallest Fragments Are

Removed. ..................................................................................................................................... 69

Figure 3.4: Simulations of Metapopulation Dynamics for Four Lemur Species Over 200 Time

Steps in a Fragmented Landscape When the Five Most Connected and Five Least Connected

Fragments Are Removed. ............................................................................................................ 70

Figure 4.1: DigitalGlobe Satellite Image of Field Site. ............................................................... 90

xiii

List of Appendices Appendix A: Cheirogaleus medius Occurrence Responses to Amount of Habitat within 10

Landscape Scales. ...................................................................................................................... 131

Appendix B: Microcebus murinus Occurrence Responses to Amount of Habitat within 10


Appendix C: Microcebus ravelobensis Occurrence Responses to Amount of Habitat within 10


1

Chapter 1: Area, Patches, Landscapes and the Determinants of Lemur Species Occurrence

1.1 Introduction We are living through the sixth great extinction event in the earth’s history (Ceballos et al.

2015). There have been 903 extinctions reported in modern history, including two primate

species (Palaeopropithecus ingens and Xenothrix mcgregori) (IUCN, 2016). While

comparatively few primates have become extinct, 263 of 423 extant primate species are

currently threatened with extinction (IUCN, 2016) and one group—lemurs—is the most

threatened group of large animal taxa in the world (Schwitzer et al. 2014). As with most other

species (Brooks et al. 2002; Hanski, 2005; Groom et al. 2006), the main threat to primates,

including lemurs, is habitat loss and fragmentation (Schwitzer et al. 2014; IUCN, 2016).

However, scientists do not understand how primate species respond to these phenomena

(Chapman et al. 2003; Marsh, 2003; Arroyo-Rodriguez & Dias, 2010). More research is needed

on a wide range of primate taxa to identify the factors that influence primate responses to habitat

loss and fragmentation, and why so many primates are currently threatened with extinction. To

complicate matters, some species appear to fare well under certain habitat loss and

fragmentation scenarios while others do not (Boyle et al. 2013). Even within some species,

habitat loss and fragmentation responses vary (Arroyo-Rodriguez & Dias, 2010). Therefore, it is

crucial to determine why some primate species are impacted more than others. Understanding

primate extinction and responses to habitat loss and fragmentation will help us understand

primate evolution, ecology, and behavior, and will ultimately help us to conserve primate

species.

1.2 Background on Madagascar and Lemurs Madagascar is an island off the south east coast of mainland Africa. The fourth largest island in

the world, at approximately 587,000 km2 in size, it is home to approximately 23.6 million

people (World Bank, 2014). The island was attached to the mainland of Africa approximately

155 million years ago, and to the Indian subcontinent approximately 87 million years ago

(Wells, 2003). Therefore, Madagascar features a mix of African and Asian flora and fauna, plus

a large complement of unique species found only on the island (Goodman & Beanstead, 2005).

2

Madagascar’s biodiversity is one of the most diverse and endemic of any country (Goodman &

Beanstead, 2005) and Madagascar is the only country where lemurs are naturally found.

Lemurs are a group of Strepsirrhine primates endemic to Madagascar, and are themselves one of

the most diverse groups of primates. There are 99 recognized lemur species and 103 taxa

(species and subspecies; Schwitzer et al. 2013). Lemurs vary dramatically: in size from the

diminutive mouse lemur (40 g) to the Indri (9.5 kg); in diet, including fauni-frugivores,

frugivores, and folivores; and in, activity pattern, such as diurnal, nocturnal and cathemeral

species (Mittermeier et al. 2010). Lemurs occupy a wide variety of habitat types including, but

not limited to, cloud forest, montane forest, moist lowland tropical forest, tropical deciduous dry

forest, and spiny forest (Mittermeier et al. 2010).

Madagascar is an ideal country in which to study the impact of habitat loss and fragmentation on

primate species. Habitats exploited by lemurs in Madagascar are fragmented and disappearing

rapidly. Since the 1950s, Madagascar has lost approximately 40% of its total forest (Harper et

al. 2007), which has been converted to agricultural fields or deforested grasslands. Therefore,

fragments of varying size surround much of the remaining continuous forest in Madagascar

(Harper et al. 2007). Forest loss and fragmentation in Madagascar are largely a result of small-

scale forest removal for rice production, and slash-and-burn agriculture to create pasture for

grazing cattle (Gade, 1996; Bloesch, 1999; Harper et al. 2007). Malagasy tropical dry forest in

particular is highly fragmented due to increased small-scale deforestation from fire along forest

edges in the 1990s (Harper et al. 2007). This pattern of deforestation in NW Madagascar has left

multiple fragments of varying sizes throughout the landscape and along the perimeter of

continuous forest tracts, such as Ankarafantsika National Park. Additionally, lemur species in

dry forest are more prone to variables that reduce their numbers than species in other forest

types (Ganzhorn, 1997). Deforestation through habitat loss and fragmentation is the leading

threat to lemurs in Madagascar (Schwitzer et al. 2014). Indeed, lemurs are the most endangered

group of animals in the world. Of the 103 taxa currently recognized, 94% are threatened with

extinction (Schwitzer et al. 2014). It is imperative that we determine how lemurs respond to

habitat loss and fragmentation in order to prevent this unique group of primates from going

extinct.

3

1.3 Habitat Loss and Fragmentation Habitat loss is simply the removal of habitat from a landscape, while habitat fragmentation

involves splitting habitat into smaller, isolated fragments (Fahrig, 2003). Habitat loss and

fragmentation effects are inextricably linked because habitat fragmentation cannot occur without

some habitat loss (McGarigal & Cushman, 2002). Habitat loss and fragmentation are landscape-

level phenomena (McGarigal & Cushman, 2002; Fahrig, 2003). Yet most studies on habitat loss

and fragmentation are conducted at the patch-level (McGarigal & Cushman, 2002; Fahrig,

2003). A landscape is an area that is heterogeneous with respect to a character of interest

(Turner et al. 2001), while a patch is a discrete measurable portion of habitat. You can observe

habitat loss at either a landscape- or patch-level. However, you can only appropriately observe

habitat fragmentation at a landscape-level (Fahrig, 2003).

Based on a great deal of research, we know that primate species richness and occurrence is

impacted by habitat loss and fragmentation (Gray et al. 2010; Lawes et al. 2000; Chapman et al.

2003; Rodriguez-Toledo et al. 2003; Anzures-Dadda & Manson, 2007; Arroyo-Rodríguez et al.

2008; Arroyo-Rodriguez & Dias, 2010; Boyle & Smith, 2010; Marshall et al. 2010; Thorton et

al. 2011; Arroyo-Rodríguez et al. 2013a). For example, Rodriguez-Toledo et al. (2003) found

that habitat loss had a strong impact on Alouatta palliata in a highly fragmented landscape in

Mexico because A. palliata occurred in only 25% of remaining habitat fragments and were only

found in all of the largest (>10 ha) fragments. In the chapters to come, I will use three different

methods—species-area relationship, metapopulation dynamics, and landscape ecology—to

identify variables impacting both lemur community and species responses to habitat loss and

fragmentation in northwestern Madagascar.

1.4 Species Richness and Occurrence Understanding the factors that determine species richness (the number of species) and

occurrence (presence/absence of a species) is critical to understanding species ecology. A

species’ ability to occur in an area will relate to many aspects of their evolution, life history,

ecology, and behavior. Knowledge of species richness and occurrence can ultimately be used to

plan and evaluate conservation measures and understand biogeography. Researchers found that

area plays a major role in determining species richness and occurrence in numerous taxa,

including primates (Lomolino, 2000; Harcourt & Doherty, 2005; Whittaker & Triantis, 2012).

4

With habitat loss and fragmentation increasing worldwide (Butchart et al. 2010), assessing the

role of area as a determinant of species richness and occurrence is crucial. It is also important to

determine how area impacts species richness and occurrence at both patch- and landscape-

levels.

1.5 Patch and Landscape Effects and Their Relation to Habitat Loss and Fragmentation

I define a patch as a discrete, measurable portion of habitat that is used by a species for

acquiring resources necessary for life. I define a fragment as a discrete portion of habitat that

has been separated from a larger whole. Patches can be considered fragments if they were

separated from a larger whole, and a fragment may be made up of more than one patch.

However, often the two terms are used synonymously in ecological literature. Research

investigating patch-level effects considers a patch or habitat fragment as the unit of analysis.

Comparing patches with different attributes can help us determine how those attributes affect

communities and species. Using a patch-level analysis is convenient because patches are easily

defined, measured, surveyed, and a useful unit for conservation management.

There is considerable debate in the ecological literature surrounding whether to use a patch- or

landscape-level study to determine the impact of habitat loss and fragmentation on species

occurrence and richness. The process of habitat loss, and habitat fragmentation are now

understood to be landscape-level phenomena (McGarigal & Cushman, 2002; Fahrig, 2003).

Therefore, many researchers have noticed the limitations of using only patch attributes to

evaluate what determines species richness and occurrence (McGarigal & Cushman, 2002;

Fahrig, 2003). More commonly, researchers are employing a landscape-level approach instead

of a patch-level approach to investigate how attributes of heterogeneous landscapes affect

species richness and occurrence. Although a landscape-level approach aligns with the scale of

the process of habitat loss and fragmentation, it introduces logistical and scale issues that must

be considered for effective study design. Logistical issues are a major factor inhibiting research

using a landscape-level approach because it is often necessary to use larger areas for analysis

using a landscape- compared to a patch-level study. To compare landscape attributes in a

landscape-level study researchers need to measure and survey multiple landscapes and

determine the appropriate measurement scale for a landscape (Turner et al. 2001; McGarigal &

Cushman, 2002; Turner, 2005). Some researchers have suggested that responses to habitat loss

5

and fragmentation at a landscape-level are species-specific (Jackson & Fahrig, 2012). More

specifically the scale by which species’ respond or the “scale of effect” is species-specific

(Jackson & Fahrig, 2012). Therefore, it can be difficult to determine the size of a landscape a

priori. Both patch- or landscape-level approaches can yield important information and the

choice of what level to use is determined by many factors including the research questions,

logistical issues, and management implications. Thus, it is appropriate to use a multi-scale

approach when logistic issues and lack of information are present.

1.6 Patch-Level Effects on Primate Communities: The Species-Area Relationship

One of the oldest and most common approaches to assess what impacts species richness is to use

the species-area relationship (Lomolino, 2000; Whittaker & Triantis, 2012). The species-area

relationship describes a pattern of increasing species richness with increasing habitat area

(Lomolino, 2000). Species-area relationships are so common that the relationship is considered

axiomatic in biogeography (Whittaker & Triantis, 2012). There are two main types of species-

area relationships. The first type includes the study of the species-area relationship within

sample areas that make up a greater whole (Lomolino, 2000; Tjørve, 2003). In sample area

studies, researchers sample and survey portions of habitat within greater and greater areas to

assess how many species occur within each size of sampled area. Sample areas can be nested

because a species can occur within multiple samples due to species movement between areas.

The second type of species-area relationship is the species-area relationship in isolates, such as

islands or habitat fragments (Lomolino, 2000; Tjørve, 2003). In isolate studies, the area

surrounding an isolate is considered hostile and limits the ability of a species to move between

isolates (Lomolino, 2000; Ricketts, 2001; Tjørve, 2003). Studying the species-area relationship

in isolates has led to large body of theory, such as the island biogeography theory (MacArthur &

Wilson 1967), and has improved our understanding of how communities respond to habitat loss.

Researchers have found the species-area relationship in most taxa including primates (Reed &

Fleagle, 1995; Cowlishaw, 1999; Cowlishaw & Dunbar, 2000; Lehman, 2004; Harcourt &

Doherty, 2005; Marshall et al. 2010). For example, Reed and Fleagle (1995) found that the

number of primate species increased with increasing area of tropical rainforest at continental

scales. Similarly, I found during my pilot study for this dissertation that lemur species richness

was linearly related to the log-area of habitat fragments (Steffens & Lehman, 2013). Looking at

the pattern and process of the species-area relationship in lemurs, a highly endangered group of

6

primates (Schwitzer et al. 2014), at a larger scale will give us a better understanding of the

community-level response to habitat loss.

1.7 Patch-Level Effects on Primate Occurrence: Metapopulation Dynamics

The species-area relationship operates at the community-level and is species irrelevant, meaning

it is concerned only with the number of species in an area, rather than species composition or

identity. But to understand how species occurrence is impacted by habitat loss and

fragmentation, we need to investigate how individual species are impacted by these factors. One

way to determine the impact is to use metapopulation dynamics (Hanski, 1994a; Hanski, 1999;

Hanski & Ovaskainen, 2003). A metapopulation is a collection of many local populations of a

species that are connected to one another through dispersal (Levins, 1969). Metapopulation

models rely on the existence of measurable, discrete habitat patches such as habitat fragments

(Hanski & Ovaskainen, 2003). Metapopulation dynamics is a theoretical framework that is used

to describe how species become spatially arranged within a landscape and is particularly useful

for understanding the impact of habitat fragmentation. Primates have been shown to form

metapopulations (Lawes et al. 2000). For example, Lawes et al. (2000) used metapopulation

dynamics to determine the minimum area requirements for Cercopithecus mitis labiatus, and to

see if other variables such as human disturbance and isolation impacted their occurrence. They

found that the minimum area required for C. m. labiatus occupancy was approximately 44 ha

and that fragment area was the greatest factor explaining C. m. labiatus occurrence.

Understanding metapopulation dynamics in primates within a community can help us determine

what drives species-area relationships and how individual species are impacted by patch area

and isolation in fragmented landscapes.

1.8 Landscape-Level Effects on Primate Occurrence While the species-area relationship and metapopulation dynamics investigate aspects of species

richness and occurrence respectively at a patch-level, habitat loss and fragmentation are

landscape-level phenomena (McGarigal & Cushman, 2002; Fahrig, 2003). Therefore, a

landscape-level approach will yield additional insights into how species respond to habitat loss

and fragmentation, and may help determine why area is an important driver of species richness

and occurrence. A landscape is an area that is heterogeneous with respect to a character of

interest (Turner et al. 2001). For example, a landscape may vary in the amount of forest within

7

its boundaries. Scale becomes increasingly important in landscape ecology studies because

species responses are typically scale dependent and are probably species-specific (Jackson &

Fahrig, 2012). Some studies on primates have found that landscape-level responses are species-

specific (Gray et al. 2010; Thorton et al. 2011; Ordóñez-Gómez, 2014). For example, Ordóñez-

Gómez (2014) found that Ateles geoffroyi showed a species-specific scale of effect to the

amount of forest cover at 126 ha for most response variables at a landscape-level. By looking at

how species respond to habitat amount in a landscape, we may be better able to understand

species responses to habitat loss and fragmentation, and to determine what factors drive species

responses to area.

1.9 A Note on the Format of the Thesis My dissertation is written as a collection of three independent manuscripts (Chapters 2–4) that

investigate how habitat loss and fragmentation affect lemur species richness and occurrence. To

reduce redundancy, in the methods section of each independent chapter, I refer to methods from

previous chapters when there is overlap. References are supplied at the end of the dissertation in

a single references section. My dissertation flows from community level relationships to species

level relationships at a patch-level and then lemur occurrence patterns at a landscape-level.

Chapter 2 investigates the role of the species-area relationship on a lemur community to

determine if area is the largest driver of species richness. Chapter 3 assesses how

metapopulation dynamics relate to lemur species occurrence and compares the impact of area

and isolation on individual species occurrence. Chapter 4 evaluates how habitat loss as

measured by the amount of habitat within species-specific landscape impacts lemur species

occurrence. In Chapter 5, I draw conclusions about how lemur species richness and occurrence

are influenced by habitat loss and fragmentation. I also explore the implications of these

findings, detail directions for future research, and provide a comprehensive review of possible

conservation actions related to the results of my research.

1.10 Dissertation Goals Habitat loss and fragmentation are proceeding rapidly in Madagascar. It is critical that we

understand what drives local lemur species extinction in landscapes that are being increasingly

fragmented. My goal is to investigate how area impacts lemur species richness and occurrence

at different scales, using a patch-level analysis within a single large landscape, and a landscape-

8

level analysis involving many small landscapes. Specifically, I will determine if lemurs show a

species-area relationship, if that relationship is sigmoidal, and if factors other than area

influence lemur species richness in a fragmented landscape. I also want to determine if

metapopulation dynamics explain lemur species occurrence and how area and isolation

independently affect lemur species occurrence. Finally, I want to evaluate if the role of area on

lemur species occurrence is the same from a landscape ecology perspective as from a patch-

level perspective. With this dissertation, I will contribute to resolving a recurring debate in

ecological theory over how species respond to habitat loss and fragmentation, demonstrating the

intertwined utility of the species-area relationship, metapopulation dynamics, and landscape

ecology.

9

Chapter 2: Species-Area Relationships of Lemurs in a Fragmented Landscape in Madagascar

2.1 Introduction The species-area relationship (SAR) occurs throughout nature (Whittaker & Triantis, 2012). The

relationship between species and area describes the pattern of increasing species richness (the

total number of species) with increasing habitat area (Lomolino, 2000) and has been

demonstrated in virtually all taxa studied, including primates (Lehman, 2004; Harcourt &

Doherty, 2005; Marshall, et al. 2010). Although well-known since the 19th century, the species-

area relationship lacked formal description as the species-area curve until the 20th century

(McGuinness, 1984; Scheiner, 2003). Species-area curves are graphical representations of

mathematical models describing the pattern of the species-area relationship.

In this chapter, I will introduce how to choose a species-area model based on various criteria,

and describe the different types of species-area curves, focusing on the nature of convex and

sigmoidal models. I will then discuss some competing hypotheses for what determines the

species-area relationship. I will assess how other variables, in addition to area, influence primate

species richness in a fragmented landscape. I will follow with a literature review of the species-

area relationship research on primates, and explore why Madagascar is an ideal locality to study

the pattern and process of the species-area relationship in primates. I will end the introduction

by presenting my objectives, hypotheses, and predictions for this study.

2.2 Choosing a Species-area Model When choosing a species-area model, it is important to consider how species-area relationship

patterns are influenced by isolate data versus sample area data, scale, and arithmetic versus

statistical space (Scheiner et al., 2000; Tjørve, 2003; Triantis et al., 2012). In isolates (such as

islands or habitat fragments surrounded by matrix), species are typically bound by the confines

of the island or habitat patch. In these situations, minimum area effects for a species will impact

whether or not certain species exist within isolates below a particular size. The “small island

effect” occurs when species richness is impacted by factors other than area in small islands, or

when the sizes of the islands are too small for the species to survive (Lomolino, 2000). There

should be a maximum number of species when there is limited movement between isolates and

10

there is no other source population (Lomolino, 2000). Therefore, in surveys of isolates, species-

area curves should be sigmoidal shaped with the lower j-portion of the curve representing the

“small island effect” and upper asymptote representing the maximum number of species (Fig.

2.1; Tjørve, 2003; Lomolino, 2000).

Figure 2.1: Sigmoidal Species-Area Model. In sigmoidal models, there is a “small island effect” in the lower j-portion of the curve where species richness is affected by factors other than area. Above the “small island effect” species number accumulate in a similar fashion to convex models. Sigmoidal models may or may not have an upper asymptote. Some sigmoidal models have flexible inflection points (the point on the curve where the slope changes from an increasing slope to a decreasing slope) while others do not. This figure is an example of a sigmoidal model with a flexible inflection point. If researchers do not consider scale when investigating the species-area relationship, then it is possible that a truncated sigmoidal model may appear to be convex (area between red arrows).

Species-area relationships measured within portions of large continuous habitat are known as

sample area relationships. In sample areas, there is no “small island effect” because species are

Area

Spe

cies

Ric

hnes

s

0 20 40 60 80 100 120 140 160 180

010

2030

4050

Island EffectSmall

Upper Asymptote

Inflection Point

11

not restricted within continuous habitat as is the case for species in isolates (Lomolino, 2000;

Tjørve, 2003). Smaller sample areas are arbitrarily chosen and do not reflect potential biological

barriers, whereas in isolates, a small island is potentially surrounded by hostile matrix (Ricketts,

2001). Species accumulation during surveys of plots of increasing size will at first increase

dramatically with the most common species being discovered rather easily (Fig. 2.2). However,

the number of new species discovered typically decreases as sample area increases (Preston,

1962). If there is a finite species pool then the most appropriate model will be a convex upward

model with an upper asymptote (Tjørve, 2003).

Scale can impact the choice and nature of the species-area curve (Scheiner et al., 2000). For

example, in isolates a sigmoidal curve is theoretically correct because of the “small island

effect” and likely upper asymptote. If a study does not include a large range of isolate sizes, then

the predicted sigmoidal nature of the relationship may be obscured (Lomolino, 2000; Tjørve,

2003; Triantis et al. 2012). A truncated sigmoidal curve may appear to be convex if the scale of

observation does not include areas that would fall within the “small island effect” (Fig. 2.1).

Alternatively, if the areas sampled are too small, then the upper asymptote may be lost. Scale

can also influence the shape of a model. When using a power model, the slope of the species-

area curve (z-values) can change depending on the scale of the study. For example, larger

islands will have lower z-values than smaller islands (Martin, 1981).

12

Figure 2.2: Convex Species-Area Models. Neither model has a “small island effect,” but instead species richness increases rapidly at first but slows down as habitat area increases indefinitely (no upper asymptote) or until a maximum of species is found (upper asymptote).

Tjørve (2003) suggests that investigating the pattern of species-area curves is most appropriately

conducted in arithmetic space. The differential use of arithmetic space versus statistical space in

species-area studies can lead to confusion and misunderstandings of different models. Most

measurements of species-area relationships occur in arithmetic space and are later transformed

into log-log or log-linear statistical space for analysis. However, the shape of a species-area

curve is not consistent across different statistical transformations (Tjørve, 2003; Triantis et al.

2012). A problem occurs when transformations are applied to species richness data in attempts

to normalize or linearize the relationships without considering how such transformations may

alter the biological interpretation of subsequent models. Transforming a species-area

Area

Spe

cies

Ric

hnes

s

0 20 40 60 80 100 120

010

2030

4050

Upper Asymptote

No Upper Asymptote

13

relationship into a linearized relationship can result in a statistically valid model, but that model

will have little biological validity (Tjørve, 2003).

Researchers should select species-area models based on existing knowledge about the system

being studied (isolates versus continuous habitat), and consider the impact of scale on species-

area models. If the study focuses on true islands that may have a “small island effect,” then a

sigmoidal model with an upper asymptote is the most appropriate. If the study focuses on

sample areas within continuous habitat, then a convex model with or without an asymptote is the

most appropriate. If there is ambiguity in the isolated nature of the study area (for example, in

habitat fragments that are not as isolated as oceanic islands), then researchers should consider

both sigmoidal and convex models with and without asymptotes. Also regardless of the system

being studied researchers should consider the effect of scale. For example, if the scale is too

small in isolate species-area relationships, a convex model may appear to be more appropriate

than a sigmoidal model (Fig. 2.1). Finally, because linearizing species-area relationships can

result in inconsistencies in how researchers interpret different models, it is more appropriate to

use arithmetic space in data analysis.

2.3 Types of Species-Area Curves Species-area curves can be categorized as those that have asymptotes versus those that do not,

and those that are of convex shape versus those that are sigmoidal. An asymptote occurs when

the distance between a line and a curve approaches zero (Fig. 2.2). In models without an

asymptote the distance between a line and a curve continually increases or decreases. Sigmoidal

models can be further divided into those that are flexible about where their inflection point

occurs (the point on the curve where the slope changes from an increasing slope to a decreasing

slope) and upper asymptote versus those that are symmetrical because their inflection point is

fixed relative to the upper asymptote (Fig. 2.1).

2.3.1 Convex Models without an Asymptote

Species-area curves are typically fitted using two classic models: the power model and the

exponential model (Connor & McCoy, 1979; Coleman, 1981; Tjørve, 2003). The power model

was introduced by Arrhenius (1921) and is sometimes called the Arrhenius model. The

exponential model was introduced as a competing model by Gleason (1922) and is sometimes

referred to as the Gleason model. Both the power and exponential models do not have an upper

14

asymptote, meaning that the number of species will increase infinitely with area. The power and

exponential models are both simplistic, relatively monotonic views of the species-area

relationship, and do not fully capture the pattern as well as other models that provide upper

asymptotes and account for possible “small island effects” (Lomolino, 2000; Tjørve, 2003).

However, when applied to sample areas, both the power and exponential models have

consistently performed well (Tjørve, 2003). The difference between the two models is that in the

exponential model species number increases rapidly at first but the rate of increase declines

sooner than in a power model. The power model is used regularly in many studies on

mammalian species richness (Lomolino, 1982; Frey et al. 2007; Triantis et al. 2012). Botanists

most frequently used the exponential model to explain species-area relationships among plant

species until the power model became the model of choice (Connor & McCoy, 1979).

2.3.2 Convex Models with an Asymptote

When considering convex models, most research has focused on the power and exponential

models (Connor & McCoy, 1979; Lomolino, 2000; Tjørve, 2003). However, there are other

convex models that may be applied to species-area relationships that have upper asymptotes,

including the Monod, negative exponential, asymptotic regression, and rational function models

(Tjørve, 2003). Of these convex models with an upper asymptote, only the negative exponential

model (Rakatowsky, 1990) always passes through the origin (Tjørve, 2003). These are potential

candidate models for species-area relationships because of their ability to reach an upper

asymptote, where researchers can infer a total species number.

2.3.3 Sigmoidal Models without an Asymptote

Although many different sigmoidal models potentially fit species-area relationships, few models

have actually been applied to real-world data (Tjørve, 2003; Lomolino, 2000). The benefit of a

sigmoidal model is most apparent in its ability to explain biological relationships between

species and area in isolates. In isolates there is an expectation that some species will exhibit a

“small island effect” (Lomolino, 2000; Tjørve, 2003). A sigmoidal model without an upper

asymptote is more appropriate in isolates where species can enter the system and no maximum

number of species is expected. The result is an s-shaped curve with no upper asymptote. The

persistence function (referred to as the persistence P2 by Tjørve, 2009) presented by Ulrich and

Buszko (2003, 2004) represents such a sigmoidal model without an upper asymptote. The

15

persistence P2 function is a parameterization of the power model to allow for either a convex or

sigmoidal shape. The persistence P2 function is very flexible because it does not have an upper

asymptote and may be useful for species-area relationships where the “small island effect”

occurs at very small scales (Tjørve, 2009).

2.3.4 Sigmoidal Models with an Asymptote

The closed nature of isolates means that an upper limit of species number should result in an

upper asymptote for a species-area curve. If species have minimum area requirements, then the

“small island effect” should occur. The result is an s-shaped curve with an upper asymptote.

Differences in models may impact their application to species-area relationships. Sigmoidal

models with an upper asymptote that are proposed to work with species-area relationships

include the logistic function, Lomolino function, the cumulative Weibull distribution, and the

extreme value function. The ability to be flexible about the inflection point is important for

species-area relationships because the “small island effect” typically only occurs in the lower

end of area (Tjørve, 2009; Tjørve & Turner, 2009). The logistic function and extreme value

function have inflection points that are fixed at 50% and 63.2% of the upper asymptote,

respectively, while the cumulative Weibull distribution and Lomolino models are flexible about

the inflection point. Therefore, if there is any asymmetry in the shape of the species-area

relationship, then the cumulative Weibull or Lomolino models may be more appropriate than the

logistic or extreme value functions.

2.4 Processes Determining the Species-Area Pattern The pattern of species-area relationships can be influenced by the type of system under study

(isolate versus sample data), scale, and by analyzing arithmetic versus statistical relationships

(Scheiner et al. 2000; Tjørve, 2003; Triantis et al. 2012). However, what are the biological

mechanisms that determine the pattern of species-area relationships? There are three main

hypotheses explaining the process behind the pattern of the species-area relationship. The first,

the habitat heterogeneity hypothesis, was developed by Williams (1964) and is invoked in

numerous studies (Mac Arthur &Wilson, 1967; Fox & Fox, 2000; Scheiner, 2003). The habitat

heterogeneity hypothesis states that an increase in area results in more habitat types (i.e. large

areas are more heterogeneous relative to smaller areas), which results in increased niche space

and type that can be filled by more species (Williams, 1964; MacArthur & Wilson, 1967;

16

Connor & McCoy, 1979; Rosenzweig, 1995; Scheiner, 2003). Given variation amongst taxa in

their niche requirements, the degree of habitat heterogeneity depends on variables that are taxa-

and scale-specific. Many studies have found a connection between species richness and habitat

heterogeneity (e.g., Ganzhorn et al. 1997; Fox & Fox, 2000; Williams et al. 2002). For example,

Fox and Fox (2000) found that habitat diversity better predicted small mammal species richness

than area from eleven sites in Myall Lakes National Park, Australia.

The second hypothesis to explain the cause of the species-area relationship is called the area per

se hypothesis. The area per se hypothesis states that species richness is the product of extinction

probabilities. Species in larger areas will have larger populations and will therefore have lower

extinction probabilities and are less prone to stochastic disturbance events than species in

smaller areas (MacArthur & Wilson, 1967). Therefore, larger areas should have more species

than smaller areas.

Finally, the third hypothesis to explain the cause of the species-area relationship is the passive

sampling or random placement hypothesis (Connor & McCoy, 1979). The passive sampling

hypothesis assumes that all individuals and species within a community are distributed

randomly. Therefore, the probability of finding a randomly distributed species within a given

sample is determined by a ratio of the size of the sample to the entire area. Thus, a species is less

likely to be detected in a smaller sample area than in a larger sample area. In this sense, the

passive sampling hypothesis is a product of sampling phenomena and not biological processes

as we see in the habitat heterogeneity or area per se hypotheses, and so the passive sampling

hypothesis may be considered a null model (Connor & McCoy, 1979). However, these three

hypotheses may not be mutually exclusive and each may be combined to explain observed

patterns of species-area relationships (Connor & McCoy, 1979; Ricklefs & Lovette, 1999).

2.5 Other Non-Area Factors Affecting Primate Species Richness Though powerful, species-area models are overly simplistic in determining what specifically

drives species richness in fragmented landscapes. Factors such as habitat structure,

anthropogenic disturbance, and isolation metrics can influence species-area relationships within

fragmented landscapes. Anthropogenic disturbances such as hunting, logging, capture for the pet

trade, fire, climate change, and disease have been shown to impact primate species (Chapman &

Peres, 2001; Estrada et al. 2006; Chapman et al. 2007; Mittermeier et al. 2010). Although there

17

is a great deal of research that has investigated the impact of human disturbance on primates,

most fragmentation research has neglected to directly measure the potential conflation of

multiple sources of anthropogenic disturbance on primate species abundance and occurrence

(Chapman et al. 2003; Arroyo-Rodriguez & Dias, 2010). Lawes et al. (2000) found no

measurable impact of anthropogenic disturbance (aside from habitat loss and fragmentation) on

primate occurrence, and found that fragment area was the strongest predictor of primate

occurrence. Similarly, using multivariate analysis, Marshall et al. (2010) found that human

disturbance variables did not add explanatory power to species-area relationship models when

applied to primates. However, Arroyo-Rodriguez et al. (2008) found that fragments further from

villages had a higher occupancy of Alouatta palliata than fragments closer to villages,

suggesting an indirect impact of human disturbance. Gillespie and Chapman (2006) noted that

increased human disturbance within fragments increased parasite risk for primates. Thus,

researchers should consider how human disturbance might impact species responses to habitat

loss and fragmentation when assessing the impacts of these processes and patterns on primates.

Like habitat heterogeneity, the structure and composition of habitat may predict species richness

in primates. For example, tree species richness and habitat structure variables (including tree

size) are correlated with lemur species richness in a positive but non-linear fashion (Ganzhorn et

al. 1997). Pyritz et al. (2010) found that understory density is negatively correlated with primate

species richness. Hanya and Aiba (2010) found that variation in fruit fall might explain some of

the variation in frugivore richness in primates. Thus, it is important to consider the potential

effects of habitat characteristics on species richness.

Isolation of fragments within a landscape can impact species richness. Island biogeography

theory states that islands that are smaller and more isolated will have lower species richness than

islands that are larger and less isolated (MacArthur & Wilson, 1967). Species richness decreases

when islands are more isolated because isolation reduces immigration rates. Inversely, species

richness increases when islands are less isolated due to increased immigration rates. Few studies

have investigated the impact of habitat isolation on primate species richness (Harcourt &

Doherty, 2005). Harcourt and Doherty found no relationship between primate species richness

and isolation. However, some researchers have investigated the impact of habitat isolation on

primate species occurrence (Lawes et al. 2000; Boyle & Smith, 2010). In a metapopulation

study, Lawes et al. (2000) found that isolation did not explain variation in primate species

18

occurrence, but did explain variation in occurrence for non-primate species. Boyle et al. (2010)

found a negative relationship between primate species occurrence and isolation, but isolation

was correlated with area. Because of the large amount of theory outside of primatology

suggesting isolation should impact species richness, it is important for studies to incorporate

isolation metrics in their analysis.

2.6 Species-Area Relationships in Primates Many researchers have demonstrated species-area relationships in primates (Reed & Fleagle,

1995; Cowlishaw, 1999; Cowlishaw & Dunbar, 2000; Lehman, 2004; Harcourt & Doherty,

2005; Marshall et al. 2010). Reed and Fleagle (1995) found that the number of primate species

is positively correlated with area of tropical rainforest at continental scales, and with rainfall at

local scales. Using data from several published studies, Cowlishaw and Dunbar (2000) found

that the species-area relationship explained 51% of the variation in primate species richness.

Cowlishaw (1999) used the species-area relationship to determine whether there was an

extinction debt for African primates. Using a power model he found that log-area significantly

explained 34% of variation in log-species richness (R2=0.34; p=0.002). Lehman (2004)

surveyed the number of primate species within different habitat types in Guyana and found that

habitat area explained 63% of the variation in primate species richness. In a global analysis of

primate species-area relationships, Harcourt and Doherty (2005) found that primate species

number was significantly greater in larger than in smaller fragments, with a few exceptions in

Africa. Marshall et al. (2010) investigated the species-area relationship for diurnal primate

species among habitat fragments in Tanzania and found that of the 17 variables measured, area

was the strongest predictor of primate species richness (R2=0.846; p<0.01). These researchers

also noted a “small island effect,” where primate species richness did not relate to area in

fragments smaller than 12–40 km2. In 2010 I conducted a pilot study for this project,

investigating the possible species-area relationship of lemur species in eight fragments of dry

deciduous forest. I found that there was a linear relationship between log-area of a fragment that

significantly explained 71.3% of the variation in log-species richness (R2=0.713; p<0.01;

Steffens & Lehman, 2013). However, none of these studies considered different candidate

models to explain species-area relationships in primates, and only Marshall et al. (2010) and

Lehman (2004) considered the impact of variables other than area that may influence species-

area relationships.

19

2.7 Testing for a Species-Area Relationship To determine the pattern of the species-area relationship it is important to avoid simple curve

fitting, and to extend studies to determine biological mechanisms to predict species-area

relationships a priori (Tjørve, 2003). For example, a species-area relationship should take the

form of a sigmoidal model in isolates, especially if there is a suspected “small island effect”

(Lomolino, 2000; Tjørve, 2003; Tjørve, 2009). Conversely, in sample areas a convex model is

more appropriate (Tjørve, 2003). Researchers can assume an asymptote in sample areas if there

are reasons to suspect a maximum number of species (Lomolino, 2000). Primates, like all

species, require a minimum area of habitat for survival (Tjørve & Turner, 2009). However, for

most species we do not know what that minimum area is (Gurd et al. 2001). The minimum area

requirement for a sexually breeding species should fall somewhere near the minimum area that

could support a breeding pair (Lambeck, 1996). The minimum area for an individual should be

positively related to its body size, with smaller individuals requiring smaller ranges (Milton &

May, 1976; Lehman et al. 2005). Additionally, researchers can expect an upper limit to species

diversity in studies that limit their investigation of species-area requirements to a particular

taxon, such as primates, because at any particular moment there is a finite maximum number of

primate species (Tjørve & Turner, 2009).

2.8 Justification Madagascar is home to approximately 99 species and 103 taxa of lemurs, many of which have

overlapping geographic ranges (Schwitzer et al. 2013). Habitats exploited by lemurs in

Madagascar are fragmented and disappearing rapidly—since the 1950s 40% of forest cover in

Madagascar has been converted to non-forest habitats (Harper et al. 2007). Forest loss and

fragmentation in Madagascar are largely a result of small-scale forest removal for rice

production, slash-and-burn agriculture, and to create pasture for grazing cattle (Gade, 1996;

Bloesch, 1999; Harper et al. 2007). Malagasy dry forest in particular is highly fragmented due to

increased incidence of small-scale deforestation from fire along forest edges in the 1990s

(Harper et al. 2007). Deforestation in northwestern Madagascar has left multiple fragments of

varying sizes throughout the landscape and along the perimeter of continuous forest tracts, such

as in Ankarafantsika National Park. Madagascar is an ideal country in which to study the impact

of habitat loss and fragmentation on primate species-area relationships because of the large

number of primate species and the highly fragmented nature of the primate habitat.

20

Although researchers conduct SAR studies on numerous taxa and in numerous situations, we are

still unclear on which SAR pattern best describes SARs in primates in fragmented landscapes.

Primates are ideal taxa in which to investigate the pattern and processes creating SARs because

many species are arboreal and require forest for travel. Therefore, discrete patches such as

habitat fragments may represent truly insular islands. If fragments occur below a particular size,

then populations of primate species may not be viable due to minimum area requirements.

Studying arboreal primates provides an opportunity to determine if a priori assumptions, such as

a sigmoidal nature of SARs in a fragmented landscape, are valid. In this study, I seek to increase

our understanding of the SAR pattern and process in primates by investigating species richness

and habitat characteristics in a highly fragmented landscape in Madagascar.

2.9 Goals The goals of my study are to determine the pattern of lemur species richness and to evaluate

how variables other than area affect species richness by investigating the following questions:

1) Which species-area model best describes the pattern of lemur species richness in a

fragmented landscape?

Hypotheses:

H0 Lemur species richness will vary randomly with respect to area.

H1 Lemur species richness will be positively related to area.

Predictions: The primate species in this study are mainly arboreal and therefore the

matrix between fragments should represent a barrier to their movement. Minimum area

requirements should limit lemur species to the largest fragments. Either the level of

isolation between fragments, the presence of minimum area requirements for lemur

species, or both should dictate whether or not I observe a small island effect and

subsequently a sigmoidal SAR.

2) Is area the main factor affecting lemur species richness, or do other variables such as

habitat characteristics also impact lemur species richness?

Hypotheses:

21

H0 Lemur species richness will vary randomly with respect to area and habitat

characteristics.

H1 Lemur species richness will be positively related to area, habitat characteristics,

decreased isolation, and decreased human disturbance.

Predictions: Lemur species richness in a fragmented landscape, in addition to area, will

be positively influenced by habitat characteristics, and will be positively associated with

decreased isolation and decreased human disturbance.

2.10 Methods

2.10.1 Study Site and Study Species

I conducted this study between June and November of 2011 in Ankarafantsika National Park,

Madagascar (Fig. 2.3). Ankarafantsika National Park is approximately 135,800 ha, and consists

of a mosaic of approximately 72,670 ha of dry deciduous forest (Alonso et al. 2002; Razafy

Fara, 2003) and grassland savannah (García & Goodman, 2003). The climate is mostly dry with

mean yearly rainfall of 1,000–1,500 mm occurring mostly in the rainy season between

November and April (Alonso et al. 2002). There are eight species of lemurs in the park (Table

2.1) including: the western wooly lemur (Avahi occidentalis), Coquerel’s sifaka (Propithecus

coquereli), the grey mouse lemur (Microcebus murinus), the golden brown mouse lemur

(Microcebus ravelobensis), Milne Edwards sportive lemur (Lepilemur edwardsi), the fat-tailed

dwarf lemur (Cheirogaleus medius), the mongoose lemur (Eulemur mongoz), and the common

brown lemur (Eulemur fulvus; Mittermeier et al. 2010). I conducted this study in the dry season

and early part of the wet season (June–November) to facilitate access, and because there is

increased visibility due to reduced foliage. All species were active during the entire study except

C. medius, which is in torpor between April and October (Dausmann et al. 2005).

22

Figure 2.3: Study Site and Distribution of Forest within Madagascar. a) Location of study site within Madagascar. b) Location of the study site within Ankarafantsika National park. c) Close up of study site showing the fragmented landscape, consisting of 42 fragments of dry deciduous forest separated by a mainly homogeneous matrix of grassland. Survey fragments are represented in dark grey and continuous forest in light grey and savannah in white.

23

Table 2.1: Primate Species in Ankarafantsika National Park.

Species Body Mass (g) Activity Pattern Diet Mean Home range

(ha) Cheirogaleus medius 120–270 Nocturnal Frugivore 1.55 ±0.42(1) Microcebus murinus 58–67 Nocturnal Fauni-frugivore 2.83 ±1.44(2)

Microcebus ravelobensis 56–87 Nocturnal Fauni-frugivore 0.59 ±0.11 (3) Propithecus coquereli 3700–4300 Diurnal Folivore 19.36(4)

Eulemur fulvus 1700–2100 Cathemeral Frugivore 13.5(5) Lepilemur edwardsi 1100 Nocturnal Folivore 1.096)

Data from: 1 Müller (1998); 2 Radespiel (2000); 3 Weidt et al. (2004); 4 McGoogan (2011); 5 Mittermeier et al. (2010); 6 Warren & Compton (1997).

I collected species richness data in 42 forest fragments and habitat characteristics data in 38 of

the 42 forest fragments in a fragmented landscape (Fig. 2.4). The forest fragments ranged in size

from 0.23 ha to 117.7 ha, and were surrounded by a relatively homogeneous matrix of grassland

savannah, mostly consisting of the species Aristida barbicollis (Bloesch, 1999). I defined habitat

fragments as patches of forest with a connected canopy, including primary and secondary

vegetation that were separated from other fragments by a clear gap in the canopy. I defined

continuous forest as forest that was not fragmented. There is no recorded history of

fragmentation within the landscape, but I investigated topographic maps from the 1950s and

found a similar spatial arrangement of fragments. Current land-use management in the

landscape allows for some degree of use and some degree of seasonal burning and subsequent

grazing by cattle of the savannah is tolerated. There are no permanent residents within the

landscape. However, during the wet season local people create temporary settlements to graze

their cattle. Although hunting is not permitted, I found evidence of traps set for lemurs and there

is evidence in other areas of the park that lemurs are hunted (García & Goodman, 2003).

24

Figure 2.4: Study Site. Close up of fragmented landscape showing fragments (dark grey) labeled in red, surrounded by savannah (white), and in proximity to continuous forest (light grey).

2.10.2 Question 1: What is the SAR Pattern?

To assess the pattern of species-area relationships, I measured the area of each fragment by

recording a track along its perimeter using a handheld global positioning device (Garmin GPS

map 60csx) and inputting the results into QGIS (2012; n=38). If obstructions prevented a

complete perimeter walk (n=4; Fragments 37, 39, 40, 42), I traced the perimeter on a high

resolution DigitalGlobe™ satellite image taken during the study (10/8/2011) from Google

Earth™ and input the polygon in QGIS to determine its area. I placed survey transects along the

25

longest axis of the fragment while going through the center of the fragment except in Fragment

12A where I placed the transect along the longest axis of the largest portion of the fragment.

To determine species richness, my team of ten and I conducted both early and late diurnal and

nocturnal surveys in each fragment. Early diurnal surveys occurred between 06:19 and 09:07

hours and late diurnal surveys occurred between 14:39 and 17:18 hours. Early nocturnal surveys

took place between 18:00 and 21:27 hours and late nocturnal surveys took place between 02:17

and 5:55 hours. To ensure temporal independence for each survey, we only conducted one of

each survey type (diurnal and nocturnal) per 24-hour period in each transect. We surveyed all

fragments at least twice during early June and between October and November to ensure an

accurate assessment of the occurrence of C. medius, who can be in torpor between April and

October (Fietz & Ganzhorn, 1999; Dausmann et al. 2005). In total, we conducted between 11

and 18 diurnal, and 11 and 21 nocturnal surveys in each fragment (Table 2.2). During survey

walks each researcher walked slowly (approximately one km/hour), scanning and listening for

all lemur species. During diurnal surveys one or two researchers scanned both sides of the

transect simultaneously. Two researchers walked together during all nocturnal surveys and each

researcher focused on one side of the transect for the entire duration of the survey. Each team

member used high-powered flashlights and headlamps during nocturnal surveys to observe eye

shine. One member of each pair of diurnal researchers was experienced in identifying each of

the eight different species within the park. When team members observed a group or individual

lemur, the team spent up to 15 minutes determining species identity. Each of the species is

easily identified by size, with the exception of the two Microcebus species. We used a suite of

characteristics to determine the species identity of the two Microcebus species. We determined a

positive identification of M. murinus when the team observed the following characteristics:

grey/brown fur, small body size, and a short tail that is thick at the base. We determined a

positive identification of M. ravelobensis when the team observed the following characteristics:

rufus fur, larger body size, and a long tail that is thin at the base. We found it difficult to identify

41% of Microcebus species sightings to the species level during surveys. For those instances, I

only assigned identification to the genus level. We easily identified all other species during

surveys. For each survey, we recorded the start and end time, the type of survey (diurnal or

nocturnal), the number of researchers conducting the survey, which side each researcher was

surveying for nocturnal surveys, direction of survey and general weather.

26

Table 2.2: Survey Data of 42 Fragments in a 3,000 ha Fragmented Landscape.

Fragment ID Transect Length (m)

Diurnal Surveys

Nocturnal Surveys Total Surveys Diurnal

Sightings Nocturnal Sightings

1 823.1 15 15 30 1 47 2 1629.9 15 15 30 3 133 3 1804.2 18 21 39 18 150 4 565.8 14 18 32 6 41 5 627.3 15 15 30 2 60 6 649.1 15 17 32 0 50 7 278 14 17 31 0 11 8 621.3 14 17 31 1 30 9 625.4 15 17 32 0 50

10 294.6 14 14 28 0 30 11 227.3 14 13 27 0 8 12a 430.7 14 13 27 0 25 12b 577 13 13 26 0 15 13 314 14 13 27 0 19 14 291.3 14 16 30 0 14 15 148.6 16 16 32 0 24 16 105 14 16 30 0 2 17 66.3 14 16 30 0 0 18 289.5 13 16 29 0 4 19 210 13 16 29 0 7 20 161.8 14 16 30 0 20 21 67.7 14 16 30 0 0 22 43.8 14 16 30 0 4 23 325.2 15 15 30 0 17 24 61.1 14 14 28 0 6 25 86.7 16 15 31 0 7 27 188.3 14 14 28 0 4 28 76.8 15 13 28 0 6 29 138.3 14 15 29 0 2 30 119.1 15 14 29 0 1 31 535.6 14 14 28 0 20 32 239.8 14 14 28 0 17 33 324.7 14 14 28 0 9 34 278.2 13 14 27 0 34 35 377.1 15 14 29 0 11 36 314.2 14 14 28 0 24 37 127.5 14 14 28 0 0 38 178.3 14 14 28 0 2 39 498.2 14 13 27 0 41 40 391 14 13 27 0 22 41 223 11 11 22 0 12 42 141.9 12 11 23 0 13

Total 15476.7 596 622 1218 31 992 The number of surveys conducted during the day (diurnal), night (nocturnal), and total. And the number of associated sightings of all lemur species.

I measured survey effort as the total area of a fragment divided by the total area surveyed. The

area surveyed was calculated using the following formula:

27

Survey Area = W(lxw)

Where l is the transect length, w is two times the mean perpendicular distance of all sightings

along a transect, and W is the number of transect walks or surveys. For the four fragments where

I did not observe any individuals, I used the mean perpendicular distance for all sightings for all

transects (M=5.07 m SD=4.15 m).

To determine the pattern of the species-area relationship in my study, I selected 10 different

potential species-area functions from the literature (Table 2.3). The 10 functions represent a

wide range of possible species-area relationship models. Five functions are convex, including

three with asymptotes (negative exponential, Monod, rational function) and two without

asymptotes (power and exponential), while five functions are sigmoidal including one without

an asymptote (persistence P2) and four with an asymptote (logistic, Lomolino, Weibull, extreme

value function). Among the sigmoidal models, some models have inflection points that are

flexible with respect to the asymptote (e.g., Weibull and Lomolino models) while others are

symmetrical with respect to the asymptote (e.g., Logistic and extreme value function).

28

Table 2.3: 10 Candidate Species-Area Models.

Name Formula1 Para-meters Shape

First parameter

(c)

Second parameter

(z)

Third parameter

(b)

Curve origin

through axis

Asymp-totic2

power S = cAZ 2 Convex Curve shape Shape* Yes No

exponential S = c + zlog(A) 2 Convex Curve

shape Shape* No No

negative exponential

S = c(1 - exp(-zA)) 2 Convex Upper

asymptote Shape Yes Yes (c)

Monod S = (cA) / (z + A) 2 Convex Upper

asymptote Shape* Depends

on parameters

Yes (c)

rational function

S = (c + zA) / (1 + bA) 3 Convex

Shape + y-axis

intersection

Curve shape + upper

asymptote

Shape + upper

asymptote

Depends on

parameters

Yes (z/b)

logistic S = c / (1 + exp(-zA+b) 3 Sigmoid

Upper asymptote + y-axis

intersection

Shape + y-axis

intersection

Shape + y-axis

intersection

Depends on

parameters

Yes (c/b)

Lomolino S = c / 1 + (zlog(b/A)) 3 Sigmoid Upper

asymptote Shape Shape Yes Yes

cumulative Weibull

distribution

S = c(1 - exp(-zAb)) 3 Sigmoid Upper

asymptote Shape Shape Yes Yes

persistence (P2) model

S = cAZ

exp(-b/A) 3 Sigmoid Upper asymptote Shape Shape Yes No

extreme value

function

S = c(1 - exp(-

exp(zA+b))) 3 Sigmoid

Upper asymptote + y-axis

intersection

Shape + y-axis

intersection

Shape + y-axis

intersection No Yes (c)

1 A represents area c, z and b are fitted constants affecting the shape of the curve. 2 Letters within parentheses under the asymptotic column represent which constants determine the asymptote of the curve. Parameters with an asterisk alter curve shape on both sides of a rotation point. In some cases, more than one constant determines the asymptote (Adapted from Tjørve, 2003; 2009).

2.10.3 Question 2: What Other Non-Area Factors Affect Lemur Species Richness?

Habitat Characteristics

We determined the habitat structure for 38 of the 42 fragments along the same transects as the

primate surveys. We measured all living stems of trees over five cm diameter at breast height

(DBH) within one meter of each side of the 38 survey transects. For each tree, we measured its

DBH (cm) using DBH tape (Forestry Suppliers, Inc., Jackson, USA), visually estimated its

height (m), and recorded its location using a handheld GPS.

29

Human disturbance and Isolation Metrics

In addition to area and habitat characteristics, we measured the amount of human impact within

each fragment, fragment proximity to settlements, and fragment isolation. We conducted human

disturbance surveys along each primate survey transect to determine human impact. We looked

for evidence of all human activities within one meter of each side of the transect. We marked

evidence locations with a GPS and recorded the number of cut trees, the number of holes local

residents dug to access a tuber called maciba (Dioscorea maciba), zebu (cattle; Bos primigenius)

dung, hunting traps, and any other human disturbance variables observed (trails, artifacts). Cut

trees represent direct damage to forest, hunting traps represent a direct threat to lemur

populations, and maciba holes, zebu dung, and other disturbances (trails, artifacts) represent

evidence of forest use by local residents. For analytical purposes, I treated all observations of

human disturbance as equivalent. I also conducted a census of all temporary settlements within

the landscape and recorded their location using a GPS. To determine the proximity of temporary

settlements to each fragment, I entered the fragments and temporary settlements within ArcGIS

software (ESRI, Redlands, USA) and I calculated the distance in meters between the nearest

edge of each fragment and the nearest temporary settlement (DNS) using the ArcGIS Spatial

Join tool. Within ArcGIS, I determined isolation metrics by measuring the nearest distance from

the edge of each fragment to the nearest edge of continuous forest (DCF) and by measuring the

distance from the center of each fragment to the center of the nearest neighboring fragment

(DNN) using the ArcGIS Spatial Join tool.

2.10.4 Statistical Analysis

Question 1: What is the SAR Pattern?

I fit each of the 10 candidate species-area functions (Tjørve, 2009) to the data using non-linear

least squares regression (NLS). I assessed all species-area relationships in arithmetic space and I

did not apply transformations to species richness because: 1) a log transformation is not

appropriate because the species richness data contains zero values (log0=infinity; Cameron &

Trivedi, 2001; O'Hara & Kotze, 2010), 2) log transformation is ineffective for count data

(O'Hara & Kotze, 2010) and may reduce biological interpretation (Lomolino, 2000; Tjørve,

2003), and 3) many statistics such as least squares regression are robust to variation in normality

30

as long as there is homoscedasticity as demonstrated in the plot of residuals against predicted

values of the model (Kundu, 1993).

During the NLS regression, I set the start value for each fitted constant using the following

values (c=1, z=0.1, b=1), which resulted in fitted models for all functions except for the

cumulative Weibull function. For the cumulative Weibull function, I assigned the first parameter

(upper asymptote) to eight, which is the total number of lemur species found in the park. This

decision resulted in a fitted model.

To determine which model was most likely among the models assessed, I selected the model

with the highest Akaike’s Information Criterion weights (wi), and using AIC that was corrected

for small sample size (AICc; Burnham & Anderson, 2002). I considered models as potential

competing models if the difference in AICc (∆i) between a particular model and the best model

was between ∆i<4–7 (Burnham et al. 2011). To determine to what extent the best model fit the

data compared with competing models (i.e. models with ∆i<4–7), I calculated the evidence

ratio:

Evidence Ratio= wj / wi

The evidence ratio represents how much more likely the most likely model (wj) is than another

model (wi). I then compared the likelihood of all models together (n=10).

Question 2: What Other Non-Area Factors Affect Lemur Species Richness?

To determine how variables in addition to area influence species richness, I conducted a variable

selection procedure using hierarchical partitioning analysis. Hierarchical partitioning is a

hierarchical form of multiple regression analysis that looks at all possible regressions

contemporaneously to select predictor variables that have a high independent (of each other)

correlation with the dependent variable (Mac Nally, 1996; Mac Nally, 2000). It differs from

normal multiple regression in that the objective is not to find the best fit model but rather which

response variables contribute the most to the variation in the dependent variable (Mac Nally,

2000). Another advantage over normal multiple regression is that hierarchical partitioning

procedure works well with highly correlated variables (Mac Nally, 2000). In the hierarchical

partitioning procedure, I included: area, five habitat variables, two human disturbance variables,

31

two isolation metrics, and survey effort (Table 2.4). I then ran a generalized linear model (GLM)

with species richness as the response variable. I selected the predictor variables, which

independently contributed greater than 10% to the variation in species richness based on the

hierarchical partitioning procedure and input them in the GLM. The hierarchical partitioning

procedure uses least squares estimation for model fitting and determines the independent causal

factors between the highly correlated variables and the dependent variable (Chevan &

Sutherland, 1991; Olea, et al. 2010). I chose to use the R-squared measure of goodness of fit for

the hierarchical partitioning procedure. Table 2.4: Potential Influence of 10 Predictor Variables on Species Richness.

Variable Category Potential Influence on Species Richness

Area Area + Tree Basal Area Habitat Characteristic +

Tree Stem Density Habitat Characteristic + Mean Tree Height Habitat Characteristic + Mean Tree DBH Habitat Characteristic +

Distance to Nearest Neighboring Fragment Isolation Metric -

Distance to Continuous Forest Isolation Metric -

Total Human Disturbance Human Disturbance - Distance to Nearest Temporary

Settlement Human Disturbance -

Survey Effort Effort Neutral + Represents a predicted positive relationship with species richness, – represents a predicted negative relationship with species richness and DBH represents diameter at breast height.

I assessed the linearity for all predictor variables by visually inspecting histograms, density

plots, and box-plots of each variable. I assessed the shape of the data by calculating skewness,

kurtosis, variance, and standard deviation. I statistically assessed normality by applying a

Shapiro Wilk’s test to each variable (Shapiro & Wilk, 1965). If a variable was not normally

distributed, I attempted to transform the variable using a natural log (ln), square root or another

appropriate transformation and retested after transformation for normality.

For the GLM model, I used a Poisson (log) link function for the model because the dependent

variable, species richness, is count data. I compared a global model with all selected predictor

variables with models consisting of all combinations of variables using AICc.

32

I conducted all statistical analyses in R statistical software (2013). I used the “stats” package for

the descriptive statistics in R (Version 3.02). For the hierarchical partitioning procedure, I used

the “heir.part” package in R (Version 1.0-4). For the GLM I used the “stats” package in R

(Version 3.02). I considered all p-values significant if p≤0.05. I conducted all spatial analyses in

QGIS (2012) and ArcGIS (2012).

2.11 Results

2.11.1 Fragment Area and Survey Results

Within the survey landscape, we found fragments had a mean size of 8.82 ± 19.6 ha. During

1,218 total surveys in fragments, we observed 1,023 individuals or groups of lemurs. We found

a total of six lemur species, including four nocturnal species (C. medius, M. murinus, M.

ravelobensis and L. edwardsi), one diurnal species (P. coquereli), and one cathemeral species

(E. fulvus). We identified 29% (n=269) of 933 Microcebus species sightings as M. murinus,

30% (n=283) as M. ravelobensis, and 41% as (n=381) unidentified individuals. The number of

species within a fragment ranged from zero to six (Table 2.5). We visually observed all lemur

species recorded in each fragment during surveys, with the exception of C. medius in Fragment

8, which we heard during a survey, then tracked to visually confirm its presence in this

fragment. We found some nocturnal species during diurnal surveys and vice-versa for diurnal

and cathemeral species. Of the 42 fragments we surveyed, we found diurnal species in only

14.3% (n=6) of the fragments. Conversely, we found nocturnal species in 92.9% (n=39) of

surveyed fragments. We did not see or hear any individuals or groups of Eulemur mongoz or

Avahi occidentalis, although these species are present elsewhere in Ankarafantsika National

Park.

33

Table 2.5: Fragment Characteristics and Species Richness.

Fragment ID

Species Richness

Area (ha)

No. Trees Measured

µTree DBH (cm)

µTree Height

(m)

Stem Density (ind/ha)

Total Dist.

DNN (m)

DCF (m)

DNS (m)

1 4 34.51 296 10.35 6.52 1798.08 172 276.04 2198.56 707.75 2 5 45.34 430 17.89 6.80 1319.10 66 527.57 636.08 1191.77 3 6 117.70 583 11.86 6.24 1615.67 116 584.68 819.02 1272.94 4 4 19.46 197 N/A 7.37 1740.90 94 474.07 1319.98 1841.82 5 4 16.01 168 5.53 6.73 1339.07 132 584.68 1153.15 1048.08 6 4 11.58 271 7.44 7.81 2087.51 81 601.36 179.74 2361.76 7 3 4.16 99 6.73 6.12 1780.58 79 462.74 1385.74 705.33 8 4 13.55 184 8.97 6.74 1480.77 111 464.75 1154.63 1170.86 9 3 15.38 277 7.93 7.68 2214.58 56 416.35 110.77 2525.35

10 2 2.78 103 10.38 5.20 1748.13 33 839.64 105.92 2441.02 11 2 2.57 23 17.42 4.04 505.94 53 471.70 210.12 1632.44 12a 2 14.22 113 11.70 5.77 1311.82 49 588.49 656.29 311.29 12b 3 5.18 N/A 7.61 N/A N/A N/A 527.57 842.41 868.53 13 2 3.75 52 10.29 4.85 828.03 39 1089.28 277.72 785.14 14 2 4.08 102 12.79 5.87 1750.77 24 510.06 2008.92 872.92 15 2 1.18 50 N/A 6.02 1682.37 8 276.04 2481.39 938.80 16 1 0.76 22 9.33 4.64 1047.62 26 442.90 2044.61 1367.79 17 0 0.42 23 12.24 5.09 1734.54 17 706.74 1831.83 1844.52 18 1 1.14 N/A 12.22 N/A N/A N/A 395.00 1058.66 1,627.80 19 2 1.40 10 10.35 6.00 238.10 29 395.00 1088.53 2002.92 20 2 1.43 47 17.89 6.26 1452.41 33 601.36 180.49 1765.76 21 0 0.38 27 11.86 5.44 1994.09 8 462.74 1090.85 783.65 22 2 0.23 28 N/A 7.50 3196.35 28 416.35 103.72 2496.82 23 2 2.48 33 5.53 5.85 507.38 87 395.16 2051.63 299.25 24 2 0.28 21 7.44 5.95 1718.49 20 730.57 452.31 1855.57 25 2 0.57 15 6.73 7.21 807.38 16 571.27 992.58 116.84 27 1 1.97 N/A 8.97 N/A N/A N/A 571.27 747.52 458.01 28 2 0.64 12 7.93 4.5 781.25 1.00 151.54 1929.84 967.99 29 2 0.71 9 10.38 4.67 325.38 13.00 151.54 2074.78 850.34 30 1 0.31 22 17.42 4.68 923.59 21.00 395.16 2478.30 443.75 31 3 17.03 85 11.70 5.09 793.50 10.00 575.17 1186.92 291.50 32 2 0.98 43 7.61 5.02 896.58 6.00 336.34 803.73 373.33 33 2 2.16 78 10.29 5.28 1201.11 9.00 336.34 433.05 695.28 34 3 1.69 91 12.79 6.57 1635.51 44.00 580.35 268.28 1026.21 35 2 5.00 65 N/A 6.09 861.84 69.00 704.14 413.70 1259.84 36 2 6.97 74 9.33 5.35 1177.59 13.00 311.38 1336.03 1456.95 37 0 0.64 19 12.24 6.53 745.10 40.00 698.22 788.00 1581.47 38 1 0.52 40 12.22 6.28 1121.70 5.00 311.38 1182.11 1483.15 39 2 5.17 N/A 10.35 N/A N/A N/a 241.78 321.61 2,092.55 40 2 3.78 126 17.89 5.70 1611.25 12.00 231.41 95.09 2229.73 41 2 1.56 28 11.86 5.25 627.80 5.00 231.41 17.06 2016.44 42 2 0.79 46 N/A 6.89 1620.86 1.00 241.78 263.28 2215.00

Landscape fragment characteristics. I measured species richness as the number of species observed within a fragment. No. Trees Measured refers to the total number of trees measured in each fragment during habitat structure surveys along primate survey transects. µTree DBH is the mean diameter at breast height of all trees measured within each fragment. µTree Height is the mean tree height. Total Dist. is the number of all human disturbance observations recorded along each transect. DNN is the distance to nearest neighbor as measured from the center of each fragment. DCF is the distance from the nearest edge of the fragment to the nearest point edge of the continuous forest. DNS is the distance from the center of each fragment to the nearest temporary settlement. I

34

report N/A for four fragments that I could not measure habitat characteristics because there were either no living stems along the transect (Fragment 27) or the transect ran along a dry riverbed (Fragments 12b, 18, and 39).


I found significant fitted parameter estimates for all models except the Lomolino and persistence

models (P2; Table 2.6). In the Lomolino model, the Z parameter was a significant contributor to

the shape of the model (Z=0.73, p<0.01), while the c and b parameters were not significant

contributors to the shape of the model (c=0.27, p=0.74; b=0.43, p=0.64). In the persistence (P2)

model, the c and Z parameters were significant contributors to the shape of the model (c=1.48,

p<0.01; Z=0.30, p<0.01) while the b parameter was not a significant contributor to the shape of

the model (b=-0.05, p=0.58). Table 2.6: Fitted Parameters of 10 Candidate Species-Area Models Using Non-linear Least Squares Regression for Primate Species in the 42 Fragments.

Name Formula Shape Asymp-totic2

Parameter Estimates and Associated p-values

power S = cAZ Convex No c=1.56; p<0.01

Z=0.28; p<0.01

exponential S = c + zlog(A) Convex No c=1.60; p<0.01

Z=0.68; p<0.01

negative exponential S = c(1 - exp(-zA)) Convex Yes (c) c=3.59;

p<0.01 Z=0.39; p<0.01

Monod S = (cA) / (z + A) Convex Yes (c) c=4.10; p<0.01

Z=2.11; p<0.01

rational function S = (c + zA) / (1 + bA) Convex Yes (z/b) c=1.33;

p<0.01 Z=0.291; p<0.01

b=0.04; p=0.03

logistic S = c / (1 + exp(-zA+b) Sigmoid Yes (c/b) c=5.29;

p<0.01 Z=0.10; p<0.01

b=0.89; p<0.01

Lomolino S = c / 1 + (zlog(b/A)) Sigmoid Yes c=0.27;

p=0.74 Z=0.73; p<0.01

b=0.43; p=0.64

cumulative Weibull distribution

S = c(1 - exp(-zAb)) Sigmoid Yes c=8.001 Z=0.21;

p<0.01 b=0.36; p<0.01

persistence (P2) model S = cAZ exp(-b/A) Sigmoid No c=1.48; p<0.01

Z=0.30; p<0.01

b=-0.05; p=0.58

extreme value function S = c(1 - exp(-exp(zA+b))) Sigmoid Yes (c) c=5.01;

p<0.01 Z=0.07; p<0.01

b=-0.99; p<0.01

S is species richness. A represents area. c, z and b are fitted constants affecting the shape of the curve. 1 For the cumulative Weibull distribution model, I manually fitted parameter c to equal eight to represent the hypothetical maximum species number. Bold Indicates p≤0.05. 2 Letters within parentheses under the asymptotic column represent which constants determine the asymptote of the curve. In some cases, more than one constant determines the asymptote.

When comparing all models together, I found that the power model had the best fit (wi=0.49)

followed by the persistence (P2: wi=0.18), Lomolino (wi=0.16), Weibull (wi=0.08), and rational

function models (wi=0.07). None of the remaining models had ∆i values within 4–7 of the

power model (Table 2.7). A scatterplot fitted with the five competing models demonstrates the

similarity between model fits (Fig. 2.5a).

35

Table 2.7: Non-linear Least Squares Regression Model Selection of Species-Area Models. Model Log Likelihood K AIC AICc Δi wi Power -41.20 2 86.41 86.72 0.00 0.49

persistence (P2) model -41.07 3 88.14 88.77 2.06 0.18 Lomolino -41.15 3 88.30 88.94 2.22 0.16

cumulative Weibull distribution -41.83 3 89.66 90.29 3.57 0.08 rational function -41.98 3 89.96 90.59 3.88 0.07

exponential -45.17 2 94.35 94.66 7.94 0.01 logistic -44.31 3 94.61 95.24 8.53 0.01

extreme value function -45.24 3 96.48 97.11 10.40 0.00 Monod -51.21 2 106.42 106.72 20.01 0.00

negative exponential -55.92 2 115.84 116.15 29.43 0.00 Non-linear least squares model selection using Akaike’s information criterion (AIC). I compared each candidate model against one another using corrected AIC values (AICc). K is the number of parameters in the model. I determined the difference between the lowest AICc values and the rest to determine the change in AICc (Δi) and the weight (wi) of that difference was calculated.

To see how much more likely the power model was over the other models, I investigated the

evidence ratios between the power model and the four competing models. I found that the power

model is 2.72, 3.06, 6.13 and 7.00 times more likely to fit the data than the persistence (P2),

Lomolino, Weibull, and rational function models, respectively. To confirm the underlying shape

of the data, I applied a lowess curve with a 50% smoothing parameter over a scatterplot of

species richness and area. The lowess curve follows a downward convex shape over the data

similar to the above competing models (Fig. 2.5b).

Figure 2.5: Five Competing Species-Area Models. a) Five competing SAR models for lemur species richness including the Persistence (P2) model, Weibull model, power model, Lomolino model and rational function model. b) 50% lowess curve fitted to species richness values.

0 20 40 60 80 100 120

01

23

45

6

(a)

Fragment Area (ha)

Lem

ur S

pece

s R

ichn

ess

Persistence (P2)WeibullPowerLomolinoRational

0 20 40 60 80 100 120

01

23

45

6

(b)

Fragment Area (ha)

Lem

ur S

pece

s R

ichn

ess

Lowess Line

36


I measured a total of 4,042 trees in fragments within the entire landscape. I measured as few as

nine trees in Fragment 29 (area=0.71 ha) and as many as 583 trees in Fragment 3 (area=117.70

ha). There is a statistically significant correlation between the number of trees measured and

fragment area (r=0.90; p<0.01). I measured 1626 incidences of human disturbance within the

landscape. Within fragments mean human disturbance was 42.79 (SD=41.02), ranging from one

to 172 incidences within fragments. There was a significant positive correlation between ln area

and ln total human disturbance (r=0.60, p<0.01). I found a mean of 477.51 m (SD=195.60m) for

distance to nearest neighbor and a mean of 994.86 m (SD=759.03 m) for distance to continuous

forest. Survey effort ranged between 0.19 and 4.95, with a mean effort of 1.18 m (SD=1.09 m)

across all fragments. There was a significant positive correlation between ln survey effort and ln

area (r=0.64, p<0.01).

Based on the hierarchical partitioning procedure, I found that area (47.84%) had the highest

independent contribution to lemur species richness followed by survey effort (27.53%), total

human disturbance (13.49%), and mean tree height (10.34%). None of the other variables

contributed greater than 10% of the variation in lemur species richness

(Fig. 2.6).

Figure 2.6: Hierarchical Partitioning Model. A hierarchical partitioning model, that shows the independent contributions of each predictor variable on species richness. ln represents natural log transformation and sqrt represents square root transformation. THD represents total human disturbance, DBH represents diameter at breast height, StDens represents stem density, THM represents mean tree height, BA represents basal area, DCF represents distance from fragment to continuous forest, DNN represents fragment nearest neighbor distance, and DCF represents distance to nearest settlement.

lnArea lnEffort lnTHD THM DNS StDens lnBA meanDBH DNN sqrtDCF

Predictor Variables

Inde

pend

ent E

ffect

s (%

)

010

2030

4050

60

37

I ran 15 GLM models including all possible combinations of the predictor variables: ln area, ln

effort, ln total human disturbance, and mean tree height. The model with the highest wi value

was the model with only ln area as a predictor (Table 2.8; wi=0.35). Ln area was the only

variable was significant in the top models (highest wi values). Ln effort, mean tree height and ln

total human disturbance variables were significant only when ln area was not included as a

variable in the model (Table 2.8). Table 2.8: Comparison of Seven Generalized Linear Models Using Poisson Link Function Predicting Lemur Species Richness.

Bold represents variables that are significant. SR represents species richness, THM represents mean tree height, THD represents total human disturbance. I transformed each predictor variable with a natural log transformation (ln) accept THM. I compared each candidate model against one another using corrected AIC values (AICc). I determined the difference between the lowest AICc values and the rest to determine the change in AICc (Δi) and calculated the weight of that difference (wi).

2.12 Discussion


I predicted that the pattern of the species-area relationship among lemurs in a fragmented

landscape would be sigmoidal with a “small island effect” and an upper asymptote. I found

multiple competing models that describe the species-area relationship in this study. Contrary to

predictions, I found that the power model, and not one of the competing sigmoidal models, was

the best-fit model. Of the remaining four competing models, three were flexible sigmoidal

models (persistence (P2), Lomolino, and cumulative Weibull distribution) that were able to look

Model Parameters Δi wi Area THM THD Effort

β SE β SE β SE β SE SR=Area 0.00 0.35 0.27 0.07

SR=Area+THM 1.42 0.17 0.25 0.07 0.12 0.12 SR=Area+Effort 2.35 0.11 0.29 0.14 0.03 0.27 SR=Area+THD 2.35 0.11 0.27 0.09 0.01 0.12

SR=Area+THM+THD 3.88 0.05 0.26 0.09 0.13 0.13 -0.03 0.12 SR=Area+THM+Effort 3.91 0.05 0.26 0.14 0.12 0.13 0.04 0.27

SR=Effort 4.27 0.04 -0.46 0.13 SR=Area+THD+Effort 4.85 0.03 0.28 0.17 0.01 0.13 0.03 0.28

SR=THM+Effort 5.02 0.03 0.15 0.12 -0.42 0.14 SR=THD+Effort 5.20 0.03 0.12 0.11 -0.38 0.15

SR=Area+THM+THD+Effort 6.51 0.01 0.28 0.17 0.13 0.13 -0.03 0.13 0.06 0.28 SR=THM+THD+Effort 6.86 0.01 0.12 0.13 0.09 0.11 -0.36 0.15

SR=THD 9.57 0.00 0.26 0.10 SR=THM+THD 10.21 0.00 0.16 0.13 0.20 0.11

SR=THM 11.73 0.00 0.26 0.11

38

convex, while the remaining model was the rational function (convex model). The two best fit

models; power and persistence (P2) did not have asymptotes. The shape of the species-area

relationship in my study was clearly convex (Fig. 2.5a). I was not able to detect a sigmoidal

shape to the species-area relationship when visually inspecting the data using a lowess curve. I

predicted a sigmoidal model because of the perceived hostility of the matrix and the small size

of the smallest fragments. However, many other studies have also found the power model to fit

SARs in other species (Lomolino, 1982; Frey, 2007; Triantis et al. 2012). Two of the competing

sigmoidal models, Lomolino and persistence (P2), appeared to be convex because these models

are very flexible in their shape. Despite the Akaike's weights indicating that the sigmoidal

models are potential competing models, my results suggest that they are not appropriate models

to assess the pattern of lemur species richness within the fragmented landscape I studied.

Detecting a sigmoidal model is difficult without a large range of fragment sizes (Triantis et al.

2012). In my study, the fragment size range was too small to detect a sigmoidal model and the

larger fragments were too small to observe an asymptote. To detect a sigmoidal pattern, Triantis

et al. (2012) suggests that studies incorporate a size difference between the smallest and largest

isolates of at least three orders of magnitude. In my study, the largest fragment (117.7 ha) was

2.5 orders of magnitude greater than the smallest (0.23 ha). Regional species richness is eight

species, so for my models to reach an asymptote, I would have needed to observe all eight

species in the fragments. However, it is possible to reach an asymptote if I included the

remainder of the park as a data point. Although I included very small fragments (<1 ha) and the

matrix is assumed to be hostile to lemur species, I was not able to detect a small island effect.

The two most likely possibilities to explain why I did not observe a small island effect are: (1) in

my sample, there is no minimum area effect for the two widest ranging lemur species

(Microcebus species), or (2) the two widest ranging species (Microcebus species) are able to

move between fragments and may only be transitory within a fragment at any given time. Life

history characteristics of Microcebus provide evidence to support the first explanation.

Microcebus species differ from the other species that occur within this landscape in their

physical size, ranging behavior, and diet. Microcebus has the smallest body mass (Rasoloarison

et al. 2000), home ranges (Radespiel, 2000; Weidt et al. 2004), and is the most insectivorous of

all the lemur species within the landscape (Table 2.1). Having a low body size and small home

range requirements may make it possible for Microcebus to survive in the smallest (<1 ha)

39

fragments. In fragments of this size, species that rely more on fruit or leaves would have

difficulty finding food, while Microcebus can rely on insects that occur with greater absolute

availability than fruits or leaves in extremely small fragments (due to edge effects; Corbin &

Schmid, 1995).

The second explanation for my inability to detect a small island effect is likely related to the

lack of isolation between fragments within the study landscape (Fig. 2.4 and Table 2.5),

suggesting that some species are able to move through the matrix between fragments. Marshall

et al. (2010) suggested that there must be enough isolation between fragments so that species

cannot move between them. In my study, there were low levels of isolation between some

fragments as measured by nearest neighbor distance, allowing for possible movement between

fragments.

The only species that I observed in the matrix were Microcebus species (n=9), found in shrubs

within the matrix, suggesting that Microcebus use the matrix to move between fragments and

that the matrix may not be as hostile as I had assumed. Even though savannah matrix is a

significant barrier for M. ravelobensis (Radespiel et al. 2008), it appears that the degree of

isolation between fragments is low within my study landscape. Low fragment isolation reduces

the insular nature of the fragments and would thus explain the convex rather than sigmoidal

nature of the observed SAR. A lack of asymptote is likely due to the largest fragments being too

small to support continual populations of the two missing lemur species (A. occidentalis and E.

mongoz). Future research should use mark-recapture methods on Microcebus species to

determine their ability to move throughout the matrix and their ability to survive within the

smallest fragments.


The hierarchical partitioning procedure showed that ln area, ln survey effort, ln total human

disturbance, and mean tree height were possible causal factors influencing lemur species

richness. I expected ln area and mean tree height to both have a positive influence on species

richness, survey effort to have a neutral impact on species richness, and human disturbance to

have a negative influence on species richness. Post hoc analysis shows that species richness is

significantly positively correlated with ln area, mean tree height, and ln total disturbance but

40

significantly negatively correlated with ln survey effort. A positive relationship with ln area and

ln tree height falls in line with expectations of the species-area relationship and the possible

reliance of arboreal primates on tall trees, which may indicate increased food and shelter

availability, and may be necessary for locomotion for large vertical clingers and leapers such as

Propithecus coquereli. A positive relationship between species richness and ln human

disturbance indicates that the relationship between human disturbance and species richness is

more complicated. A negative relationship between survey effort and species richness suggests

that I over-surveyed smaller fragments that contained the fewest species. However, the GLM

analyses found that only the ln area only model was the most likely model. The GLM models

suggest that biogeographic factors such as area have a stronger positive influence than the

possible negative influence of human disturbance. However, during my study I did not find two

species of lemur (A. occidentalis and E. mongoz) that exist within the park in any of my survey

fragments. Although it is difficult to determine absence for any species, it is unlikely that I did

not observe individuals of E. mongoz or A. occidentalis due to methodological considerations

because both species are very conspicuous primates and have loud, distinct vocalizations.

During a shorter pilot study for this project between June and August 2010, we found one

individual A. occidentalis within the Fragment 3 (Figure 2.4) after only seven surveys, and

heard repeated vocalizations from the same location. However, during the 2011 seven-month

study, we conducted 21 nocturnal surveys in the same fragment and we failed to detect (observe

or hear) any A. occidentalis individuals, suggesting that they were recently extirpated in my

study site or are transient within the site. As a cautionary note, determining absence is extremely

difficult and other studies have found species within survey sites only after years of survey

effort. For example, Lehman (2014) unexpectedly observed a single Indri indri individual

walking to camp but not during 1,318 km of surveys in the wet forests of Vohibola III.

Minimum area requirements, limited numbers of large trees, and hunting pressure may also

relate to the absence of Avahi in the forest fragments. In eastern littoral forests, Norscia (2008)

only found Avahi meridionalis in forest fragments larger than 75 ha and no correlation between

fragment area and Avahi density (Norscia, 2008). Although A. occidentalis have small home

ranges and rely on low quality abundant leaves, suggesting that they would be able to inhabit

relatively small fragments, it is possible that the number and distribution of large trees impacts

Avahi occurrence (Norscia, 2008). Norscia (2008) found that the percentage of large trees above

41

3.2 cm DBH was significantly positively related to Avahi density. Illegal hunting does occur

within the park, and the although the most commonly hunted species are the larger-bodied and

more common P. coquereli and E. fulvus, García and Goodman (2003) did identify remains of

A. occidentalis from a hunt within the park.

In a rapid assessment, researchers only found E. mongoz in one of four survey sites within

Ankarafantsika National Park (Schmid & Rasoloarison, 2002). Shrum (2008) only found E.

mongoz in fragments larger than 250 ha. In my study, the largest fragment was only 117.7 ha.

Hunting may also contribute to my failure to detect E. mongoz. For example, local informants

indicated that E. mongoz is a preferential food source (Steffens, personal communication). Thus,

the absence of E. mongoz and A. occidentalis was likely due to hunting pressure, that some

aspect of the habitat structure being unsuitable, and/or that there were too few large fragments to

support populations of either species.

Considering lemurs are mainly arboreal, it is surprising that most of the habitat variables did not

appear to influence lemur species richness. Besides ln area and total human disturbance, the

hierarchical partitioning indicated that the only habitat variable that influenced lemur species

richness was mean tree height. Either fragment area is too great an influence on species richness

in this landscape that it overrides the possible influence of habitat characteristics, or my

measures were poor indicators of lemur habitat. Future research should look at other variables

such as food tree species diversity and presence of possible competitors and predators.

2.12.3 Suggestions for Conservation

Area is the main driver of lemur species richness at a patch-level. Preserving larger fragments

will help ensure more species of lemur are protected from extinction. Although I did not find

many habitat variables that appeared to influence lemurs at a community level, conservation

managers should continue to consider habitat quality and especially measures to maintain mean

tree height in a fragment. Two species are currently presumed locally extinct in fragments

within my study site. Measures that increase fragment size, protect existing fragment size, and

discourage anthropogenic use of large trees are needed to help ensure future local extinctions are

avoided.

42

2.13 Conclusion Lemur species do show a species-area relationship in a fragmented landscape in Ankarafantsika

National Park, Madagascar. Contrary to predictions, I found that the lemur SAR was convex

rather than sigmoidal and I did not detect a small island effect for primates in my study

landscape. The lack of a small island effect may be due to the movement of lemurs between

fragments and/or the small body size, small habitat requirements, and fauni-frugivorous diet for

the two most widely distributed species (Microcebus species) found within the landscape. Of the

15 GLM models I ran, the ln area only model was the most likely considering my data. I found

that mean tree height and ln human disturbance appeared to have little effect on lemur species

richness in this landscape. Therefore, the SAR found among lemurs within this landscape is

driven mainly by the biogeographic factor: area. However, the possible local extinction of A.

occidentalis and E. mongoz may be the result of both reduced fragment size and hunting

pressure. No measures of habitat characteristics or fragment isolation appear to influence the

SAR I observed among the lemur species in my study. Future research should investigate other

habitat measures, such as food species diversity, habitat structural heterogeneity, the ability of

lemurs to move throughout the matrix, the presence of competitors, and potential predators to

develop a more comprehensive understanding of how lemurs respond to habitat loss and

fragmentation.

43

Chapter 3: Population Dynamics of Lemurs in a Fragmented Landscape in Madagascar

3.1 Introduction In Chapter 2, I investigated the community level effects of habitat loss and fragmentation on

primate species richness. I found that even in the context of high human disturbance, lemur

species at the community level continue to follow a classic species-area relationship. But how

are individual lemur species affected by habitat loss and fragmentation?

There are eight species of lemurs within Ankarafantsika National Park and six occur within my

study site including: the fat-tailed dwarf lemur (Cheirogaleus medius), the grey mouse lemur

(Microcebus murinus), the golden brown mouse lemur (Microcebus ravelobensis), Coquerel’s

sifaka (Propithecus coquereli), the common brown lemur (Eulemur fulvus), and Milne Edwards

sportive lemur (Lepilemur edwardsi; see Chapter 2; Mittermeier et al. 2010). Of the six species

found within the study site, two are considered least concern (M. murinus and C. medius), one is

considered near threatened (E. fulvus) and three are considered in endanger of extinction (M.

ravelobensis, P. coquereli, and L. edwardsi; IUCN Red List, 2016).

The main reason lemur species are threatened with extinction is habitat loss and fragmentation

of the forests in Madagascar (Schwitzer et al. 2014). The high degree of habitat disturbance in

Madagascar even affects protected forests such as those within Ankarafantsika National Park.

The processes of habitat loss and fragmentation create landscapes with discrete patches of

habitat (Fahrig, 2003). Metapopulation dynamics is a useful approach to find out how individual

species respond to these processes (Hanski, 1994a; Hanski, 1999; Hanski & Ovaskainen, 2003).

A metapopulation is a population of populations (Levins 1969, 1970). Within a metapopulation

local populations of breeding individuals live within discrete definable patches of habitat. Each

local population within each patch has a probability of going extinct. The probability of

extinction is a function of the area of a patch, because smaller areas have fewer individuals and

are more prone to extinction, while larger areas support a greater number individuals and are

less prone to extinction (MacArthur & Wilson 1967). Local populations within a

metapopulation are connected via dispersal. Dispersal is the movement of an individual from

their natal home range to a breeding range (natal dispersal) or from their breeding home range to

a new breeding range (secondary or breeding dispersal; Matthysen, 2012). Patch colonization

44

potential for dispersers is a function of the connectivity or, inversely, isolation, between patches

and the species-specific ability to disperse among patches (Hanski, 1999).

A metapopulation system is dynamic (Hanski & Gilpin, 1991). Over time, some patches may

go extinct while others may not. The patches that continue to have populations provide migrants

to colonize extinct patches. Occasionally, the number of individuals within a small patch may

decline and near extinction. However, colonists from a larger patch may arrive to the smaller

patch and “rescue” it from extinction. This process is termed the rescue effect (Hanski 1999). If

there is no change in the area or isolation of the patches within a landscape, then a dynamic

equilibrium should emerge where there is a balance between extinction and colonization within

the metapopulation (Hanski 1994a).

3.2 Types of Metapopulations There are different types of single species metapopulations with variable characteristics,

including but not limited to: the Levins metapopulation (Levins 1969, 1970), mainland-island

metapopulation (Hanski, 1994abc), patchy population metapopulation (Harrison & Taylor,

1997), non-equilibrium metapopulation (Harrison & Taylor, 1997), and the intermediate

metapopulation (Harrison & Taylor, 1997). The Levins metapopulation (Levins 1969, 1979) is

often referred to as the classic metapopulation (Harrison, 1991; Hanski, 1997; Baguette, 2004)

and occurs where there are a large number of small patches with local populations that are

equally prone to extinction with low levels of migration between patches (Fig. 3.1a; Hanski,

1997). Hanski (1997) identifies four conditions that need to be met for a Levin’s metapopulation

model to exist: suitable habitat is in discrete patches, all populations large and small have a high

extinction risk, habitat patches are connected enough to allow for recolonization, and population

dynamics are not synchronous in local populations. This is a simple model and it applies poorly

in fragmented habitats because fragments are not all the same size and are not equally isolated.

As a result, local populations do not have similar levels of extinction risk. Therefore, a more

complicated metapopulation occurs when patches vary in size and distance.

45

Figure 3.1: Five Types of Metapopulations. Filled circles represent occupied patches, empty circles represent un-occupied patches, dotted lines represent local population boundaries, arrows represent dispersal distance and direction: a. Levins (classic); b. mainland-island; c. patchy population; d. non-equilibrium population; and e. intermediate metapopulation. (Adapted from Harrison, 1991).

A mainland-island metapopulation occurs where a patch or population within a fragmented

landscape is particularly large (mainland) and is surrounded by smaller patches. The large

mainland has a large population of individuals that is unlikely to become extinct (MacArthur &

Wilson 1967). However, patches vary in their extinction risk based on their size (Hanski 1994a),

with smaller patches at relatively higher extinction risk than larger patches. Because of its large

size the mainland produces an unlimited supply of migrants called propagule rain (Hanski,

1994ab). The mainland’s unlimited supply of migrants is independent on the number of patches

occupied within the system (Hanski, 1994ab). The colonization potential or isolation of island

patches is related to their distance from the mainland (Hanski, 1994ab). A mainland-island

metapopulation may help explain the source-sink dynamics observed in some metapopulations

(Harrison, 1991). Source-sink dynamics are situations where reproduction among individuals is

lower than mortality (sink) and a population is maintained through migration from a nearby

population that is growing (source; Pulliam 1998).

a. b. c.

d. e.

46

A patchy metapopulation occurs when the habitat patches are not separate enough to create

effective isolation between patches (Harrison, 1991; Fig. 3.1c). In patchy populations, local

populations are so close that it is unlikely that any will become extinct because they are

connected to one another by as much movement between patches as within patches (Harrison,

1991). This high degree of movement between patches violates the assumption of the classical

metapopulation of low levels of colonization between patches and is not generally considered a

true metapopulation (Hanski, 1997; Harrison & Taylor, 1997).

Non-equilibrium metapopulations occur when the assumption of classic metapopulation

dynamics - that extinction and colonization will balance over time - fails (Hanski, 1997). In non-

equilibrium metapopulations, local extinctions occur more than recolonization, resulting in a

declining metapopulation (Harrison, 1991). The classic metapopulation model assumes that

local extinctions within a patch create a patch that can be potentially recolonized. If

deterministic processes such as human disturbance reduce habitat quality or size below a species

threshold, then some patches may be unavailable for recolonization (Harrison, 1991; Fig. 3.1d).

Intermediate metapopulations combine elements of some or all of the above metapopulations

(Harrison, 1991; Fig. 3.1e). The timescale under consideration and the spatial arrangement of

patches make an intermediate metapopulation likely (Harrison, 1991). For example, a

metapopulation is intermediate if it has one large patch that produced an indefinite supply of

migrants (a mainland) but there is some level of migration between the remaining smaller

patches (Harrison, 1991). In such cases, rescue effects may result in patches with a higher

probabilities of being occupied because they are closer to other occupied patches (Harrison,

1991).

3.3 Metapopulation Models The three types of metapopulation models used to describe metapopulations (Hanski, 1994c)

are: spatially implicit models, spatially explicit models, and spatially realistic models.

Researchers can chose which model to use depending on their question, the data available, and

the nature of the landscape configuration within a study.

Spatially implicit models are simple models with fewer variables that incorporate assumptions

about spatial structure in a landscape, such as all patches are the same size (Hanski, 1994c).

47

Spatially implicit models do not consider spatial elements such as patch size or isolation within

a landscape. The earliest type of metapopulation described, the classic Levin’s metapopulation,

is an example of a spatially implicit model. The assumptions of the classic Levin’s

metapopulation model include: extinction risk is equal between each patch because all patches

are the same size, (Hanski, 1994c), local populations are equally connected, population

dynamics within a patch are independent from those among patches, and spatial relationships

among patches are not considered. In the classic Levin’s metapopulation, the proportion of

patch occupancy at any one time is the variable of interest, not patch size or isolation (Hanski,

1997). Applying a classic metapopulation approach to real landscapes is problematic because

the actual configurations and sizes of patches often do not fall within the assumptions of a

classic metapopulation where patches are similar in size with similar levels of isolation.

Spatially explicit models are more complicated with more variables than spatially implicit

models and can include the spatial location and therefore isolation of patches relative to one

another, but they do not incorporate size or quality of patches into the model (Hanski, 1994c).

Spatially explicit models arrange multiple patches as cells in a lattice. Patches are either

occupied or not with no variation in patch size. The distance to adjacent patches is equivalent

but migration between occupied and unoccupied patches are distance dependent (Hanski,

1994c).

Spatially realistic models can incorporate more spatial information about a landscape including

the size, shape, and location of patches and the distances between them (Hanski, 1994c).

Spatially realistic models can even incorporate patch quality (Hanski, 2001). Therefore, they are

more representative than spatially implicit or explicit models. Because landscapes do not always

look like classic metapopulation assumptions (Fig. 3.1a), Hanski (1994a) developed a spatially

realistic stochastic-patch occupancy model, called the incidence function model (IFM), to test

species vulnerability to habitat fragmentation in a realistic context. In an IFM, incidence is a

measure of the probability of occurrence of a species within a patch and is a function of patch

extinction probability and colonization potential (Hanski, 1999). Measuring patch extinction

probability and colonization potential directly is difficult (Hanski, 1999). To determine the

extinction probability of a species within a patch or patch network, it is necessary to acquire

long-term data on mortality of individual primates, who are long lived, within patches of

varying sizes. To determine colonization potential of a patch or patch network, it is necessary to

48

know primate species dispersal abilities between patches over more than one year. For primates

this requires long-term data on species dispersal patterns, which is difficult to measure.

Researchers can determine incidence in a metapopulation model without data on species

extinction or colonization rates. Fortunately, using an IFM, we can infer the extinction

probability of a patch and its colonization potential using simple occurrence data

(presence/absence) gathered from a single-survey period among patches within a fragmented

landscape (Hanski, 1994ab). An IFM uses area as a proxy for extinction risk and isolation as a

proxy for colonization potential. Thus, it describes the probability that a species occurs

(incidence) within a patch as a function of both the area (extinction risk) and isolation

(colonization potential) of that patch (Hanski, 1994ab). The only additional data required are the

sizes and locations of each patch and knowledge of the median-dispersal range of a species

within the landscape. The benefit of an IFM is that it is more realistic because it incorporates

patch area and isolation directly measured from the landscape and it can be easily parameterized

based on occurrence data of species within a fragmented landscape at one particular point in

time.

You can use a mainland-island IFM if you apply three simplifying assumptions. First, in

mainland-island dynamics the mainland can never go extinct (Hanski, 1994b). Second, migrants

only come from the mainland; there is no movement between island patches (Hanski, 1994b).

Third, the colonization potential of a patch is a negative exponential function of the distance

from an island to the mainland (Hanski 1994b).

3.4 Dispersal Like patch extinction probability and colonization potential, gathering dispersal information on

primate species is quite difficult. Understanding dispersal requires long-term data on the

movements of multiple individuals within a population within the landscape of interest

(Sutherland et al. 2000; Bowman et al. 2002; Schliehe-Diecks et al. 2012). We know little about

dispersal for many of the lemur species that occur in Ankarafantsika National Park. However,

researchers have studied dispersal in Microcebus murinus (Radespiel et al. 2003; Schliehe-

Diecks et al. 2012) and M. ravelobensis (Radespiel et al. 2009). Researchers found greater

dispersal distances for M. murinus than M. ravelobensis. Using mark recapture methods on M.

murinus, Radespiel et al. (2003) found that the male median-dispersal range was 251 m and the

49

female median-dispersal range was 63 m. Looking at dispersal movements in M. murinus in

Kirindy Forest, Schliehe-Diecks et al. (2012) found that successful male dispersal events ranged

between 180 m to 960 m. Radespiel et al. (2009) found that M. ravelobensis male natal dispersal

ranged between 7 m to 193 m and female natal dispersal ranged from 11 m to 193 m in

Ankarafantsika National Park.

Although it is difficult to determine a value for primate species median-dispersal ability, it is an

important parameter (α) within an IFM. Fortunately, IFMs are not overly sensitive to variations

in the value of α (Hanski, 1994a). There are also solutions to determining 𝛼 when dispersal data

are lacking. One solution is to parameterize α using the occurrence data from a single-survey

period. This method involves inputting different values of α into the IFM and selecting the best-

fitting value (Hanski, 1999). The effect of using occurrence data to generate one of the

parameters used in the IFM also means that the standard errors of the model are biased

(Oksanen, 1994). Another solution is to use a proxy for dispersal. Mammal vagility affects both

home range size and dispersal distance, independent of body size (Bowman et al. 2002). In a

study on terrestrial and arboreal North American mammals, Bowman et al. (2002) demonstrate

that dispersal distance is related to a single constant value, the linear dimension of home range

size. Whitmee & Orme (2013) also found that home range was the most important predictor

variable related to median-dispersal distance in mammals. Although similar to Bowman et al.

(2003), Whitmee & Orme (2013) used mostly mammals from the Nearctic and Palearctic, they

also included five arboreal primate species. From a biogeographic perspective, Beaudrot and

Marshall (2011) demonstrate that dispersal limitation determines community structure in

primates including lemurs. Bannar-Martin (2014) also found spatial and dispersal limitations

determine lemur community assembly while other researchers found that lemurs were

ecologically limited (Kamilar, 2009). If lemurs are dispersal limited like other primates, then the

results from Bowman et al. (2002) may not hold for this group. However, the constant Bowman

et al. (2002) created allows for the use of home range data, which is readily available for many

primate species, to determine dispersal ability.

3.5 Effect of Additional Variables within Incidence Function Models

One potential benefit of IFMs is their ability to incorporate other variables that could impact the

extinction risk or colonization potential of a patch (Moilanen & Hanski, 1998). It is

50

straightforward to add an additional variable to the regression form of a metapopulation model

but it is very difficult to tease out how the additional variable affects either or both extinction

risk and colonization potential of a patch within the model (Oksanen, 2004). To do so requires

in-depth knowledge on how the additional variables impact extinction risk and colonization

potential a priori (Moilanen & Hanski, 1998).

The results of incorporating additional variables into an IFM are mixed (Moilanen & Hanski,

1998; Lawes et al. 2000; Jaquiéry et al. 2008). Incorporating additional variables that affect

species occurrence, such as habitat quality and landscape structure, did not improve the fit over

metapopulation models that only incorporated area and distance metrics (Moilanen & Hanski,

1998). Jaquiéry et al. (2008) found that population size of greater white-toothed tree shrews was

better explained by using habitat quality (factoring in human disturbance) over patch size.

Lawes et al. (2000) did not find any measures of human disturbance that improved the fit of

metapopulation models on Cercopithecus mitis labiatus (samango monkey). However, they did

find that the addition of human disturbance variables improved the fit of metapopulation models

for Philantomba monticola (blue duikers) and Dendrohyrax arboreus (tree hyraxes). I found

that area was the main factor influencing lemur species at the community level and

incorporating additional variables into metapopulations met with mixed results (Chapter 2).

Therefore, I will not add additional variables to metapopulation models in this chapter.

3.6 Simulating Metapopulation Dynamics Over Time Metapopulation dynamic models produce two results: species-specific extinction and

colonization potentials for animals in each patch within a set of patches. Once the extinction and

colonization probabilities of a patch are known, simulations can be run to model species

occurrence over time (Hanski, 1998). Simulating metapopulation dynamics is very useful to

determine whether a metapopulation is in decline. These simulations are also useful for

artificially altering the system through the addition or removal of patches to determine how

changes in landscape composition and structure affect species occupancy over time.

3.7 Metapopulation Dynamics: Forest Loss and Fragmentation Effects on Primate Occurrence

There is no research on metapopulation dynamics in lemurs and few studies on primates (Lawes

et al. 2000; Chapman et al. 2003; Mandujano & Escobedo-Morales, 2008). Some researchers

51

have used metapopulation theory as a conservation tool to: predict species persistence in a

fragmented landscape under varying conservation strategies (Swart & Lawes, 1996; Chapman et

al. 2003), assess extinction risk (Zeigler et al. 2013), determine factors impacting species

occurrence (Lawes et al. 2000), determine species minimal critical patch size (Lawes et al.

2000), and for population viability analyses (Mandujano & Escobedo-Morales, 2008). Other

studies have investigated the impact of patch size and isolation within suspected

metapopulations but they did not use metapopulation dynamics per se (Rodriguez-Toledo et al.

2003). In forest fragments along the periphery of Kibale National Park, Uganda, Chapman et al.

(2003) fitted a mainland-island incidence function model to occurrence data on four primate

species. The metapopulation models accounted for a substantial amount of variation in each

species occurrence. However, they found low confidence in the estimated coefficients for the

models. For both Procolobus badius (red colobus) and Colobus guereza (black and white

colobus), Chapman et al. (2003) found a strong area effect on occurrence but little influence of

connectivity on each species occurrence. For Cercopithecus ascanius (redtail monkey) fragment

size or distance did not affect occurrence while Chapman et al. (2003) found Pan troglodytes

(chimpanzees) were an unsuitable species for the application of metapopulation dynamics

because of their highly mobile nature. In a fragmented portion of Podocarpus forest in

KwaZulu-Natal Province, South Africa, Lawes et al. (2000) applied a mainland-island incidence

function model to C. m. labiatus occurrence and additional land use and environmental factors.

Lawes et al. (2000) found that the best-fit model incorporated only area as a factor determining

C. m. labiatus occurrence. Model fit was not improved by the inclusion of isolation, land use, or

other environmental factors.

The effects of forest loss and fragmentation on primate species occurrence are well studied

(Chapman et al. 2003; Rodriguez-Toledo et al. 2003; Arroyo-Rodriguez & Dias, 2010; Raboy et

al. 2010; Arroyo-Rodríguez et al. 2013). Typically, primate species occurrence decreases with

increased forest loss (Anzures-Dadda & Manson, 2007; Arroyo-Rodríguez et al. 2008; Boyle &

Smith, 2010; Raboy et al. 2010). For example, Anzures-Dadda & Manson (2007) show that for

Alouatta palliata, the probability of occurrence in a fragment was positively related to area.

Similarly, Boyle et al. (2010) found that for five of six primate species, occurrence was also

positively related to fragment size in the Brazilian Amazon.

52

Forest separated into smaller and more isolated patches reduces the amount of habitat and can

impact fragment connectivity. Several studies have focused on how habitat connectivity (or

isolation) between fragments affects primates (Lawes et al. 2000; Cristobal-Azkarate et al. 2005;

Marshall et al. 2010; Raboy et al. 2010). Although community structure may be determined by

geographic distance across the island of Madagascar (Beaudrot & Marshall, 2011), isolation has

so far been observed to have little to no effect on individual primate occurrence when compared

to fragment area (Lawes et al. 2000; Cristobal-Azkarate et al. 2005; Raboy et al. 2010). Thus,

scale dependence in biogeographic research represents an important consideration when seeking

to apply models from patches through landscapes to regions. For example, Cristobal-Azkarate et

al. (2005) found no relationship between isolation, as measured by the shortest distance to

nearest fragment and shortest distance to nearest fragment with monkeys, and monkey

occurrence. Marshall et al. (2010) did find that isolation of forests by farmland had a significant

effect on species richness. However, they suggest that the type of matrix between fragments was

more important than isolation distance. Raboy et al. (2010) found that patch isolation did not

predict golden-headed lion tamarin (Leontopithecus chrysomelas) occurrence. The effect of

habitat fragmentation separate from habitat loss on primate occurrence is not well understood

(Chapman et al. 2003; Marsh, 2003; Cardillo et al. 2008; Arroyo-Rodríguez et al. 2013b). In

order to better understand how fragmentation impacts primate species, researchers need to

further assess how connectivity, independent of habitat loss, affects primate occurrence (Hanski

& Ovaskainen, 2003; Arroyo-Rodríguez et al. 2013b).

3.8 Justification Although lemurs represent the largest biomass and biodiversity of mammals in Madagascar

(Garbutt, 2007), the forested habitats preferred by lemurs have decreased and continue to

decrease dramatically (Schwitzer et al. 2014). Forty percent of forest cover was converted to

non-forested habitat in Madagascar between the 1950s and 2000s (Harper et al. 2007; Schwitzer

et al. 2014), mostly due to human-induced disturbance (Schwitzer et al. 2014). As a result of

this, lemurs are currently considered the most endangered mammal group in the world, with

94% of the species threatened with extinction (Schwitzer, 2013).

In western Madagascar, the forest is mostly tropical dry deciduous forest, which is extremely

sensitive to fire (Bloesch, 1999). Fire has resulted in a high degree of forest loss and increased

53

habitat fragmentation (DeFries et al. 2005; Harper et al. 2007). There are a few bastions of intact

dry forest in the west and Ankarafantsika National Park contains the second largest portion of

continuous dry forest. However, habitat loss and fragmentation continue along the periphery of

these continuous forest patches (DeFries et al. 2005). Therefore, lemurs along the forest edge in

protected areas such as Ankarafantsika could be subject to increased habitat fragmentation.

Most researchers conduct their primate research near the center of Ankarafantsika at Ampijoroa

(Warren & Compton 1997; Radespiel 2000; Radespiel et al. 2001; Radespiel et al. 2003a;

Radespiel et al. 2003b; Rendigs et al. 2003; Weidt et al. 2004; Braune et al. 2005; Radespiel et

al. 2006; Olivieri et al. 2008; Radespiel et al. 2009; Thorén et al. 2010; Thorén et al. 2011;

Crowley et al. 2011,2012; Eichmueller et al. 2013; Kun-Rodrigues et al. 2014;) with some

exceptions (García & Goodman, 2003; Radespiel et al. 2008; Craule et al. 2009;

Rakotondravony & Radespiel 2009). Preliminary research indicates considerable differences in

lemur biogeography between continuous and fragmented forests (Craule et al. 2009; Crowley et

al. 2013; Steffens & Lehman, 2016). For example, Steffens and Lehman (2016) found that

contrary to a previous study of M. murinus and M. ravelobensis in nearby continuous forest

(Rakotondravony & Radespiel 2009), there were significant, positive correlations between

density and abundance for both species in fragments. Knowing that there are differences in

biogeographic patterns in some species in continuous versus fragmented habitat the question

arises as to how other species will respond to increased habitat fragmentation? Thus,

understanding spatial variations in lemur responses to forest fragmentation is critical to a more

informed understanding of lemur conservation biogeography.

My study is important because it is the first of its kind to use metapopulation dynamics to

determine how six lemur species (including three endangered species) respond to reduction in

habitat and changes in habitat connectivity. Moreover, it is imperative that we determine how

lemurs are affected by habitat loss and fragmentation because they face high extinction

probabilities, are the most endangered animal group in the world, and their habitat is

increasingly fragmented and disturbed.

3.9 Goal The goal of this study is to apply a spatially realistic metapopulation approach to investigate the

vulnerability to habitat loss and fragmentation of six lemur species living in a fragmented

54

landscape in Ankarafantsika National Park. Using metapopulation dynamics, I will answer the

following questions:

1) Do lemur species form metapopulations?

Hypotheses:

H0 Individual lemur species occurrence will vary randomly with respect to patch area

and isolation.

H1 A spatially realistic metapopulation model (incorporating area and connectivity) will

predict lemur species occurrence.

2) What are the separate effects of area (extinction risk) and connectivity/isolation

(colonization potential) within a lemur metapopulation?

Hypotheses:

H0 Incidence of occurrence and extinction probability will not differ between large and

small patches.

H1 Incidence of occurrence and extinction probability will be higher and lower

respectively in large versus small patches.

H0 Incidence of occurrence and colonization probability will not differ between more

and less connected patches.

H1 Incidence of occurrence and colonization probability will be higher in more

connected versus less connected patches.

3) Within simulated metapopulations over time, how do area and connectivity/isolation

affect occurrence? What are the conservation implications?

55

3.10 Methods


I conducted this study in Ankarafantsika National Park, Madagascar. I describe the study site in

detail in Chapter 2. The study landscape is entirely within the park boundaries. There are eight

species of lemurs within four families in the park (Table 3.1): two Indriidae: the western wooly

lemur (A. occidentalis) and Coquerel’s sifaka (P. coquereli); three Cheirgaleidae: the grey

mouse lemur (M. murinus), the golden brown mouse lemur (M. ravelobensis) and the fat-tailed

dwarf lemur (C. medius); one Lepilemuridae: Milne Edwards sportive lemur (L. edwardsi); and

two Lemuridae: the mongoose lemur (E. mongoz), and the common brown lemur (E. fulvus)

(Mittermeier et al. 2010). Table 3.1: Primate Species in Ankarafantsika National Park Found within Study Site.

Species Body mass (g)

Activity pattern Diet Mean home

range (ha)

Median-dispersal

distance (m) based on

home range*

Median/ Mean- dispersal

distance reported in

literature (m) Cheirogaleus

medius 120-270 Nocturnal Frugivore 1.55 ±0.42(1) 873 N/A

Microcebus murinus 58-67 Nocturnal Fauni-

frugivore 2.83 ±1.44(2) 1177 Median=251(7)

Microcebus ravelobensis 56-87 Nocturnal Fauni-

frugivore 0.59 ±0.11 (3) 530 Mean=54(8)

Propithecus coquereli

3700-4300 Diurnal Folivore 19.36(4) 3080 N/A

Eulemur fulvus 1700-2100 Cathemeral Frugivore 13.5(5) 2572 N/A

Lepilemur edwardsi 1100 Nocturnal Folivore 1.09(6) 731 N/A

*Median-dispersal distance was calculated as seven times the square root of the mean reported home range from each study. Data from: 1 Müller 1998; 2 Radespiel (2000); 3 Weidt et al. (2004); 4 McGoogan (2011); 5 Mittermeier et al.2010; 6 Warren & Compton (1997); 7 Radespiel et al. (2003); 8 Radespiel et al. (2009).

Cheirogaleidae

The Cheirogaleidae are characterized by their nocturnal behavior and small body size. Most

species are omnivorous but many, such as Microcebus species, rely mainly on fruit and insects

(Lahann, 2007). In the study landscape, two species of the same genus potentially occur: M.

murinus and M. ravelobensis. Morphological differences between the two Microcebus species

include color of pelage in that M. murinus has greyish brown pelage, while M. ravelobensis has

golden brown pelage (Zimmermann et al. 1998). Differences also include the size and thickness

of tail with M. murinus having a shorter, thicker tail than M. ravelobensis (Zimmermann et al.

56

1998). There are also interspecific differences in diet and behavior between the two Microcebus

species. Although the there are similarities in both Microcebus species’ diet, M. ravelobensis

appears more generalized than M. murinus (Radespiel et al. 2006). M. murinus exhibits

behavioral torpor during the dry season while M. ravelobensis is more active (Thorén et al.

2011). M. murinus prefers higher elevation drier forest and M. ravelobensis prefers lower

elevation forest closer to water (Rakotondravony & Radespiel, 2009).

The single Cheirogaleus species C. medius is larger than the two Microcebus species, ranging in

mass from 120 g to 270 g (Mittermeier et al. 2010). This species consumes greater amounts of

fruit than either Microcebus species, and hibernates during much of the dry season from April to

October (Fietz & Ganzhorn, 1999; Dausmann et al. 2005). To survive these long bouts of

hibernation, C. medius stores fat in its tail, thus increasing its body mass (Fietz & Ganzhorn,

1999). Prior to hibernation, C. medius reduces its activity while increasing high sugar fruit

consumption (Fietz & Ganzhorn, 1999). However, following hibernation C. medius increases its

activity level (Fietz & Ganzhorn, 1999).

Indriidae

There are two species from the Indriidae family that occur within the study landscape: P.

coquereli and A. occidentalis. P. coquereli is the largest species found in Ankarafantsika

(between 3700 g and 4300 g). It is diurnal and social, living in groups of between 2 and 10

individuals (McGoogan, 2011). This species is mainly folivorous, but also consumes fruit

(McGoogan, 2011). A. occidentalis is nocturnal, highly folivorous, and lives in adult pairs with

offspring (Mittermeier et al. 2010). This species is smaller in size than P. coquereli but similar

in size to Lepilemur edwardsi. A. occidentalis are distinguished from L. edwardsi by the white

patches on the back of their legs (Mittermeier et al. 2010).

Lemuridae

Two Lemuridae species are known to occur within the park: E. mongoz and E. fulvus. Both

Eulemur species are highly frugivorous, but they supplement their diet with leaves and flowers

during periods of fruit scarcity (Mittermeier et al. 2010). Both species are cathemeral in that

they change their activity pattern seasonally (Curtis et al. 1999; Rasmussen, 1999). E. fulvus

forms groups of 3 to 12 individuals while E. mongoz groups are smaller, usually consisting of

57

one breeding pair and their offspring. E. mongoz males are easily distinguished from E. fulvus

by their orange-colored chins. Female E. mongoz are greyer than E. fulvus females.

Lepilemuridae

There is only one member of the Lepilemuridae family found in the park: L. edwardsi. This

species is mainly folivorous, nocturnal, solitary, and is similar in size to A. occidentalis, but

darker in color with a more pointed face (Mittermeier et al. 2010).

3.10.2 Question 1: Do Lemur Species Form Metapopulations?

Although there are some differences in meaning between the terms patch and fragment they are

often used interchangeably. I define patches as discrete and measurable portions of habitat that

differ from neighboring portions, while I define fragments as discrete and measurable portions

of habitat that have been separated from a larger whole. A fragment may contain many patches

and a patch may not necessarily be a fragment. In this chapter, I will use the term patches when

referring to the metapopulation literature and fragments when a referring to my study site where

each fragment is considered a single habitat patch.

The data needed for a metapopulation model include at least a single survey of patch occupancy

within a network of patches, the x and y coordinates of each patch to determine the distance

between each patch, patch area, and the species-specific dispersal ability within the landscape

(Hanski, 1999). It is relatively easy to gather occurrence data for primates compared to

determining their population density. It is also easy to measure patch area and distances using

current GPS and GIS technology. However, it is very difficult to know the dispersal ability of

primates within a landscape.

To test the hypothesis that metapopulation dynamics explain occupancy patterns in lemur

species, I compared five or six potential metapopulation models including a null model for each

of the six species found in the study site. The metapopulation models include four incidence-

function models with rescue effect (IFM) and one mainland-island incidence function model

without rescue effect (MI-IFM). I applied the MI-IFM and null models to all six lemur species

occurrence data. The remaining four IFMs are similar but differ in the way the dispersal

parameter (α) was defined. In the first IFM, α is parameterized based on occupancy data from

one survey period (IFM). I applied the IFM model to all six lemur species occurrence data. The

58

next two IFMs, use a proxy for the dispersal parameter (α) based on the home range reported for

each species in the literature (IFMproxy and IFMprox2). I calculated the first proxy for median-

dispersal ability as the square root of the mean home range multiplied by seven (Equation (17);

IFMproxy) and I calculated the second as the square root of the mean home range (IFMproxy2).

I applied the IFMproxy and IFMproxy2 models to all six species occurrence data. For the final

IFM model, the dispersal parameter (α) is based on actual dispersal data from the literature

(IFMlit). I applied the IFMlit model only to those species where there was data on dispersal

ability (M. murinus and M. ravelobensis).

For each model, I input data on patch occupancy for each species, a measure of species-specific

dispersal distance (α), patch area, and isolation. I collected data on patch occupancy, patch area,

and isolation using the methods outlined in Chapter 2. I then took a linearized version of the

IFM models (see equation (9) below) and the MI-IFM model (see equation (15) below) and ran

binomial generalized linear models (GLM) with a logit link function using the glm{stats}

function in R (version 3.02), for each species. For each GLM I ran I input species occurrence as

the response variable and species-specific dispersal distance (α), patch area, and isolation as the

predictor variables.

Model Comparison

To determine the relative strength of the six fitted models (IFM, IFMproxy, IFMproxy2, IFMlit,

MI-IFM, and Null), I calculated corrected Akaike's information criterion (AICc) and then

performed a model averaging procedure to produce AIC weights (wi; Burnham, Anderson, &

Huyvaert, 2011). I considered the model with the highest wi as the model with the highest

likelihood of being selected among the six models (Burnham et al. 2011). I considered models

with AICc values within two of the model with the lowest AICc as potential candidate models. I

did not consider any models with AICc values greater than two of the lowest AICc as candidate

models.

Incidence Function Model with Rescue Effect (IFM; IFMproxy; IFMproxy2; IFMlit):

The IFM with rescue effect takes the following form (Hanski, 1999):

𝑱𝒊 =𝑪𝒊

𝑪𝒊&𝑬𝒊(𝑪𝒊𝑬𝒊(1)

59

𝐸+ =𝑒𝐴+. , 𝑓𝑜𝑟𝐴 ≥ 𝑒

4.(2)

𝑀+ = 𝛽𝑆 = 𝛽(3)

𝐶+ =𝑀+=

𝑀+= + 𝑦=

=𝑆+=

𝑆+= + 𝑦,𝑤ℎ𝑒𝑟𝑒𝑦𝑎𝑏𝑠𝑜𝑟𝑏𝑠𝛽(4)

𝐽+ =𝑆+=𝐴+.

𝑆+=𝐴+. + 𝑒𝑦=

1

1 + 𝑒𝑦𝑆+=𝐴+.

= 1 +𝑒𝑦𝑆+=𝐴+.

(4

(5)

where Ji is patch incidence in patch i, Ei is the extinction probability, Ci is the colonization

probability, Si is a measure of connectivity for patch i, and Ai is the area of patch i. It is difficult

to estimate extinction and colonization directly (Hanski, 1999), however, it is possible to

calculate e and y with data on patch occupancy 𝒑𝒋, patch size A, and connectivity S collected

during a single time period survey. Connectivity is estimated using the following:

𝑆+ = exp −𝛼𝑑+P 𝑝PR

PS+𝐴P(6)

where 𝜶 is inverse of the median species-specific dispersal distance and 𝒅𝒊𝒋 is a distance matrix

among patches. It is possible to change equation (5) by applying a linear model for the log-odds

of incidence:

𝐽+ 1 +𝑒𝑦𝑆+=𝐴+.

(4

=1

1 + 𝑒𝑥𝑝 𝑙𝑜𝑔 𝑒𝑦 − 2 𝑙𝑜𝑔 𝑆+ − 𝑥𝑙𝑜𝑔 𝐴+(7)

𝑙𝑜𝑔𝐽+

1 − 𝐽+= − 𝑙𝑜𝑔 𝑒𝑦 + 2 𝑙𝑜𝑔 𝑆+ + 𝑥𝑙𝑜𝑔 𝐴+ (8)

A logit transformation results in:

𝑙𝑜𝑔𝑖𝑡 𝐽+ = 𝛽^ + 2𝑙𝑜𝑔𝑆 + 𝛽4×𝑙𝑜𝑔𝐴(9)

The mainland-island incidence function model without rescue effect takes the following form

(MI-IFM; Hanski, 1994b):

60

𝐽+ =𝐶+

𝐶+ + 𝐸+(10)

𝐶+ = 𝑞𝑒𝑥𝑝 −𝛽𝐷+ 11

where q and 𝛽 are two parameters. Assuming that all the species are common on the mainland,

where 𝐶+ approaches one when 𝐷+ approaches zero than q=1 and equation (11) can be simplified

further (Hanski, 1994b):

𝐸+ =𝑒𝐴+. , 𝑓𝑜𝑟𝐴 ≥ 𝑒

4. 12

𝐽+ = 1 +𝜇efg𝐴+.

(4

13

It is possible to linearize equation (13) by applying linear model for the log-odds of incidence:

𝑙𝑜𝑔(𝐽+) = −log 𝜇 − 𝛽𝐷+ + 𝛽= log 𝐴k (14)

A logit transformation results in:

𝑙𝑜𝑔𝑖𝑡(𝐽+) = 𝛽^ − 𝛽4𝐷 + 𝛽= log(𝐴) (15)

Patch Occupancy

I determined patch occupancy of each fragment using standard line-transect methods (see

Chapter 2 for a full description). If we did not observe a species during any survey, I considered

it absent within a fragment. I conducted between 11 and 18 diurnal and 11 and 21 nocturnal

surveys of each fragment.

Dispersal distance

To determine connectivity within each IFM, I needed to determine the dispersal parameter (α).

However, there is limited data on dispersal ability in most primates, especially lemurs. For the

species in this study, previous researchers reliably measured dispersal distance only in M.

murinus (Radespiel et al. 2003; Schliehe-Diecks et al. 2012) and to a lesser degree M.

ravelobensis (Radespiel et al. 2009). Therefore, I ran multiple IFM models incorporating

61

different dispersal parameters. For each species, I ran three IFMs using different measures for

dispersal (α), except for M. murinus and M. ravelobensis for which I ran four IFMs with

different measures for dispersal (α). For the first IFM, I determined which α fit the survey data

by running a function in R that ran all possible values for α and selected the one that provided

the lowest deviance (IFM; Oksanen, 2004). For the second IFM, I used a proxy for median-

dispersal distance based on a function of the home range size of each species (IFMproxy) from

Bowman et al. (2002). Bowman et al. (2002) found that median-dispersal distance could be

calculated as the linear dimension (square root) of the mean home range multiplied by a factor

of 7. Therefore, I took the mean reported home range for each species and determined its

dispersal ability with the following formula:

ℎ𝑜𝑚𝑒𝑟𝑎𝑛𝑔𝑒(𝑘𝑚=)×7 = 𝑚𝑒𝑑𝑖𝑎𝑛𝑑𝑖𝑠𝑝𝑒𝑟𝑠𝑎𝑙𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒 𝑘𝑚 (16)

Alpha (𝛼) is calculated as:

4pqr+str+uvqwusxr+uystzq

(17)

The proxy for dispersal using the formula from Bowman et al. (2002) overestimated the median-

dispersal of the two known species (M. murinus and M. ravelobensis; Table 3.1). For the third

IFM (IFMproxy2), I created a second proxy where I took the linear dimension of the mean

reported home range (i.e. ℎ𝑜𝑚𝑒𝑟𝑎𝑛𝑔𝑒(𝑘𝑚=)), because lemurs like many arboreal primates

may be dispersal limited (Bannar-Martin, 2014) and the value derived from formula (16) may

overestimate median-dispersal distance. The values derived using the linear dimension of the

home range better fit the known home range and dispersal distances for M. murinus and M.

ravelobensis (Table 3.1). For the final IFM (IFMlit: M. murinus and M. ravelobensis only), I

used the largest median-dispersal distance reported for each species regardless of sex (M.

murinus = 251 m (Radespiel et al. 2003); M. ravelobensis = 54 m (Radespiel et al. 2009).

Patch Area and Isolation

I measured the area and isolation of each fragment using the methods detailed in Chapter 2.

62

3.10.3 Question 2: What are the Separate Effects of Area (Extinction Risk) and Connectivity/Isolation (Colonization Potential) within a Lemur Metapopulation?

To test the hypothesis that incidence probability was higher in larger fragments than smaller

fragments, I determined the incidence probability for each patch (Ji) based on equation (1;

among patch models) or (10; mainland-island model) depending on which candidate models I

selected in section 2.2. However, in order to calculate incidence probability (Ji), I needed to

determine the extinction probability (Ei) and colonization probability (Ci) for each patch. For

the among patch models, with known patch sizes, patch occupancies, and connectivity, I used a

generalized linear binomial model with logit link function (equation (9)) to determine the

coefficients 𝛽{ and 𝛽4 = 𝑥 and to separate e from y to calculate the extinction and colonization

probabilities for each patch using the following steps:

𝑒𝑦 = exp −𝛽{ (18)

𝑒 = minvS{

𝐴. (19)

𝑦 = 𝑒𝑦/𝑒 (20)

Once I separated e and y, I calculated the extinction probability (𝐸+) and colonization probability

(𝐶+) and subsequently incidence probability (Ji) of each patch using equations (2) and (4)

respectively.

For the mainland-island model, I also used a generalized linear binomial model from a single

survey of patch occupancy on equation (15). Using this equation, I was able to determine the

coefficients 𝛽{ = 𝜇, 𝛽4 = 𝛽 and 𝛽= = 𝑥 to calculate the colonization and extinction

probabilities using equations (11) and (12) respectively. I then input the values from the

colonization and extinction probabilities into equation (10) to determine the incidence

probability for each patch.

For both the among patch and the mainland-island model, I sorted fragments by descending size

and divided them in half: the top half represented the largest fragments and the bottom half

represented the smallest fragments. To test the hypothesis that incidence probability was higher

in larger fragments than smaller fragments, I compared the mean incidence probability for the

63

largest fragments with the mean incidence probability for the smallest fragments using a one-

tailed t-test for each species in R.

Using the incidence probability calculated above, I then determined whether incidence

probability was higher in more connected/closer fragments than less connected/further

fragments. In the among patch models, fragment connectivity (Si; equation (6)) incorporates a

species-specific dispersal parameter. Therefore, patch connectivity was species-specific. For

each species, I sorted connectivity (Si) in descending order and divided Si in half —the top half

representing the most connected fragments, and the bottom half representing the least connected

fragments. In the mainland-island model, I sorted connectivity (the distance from the mainland)

and divided this distance in half - the top half representing the closest fragments (most

connected), and the bottom half representing the furthest fragments (least connected). To test the

hypothesis that incidence probability was higher in more connected than in less connected

fragments, I compared the mean incidence probability for the most connected/closer fragments

with the mean incidence probability of the least connected fragments using a one-tailed t-test for

each species in R.

3.10.4 Question 3: Within Simulated Metapopulations Over Time, How Do Area and Connectivity/Isolation Affect Occurrence? What are the Conservation Implications?

To see if there was a difference in how area affected occurrence compared to connectivity, I

simulated metapopulation dynamics for each species over time based on the extinction and

colonization probabilities derived from the IFM in R. I simulated the metapopulation dynamics

in the following manner: If a patch is empty (pi = 0) at time (t) then the probability that it will

become occupied (pi=1) is a function of the colonization probability (𝐶+(𝑡) =�g�(y)

�g�(y)&�

) for that

patch. If the patch is occupied (pi=1) at time (t) the probability that it will become extinct (pi =

0) is a function of the extinction probability for that patch (Ei(t)=𝐸+(𝑡) = 𝑒𝐴+(.).

One can run simulations as a single-time step or multiple-time steps. However, in multiple time

steps patch occupancy is based on the previous step only. I ran two sets of simulations. The first

set represents a worst case scenario where I ran a simulation separating out the five largest

fragments and a second simulation where I separated out the five most connected fragments.

The second set represents the opposite scenario where fragments that are smaller and least

64

connected were removed from the simulation. For each species, I then ran the two sets of two

simulations for 200 time steps (equivalent of 200 years). In the first half of the simulation, for

the first 100 steps all 42 fragments contribute to the metapopulation. At time step 101, I

separately continue simulations on the five removed fragments (i.e. five largest, five smallest,

five most connected, five least connected) from the remaining 37 fragments for 99 more time

steps. This method allows the ability to look at what happens to lemur species occupancy in the

remaining 37 fragments when the five largest/smallest and five most/least connected fragments

are removed. In addition, this method gives the ability to look at what happens to lemur species

occupancy in the five largest/smallest and five most/least-connected fragments independent of

the remaining 37.

3.11 Results


There were 42 fragments within the study landscape with a mean size of 0.09 ± 0.20 km2. The

mean distance between centroids of each fragment was 2.82 ± 1.36 km with a range of 0.15 km

to 6.48 km. The proxy for median-dispersal distance ranged from 700 m (M. ravelobensis and L.

edwardsi) to 3357 m (P. coquereli; Table 3.1). Patch occupancy differed among species.

Smaller-bodied Cheirogaleids occupied the largest number of fragments while the remaining

three larger species occupied the fewest (Table 3.2). However, a linear regression of occupancy

versus body size yielded no relationships. I did not observe any A. occidentalis or E. mongoz

individuals during this study. Therefore, I did not include these species in the analysis. Table 3.2: Lemur Patch Occupancy in a Fragmented Landscape.

Species Number of occupied patches Number of unoccupied patches

Cheirogaleus medius 12 30 Microcebus murinus 35 7

Microcebus ravelobensis 34 8 Propithecus coquereli 3 39

Eulemur fulvus 7 35 Lepilemur edwardsi 2 40

The probability of occurrence differed among species (Fig. 3.2). Both Microcebus species had

the highest probability of occurrence in the landscape followed by C. medius. E. fulvus had the

lowest probability of occurrence within the landscape.

65

Figure 3.2: Probability of Occurrence Among Patches for Four Lemur Species in a Fragmented Landscape. Colors represent the probability of occurrence: red reflects the highest probability of occurrence for a species within a fragment and white the lowest. The probability of occurrence is based on the fitted incidence function model with α parameterized from the data (IFM) for C. medius, M. murinus, and E. fulvus and the IFM with α determined from the literature (IFMlit) for M. ravelobensis. The size of each circle represents the size of each fragment relative to one another. The position of fragments is based on Universal Transverse Mercator (UTM) coordinate system. Northing is equivalent to latitude and easting is equivalent to longitude.

I found differences in model selection results between species (Table 3.3). For C. medius, two

models had nearly identical low AICc values. The first was the IFM (model where I determined

dispersal (α) based on the survey data) and the second was the IFMproxy2 (model where I

determined dispersal (α) based on the linear dimension of the mean reported home range). For

M. murinus, the model with the lowest AICc value was the MI-IFM and the only other model

within two AICc values was the IFM. I found a different model for M. ravelobensis to have the

lowest AICc value, IFMlit (where I determined dispersal (α) from the literature), followed by

three other models within two AICc values, MI-IFM, IFMproxy2, and IFM. In fact, all four

66

models were roughly equivalent. For both P. coquereli and L. edwardsi, there were no models

with values lower than or within two of the null model. For E. fulvus, the model with the lowest

AICc value was IFM with IFMproxy (where I determined dispersal (α) a proxy based on home

range) as the only model within two AICc values. Therefore, I was able to reject the null

hypothesis that occurrence varied randomly with respect to area and connectivity for all species

except for P. coquereli and L. edwardsi. Table 3.3: Metapopulation Models of Six Lemur Species in 42 Fragments in a Fragmented Landscape.

Species Modela Null Deviance

Residual Deviance DoF P-value AICc ΔAICc wi

Cheirogaleus medius

IFM 52.78 26.69 41 <0.01 31.00 0.00 0.37 IFMproxy2 52.98 26.74 41 <0.01 31.05 0.05 0.36 IFMproxy 57.72 29.28 41 <0.01 33.00 2.00 0.14 MI-IFM 50.25 26.49 41 <0.01 33.12 2.12 0.13

Null 8.57 8.57 41 N/A 56.75 25.75 0.00

Microcebus murinus

MI-IFM 37.85 25.25 41 <0.01 31.88 0.00 0.53 IFM 39.86 29.35 41 <0.01 33.66 1.78 0.22

IFMproxy2 40.76 30.07 41 <0.01 34.38 2.50 0.15 IFMlit 41.94 31.00 41 <0.01 35.30 3.43 0.10 Null 5.83 5.83 41 N/A 40.59 8.71 0.01

IFMproxy 56.62 41.77 41 <0.01 46.07 14.20 0.00

Microcebus ravelobensis

IFMlit 42.29 28.94 41 <0.01 33.25 0.00 0.25 MI-IFM 40.9 26.65 41 <0.01 33.28 0.03 0.25

IFMproxy2 42.48 20.07 41 <0.01 33.38 0.13 0.24 IFM 42.71 29.22 41 <0.01 33.53 0.28 0.22

IFMproxy 48.04 32.59 41 <0.01 36.90 3.65 0.04 Null 6.48 6.48 41 N/A 44.98 11.73 0.00

Propithecus coquereli

Null 2.77 2.77 41 N/A 9.55 0.00 0.56 IFM 27.0 7.54 41 <0.01 11.84 2.30 0.18

IFMproxy2 27.65 7.76 41 <01 12.06 2.52 0.16 MI-IFM 21.61 6.32 41 <0.01 12.95 3.41 0.10

IFMproxy 47.72 13.32 41 <0.01 17.63 8.08 0.01

Eulemur fulvus

IFM 40.18 8.4 41 <0.01 12.71 0.00 0.48 IFMproxy 41.65 8.44 41 <0.01 12.74 0.04 0.47

IFMproxy2 37.36 14.22 41 <0.01 18.53 5.82 0.03 MI-IFM 37.85 12.9 41 <0.01 19.53 6.82 0.02

Null 5.83 5.83 41 N/A 40.59 27.88 0.00

Lepilemur edwardsi

Null 1.91 1.91 41 N/A -6.42 0.00 1.00 MI-IFM 16.08 6.17 41 <0.01 12.80 19.22 0.00

IFM 25.96 14.87 41 <0.01 19.17 25.59 0.00 IFMproxy2 25.97 14.87 41 <0.01 19.18 25.60 0.00 IFMproxy 29.03 16.41 41 <0.01 20.72 27.14 0.00

a IFM= incidence function model where α was parameterized based on occupancy data from one survey period; MI-IFM= mainland-island incidence function model; IFMproxy= IFM where α was calculated as a proxy for dispersal ability based on the square root of the mean home range multiplied by seven reported for each species in the literature; IFMproxy2= IFM where α was calculated as a proxy for dispersal ability based on the square root of the mean home range reported for each species in the literature IFMlit= IFM where α was based on the literature for species where data has been reported on dispersal ability; DoF= Degrees of freedom.

67

3.11.2 Question 2: What are the Separate Effects of Area (Extinction Risk) and Connectivity/Isolation (Colonization Potential) on a Lemur Metapopulation?

I found that the mean probability of occurrence for all species was significantly greater in large

fragments compared to smaller fragments (Table 3.4). However, there were species- and model-

specific differences among the study species in the probability of occurrence between the least

connected/further and more connected/closer fragments (Table 3.5). For C. medius, there was no

significant difference in the probability of occurrence in more or less connected fragments. For

M. murinus, there was a significant negative difference in the probability of occurrence in

further rather than closer fragments within the MI-IFM. However, I found no significant

difference in the IFM for M. murinus. For M. ravelobensis, for both the IFMlit and MI-IFM the

probability of occurrence was significantly lower in more connected/closer fragments than less

connected/further fragments. However, I found no significant difference in probability of

occurrence in the IFM for M. ravelobensis. For E. fulvus, I found no significant difference in the

probability of occurrence in the IFM but did find a significantly higher probability of occurrence

in more connected rather than less connected fragments. Therefore, area contributes more to

lemur species occurrence than connectivity and connectivity for some species (M. murinus and

M. ravelobensis) has a negative effect on species occurrence. Table 3.4: T-test Results for the Mean Probability of Occurrence (Ji) for Four Lemur Species in the Largest Versus Smallest Fragments.

Species Modela Ji Mean Largest N Ji Mean Smallest N T-statistic p-value

Cheirogaleus medius IFM 0.54 21 0.09 21 6.52 <0.01

IFMproxy2 0.54 21 0.09 21 6.70 <0.01 IFMproxy 0.53 21 0.11 21 5.30 <0.01

Microcebus murinus MI-IFM 0.94 21 0.55 21 6.52 <0.01 IFM 0.97 21 0.70 21 7.38 <0.01


IFMlit 0.97 21 0.64 21 7.46 <0.01 MI-IFM 0.97 21 0.65 21 6.47 <0.01

IFMproxy2 0.97 21 0.65 21 7.4 <0.01 IFM 0.97 21 0.87 21 14.59 <0.01

Eulemur fulvus IFM 0.33 21 0.004 21 3.91 <0.01 IFMproxy 0.33 21 0.06 21 3.00 <0.01

a See Table 2.3 for definitions.

68

Table 3.5: T-test Results for the Mean Probability of Occurrence (Ji) for Four Lemur Species in the Most Versus Least Connected Fragments.

Species Model Ji Mean most connected N Ji Mean least

connected N T-statistic

p-value

Cheirogaleus medius IFM 0.29 21 0.33 21 -0.44 0.66

IFMproxy2 0.24 21 0.38 21 -1.50 0.10 IFMproxy 0.37 21 0.27 21 -1.00 0.30

Microcebus murinus MI-IFM 0.63 21 0.86 21 -2.98 <0.01

IFM 0.80 21 0.87 21 -1.29 0.21


IFMlit 0.69 21 0.93 21 -4.22 <0.01 MI-IFM 0.72 21 0.89 21 -2.47 0.02

IFMproxy2 0.76 21 0.86 21 -1.50 0.20 IFM 0.91 21 0.93 21 -0.91 0.37

Eulemur fulvus IFM 0.28 21 0.06 21 2.44 0.02

IFMproxy 0.35 21 0.05 21 3.43 <0.01 a See Table 2.3 for definitions.

3.11.3 Question 3: Within Simulated Metapopulations Over Time, How Do Area and Connectivity/Isolation Affect Occurrence?

Fragment area has a greater influence on overall species occurrence than fragment isolation

does. Removal of the five largest fragments via simulation caused all species to decline in

occurrence (Fig. 3.3A). The most extreme example was C. medius, which became extinct in the

remaining fragments. None of the species showed any appreciable change in occurrence among

the five largest fragments when they were separated from the remaining 37, although occurrence

of C. medius did vary between three and five in the five largest fragments. Removing the five

smallest fragments (Fig. 3.3B) caused no noticeable decline in species occurrence in the 37

remaining fragments. Separation of the five smallest fragments from the remaining 37 caused a

decline in occurrence of all species in the five smallest fragments. Removal of the five most

connected fragments via simulation resulted in no obvious changes in occurrence for any

species (Fig. 3.4A). However, occurrence for all four species declined within the five most

connected fragments following separation from the remaining 37. Removal of the five least

connected fragments caused no change in occurrence for any of the species except C. medius,

which saw a declining trend in occurrence following removal of the five least connected

fragments. There was a decline to zero occurrence for all species, accept M. murinus, in the five

least connected fragments when they were separated from the remaining 37.

69

Figure 3.3: Simulations of Metapopulation Dynamics for Four Lemur Species Over 200 Time Steps in a Fragmented Landscape When the Five Largest and Five Smallest Fragments Are Removed. Simulated species occurrence over time using a Markov chain process. The five largest (A) and five smallest (B) fragments (black lines), respectively are removed from the rest of the fragments (n=37; red lines) at time period 101(vertical line). After this point I ran simulations, to time period 200, separately to demonstrate the impact of either removing the largest (A) or smallest (B) fragments (black). I ran simulations using the IFM with α parameterized from the data (IFM) for C. medius, M. murinus, and E. fulvus and the IFM with α determined from the literature (IFMlit) for M. ravelobensis.

0 50 100 150 200

05

1015

20

Time

Occ

upie

d pa

tche

s

Cheirogaleus medius

0 50 100 150 200

05

1015

20

Time

Occ

upie

d pa

tche

s

Cheirogaleus medius

0 50 100 150 200

05

1015

2025

3035

Time

Occ

upie

d pa

tche

s

Microcebus murinus

0 50 100 150 200

05

1015

2025

3035

TimeO

ccup

ied

patc

hes

Microcebus murinus

0 50 100 150 200

05

1015

2025

3035

Time

Occ

upie

d pa

tche

s


0 50 100 150 200

05

1015

2025

3035

Time

Occ

upie

d pa

tche

s


0 50 100 150 200

02

46

810

Time

Occ

upie

d pa

tche

s

Eulemur fulvus

0 50 100 150 200

02

46

810

Time

Occ

upie

d pa

tche

s

Eulemur fulvus

5 Removed Rest

A. B.

70

Figure 3.4: Simulations of Metapopulation Dynamics for Four Lemur Species Over 200 Time Steps in a Fragmented Landscape When the Five Most Connected and Five Least Connected Fragments Are Removed. Simulated species occurrence over time using a Markov chain process. I removed the five most connected (A) and five least connected (B) fragments, respectively (black lines) from the rest of the fragments (n=37; red lines) at time period 101(vertical line). After this point I ran simulations, to time period 200, separately to demonstrate the impact of either removing the most (A) or least (B) connected fragments (black). I ran simulations using the IFM with α parameterized from the data (IFM) for C. medius, M. murinus, and E. fulvus and the IFM with α determined from the literature (IFMlit) for M. ravelobensis.

0 50 100 150 200

05

1015

20

Time

Occ

upie

d pa

tche

s

Cheirogaleus medius

0 50 100 150 200

05

1015

20

Time

Occ

upie

d pa

tche

s

Cheirogaleus medius

0 50 100 150 200

05

1015

2025

3035

Time

Occ

upie

d pa

tche

s

Microcebus murinus

0 50 100 150 200

05

1015

2025

3035

TimeO

ccup

ied

patc

hes

Microcebus murinus

0 50 100 150 200

05

1015

2025

3035

Time

Occ

upie

d pa

tche

s


0 50 100 150 200

05

1015

2025

3035

Time

Occ

upie

d pa

tche

s


0 50 100 150 200

02

46

810

Time

Occ

upie

d pa

tche

s

Eulemur fulvus

0 50 100 150 200

02

46

810

Time

Occ

upie

d pa

tche

s

Eulemur fulvus

5 Removed Rest

A. B.

71

3.12 Discussion In the first study of its kind on lemur biogeography, I applied a metapopulation approach to

determine how lemur species occurrence was affected by habitat loss and fragmentation. Using

a single-season survey of patch occupancy, I found that lemur species respond differently to

area and connectivity/isolation in a fragmented landscape. Simulations of metapopulation

dynamics provide support that lemur species occurrence was more affected by area than

isolation. Based on the simulation results, I suggest that the most connected fragments are not as

important to the maintenance of the metapopulation as has been previously predicted. For some

species separating the most connected fragments from the remaining fragments actually resulted

in declines in occurrence within the most connected fragments.


A metapopulation approach is appropriate when habitat is in discrete patches, when ecological

processes occur at the local and metapopulation scale, when habitat within the discrete spatial

unit is large enough for local breeding populations, and when the patches are relatively

permanent (Hanski, 1999). The level of habitat loss and fragmentation in Madagascar provides

an ideal situation in which to apply a metapopulation approach to studying with lemur

populations: there is only a fraction of habitat remaining, habitat patches (fragments) are

discrete units separated by non-habitat, many patches are large enough to maintain local

breeding populations, and local populations are connected to one another through dispersal, thus

creating metapopulations.

Two of the species that I studied, P. coquereli and L. edwardsi, did not form a metapopulation

within my study site. Both of these species only occurred within a small number of fragments

within the landscape (three for P. coquereli and two for L. edwardsi). There are three scenarios

that could explain why neither P. coquereli nor L. edwardsi formed metapopulations: area

effects, dispersal ecology and anthropogenics. First, the existing populations of the two species

within the fragments declined and became locally extinct over time because fragment size was

too small to support populations. McGoogan (2011) found that P. coquereli home ranges can be

as large as 23.34 ha (using 95% kernel density methods), which is much larger than the majority

of fragments occurring within my study landscape. Warren and Crompton (1997) found that for

L. edwardsi, yearly home range size is 0.81-1.70 ha, which is much smaller than many of the

72

fragments where L. edwardsi was found absent in my study. This home range size suggests that

L. edwardsi should be able to tolerate smaller fragments, however their mean horizontal

distance travelled per day was quite large. Warren and Crompton (1997) found that L. edwardsi

could travel as much as 463 m (horizontal travel distance) per day. Few fragments in my study

site had linear dimensions greater than 463 m. L. edwardsi occurrence appears to differ in

continuous versus fragmented habitats. For example, Craul et al. (2009) found L. edwardsi to

occur in 13 of 17 continuous forest sites surveyed but only two of six habitat fragments, a

finding that supports the hypothesis that this species is not tolerant to habitat loss and

fragmentation. I found both P. coquereli and L. edwardsi to occur only in larger fragments (the

three largest for P. coquereli and the largest and fourth largest for L. edwardsi), and absent in all

fragments smaller than 11.58 ha. Therefore, the explanation of fragments being too small for

survival is plausible for both P. coquereli and L. edwardsi. The second scenario is that these

species have limited ability to disperse through the matrix and therefore cannot colonize

fragments. Both species have large day ranges between 260 m and 1909 m (McGoogan, 2011;

Warren & Crompton, 1997) and both are highly arboreal vertical clingers and leapers. It’s likely

that these species would avoid matrix habitat that has few or no trees as is in my study area. It is

also likely that the distances between fragments are too large for P. coquereli and L. edwardsi to

attempt crossing. I found P. coquereli in three adjacent and nearby fragments (Fragment 1, 3,

and 4; Figure 2.4). I found L. edwardsi in the largest fragment (Fragment 3) and another

fragment adjacent and close to the continuous forest (Fragment 6; Figure 2.4). The third

scenario is anthropogenics. Both P. coquereli and L. edwardsi are preferred species to hunt in

Ankarafantsika National Park (García & Goodman, 2003). It is possible that these species may

be hunted out of smaller fragments or they avoid moving between fragments due to hunting

pressure. I suggest that these species exist within a declining non-equilibrium metapopulation.

Without intervention it is likely that these populations will become locally extinct within the

area similar to E. mongoz and A. occidentalis.

For the four lemur species that I found that had formed a metapopulation, the type of

metapopulations differed between species. Although the landscape formed a classic mainland-

island metapopulation, I only was able to select this model as a potential candidate model in M.

murinus and M. ravelobensis. However, there were differences in model selection between even

these two similar species. For M. murinus, the mainland-island model was the most likely model

73

followed by the IFM where I estimated the dispersal parameter (α) using the survey data (IFM).

For M. ravelobensis, four models were equivalently likely, including the IFM where I

determined the dispersal parameter (α) based on the literature (IFMlit), the mainland-island

model (MI-IFM), the IFM where I determined dispersal as the linear dimension of the reported

home range (IFMproxy2), and the IFM where I estimated the dispersal parameter (α) using the

survey (IFM). Differences in the habitat preferences and dispersal abilities of Microcebus

species may explain why these similarly related species appear to form different

metapopulations. M. murinus prefers higher elevation and dryer forests than M. ravelobensis

does (Rakotondravony & Radespiel, 2009). There was very little difference in elevation between

fragments within this study and I selected the landscape due to its relatively homogeneous forest

separated by relatively homogeneous grassland savannah. There were some differences in

habitat between fragments. For example, some fragments had dry riverbeds, which may indicate

a difference in habitat type. Although there were dry riverbeds within some of the fragments,

both species were present in all fragments that had a dry riverbed. In fact, both species had

similar occupancy rates within the landscape with M. ravelobensis occurring in one fewer

fragment than M. murinus. Steffens and Lehman (2016) found that abundance in both M.

murinus and M. ravelobensis were related to similar factors including: dendrometrics, fragment

area, and isolation within the same fragmented landscape. In continuous forest, Burke and

Lehman (2015) found differences between M. murinus and M. ravelobensis in their capture rates

of each species and the body mass of female M. ravelobensis between the edge and interior

habitat. They captured more M. ravelobensis and fewer M. murinus along the edge than in the

interior habitat. They found female M. ravelobensis along the edge had greater body mass than

those in the interior habitat. Therefore, preference for different microhabitats may not explain

why models differed between these species but edge effects might.

Dispersal ability differs between M. murinus and M. ravelobensis in continuous forest

(Radespiel et al. 2003; Radespiel et al. 2009; Schliehe-Diecks et al. 2012) and may explain their

different metapopulation dynamics in a fragmented landscape. Although both species appear to

be dispersal limited, in continuous forest M. ravelobensis may be more dispersal limited than M.

murinus with 0.05 km (Radespiel et al. 2009) and 0.25 km (Radespiel et al. 2003) median-

dispersal distances respectively. I was able to determine median-dispersal ability for both

species using metapopulation dynamics in a fragmented landscape. Based on metapopulation

74

dynamics, I found that both Microcebus species had the same median-dispersal ability of 0.10

km. A median-dispersal distance of 0.10 km is less than half the distance reported from the

literature for M. murinus (0.25 km; Radespiel et al. 2003) and double the distance reported for

M. ravelobensis (0.05 km; Radespiel et al. 2009). The most likely model for M. ravelobensis

contained the smaller median-dispersal distance reported in the literature (0.05 km; Radespiel et

al. 2012) and the most likely model for M. murinus was the mainland-island model. In a

mainland model, immigrants continually spread from the continuous forest to the fragments

(MacArthur & Wilson, 1967). M. murinus is less dispersal limited and edge intolerant than M.

ravelobensis in continuous forest. Yet neither species showed major differences in occurrence in

a fragmented landscape. A mainland-island metapopulation model may be more suitable for M.

murinus because of their greater ability to disperse from continuous forest while M. ravelobensis

may not be as dispersal limited as thought and is more edge tolerant and therefore capable of

surviving in fragments once established.

For both C. medius and E. fulvus, the most likely model was the among patch incidence function

model where I estimated the dispersal parameter (α) using the survey data (IFM). Based on this

model, I suggest that their ability to move among patches exceeded their ability to move from

the continuous forest to occupied fragments. For both species, I selected additional candidate

models: the IFM where I estimated the dispersal parameter (α) using home range as a proxy (E.

fulvus and C. medius; IFMproxy), and IFM where I estimated (α) using the linear dimension of

home range (C. medius, IFMproxy2). The dispersal parameter suggests E. fulvus has the highest

dispersal ability of all the species in the study. In fact, both species have large median-dispersal

abilities, although I found E. fulvus to have a roughly 10 times greater median-dispersal distance

(2.35 km) than C. medius (0.21 km), and E. fulvus has roughly 10 times the body mass of C.

medius (Table 3.1). Moreover, C. medius and E. fulvus were the only two frugivores in the study

site. It is predicted that larger species have greater dispersal ability than smaller species

(Sutherland et al. 2000) and frugivorous primates have larger home ranges than folivorous

primates (Clutton-Brock & Harvey 1977; Richard 1985). Thus, home range size may predict

dispersal ability (Bowman et al.2002) and frugivores have greater dispersal ability than

folivores. However, E. fulvus occupied fewer fragments than C. medius. Within this landscape

there were likely few fragments large enough for E. fulvus to live in, which required E. fulvus to

move between fragments. E. fulvus may be transient within patches that are smaller than they

75

would normally need to be able to survive. For example, I found E. fulvus in one fragment that

was smaller (4.16 ha) than their reported home range yet absent in four fragments that were

within their reported home range. Although E. fulvus is smaller than P. coquereli they both have

similar home range sizes (Table 1; McGoogan, 2011; Mittermeier et al. 2010). However, E.

fulvus occupied seven fragments and P. coquereli only two fragments. P. coquereli may be

limited by edge effects that reduce habitat suitability (McGoogan, 2011; Kun-Rodriguez et al.

2014) where E. fulvus may be more edge tolerant. Lehman et al. (2006a; 2006b) found that

contrary to predictions Eulemur rubriventer was edge tolerant. Lehman (2006b) suggests that E.

rubriventer behaved more like a folivore/frugivore than a strict folivore. It is possible that E.

fulvus behaved the same way. However, we need further study to determine if E. fulvus is more

folivorous or frugivorous within the fragments and to determine what is the availability of

fruiting trees within the fragments.

E. fulvus is an important seed disperser and so its ability to disperse has secondary benefits to

habitat maintenance (Sato, 2012; Ganzhorn et al. 1999). It is important to understand this

species’ response to habitat fragmentation as losing an important seed disperser may have

cascading negative effects throughout the landscape (Valenta et al. 2014; Sato, 2012; Ganzhorn

et al. 1999). C. medius is also an important seed disperser but with its small body mass and

hibernation patterns, may be better suited to survive in more fragments than E. fulvus. I found C.

medius in one fragment that was 1.69 ha but the remaining fragments where it occurred were

larger than 11.58 ha. Unlike any of the other species observed within the fragments, C. medius is

capable of extended hibernation (Dausmann et al. 2005). Like other Cheirogaleus species, C.

medius consumes large amounts of high-sugar fruits prior to hibernation in order to build up fat

reserves (Fietz & Ganzhorn, 1999). Therefore, C. medius may be able to survive only in

fragments that have a high availability of fruit during this crucial period. Tree holes used by C.

medius must be carefully selected in order to allow sufficient maintenance of body temperature

during the months that they hibernate (Dausmann et al. 2005). Like fruit availability, tree holes

may be a limiting resource for C. medius in the fragments.

76

3.12.2 Question 2: What are the Separate Effects of Area (Extinction Risk) and Connectivity/Isolation (Colonization Potential) on a Lemur Metapopulation?

Area

I found that larger fragments had a higher probability of occurrence for all four species that I

conducted the analysis on, regardless of the selected candidate model. In Chapter 2, I found that

area was the strongest predictor of species richness within the same landscape. Area not only

predicts species richness but individual species occurrence whereas connectivity seems to have

little effect on lemur species occurrence; with some exceptions (see below). The pattern of area

predicting lemur occurrence at the species level may be driving the same pattern at the

community level. Hanski (2010) suggests that species-area relationships can be derived from

single species metapopulation models by summing the predicted incidences for each species. In

metapopulation dynamics as local populations grow and reach their carrying capacity

individuals are forced to leave the patch to find a new suitable habitat patch (Hanski 1991). An

empty patch is considered colonized when an immigrant arrives and subsequently survives in

that patch. The quality and size of the habitat determines survival of an individual in a

previously unoccupied patch (Hanski, 2010). However, assessing habitat quality is more

difficult than measuring area of a patch. Hanski (2010) argues that how much habitat quality

versus area contributes to species occurrence is dependent on specific circumstances. Many

studies on primates found that area was the largest predictor of primate species occurrence

(Lawes et al. 2000; Chapman et al. 2003; Rodriguez-Toledo et al. 2003; Arroyo-Rodriguez &

Dias, 2010; Marshall et al. 2010). For example, Lawes et al. (2000) found that area was the only

factor that impacted occurrence in Cercopithecus mitis in a fragmented landscape even when

considering other factors such as isolation and habitat disturbance. We need future research to

evaluate the relative contribution of habitat quality versus area to lemur species occurrence.

Connectivity/Isolation

I hypothesized that connectivity would have a significant positive effect on lemur species

occurrence. The only species to show a positive relationship in occurrence probability and

connectivity was E. fulvus using the IFMproxy model but not with the IFM model. Contrary to

predictions, occurrence probability was the same between more and less connected fragments

for one species (C. medius using all candidate models) and negative between more and less

77

connected fragments in M. murinus (MI-IFM model) and M. ravelobensis (IFMlit model). For

E. fulvus there was a significant positive difference in occurrence between more and less

connected fragments. This result means that although E. fulvus had the greatest dispersal ability

among the six species in this study, it is more likely to occur in more versus less connected

fragments. Migration between fragments is risky for arboreal lemurs. For example, species

travelling through the matrix have increased predation risk (Irwin, 2009) and there is the

possibility of arriving at an unsuitable fragment requiring further migration. Therefore, E.

fulvus, although capable of migrating to any fragment, appears to stay within the largest and

most connected fragments. Although occurring in multiple fragments, C. medius tended to occur

near the largest fragment (Fragment 3; Figure 2.4 and Table 2.5) or the continuous forest.

Therefore, they are either not limited by dispersal or they form an intermediate metapopulation.

In an intermediate metapopulation, they would be able to move between fragments but one or

more of the larger fragments would act as a mainland source of more colonists. If Fragment 3

acted as a second mainland source, this pattern would explain a lack of difference in occurrence

probability between more and less connected fragments (Harrison, 1991). For both Microcebus

species, the differences in occurrence probability were negative, meaning that the probability of

occurrence was lower in more connected than in less connected fragments. One explanation for

this negative probability is that the two species of Microcebus have higher occurrence

probability in more isolated fragments because they are not area limited and are able to survive,

possibly in the long-term, within the smallest fragments regardless of isolation. Other studies

have recorded Microcebus species in all but the smallest (<1 ha) fragments (Ganzhorn, 2003;

Schad et al. 2004; Olivieri et al. 2008) However, these studies did not include as many small (<1

ha) fragments as my study. I found Microcebus to occur in fragments as small as 0.23 ha. It is

possible that there are source-sink dynamics occurring within the study landscape, where large

patches and possibly the nearby continuous forest provide a constant source of potential

immigrants for smaller patches (Harrison 1991). For example, I only observed one Microcebus

individual in the smallest fragment, suggesting that occupancy in this patch is ephemeral and

thus maintained through colonization. Ganzhorn and Schmid (1998) also found a potential

source-sink relationship occurring for M. murinus in secondary forests within a fragmented

landscape. They observed poorer conditions (smaller, fewer trees and warmer temperatures)

within the secondary versus primary forest and within the secondary forest they never re-

captured any individuals but were able to recapture seven within the continuous forest.

78

Habitat loss and fragmentation

We know little about the independent effects of habitat loss and fragmentation on primates

(Arroyo-Rodríguez et al. 2013b). Most studies research habitat loss and fragmentation at the

patch-level (Arroyo-Rodríguez et al. 2013b). However, habitat fragmentation is a landscape-

level process, and the combined effects of habitat loss and fragmentation are hard to separate at

a patch-level (Fahrig, 2003). In metapopulation dynamics, area affects a species extinction

probability and connectivity/isolation impacts the colonization potential of a patch (Hanski,

1999). Although a metapopulation approach uses data from a patch-level, it is in essence a

pseudo-landscape approach because it incorporates landscape-level connectivity measures. It

also allows for the analysis of how area versus connectivity/isolation impact species occurrence.

My study shows that area strongly affects species occurrence and that isolation has species-

specific neutral, negative and positive impacts on lemur species occurrence.

3.12.3 Question 3: Within Simulated Metapopulations Over Time, How Do Area and Connectivity/Isolation Affect Occurrence? What are the Conservation Implications?

One of the advantages of a metapopulation approach is that it is possible to model extinction

(area) and colonization (connectivity/isolation) probability, which allows for the simulation of

their effects on occurrence over time. When I removed the five largest fragments from the

metapopulation, all four species in the remaining fragments collapsed. Conversely, when I

removed the five smallest fragments no species populations in the remaining fragments

collapsed. When I removed the five most connected fragments, the remaining metapopulation

was marginally or not at all affected. When I removed the five least connected fragments, C.

medius showed a marked decline in occurrence. However, none of the other species showed any

appreciable decline. E. fulvus was the only species to not show extinction in the five most

connected fragments separated from the remaining 37 fragments. Lemur occurrence collapsed

for all but one species (M. murinus) when I separated the five least connected fragments from

the rest. It is possible that large yet more isolated fragments help maintain metapopulation

dynamics, in essence creating an intermediate metapopulation with multiple mainland sources

for colonists.

The conservation implications of altering the amount of habitat in this landscape are clear. Large

fragments are crucial to maintaining lemur species metapopulations. What is less clear is the

79

role of connectivity within the landscape. Simulation models suggest that removing the most

connected fragments will not affect the remaining metapopulation. However, lemurs occupying

the most connected fragments are unlikely to survive when separated from the remaining

metapopulation. Therefore, connectivity has a reduced overall effect on the metapopulation

dynamics of the entire system but matters when particular fragments are removed. These

patterns are important to consider if decisions are being made on the fate of particular fragments

within the system. Loss of fragments due to un-regulated anthropogenic disturbance may

remove fragments that are the most valuable to maintain each lemur species’ metapopulation

dynamics. For example, my simulations suggest that maintaining the size and number of large

fragments within the population is most important and care must be taken when considering

fragment connectivity. The loss of fragments with high connectivity values will have little

impact on the lemur occurrence over the entire system but contrary to previous thinking

removing less connected fragments may have dramatic negative impacts on lemur species

occurrence.

3.13 Suggestions for Conservation Ankarafantsika National Park contains different zones of protection including multiuse zones,

regeneration zones and fully protected zones. My study area was within one of the multiuse

zones where burning of grass for grazing cattle is tolerated. This activity has resulted in habitat

loss and fragmentation, and has created a fragmented landscape near a large portion of

continuous forest. My simulation models suggest that increased attention should be paid to

maintaining fragment size. I would recommend measures that reduce the impact of fire on

fragment size. Currently, fires are lit within the zone and forest is incidentally burned (personal

observation). I suggest the creation of a fire management plan following Bloesch (1999). If

source-sink dynamics are occurring with the landscape for species like M. murinus and M.

ravelobensis, then I also recommend considering methods to improve connectivity between

fragments, although connectivity does not appear to be influencing occurrence within the greater

metapopulation.

3.14 Conclusion A metapopulation approach is useful for determining the effect of area and

connectivity/isolation on lemur species occurrence in a fragmented landscape. This study shows

80

that lemurs form metapopulations in fragmented landscapes. Within their metapopulation,

lemurs are impacted by both habitat area and isolation. However, fragment area has a larger

impact on lemur occurrence than isolation. I identified dispersal ability as potentially explaining

differences in model selection between lemur species. P. coquereli and L. edwardsi do not form

traditional metapopulations and likely form a non-equilibrium metapopulation that is declining

toward local extinction. It is possible that source-sink dynamics are impacting populations of M.

murinus and M. ravelobensis within the landscape. The two most frugivorous lemurs, E. fulvus

and C. medius, are able to maintain stable metapopulations likely through dispersal and their

ability to survive within the largest fragments. To perpetually maintain metapopulations for each

species, I recommend a fire management strategy that reduces further habitat loss and isolation

among fragments in the landscape. We should pay special attention to connecting the largest

fragments to reduce the likely local extinction of P. coquereli and L. edwardsi, as well as

allowing seed dispersers of large fruits such as E. fulvus and C. medius to increase the area of

potential seed deposition.

81

Chapter 4: Lemur Species-Specific Scale Responses to Habitat Loss in Fragmented Landscapes in NW Madagascar

4.1 Introduction In Chapters 2 and 3, I investigated the patch-level effects of habitat loss and fragmentation on

lemur species richness and individual lemur species occurrence, respectively. In these two

chapters, I found that area was a major determinant of both species richness and individual

species occurrence. However, I did not determine the response of lemur species to habitat loss

and fragmentation at a landscape-level. In this chapter, I investigate the impact of habitat loss

and fragmentation on lemur species using a landscape-level approach derived from landscape

ecology.

Landscape ecology is concerned with the study of the relationship between spatial pattern and

ecological processes at various scales (Turner et al. 2001). A landscape is defined as “an area

that is spatially heterogeneous in at least one factor of interest” (Turner et al. 2001:7), for

example the amount of forest within an area. Landscape ecology focuses on larger spatial scales

than typical ecological research. However, the choice of the scale for a study is both dependent

on the question asked and the species of interest (Wiens, 1989; Turner et al. 2001). Researchers

mainly conduct research on the impact of habitat loss and fragmentation at a patch- versus a

landscape-level, despite habitat loss and fragmentation being landscape-level processes

(McGarigal & Cushman, 2002; Fahrig, 2003; Arroyo-Rodriguez et al. 2013b; Arroyo-Rodriguez

& Fahrig, 2014).

Although it is recognized that habitat loss and fragmentation effects are landscape-level

phenomena (McGarigal & Cushman, 2002; Fahrig, 2003), including within the field of

primatology (Arroyo-Rodriguez et al. 2013b; Arroyo-Rodriguez & Fahrig, 2014), few studies

have actually measured habitat loss and fragmentation at a landscape-level. Researchers

continue to use a patch-level analysis because of the ease of data collection and its historical use

in ecological theory, such as the island biogeography theory (MacArthur & Wilson, 1967) and

metapopulation dynamics (Hanski, 1999). One of the major issues with applying a landscape-

level approach is scale, which involves choosing what size of landscapes to study. The choice of

82

scale depends on the research question, the species studied, and the process of interest (Turner et

al. 2001; McGarigal & Cushman, 2002; Turner, 2005). Although fragmentation research should

be conducted at relevant species-specific scales, such as the area of a home range (McGarigal &

Cushman, 2002), few studies have taken steps to determine the species-specific responses to

habitat loss and fragmentation.

4.2 Landscape-Level Effect of Habitat Amount on Species Occurrence.

Landscape ecology is a sub-discipline of ecology, studying how landscape structure impacts

species abundance and distribution (Fahrig, 2005). One can divide landscape structure into two

elements: landscape composition and configuration (Fahrig, 2005; McGarigal & McComb,

1995). Landscape composition refers to the types and amount of habitat within a landscape

(Fahrig, 2005), while landscape configuration refers to the arrangement of habitat within a

landscape (McGarigal & McComb, 1995). Habitat loss and fragmentation are two separate yet

related landscape-level phenomena (Fahrig, 2003). Habitat loss is simply the removal of habitat

from a landscape (McGarigal & Cushman, 2002; Fahrig, 2003). Habitat fragmentation is the

separation of habitat into smaller less connected portions (McGarigal & Cushman, 2002; Fahrig,

2003). The impact of habitat loss on species richness and occurrence is well studied (Fahrig,

2003). However, fewer studies have focused on the impact of habitat fragmentation independent

of habitat loss (Fahrig, 2003).

We now understand that habitat loss and fragmentation are landscape-level processes where the

size and shape of patches within a landscape are altered to change the composition and

configuration of habitat (McGarigal & Cushman, 2002; Fahrig, 2005; Arroyo-Rodriguez &

Fahrig, 2014). Although habitat loss and fragmentation are landscape-level phenomena,

researchers still study these processes using fragments as the unit of analysis in patch-level

studies. Patch-level studies can provide important insight into the mechanisms that result in

landscape patterns, such as patch occupancy (Arroyo-Rodriguez & Fahrig, 2014). In Chapter 3,

I introduced many examples of patch-level research on the impact of habitat loss and

fragmentation on primate species occurrence. The research clearly showed that species

occurrence is strongly related to patch/fragment area (Anzures-Dadda & Manson, 2007; Arroyo-

Rodríguez et al. 2008; Boyle & Smith, 2010). Fragment isolation appears to have a lesser effect

on primate species occurrence (Lawes et al. 2000; Cristobal-Azkarate et al. 2005). In Chapter 2,

83

I investigated the species-area relationship on a lemur community. Regardless of the other

variables in the analysis, I found that area was the main driver of species richness. In Chapter 3,

I investigated how habitat loss and fragmentation affect individual lemur species occurrence at a

patch-level. I found that fragment area was the primary driver of lemur species occurrence and

that habitat fragmentation (isolation) per se did not affect lemur species as much as habitat loss

(fragment area). A major issue with using just a patch-level approach is that patch-level studies

typically ignore the context within which the patch exists. For example, these studies may

ignore the matrix (Ricketts, 2001) or the amount of continuous forest near a fragment (Arroyo-

Rodriguez & Fahrig, 2014). Therefore, patch-level studies provide a very localized view of the

impact of habitat loss on primate species occurrence.

A landscape-level approach provides a broader view of the impact of habitat loss on primate

species occurrence (Arroyo-Rodriguez & Fahrig, 2014). Landscape-level studies investigating

the impact of primate species richness, occurrence, and abundance are increasing in number. A

landscape-level approach is appropriate when one is interested in determining the effect of

landscape composition and configuration on a response variable (Fahrig, 2005). Arroyo-

Rodriguez & Fahrig (2014) suggest two approaches for landscape studies in primatology. The

first is a patch-landscape scale study where habitat variables and species responses are measured

in focal patches and their surrounding landscape (Arroyo-Rodriguez & Fahrig, 2014). The

second is a pure landscape-scale study where habitat variables and species responses are

measured within entire landscapes (Arroyo-Rodriguez & Fahrig, 2014).

A few primate researchers have used a landscape-level approach to study the impact of habitat

loss and fragmentation on primate occurrence (Anzures-Dadda & Manson, 2007; Thornton et al.

2011; Arroyo-Rodriguez et al. 2013b; Benchimol et al. 2014; Sales et al. 2015). However, there

were few consistent results among these studies even within similar species. Anzures-Dadda and

Manson (2007) found that patch occupancy in mantled howler monkeys (Alouatta palliata) was

positively related to both forest cover and fragmentation, whereas Thornton et al. (2011) found

that patch occupancy in Central American black howlers (A. pigra) was not related to forest

cover or fragmentation. Sales et al. (2015) found that models containing local predictors, such as

canopy height and canopy openness, better explained primate occupancy than models containing

landscape metrics, such as habitat amount and fragmentation.

84

The apparent differences in primate responses to landscape-level variables between studies may

be the result of actual differences among species or differences in study design. Although scale

effects are apparent and well documented in the ecological literature (Wiens, 1989; Holland et

al. 2004; Bradter et al. 2013) the scale of effect for primate species is poorly studied (Arroyo-

Rodriguez & Fahrig, 2014).

4.3 Species-Specific Scale Responses to Habitat Amount.

4.3.1 Scale and Landscapes.

All ecological patterns and processes occur in various temporal and spatial scales (Hutchinson,

1965). However, spatial scale is frequently ignored in primatology (Arroyo-Rodriguez & Fahrig,

2014). Recently, researchers have applied landscape ecological methods and theory to studies of

primate behavior and ecology, and biogeography (Gray et al. 2010; Thorton et al. 2011;

Ordóñez-Gómez et al. 2014). These studies have begun to recognize the need to look at

ecological patterns and processes at the appropriate scale.

4.3.2 “Scale of Effect.”

Researchers undertaking landscape-level studies are interested in finding how patterns and

processes affect species responses (richness, occurrence, abundance; Turner et al. 2001).

Although a landscape-level approach is suitable to look at questions concerning how landscape-

level patterns and processes, such as habitat loss and fragmentation, affect species responses,

what is less clear is at what scale or size of landscape do species respond to habitat loss and

fragmentation (Jackson & Fahrig, 2012). This “scale of effect” occurs when the landscape

structure is identified and measured at scales relevant to the species. Studies on numerous taxa

demonstrate species-specific scale responses to habitat loss and fragmentation including

amphibians (Eigenbrod et al. 2011), birds (Bradter et al. 2013; Gilroy et al. 2015; Thorton &

Fletcher, 2014), insects (Holland et al. 2004; Bellier et al. 2007), arboreal mammals (Patterson

& Malcolm, 2010), terrestrial mammals (de Knegt et al. 2011), and plants (Borcard et al. 2004).

A few studies have assessed the “scale of effect” in primate species (Gray et al. 2010; Thorton et

al. 2011; Ordóñez-Gómez et al. 2014). Gray et al. (2010) investigated different scale responses

of red-cheeked gibbon (Nomascus gabriellae) habitat preferences in a fragmented landscape.

They found that gibbon occurrence responded to different predictor variables at different scales.

For example, semi-evergreen forest at a small scale (1500 m radius), dipterocarp forest at an

85

intermediate scale (3000 m radius), and evergreen forest at a large scale (5000 m radius) best-

explained variance in gibbon occurrence. In a landscape-scale study, Thorton et al. (2011) found

that spider monkeys (Ateles geoffroyi) responded negatively and strongly to habitat

fragmentation, particularly at a 500 m scale, and negatively to habitat loss, while the Central

American black howler monkeys (Alouatta pigra) did not respond to habitat fragmentation or

habitat loss at any spatial scale. Ordóñez-Gómez et al. (2014) investigated the “scale of effect”

in how different landscape attributes affected spider monkey diet and behavior. These

researchers found that most response variables were related to landscape variables at a 126 ha

landscape-level. However, they found that the “scale of effect” differed between response

variables and landscape characteristics, suggesting a multi-scale approach may be the most

appropriate.

Most researchers typically determine the “scale of effect” post hoc by assessing species

responses at multiple scales and selecting the scale with the strongest effect (Holland et al. 2004;

Jackson & Fahrig, 2012; Bradter et al. 2013). The problem with this approach is that it results in

a less efficient study design (Jackson & Fahrig, 2012) and may not actually detect the actual

“scale of effect” due poor scale selection (Jackson & Fahrig, 2014). However, until we

understand why certain scales produce a “scale of effect” for a species we are limited to

determining the “scale of effect” post hoc. Jackson and Fahrig (2014) suggested that one way of

dealing with this issue is by selecting a large ranges of scales ranging from the size of a single

territory/home range to greater than the average dispersal distance.

4.3.3 Species-Specific “Scale of Effect”

We do not fully understand what determines species-specific “scale of effect” (Jackson &

Fahrig, 2012). In the first study of its kind, Jackson & Fahrig (2012) determined that a species

dispersal distance was strongly positively related to “scale of effect,” and in a meta-analysis of

avian body size, Thorton & Fletcher (2015) found that body size mediated the “scale of effect.”

For primates, what determines species-specific “scale of effect” is still unknown. However, as

indicated by the above research, dispersal ability and body size, which are correlated with

mobility (Bowman et al. 2002; Whitmee & Orme, 2013), likely mediate the “scale of effect.”

86

4.4 Justification I described in Chapter 3 that lemur species are at high risk of extinction mainly due to habitat

loss and fragmentation (Schwitzer et al. 2014). All researchers have applied a patch-level

approach to determine how lemurs respond to habitat loss and fragmentation, even though

habitat loss and fragmentation are landscape-level phenomena. To date, no research on lemurs

has explicitly addressed how lemurs respond to habitat loss and fragmentation using a

landscape-level approach.

Preliminary studies on primates suggest that primate responses to area are species-specific and

that there is no typical primate response to the amount of habitat in a landscape (Gray et al.

2010; Thorton et al. 2011; Ordóñez-Gómez et al. 2014). My study is important because it is the

first of its kind to determine the species-specific scale responses of lemurs to habitat loss and

fragmentation using a landscape-level approach. My study has the added benefit of determining

whether two sympatric species from the same genus (Microcebus) respond to habitat loss and

fragmentation in the same way. By using a landscape-level approach, I will be able to not only

determine how habitat loss and fragmentation impacts lemur species occurrence but, more

importantly, at what scale, all while accounting for spatial autocorrelation in the data.

4.5 Goal The goal of this study is to assess how lemur species occurrence is related to habitat amount at a

landscape-level. To solve how lemur species occurrence is related to habitat amount I must first

determine at what scale each species responds to habitat amount. Therefore, I will determine

how lemur species occurrence is related to habitat amount within species-relevant scales while

testing the following questions and hypotheses:

1) At what scale do species respond to habitat amount?

Hypotheses:

H0 Lemur species occurrence with respect to habitat amount will vary randomly with

respect to scale.

H1 Scales relating to home range or dispersal ability will predict lemur occurrence with

respect to habitat amount.

87

2) What is the effect of habitat amount on species-specific occurrence of lemurs at the

landscape-level?

Hypotheses:

H0 The probability of occurrence for each lemur species will vary randomly with respect

to habitat amount at a species-relevant scale.

H1 The probability of occurrence for each lemur species will increase with increasing

habitat amount at a species-relevant scale.

4.6 Methods


My study site and study species are described in more detail in Chapters 2 and 3. In this chapter,

I analyzed landscape-level data on three of the six species found in the study site (Cheirogaleus

medius, Microcebus murinus, and Microcebus ravelobensis). I excluded three of the species

because I had either observed too few occurrences to make a meaningful statistical assessment

(Propithecus coquereli and Lepilemur edwardsi), or because the nature of the spatial

autocorrelation among landscape sizes was too complex to make a meaningful determination as

to what scale a species responded to habitat amount (Eulemur fulvus).

4.6.2 Question 1: What is the Scale of Species Response to Habitat Amount?

Species Presence and Pseudo-Absence

For each species, I determined presence along line transects using the same methods that I

described in Chapter 3. I then plotted each observation location within ArcGIS. Using the

observer to animal sighting distance, I determined the effective detection width for each species

following Müller et al. (2000) and Steffens and Lehman (2016). I then created an effective

detection buffer by creating a buffer along both sides of each transect equivalent to the effective

detection width (m) for each species within ArcGIS. I removed any presence points located

outside of the transect detection buffers from the analysis. To create pseudo-absence points for

logistic regression analysis, I buffered each presence point with a circle representing one home

range size for each species; clipped the effective transect buffers by the home range buffers,

88

leaving survey areas where no individuals were observed; and randomly placed approximately

the same number of points as the number of observed presence points within the clipped

effective detection buffers. I searched for duplicate points using the dup.coords function in R

and I manually deleted duplicate points and their associated variables for all analyses. I

employed this method to better reflect actual absence points rather than creating pseudo-absence

points using a simple random sample of the landscape.

Landscape-Level

To determine the scale at which each species responds to habitat loss I created landscapes of

varying sizes in ArcGIS for each species. For each species presence and pseudo-absence points,

I created 10 arbitrary buffers representing approximately ¼, ½, 1, 2, 4, 8, 16, 32, 64 and 128

times each species’ mean home range (Table 4.1; Landscape Scale Sizes). Jackson and Fahrig

(2014) recommended using a large range size of landscape scales, from the size of a single

home range/territory to greater than the average dispersal distance for a species. I went further

by including two scales smaller than a single home range for each species. To make the results

easier to compare between species, I rounded the home range for C. medius and M. ravelobensis

up from 1.55 to 2.00 ha and from 0.59 to 1.00 ha respectively and rounded the home range for

M. murinus down from 2.83 to 2.00 ha. I then measured the amount of forest within each

species occurrence buffer using the below methods. Table 4.1: Study Species Characteristics and Landscape Scale Sizes

Species Body Mass (g)

Activity Pattern Diet Mean Home

Range (ha) 10 Landscape Scale

Sizes (ha) Cheirogaleus

medius 120–270 Nocturnal Frugivore 1.55 ±0.42(1) 0.5, 1, 2, 4, 8, 16, 32, 64, 128, 256

Microcebus murinus 58–67 Nocturnal Fauni-frugivore 2.83 ±1.44 (2) 0.5, 1, 2, 4, 8, 16,

32, 64, 128, 256 Microcebus ravelobensis 56–87 Nocturnal Fauni-frugivore 0.59 ±0.11(3) 0.25, 0.5, 1, 2, 4, 8,

16, 32, 64, 128 Data from: 1 Müller (1998); 2 Radespiel (2000); 3 Weidt et al. (2004)

Amount of habitat

I determined the amount of forest from a DigitalGlobe four band, 2x2 m resolution satellite

image within an extent within 8199684.36 and 8191497.94 northing and 679451.29 and

685941.31 easting UTM coordinates. The size of the image is approximately 8 km north to

south and 6.2 km east to west (Fig. 4.1). I added the DigiGlobe image as a raster to ArcGIS and

georeferenced the image using the georeference tool and 16 reference points taken from the base

map within ArcGIS. I classified the raster image into forest and non-forest using an

89

unsupervised classification and checked this classification against the original RGB raster for

accuracy. I removed all non-forest values from the classified raster to create a raster consisting

of only forest cells. I measured the amount of forest within all 10 species occurrence buffers

using zonal statistics tool in ArcGIS. Some of the larger buffers exceeded the extent of the raster

image. I retained sightings for species where all buffers had at least 80% of their area within the

extent of the raster. Both Microcebus species had 20 sightings each that had more than 20% of

their landscape buffers outside of the extent of the raster image. I removed these 40 sightings

from analyses.

90

Figure 4.1: DigitalGlobe Satellite Image of Field Site.

91

Analysis

I used logistic regression analysis and Akaike’s Information Criterion (AIC) to assess how

habitat amount affects species occurrence. Logistic regression is useful for analysis of

presence/absence data because it does not expect a linear relationship between the response and

independent variables like linear regression does (Quinn and Keough, 2002). Instead, the error

term can be modeled using a binomial logit link identity structure (Quinn & Keough, 2002).

Like most regression models, logistic regression assumes that the error terms are independent

(Quinn & Keough, 2002). Although each observation must be independent from one another,

ecological data are rife with spatial autocorrelation, resulting in a violation of the assumption

that the error terms are independent because of residual spatial autocorrelation (de Knegt et al.

2010; Zuckerberg et al. 2012). For example, sampling landscapes surrounding observation

points may result in large amounts of overlap between sample landscapes thereby increasing the

likelihood of residual spatial autocorrelation. To solve the problem of overlapping landscapes,

Holland et al. (2004) developed a method to determine the largest number of spatially

independent sites (non-overlapping) from occurrence data at various scales. However, non-

overlapping sites do not necessarily eliminate or even reduce residual spatial autocorrelation

(Zuckerberg et al. 2012). For example, Zuckerberg et al. (2012) found that overlapping and non-

overlapping sites show similar levels of spatial autocorrelation. These researchers suggest that

we should not be as concerned with finding spatially independent sites, but rather should look at

increasing independence in the errors in regression analysis (i.e. reducing residual spatial

autocorrelation). One technique to reduce residual spatial autocorrelation in GLMs is to use

Moran’s Eigenvector (ME) filtering, an extension of principal coordinate analysis of neighbor

matrices PCNM (Dray et al. 2006; Griffith & Pere-Neto, 2006). Bocard and Legendre (2002)

first developed PCNM to investigate the spatial structure of data. Later other researchers

developed a filtering technique to select a subset of eigenvectors for inclusion in linear models

and GLMs (Dray et al. 2006; Griffith & Pere-Neto, 2006).

I applied an ME filtering approach to reduce residual spatial autocorrelation in my data using

the ME function in the spdep 0.5-88 package in R (Bivand et al. 2013; Bivand & Piras, 2015).

ME filtering in the spdep package uses a brute force selection technique to determine which

eigenvectors reduce spatial autocorrelation within a GLM below a particular alpha value of

Moran’s I (Bivand et al. 2013; Bivand & Piras, 2015). Then you add the selected fitted values of

each vector as independent variables to the GLM. The resulting spatial GLM will have either no

92

or reduced spatial autocorrelation among the error terms. I ran each ME filtering procedure

using 1000 simulations and a 0.20 Moran’s I alpha value for inclusion. In some cases, the

filtered eigenvectors created near perfect separation among the data and resulted in poorly

specified although spatially un-correlated models. For situations where the ME filtering

procedure produced poorly specified models, I re-ran the ME filtering step with 0.01 and 0.001

Moran’s I alpha values. I added the fitted values from the selected eigenvectors to create a

spatial GLM and selected among the different alpha values based on which resulting GLM had

the lowest AIC value. I tested for spatial autocorrelation in the residuals of the spatial models

using moran.test function in the spdep package in R (Bivand et al. 2013; Bivand & Piras, 2015)

to determine if they met the assumption of independence of errors. I considered data to be

independent if there was no significant spatial autocorrelation or if the spatial autocorrelation

was significant but low (i.e. P-value <0.05 and Moran’s I value <0.33). I ran all GLMs using the

glm function with a binomial “logit” link function in R (R Core Team, 2015).

To determine at which scale species respond to habitat amount, I ran the above ME filtering

procedure on species that showed residual spatial autocorrelation, and ran the resulting spatial

GLMs (which included the fitted Moran’s eigenvectors as variables in the model) on all 10

species occurrence buffer sizes, including a null model (with spatial filtering applied). The one

exception was for C. medius, which showed no residual spatial autocorrelation. For C. medius, I

ran non-spatial GLM’s without the added step of the ME filtering procedure. I selected the

models with the smallest AIC to determine at which scale each species responds to habitat

amount. If for any species analysis the null model had the lowest AIC value of all the models, I

determined that there was no scale of response to habitat amount for that species.

4.6.3 Question 2: What is the Landscape-Level Effect of Habitat Amount on Species Occurrence?

To assess if the amount of habitat determined species occurrence, I ran non-spatial (C. medius)

and spatial GLMs (Microcebus spp.; including the fitted Moran’s eigenvectors as variables in

the model) from the scale that was selected based on AIC values. I assessed the sign and

magnitude of the coefficients and tested the significance of the model using the ANOVA

function in R (R Core Team, 2015), thereby determining if the amount of habitat influenced

occurrence. I then calculated McFadden’s R2 as a measure of pseudo R2 for each GLM

regression model.

93

4.7 Results

4.7.1 Description of Spatial Autocorrelation Among Landscapes.

I found significant positive spatial autocorrelation in all non-spatial models for all species

except C. medius. For C. medius there was no spatial autocorrelation among any of the non-

spatial models (Table 4.2), likely due to the fact that there were fewer and more spatially

dispersed (less overlap among landscapes) presence/pseudo-absence points for this species than

the other species. Therefore, for C. medius I was able to report on non-spatial GLMs without

using the ME filtering approach. I found higher (>0.67) significant positive spatial

autocorrelation among presence/pseudo-absence points in non-spatial models and lower (<0.33)

significant spatial autocorrelation among presence/pseudo-absence points in spatial models for

both M. murinus and M. ravelobensis (Table 4.2). For M. murinus, I had to use Moran’s I alpha

values of 0.01 and 0.001, instead of 0.2 for the ME spatial filtering procedure to produce

properly specified models for two scales: 2 ha and 4 ha respectively. The resulting spatial GLMs

incorporated fewer parameters (fewer fitted Moran’s eigenvectors) and had more spatial

autocorrelation than models based on 0.2 alpha but the spatial autocorrelation was still low

(<0.33). Table 4.2: Results of Residual Spatial Autocorrelation Tests.

Species Model Residuals (Landscape Scale in ha) Moran’s I Expected P-value

Cheirogaleus medius

P = A(0.5) -0.03 -0.02 0.51 P = A(1) -0.04 -0.02 0.53 P = A(2) -0.01 -0.02 0.48 P = A(4) 0.03 -0.02 0.39 P = A(8) 0.07 -0.02 0.32

P = A(16) 0.10 -0.02 0.27 P = A(32) 0.10 -0.02 0.25 P = A(64) 0.11 -0.02 0.23

P = A(128) 0.15 -0.02 0.18 P = A(256) 0.18 -0.02 0.14

Null 0.18 -0.02 0.15

Microcebus murinus

P = A(0.5) 0.90 -0.002 <0.01 P = A(1) 0.91 -0.002 <0.01 P = A(2) 0.91 -0.002 <0.01 P = A(4) 0.91 -0.002 <0.01 P = A(8) 0.91 -0.002 <0.01

P = A(16) 0.91 -0.002 <0.01 P = A(32) 0.92 -0.002 <0.01 P = A(64) 0.92 -0.002 <0.01

P = A(128) 0.91 -0.002 <0.01 P = A(256) 0.91 -0.002 <0.01

Null 0.91 -0.002 <0.01

94

Microcebus murinus

P = A(0.5)+21 Fitted ME 0.10 -0.002 0.03 P = A(1) +21 Fitted ME 0.10 -0.002 0.03 P = A(2) +17 Fitted ME 0.20 -0.002 <0.01 P = A(4) +15 Fitted ME 0.24 -0.002 <0.01 P = A(8) +21 Fitted ME 0.09 -0.002 0.05

P = A(16) +21 Fitted ME 0.13 -0.002 0.01 P = A(32) +21 Fitted ME 0.10 -0.002 0.03 P = A(64) +21 Fitted ME 0.10 -0.002 0.03

P = A(128) +20 Fitted ME 0.11 -0.002 0.02 P = A(256) +20 Fitted ME 0.11 -0.002 0.02

Null+20 Fitted ME 0.11 -0.002 0.02


P = A(0.25) 0.81 -0.002 <0.01 P = A(0.5) 0.81 -0.002 <0.01 P = A(1) 0.81 -0.002 <0.01 P = A(2) 0.82 -0.002 <0.01 P = A(4) 0.81 -0.002 <0.01 P = A(8) 0.82 -0.002 <0.01

P = A(16) 0.82 -0.002 <0.01 P = A(32) 0.82 -0.002 <0.01 P = A(64) 0.83 -0.002 <0.01

P = A(128) 0.83 -0.002 <0.01 Null 0.83 -0.02 <0.01


P = A(0.25)+21 Fitted ME 0.13 -0.002 <0.01 P = A(0.5)+21 Fitted ME 0.12 -0.002 <0.01 P = A(1)+22 Fitted ME 0.12 -0.002 <0.01 P = A(2)+22 Fitted ME 0.09 -0.002 0.04 P = A(4)+22 Fitted ME 0.13 -0.002 <0.01 P = A(8)+21 Fitted ME 0.13 -0.002 <0.01

P = A(16)+11 Fitted ME 0.13 -0.002 <0.01 P = A(32)+22 Fitted ME 0.13 -0.002 <0.01 P = A(64)+22 Fitted ME 0.10 -0.002 0.03 P =A(128)+22 Fitted ME 0.13 -0.002 <0.01

Null +21 Fitted ME 0.10 -0.002 0.03 P= Occurrence, A= Amount of Forest, ME= Moran’s Eigenvectors


Species occurrence buffers ranged in size from 0.5 to 256 ha for C. medius and M. murinus, and

from 0.25 to 128 ha for M. ravelobensis (Table 4.3). Scale responses differed between species.

For C. medius, the scale of effect to habitat amount was between 1 ha (ΔAIC=0.35) and 4 ha

(ΔAIC=0.35), with 2 ha having the lowest ΔAIC (0.00) which is an area slightly larger than their

mean home range size. Therefore, for C. medius I was able to reject the null hypothesis that

lemur species occurrence with respect to habitat amount will vary randomly with respect to

scale. For M. murinus, the scale of effect to habitat amount was 8 ha (Table 4.3), which is

approximately three times the size of their reported mean home range size but within the

95

median-dispersal distance (251 m) (Radespiel et al. 2003). Therefore, I was able to reject the

null hypothesis for this species as well. For M. ravelobensis, I was not able to reject the null

hypothesis because it appears there was no scale of effect to habitat amount for this species

because the null spatial model (accounting for spatial autocorrelation) had the lowest ΔAIC

value (Table 4.3). The scales with the next lowest ΔAIC values were at 64 ha (ΔAIC=2.14). Table 4.3: Lemur Species Scale Responses to Habitat Amount.

Species Model (Landscape Scale in ha) Null Deviance (DoF)

Residual Deviance

(DoF) ΔAIC McFadden’s

R2

Cheirogaleus medius

P = A(2) 61.00(43) 46.19(42) 0.00 0.24 P = A(4) 61.00(43) 46.54(42) 0.35 0.24 P = A(1) 61.00(43) 47.16(42) 0.97 0.23 P = A(8) 61.00(43) 48.68(42) 2.49 0.2

P = A(16) 61.00(43) 48.96(42) 2.77 0.2 P = A(0.5) 61.00(43) 49.29(42) 3.09 0.19 P = A(32) 61.00(43) 49.62(42) 3.43 0.19 P = A(64) 61.00(43) 50.90(42) 4.71 0.17

P = A(128) 61.00(43) 54.00(42) 7.81 0.11 Null 61.00(43) 61.00(43) 12.81 0

P = A(256) 61.00(43) 59.10(42) 12.91 0.03

Microcebus murinus

P = A(8)+21 Fitted ME 720.87(519) 99.92(497) 0.00 0.86 P = A(1)+21 Fitted ME 720.87(519) 113.26(497) 13.34 0.84

P = A(0.5)+21 Fitted ME 720.87(519) 123.52(497) 23.60 0.83 Null+20 Fitted ME 720.87(519) 139.94(499) 36.02 0.81

P = A(2)+17 Fitted ME 720.87(519) 144.11(501) 36.19 0.8 P = A(32)+21 Fitted ME 720.87(519) 136.72(497) 36.80 0.81

P = A(256)+20 Fitted ME 720.87(519) 139.11(498) 37.19 0.81 P = A(128)+20 Fitted ME 720.87(519) 139.84(498) 37.92 0.81 P = A(64)+21 Fitted ME 720.87(519) 138.32(497) 38.40 0.81 P = A(16)+21 Fitted ME 720.87(519) 144.25(497) 44.33 0.8 P = A(4)+15 Fitted ME 720.87(519) 161.71(503) 49.79 0.78


Null+21 Fitted ME 774.92(558) 228.67(536) 0.00 0.7 P = A(64)+22 Fitted ME 774.92(558) 228.61(535) 2.14 0.7

P = A(0.25)+21 Fitted ME 774.92(558) 240.11(536) 11.64 0.69 P = A(0.5)+21 Fitted ME 774.92(558) 242.01(536) 13.54 0.69 P = A(8)+21 Fitted ME 774.92(558) 242.63(536) 14.16 0.69

P = A(16)+11 Fitted ME 774.92(558) 242.83(546) 14.36 0.69 P = A(2)+22 Fitted ME 774.92(558) 242.07(535) 15.60 0.69 P = A(4)+22 Fitted ME 774.92(558) 242.18(535) 15.71 0.69

P = A(32)+22 Fitted ME 774.92(558) 243.08(535) 16.61 0.69 P =A(128)+22 Fitted ME 774.92(558) 243.08(535) 16.61 0.69 P = A(1)+22 Fitted ME 774.92(558) 247.55(535) 21.08 0.68

P= Occurrence, A= Amount of Forest, and ME= Moran’s Eigenvectors; DoF= Degrees of Freedom.

96

4.7.3 Question 2: What is the Landscape-Level Effect of Habitat Amount on Species Occurrence?

The amount of forest in a landscape was a significant positive contributor to C. medius

occurrence in all of the models except the largest scale (256 ha) measured (Appendix A), where

amount of forest was positive but not a significant contributor to C. medius occurrence. C.

medius occurrence was most influenced by the amount of forest at landscape scales of 2 ha and

4 ha because the associated McFadden’s R2 values were the highest of all scales measured

(Table 4.3). The amount of forest in a landscape was a significant positive contributor to M.

murinus occurrence in the six smallest landscape scales. At the three larger scales (32, 64, and

128 ha) the amount of forest was a non-significant but positive contributor to M. murinus

occurrence (Appendix B). In the largest landscape scale (128 ha) the amount of forest was a

non-significant negative contributor to M. murinus occurrence (Appendix B). As with C. medius

the scale of effect selected based on ΔAIC for M. murinus (8 ha) had the highest McFadden’s R2

value (Table 4.3). The 8 ha landscape scale is where amount of forest had the greatest effect on

M. murinus occurrence. The amount of forest was a significant positive contributor to M.

ravelobensis occurrence at the three smallest scales (Appendix C). However, since the null

model had the lowest ΔAIC and highest McFadden’s R2 value, the GLM results for M.

ravelobensis suggest that they do not respond to habitat amount at any scale measured (Table

4.3).

4.8 Discussion In the first study of its kind on lemur biogeography, I modeled species-specific scale responses

to habitat amount using a landscape-centered approach accounting for the effect of spatial

autocorrelation from overlapping landscapes. Patterns and processes in ecological data are

related to spatial features of landscapes resulting in a high degree of spatial autocorrelation. Not

accounting for this spatial autocorrelation in presence/absence data may result in erroneous

results because of the assumption requiring independence of errors for logistic regression

(Quinn & Keough, 2002). For Microcebus spp., I found that spatial models (which incorporate

21 additional variables) had lower AIC values compared to univariate models that did not

account for spatial autocorrelation. I found that even related species responded differently to

habitat amount at a landscape-level. For two species (C. medius and M. murinus), the amount of

habitat predicted occurrence, but for M. ravelobensis, I found no effect of amount of habitat on

97

species occurrence. Below, I will suggest that the species-specific scale of effect to habitat

amount was related to home range size (C. medius) or potential dispersal ability (M. murinus),

and was explained by population density (C. medius) and edge effects (M. murinus).


Scale responses to landscape characteristics are an important consideration in landscape ecology

(Hutchison, 1965; Turner et al. 2001). Species-specific scale responses to landscape

characteristics have been demonstrated in numerous taxa (Borcard et al. 2004; Holland et al.

2004; Bellier et al. 2007; Patterson & Malcolm, 2010; de Knegt et al. 2011; Eigenbrod et al.

2011; Bradter et al. 2013; Gilroy et al. 2015; Thorton & Fletcher, 2015), including primates

(Gray et al. 2010; Thorton et al. 2011; Ordóñez-Gómez, 2014). However, researchers have not

inspected species-specific scale responses in lemur biogeography.

C. medius and M. murinus showed species-specific scale responses to habitat amount and M.

ravelobensis did not. For C. medius, the scale of effect was similar to their reported mean home

range size. The greater the scale, the higher the ΔAIC value was for this species, suggesting that

habitat amount had less of an effect on C. medius occurrence at larger than smaller scales. For

M. murinus, the scale of effect was greater than their mean home range but within the reported

median-dispersal distance for this species (Radespiel et al. 2009). I found no scale responses for

M. ravelobensis to habitat amount. For this species, something other than area must be

contributing to their occurrence patterns in the landscape.

Thorton et al. (2011) found a scale of effect of 500 m (radius) for the negative response of

habitat fragmentation on Ateles geoffroyi occurrence in a fragmented landscape within Petén

region of Guatemala. A 500 m radius would be roughly equivalent to 78.54 ha and well within

reported home ranges sizes for this species (Chapman et al. 1995), a similar pattern of response

to C. medius. However, like my results for M. ravelobensis, Thorton et al. (2001) did not find a

species-specific scale response for habitat loss or fragmentation for A. pigra. Investigating the

scale of effect on A. geoffroyi diet and behavior, Ordóñez-Gómez et al. (2014) found that the

amount of forest cover influenced behaviors, such as resting and time spent feeding on leaves, at

a 126 ha landscape-level. Again, this landscape-level is well within the home range size reported

for this species. Conversely, Gray et al. (2010) found that gibbon (N. gabriellae) occurrence was

positively influenced by the amount of evergreen forest at large scales (3000 m), which is much

98

larger than the home range in evergreen forest reported for this species (Traeholt et al. 2005).

These results differ from my models for C. medius and for those reported for A. geoffroyi

(Thorton et al. 2011; Ordóñez-Gómez et al. 2014). Thus, the scale of effect for C. medius was

similar to those reported for A. geoffroyi (Thorton et al. 2011; Ordóñez-Gómez et al. 2014), but

both are much smaller than what was reported for N. gabriellae (Gray et al. 2010). Home range

is and locomotor ability may explain why these three frugivores have such disparate scale

responses. Home range sizes are much larger in N. gabriellae (Kenyon, 2007) and A. geoffroyi

(Chapman et al. 1995) than in C. medius (Müller, 1998), although C. medius and A. geoffroyi

share a similar scale response (when accounting for home range size). Also in terms of mobility,

there is a gradient of ability and agility from the slow quadrupedal locomotion of C. medius

(Young et al. 2007) to the intermediate brachiation of A. geoffroyi (Cant, 1986) to the fast and

efficient richochetal brachiation found in N. gabriellae (Usherwood et al. 2003; Rawson &

Traeholt, 2011). Therefore ranging behaviour and locomotor ability are likely to explain why N.

gabriellae has a greater scale response than A. geoffroyi and C. medius.

Jackson and Fahrig (2012) suggested that species-specific responses to landscape characteristics

are related to some element of a species’ life history such as mobility, body size, and

reproductive rate. These researchers found that dispersal distance has a positive impact on scale

of effect and they suggested that landscape radius should be assessed at sizes four to nine times

greater than the median-dispersal distance for a species. While this landscape size may be

accurate for large-bodied, highly mobile species, my results and studies on other primate species

suggest that a smaller landscape size may be more appropriate (Gray et al. 2010; Thorton et al.

2011; Ordóñez-Gómez, 2014). Most primates are dispersal limited due to their highly arboreal

nature (Beaudrot & Marshall, 2011). Smaller landscapes would make sense for dispersal-limited

species because dispersal costs are greater for dispersal-limited species and would therefore

limit their ability to move long distances and may result in smaller home range sizes (Isbell &

van Vuren, 1996). Alternatively, researchers invoked body size to explain species differences in

scale of effect response to habitat loss and fragmentation, where some species showed a relation

to body size while others did not (see review in Jackson & Fahrig, 2012). However, body size

differences do not explain why N. gabriellae had a larger scale response than A. geoffroyi or C.

medius because both N. gabriellae and A. geoffroyi are similar in body size (Ford & Davis,1992;

Rawson & Traeholt, 2011) but both are larger than C. medius (Fietz, 1999).

99

Habitat heterogeneity may impact the scale of effect for frugivorous primates (Ordóñez-Gómez

et al. 2014). For example, both A. geoffroyi and C. medius respond to habitat fragmentation at

scales well below their maximum home range size (Thorton et al. 2011; Ordóñez-Gómez et al.

2014) and these highly frugivorous species may respond at a smaller scale because they may be

sensitive to differences in habitat quality as a result of their diet, limiting its scale of effect. In

my study, highly frugivorous C. medius had a smaller scale response than the more diminutive

fauni-frugivore, M. murinus. However, frugivorous N. gabriellae had a much larger scale

response than would be predicted based on the research done on A. geoffroyi (Thorton et al.

2011; Ordóñez-Gómez et al. 2014) and my results on C. medius. Differences in diet among

frugivores may explain why N. gabriellae has a larger scale response than A. geoffroyi or C.

medius. Both A. geoffroyi and C. medius consume a greater proportion of fruit (A. geoffroyi fruit

is >70% of diet, Chapman et al. 1995; C. medius fruit is >60% of diet, Fietz & Ganzhorn, 1999)

than N. gabriellae (fruit is =30.31% of diet, Rawson & Traeholt, 2011). However, A. pigra, a

facultative folivore that eats as much fruit as is available but will switch to leaves if needed

(Pavelka & Knopff, 2004), showed no species-specific scale response (Thorton et al. 2011).

Therefore, it is unclear if diet is a major determining factor of species-specific scale of responses

to habitat amount.

Fahrig (2001) suggests reproductive rate as a potential mechanism to determine minimum

habitat requirements (Fahrig, 2001). In birds, species with higher reproductive rates required

less habitat (Vance et al. 2003). Although primates are known for their slow reproductive rates

(Jones, 2011), C. medius has a relatively high reproductive rate that can change with geography,

while other life history characteristics such as home range size remain static (Lahann &

Dausmann, 2011). However, C. medius, A. geoffroyi, and N. gabriellae all have similar birth

intervals of about 20–24 months (A. geoffroyi, Difore & Campbell, 2007; C. medius, Lahann &

Dausmann, 2011; N. gabriellae, Rawson & Traeholt, 2011). Therefore, reproductive rate does

not explain the large-scale response for N. gabriellae.

In terms of group composition and social behavior, both N. gabriellae and C. medius form

mostly monogamous pairs (Fietz, 1999), while A. geoffroyi form large fission/fusion groups

(Chapman et al. 1995). Although both N. gabriellae and C. medius form monogamous pairs, N.

gabriellae is highly territorial (Traeholt et al. 2006) while C. medius is not (Müller, 1998).

100

Therefore, it is possible that territoriality will affect a species scale of effect to habitat loss and

fragmentation.

Another possible explanation for why A. geoffroyi and C. medius have smaller scale responses

than N. gabriellae may be related to their respective population densities. Both A. geoffroyi and

C. medius occur at much higher densities (Ganzhorn & Kappeler, 1996; Estrada et al. 2004) than

N. gabriellae (Rawson et al. 2009). For example, Estrada et al. (2004) found that A. geoffroyi

density ranged from 17.0 to 56.4 individuals/km2 and Schäffler and Kappeler (2015) found C.

medius density can occur at 180 individuals/km2 but, Rawson et al. (2009) found that N.

gabriellae occurs at a much lower density of 0.71 groups/km2. Because of low population

densities and their territorial nature, the smallest area where a breeding population of N.

gabriellae could occur would be much larger than it would be for A. geoffroyi and C. medius

irrespective of reproductive rate, diet, or group composition.

Even though M. murinus is much smaller than C. medius, it showed a larger scale response.

However, the closely related and similarly sized M. ravelobensis showed no scale of effect to

habitat loss and fragmentation. What differences between M. murinus and M. ravelobensis

would explain why one species shows a scale of effect while the other does not? M. murinus has

larger home ranges and greater dispersal ability in continuous forest (Radespiel, 2000; Radespiel

et al. 2003) than M. ravelobensis (Weidt et al. 2004; Radespiel et al. 2008). In Chapter 3, I

found evidence to suggest that M. ravelobensis may be more dispersal limited than M. murinus

in the same fragmented landscape. The two species are similar in size but M. murinus stores fat

in its tail seasonally (Zimmerman et al.1998). Although their diet is similar (Radespiel et al.

2006), one study showed that they have different microhabitat preferences in continuous forest

(Rakotondravony & Radespiel, 2009), with M. murinus preferring higher elevation drier forest

than M. ravelobensis. Reproductive rate between each species is similar in that they are both

seasonal breeders (Radespiel, 2000; Randrianambinina et al. 2003) and they are both solitary

foragers (Radespiel et al. 2003; Randrianambinina et al. 2003; Weidt et al. 2004). One

difference is that M. murinus typically sleeps in protected tree holes with only females sleeping

with conspecifics (Radespiel et al. 2003) while both sexes of M. ravelobensis may sleep with

conspecifics in less protected sites (Radespiel et al. 2003) and self-constructed leaf nests

(Thorén et al. 2011). Although I suggest that differences in scale responses between more

frugivorous species appear to be related to population density, this does not appear to be the

101

case for M. murinus and M. ravelobensis. Steffens and Lehman (2016) found that there were no

significant differences in abundance or density between each Microcebus species within the

same study landscape. We found similar factors (dendrometrics and fragment size and distance

metrics) explained abundance for both species. However, we found that potential edge effects

explain density in M. ravelobensis. In continuous forest near my research landscape, Burke and

Lehman (2015) found that M. ravelobensis was more edge tolerant than M. murinus. A need for

more protected sleeping sites may preclude M. murinus from occupying edge habitats that are

characterized by having smaller trees (Murcia, 1995). Conversely, M. ravelobensis does not

require protected sleeping sites, even following birth (Thorén et al. 2011). Other primate species

that show no species-specific scale of effect to habitat loss and fragmentation such as A. pigra

(Thorton et al. 2011) are also thought to be edge tolerant (Arroyo-Rodríguez & Dias, 2010).

Thus, edge effects may play a crucial but largely unstudied role in determining a species scale of

effect in fragmented landscapes.

I have shown that for three nocturnal lemur species, the scale of effect for habitat loss and

fragmentation in a fragmented landscape is species-specific and may be moderated by things

other than typical life history characteristics (Jackson & Fahrig, 2012). Factors such as home

range size, locomotor ability, population density, lack of territoriality, and edge effects may

influence each species’ scale of effect. Future studies should incorporate additional variables to

determine at precisely what scale lemur species respond to habitat amount.

4.8.2 Question 2: What is the Effect of Habitat Amount on Species-Specific Occurrence of Lemurs at the Landscape-Level?

Many studies have looked at habitat amount (area) and its relation to species richness (e.g., the

species-area relationship; Chapter 2) and occurrence (e.g., metapopulation dynamics; Chapter

3). As I demonstrated in Chapter 2, area is the factor with the greatest influence on lemur

species richness in a fragmented landscape. In Chapter 3, I demonstrated that area had a greater

impact on individual lemur species occurrence than isolation/connectivity within a fragmented

landscape. What about when landscapes are used as the unit of analysis? Does the area pattern

continue to hold? For both C. medius and M. murinus, the amount of habitat (a measure of area)

within a landscape is a determinant of species occurrence at species relevant scales. However,

the amount of forest was not a determinant of species occurrence at larger scales. For M.

ravelobensis, the amount of habitat within a landscape had a positive effect at only smaller

102

scales. However, I can accept the null hypothesis that area randomly affects M. ravelobensis

occurrence as the null model had the lowest ΔAIC values. This result is likely due to M.

ravelobensis’s tolerance for edge effects, and because, regardless of scale, occurrence in this

lemur species is not determined by how much habitat occurs within a landscape. Therefore, I

suggest that research is conducted investigating other landscape and biogeographic factors to

determine what impacts occurrence for this species.

4.9 Suggestions for Conservation Protecting sufficient area to maintain populations is important as I suggest in Chapter 3, based

on metapopulation dynamics. However, using a landscape-level approach, I found that

responses of three nocturnal lemur species occurred at small scales, suggesting that there is a

need to consider how small-scale disturbances may affect species occurrence. For example, tavy

(slash-and-burn agriculture) and selective logging may not negatively impact an area at scales

relevant to humans in the short term, but may have detrimental impacts for C. medius and M.

murinus at small spatial scales. For M. ravelobensis, whose occurrences was not impacted by

the amount of habitat within a landscape, we should assess other variables to determine what

drives occurrence in this species in fragmented landscapes. From a conservation perspective, it

appears that M. ravelobensis, the only species in this study considered endangered, is tolerant of

habitat loss and fragmentation. Although it is edge tolerant in continuous forests, this species

may be at greater risk of predation from both endemic and non-endemic predators due to its use

of unprotected sleeping sites and existence in very small fragments. I recommend a dual

approach that helps maintain or increase habitat amount at large and small scales to protect

lemur species with greater efficacy.

4.10 Conclusion Although both species-area relationships and metapopulation dynamics have uses in

understanding lemur biogeography, landscape ecology is very useful for determining how

individual species respond to habitat loss and fragmentation at species relevant scales. Area was

the main driver of species richness and occurrence in three nocturnal lemurs using patch-level

analyses, but I was able to demonstrate that this result is not always true when using a

landscape-level approach even among closely related species (e.g., two species of Microcebus).

Frugivorous species at higher densities, with more limited mobility, which are not territorial,

have small-scale responses to habitat amount. More versatile fauni-frugivores have

103

comparatively larger scale responses (M. murinus) or no response at all (M. ravelobensis) to

habitat amount on species occurrence. My study demonstrates that the amount of habitat (a

measure of area) is not always a determinant of species occurrence despite its nearly axiomatic

use in the biogeographic literature (Whittaker & Triantis, 2012), suggesting that we still do not

fully understand what determines species occurrence. Therefore, further research that

incorporates variables in addition to the amount of habitat using a landscape-level approach

while considering scale and spatial autocorrelation are needed to determine exactly how primate

species, including the diurnal species sympatric with the three nocturnal species described in

this dissertation, respond to habitat loss and fragmentation.

104

Chapter 5: Patch and Landscape Determinants of Lemur Species Richness and Occurrence

5.1 Conclusion My study aimed to investigate how area impacts lemur species richness and occurrence at

different scales, such as using a patch-level analysis within a single large landscape and a

landscape-level analysis using many small landscapes. I applied a patch-level approach to

investigate the nature of species-area relationships on primate communities (Chapter 2); I

applied metapopulation dynamics to look at how individual species occurrence is impacted by

patch area and isolation (Chapter 3); and I looked at landscape-level patterns of lemur species

occurrence with relation to habitat area in species-specific landscapes (Chapter 4). I found that

area was the main driver of species richness and occurrence at a patch-level, but that area did

not always determine species occurrence at a landscape-level.

5.2 Summary of Results

5.2.1 Species-Area Relationships in a Lemur Community

Species-area relationships in a fragmented landscape should follow a sigmoidal pattern

(Lomolino, 2000; Tjørve, 2003; Tjørve, 2009). However, I found that lemur species richness

patterns followed a convex pattern (power model) within a fragmented landscape. Sigmoidal

models may form s-shaped curves (Chapter 2, Fig. 2.1). In a sigmoidal species-area relationship

the lower, straight portion of the s-curve is hypothesized to be relatively flat due to the “small

island effect” (Lomolino, 2000). The “small island effect” occurs when something other than

area is predicting species richness or the area is too small to support a population for a particular

species (Lomolino, 2000). The curve steepens as area becomes the main predictor of species

richness, and eventually the curve plateaus as the total number of species is reached in a finite

species pool (Lomolino, 2000). On the other hand, power models form convex curves that are

relatively monotonic compared to sigmoidal models, and they do not have an upper asymptote

(Chapter 2, Fig. 2.2; Lomolino, 2000; Tjørve, 2003). With respect to the species-area

relationship forming a power model rather than a sigmoidal model in my study, either there is a

high degree of movement between fragments or some species are able to survive in even the

smallest habitat fragments. The results from Chapter 3, summarized below, demonstrate that

some species (C. medius and E. fulvus) are able to move between fragments while others (M.

105

murinus and M. ravelobensis) may move but are more tolerant of edge effects (specifically M.

ravelobensis) and can survive in even the smallest fragments.

Determining which model best describes the species-area relationship in my study site is

impacted by certain limitations of the study landscape. The study landscape contains 42

fragments ranging from 0.23 to 117.7 ha and I only observed six of the eight potential lemur

species (I suggest two have gone locally extinct). Triantis et al. (2012) suggest a fragment range

size of three orders of magnitude (compared to the 2.5 in my study) to facilitate the detection of

a sigmoidal pattern in the species-area relationship. Therefore, there is some uncertainty if the

power model is more appropriate than a sigmoidal model, as I found in my study. However, my

study included the largest number of fragments investigating the species-area relationship on

lemurs in Madagascar and is one of the largest habitat fragmentation studies on primates in the

world. My study landscape contained too few large fragments to support all eight possible lemur

species. My study demonstrates that local extinction of certain larger-bodied, preferentially

hunted species (e.g., E. mongoz and A. occidentalis) occurred in fragments that continue to

support other large, preferentially hunted species (e.g., P. coquereli). As noted above, both

Microcebus species are able to survive in even the smallest fragments in my study site (e.g.,

0.23 ha), which would obscure the detection of a sigmoidal curve unless even smaller (< 0.23

ha) fragments occurred within my study landscape. Although other studies have found lemurs in

fragments greater than 1 ha (Ganzhorn, 2003; Schad et al. 2004; Olivieri et al. 2008), my study

is the first to find lemurs in fragments smaller than 1 ha.

I found that area was the strongest predictor of lemur species richness when comparing different

Generalized Additive Models. However, using a hierarchical partitioning procedure, both mean

tree height and the total amount of human disturbance contributed a combined 28.04% of the

variation in lemur species richness compared to 60.82% for ln area. It is unclear if mean tree

height or human disturbance had a positive or negative influence on lemur species richness but

human disturbance was positively related to fragment area. Additionally, the absence of E.

mongoz and A. occidentalis may be related to human disturbance factors. For example, I found

lemur traps located within fragments in the study site and local residents report that they

preferentially hunt E. mongoz. The results of this chapter demonstrate that at a patch-level, area

appears to be a driving factor determining lemur species richness, but future studies should

further investigate the potential role of human disturbance and mean tree height.

106

5.2.2 Metapopulation Dynamics of Lemur Species

Not all lemur species formed stable metapopulations. Both P. coquereli and L. edwardsi

appeared to form non-equilibrium metapopulations, and these two species may be declining

towards extinction within my study site. The low occurrence of both P. coquereli and L.

edwardsi in the study site precluded my ability to analyze their metapopulation structure in more

detail. This raises the issue that the two most apparently vulnerable species in my study site

have populations below levels that could be analyzed to determine how they are impacted by

habitat loss and fragmentation. Two other species, E. mongoz and A. occidentalis, already

appear locally extinct in the landscape. Immediate conservation action is warranted and we need

further research in landscapes with intermediate stages of loss containing larger and more

connected fragments. This research and conservation action will help us better understand the

impact of habitat loss and fragmentation on P. coquereli and L. edwardsi occurrence to prevent

the local extinction of these species. The two frugivorous species (E. fulvus and C. medius)

formed stable metapopulations, but C. medius may be more sensitive to increased isolation than

E. fulvus. The two smaller-bodied, fauni-frugivores, M. murinus and M. ravelobensis are

virtually ubiquitous in fragments throughout the landscape inhabiting 35 and 34 of the 42

fragments, respectively. Although both species occupied even the smallest fragments, I found

small differences in model selection between each species. The most likely models for M.

murinus were the mainland-island model followed by the incidence function model (IFM),

where dispersal was estimated from the survey data. For M. ravelobensis, four of the five

models were roughly equivalent including the mainland-island model, IFMs where dispersal

was determined by the literature, occupancy data, and based on the square root of the mean

home range reported in the literature. Both species appear to form source-sink mainland-island

metapopulations within an intermediate metapopulation. For both species, I hypothesize that the

larger fragments and continuous forest likely act as sources and the smallest fragments act as

sinks. Moreover, some movement is likely occurring between all the fragments for M. murinus,

as evidence by their greater dispersal ability, while M. ravelobensis are more edge tolerant and

capable of surviving within small fragments. We should test these hypotheses using

metapopulation dynamics combined with mark recapture methods.

Area was the main driver of lemur species occurrence in my study site, while the relationship

between isolation/connectivity and lemur occurrence was less clear. E. fulvus was the only

species to show a positive relationship between connectivity and lemur occurrence. There was a

107

neutral response to isolation/connectivity for C. medius. For both Microcebus species, the

probability of occurrence was actually lower in more connected than less connected fragments.

Differences in lemur species dispersal ability and edge tolerance may be driving differences in

metapopulation dynamics among lemur species. For both Microcebus species, and especially M.

murinus, source-sink dynamics may be occurring within my study site. The results of my study

on lemur species occurrence using metapopulation dynamics help explain why lemurs in this

fragmented landscape form non-sigmoidal species-area relationships. The results of this chapter

help inform the species-area relationship patterns observed in Chapter 2 and show that at a

patch-level lemur species occurrence in a fragmented landscape is mainly driven by area.

5.2.3 Landscape Effects on Lemur Species

I investigated the species-specific landscape-level effects of habitat loss on C. medius, M.

murinus, and M. ravelobensis within my study landscape. Although I found that area is a major

factor determining lemur species richness and occurrence at a patch-level, at a landscape-level

area was not as important for all lemur species. The results from Chapter 4 provide further

evidence to support my suggestion in Chapter 3 that the different metapopulation models

selected for each Microcebus species were the result of greater dispersal ability (M. murinus)

and edge tolerance (M. ravelobensis). M. murinus had a relatively large scale response to habitat

amount, which I suggest is due to its greater modeled and reported dispersal ability (Radespiel et

al. 2003; Schliehe-Diecks et al. 2012), while M. ravelobensis had no response to habitat amount

within a landscape, which I suggest is due to its reported greater edge tolerance (Burke and

Lehman 2015). C. medius had small scale responses to habitat amount within a landscape,

similar to reports for Ateles geoffroyi (Thorton et al. 2011), but contrary to results for Nomascus

gabriellae (Ordóñez-Gómez et al. 2014), all highly frugivorous species. I suggest that for

frugivorous species, population density, home range size, mobility, and territoriality may

influence species responses to habitat amount using a landscape-level of analysis. Using a

landscape approach, I was able to show that lemurs have species-specific responses to habitat

loss and fragmentation. Together with Chapter 3, the results of Chapter 4 help explain why a

sigmoidal SAR was not observed in Chapter 2. The results from Chapter 4 of my study are

important because they clearly show that area is not always the driving factor determining

species occurrence, and that other variables such as edge tolerance, dispersal ability, population

108

density, and territoriality may play important roles in determining how species respond to

habitat loss and fragmentation.

5.3 Implications

5.3.1 Species-Area Relationship

Area has been the focus of biogeographic research since Watson (1859) published the first

species area curve. Since his work on the species-area relationship, countless other studies have

continued to find a re-occurring pattern of increasing species richness with increasing area

(Whittaker & Triantis, 2012). Whether areas are larger portions of continuous habitat or

fragments of habitat, the pattern of increasing species richness with increasing area has held.

Although the shape of the species-area relationship is still debated (Tjørve, 2003; Tjørve, 2009),

the species-area relationship itself is so strong that it is considered axiomatic in biogeography

(Whittaker & Triantis, 2012). However, what causes the species-area relationship is still poorly

understood. My work continues to demonstrate the ability of area to overwhelm other variables

that potentially impact species richness. I weigh in on the debate regarding the shape of species

area relationships. Although I predicted I would find a sigmoidal model to represent lemur

species-area relationships in a fragmented landscape with potentially hostile matrix, I found that

a model more suitable to continuous forest was the most likely. Other research on species-area

in primates also demonstrates that area is the main factor determining primate species richness

(Reed & Fleagle, 1995; Cowlishaw, 1999; Cowlishaw & Dunbar, 2000; Lehman, 2004;

Harcourt & Doherty, 2005; Marshall et al. 2010). However, only a few of these studies (Reed &

Fleagle, 1995; Lehman, 2004; Marshall et al. 2010) incorporated elements other than area.

Although my study confirms that area is an important contributor to primate species richness, it

does suggest that other variables may be involved, such as the amount of human disturbance and

mean tree height.

5.3.2 Metapopulation Dynamics

Following work on community level patterns of species distribution, researchers from

population biology such as Levins (1969, 1970) and later Hanski (1994abc) sought to explain

how individual species distribute themselves within patchy environments. Their work led to the

creation of metapopulation dynamics. Incorporating both area and isolation, metapopulation

models are useful to determine how individual species are impacted by habitat loss and

109

fragmentation (Hanski 1994b). Metapopulation dynamics have been applied to primate species

but never to lemur species. As the first researcher to apply metapopulation dynamics to lemur

species within a fragmented landscape, I was able to determine how lemur species are affected

by habitat loss and fragmentation from a patch perspective. I show that area is a greater

determinant of primate occurrence than isolation. Other primate studies have found similar

results showing a limited isolation impact on primate occurrence (Lawes et al. 2000; Chapman

et al. 2003; Cristobal-Azkarate et al. 2005; Raboy et al. 2010).

An interesting use of metapopulation dynamics is in determining dispersal patterns of a species

within fragmented landscapes. I found that my metapopulation models generated 101 m

dispersal parameter (α) estimates for both Microcebus species that underestimate reported

dispersal distances for M. murinus (251 m; Radespiel et al. 2003) and overestimate reported

dispersal distances for M. ravelobensis (54 m; Radespiel et al. 2009). I generated my dispersal

estimates within a fragmented landscape while the reported dispersal distances were from a

continuous forest landscape. This suggests that dispersal patterns in continuous forest may differ

from dispersal patterns in fragmented forest landscapes. My use of the dispersal index, square

root of each species home range, provided a closer approximation to reported dispersal distances

(M. murinus = 168.2 m; M. ravelobensis = 76 m). We need further research to determine if

dispersal ability in primates is different in continuous forest versus fragmented forest

landscapes.

5.3.3 Landscape Effects

Both the species-area relationship and metapopulation dynamics look at community and species

level patterns, respectively, from a patch perspective. Meanwhile, since the early 2000s

researchers from landscape ecology, such as McGarigal and Cushman (2002) and Fahrig (2003),

have argued that viewing species richness and occurrence patterns from a patch perspective can

mask underlying trends that may explain the cause of the patterns themselves. Specifically,

work by Fahrig (2003) demonstrates how using landscapes as units of analysis as opposed to

patches yield important insights into how species richness responds to habitat loss and

fragmentation, as the latter process is a landscape-level phenomenon. A shift from a patch

perspective to a landscape perspective introduces some new problems, such as determining what

landscape size is appropriate and relevant to the species of interest (Jackson & Fahrig, 2012;

Arroyo-Rodriguez & Fahrig, 2014). In the first study of its kind on lemurs, I investigated

110

species-specific landscape scales assessing lemur species occurrence in a fragmented landscape.

My study affirms the need to use a landscape-level approach when assessing the impact of

habitat loss and fragmentation on species occurrence. Using a landscape approach, I found that

for C. medius and M. murinus area was still the main variable impacting occurrence, but for M.

ravelobensis something other than area explains occurrence. As with similar studies on primates

(Gray et al. 2010; Thorton et al. 2011; Ordóñez-Gómez et al. 2014) and other arboreal mammals

(Patterson & Malcolm, 2010), I found differing scale responses among the species of my study.

By looking at species occurrence at a landscape-level, I demonstrated that area is not always the

main determinant of species occurrence, even in closely related species. Thus, other factors must

be influencing the ability of the lemur species I studied to occur in a fragmented landscape that

are partially masked at a patch-level, such as dispersal ability, edge tolerance, population

density, and territoriality in primates. My research also demonstrates the importance of looking

at species-specific landscape scale responses to habitat loss and fragmentation. Without

determining relevant scale responses, researchers may miss important information into how

species respond to habitat loss and fragmentation. My dissertation fits within the greater

discipline of biogeography by both reaffirming the role of area in determining primate richness

and occurrence at a patch-level, while also highlighting the importance of landscape-level

analysis to assessing species occurrence in a fragmented landscape.

5.4 Directions for Future Research

5.4.1 Species-Area Relationships

Future research on species-area relationships on lemurs should incorporate more variables and

should focus strongly on anthropogenic factors that could influence lemur species richness. A

study is needed with a greater range of both larger and smaller fragments than those used in my

current study in order to determine if a sigmoidal relationship is possible for lemur species.

Comparing species-area relationships among fragments of habitat to those within continuous

forest in Madagascar may help resolve some of the questions as to what species-area patterns

lemurs are expected to form under different conditions. Although a patch-level analysis, species-

area relationships are still a useful tool for conservation managers with limited time and budget

to determine the potential impacts of habitat loss and fragmentation on species richness.

111

5.4.2 Metapopulation Dynamics

Other metapopulation dynamics research on primates has focused on determining: species

persistence in fragmented landscapes (Chapman et al. 2003; Swart & Lawes, 1996), occurrence

(Lawes et al. 2000), extinction risk (Zeigler et al. 2013), minimum critical patch size (Lawes et

al. 2000), and population viability (Mandujano & Escobedo-Morales, 2008). The results of my

study will provide a starting point for other researchers to use metapopulation dynamics on other

primate species, in other regions, and for other purposes. For example, the results of my study

contribute to the single large reserve or several small reserves (SLOSS) debate (Diamond,

1975). I found the same species richness in the largest habitat fragment as within a combination

of other fragments, but not including the largest. This fact suggests that either single large or

several small would be a suitable management choice for this landscape. However, when you

look at the simulated metapopulation dynamics over time, if the largest fragments are removed

then species occurrence collapses for all four species. Meanwhile species occurrence among the

five remaining fragments remained relatively constant. The collapse in the remaining fragments

and subsequent stability in the five largest provides evidence that protecting a single (or a few)

large fragments may protect more species than several small fragments. By using similar

metapopulation dynamics and simulation techniques as I have in this study, other researchers

will be able to expand on the SLOSS debate and other topics in primate population ecology and

conservation biogeography.

An obvious direction for future metapopulation research would be to look at the nature of

source-sink relationships in lemurs and other primates found in fragmented forests near

continuous forest. My study suggests that source-sink dynamics may be occurring for

Microcebus species. Other research on lemurs has shown that some lemurs may be attracted to

edge habitats (Ganzhorn, 1995; Lehman et al. 2006a; 2006b). It is unclear if species can persist

in small habitat fragments that may appear attractive in the short-term due to certain benefits of

edge effects (e.g., increased leaf and fruit productivity), but that may be detrimental in the long

term due to negative effects of small fragment size, such as increased predation pressure or

reduced mating opportunity. We need to understand how source-sink dynamics are operating on

primates and what impact they have on primate persistence in fragmented landscapes.

Knowledge of dispersal is limited for many primate species, especially in fragmented

landscapes. Future research should investigate dispersal patterns in primates within fragmented

112

landscapes. Metapopulation dynamics can be used to generate fitted dispersal estimates.

However, dispersal can also be measured directly by incorporating mark recapture techniques

(Hanks 1994a), something that is possible for small-bodied primates such as Microcebus species

(Radespiel et al. 2003). Determining dispersal ability in primates under varying conditions is a

critical factor in assessing how species respond to habitat loss and fragmentation.

5.4.3 Landscape Ecology

There has been a recent acknowledgment of the role of landscape ecology in the field of

primatology. Influenced by Farhrig (2003) primate researchers have expressed the need for

more research using a landscape perspective and considering species-specific landscape

responses (Arroyo-Rodriguez & Fahrig, 2014). Those studies that have incorporated species-

specific landscape responses in primates have found that there are indeed differences (Gray et

al. 2010; Thorton et al. 2011; Ordóñez-Gómez et al. 2014). Building on the work of these early

pioneers in primate landscape ecology, I also found that lemur responses are species-specific at

the landscape-level. I suggest that future research continue to investigate how landscape-level

effects of habitat loss and fragmentation at species relevant scales impact primate species

richness and occurrence. Research on other mammals suggests that scales as large as four to

nine times the median-dispersal distance are appropriate (Jackson and Fahrig, 2012). However,

my research suggests that since many primates are dispersal limited, smaller scales should be

considered. My results for Microcebus species suggest that responses to habitat loss and

fragmentation reflect differences in dispersal ability and edge tolerance. Future studies should

measure dispersal ability in these species within fragmented habitats. For example, looking at

mark/recapture techniques using both metapopulation dynamics and landscape-level studies will

help determine how much dispersal ability influences these species occurrence. The influence of

edge effects on species occurrence is another avenue for future research. Future research should

look at how edge responses impact occurrence of species within habitat fragments and to

compare those results in edge habitats of nearby continuous forest. It is important to determine if

the results of my study are dependent on the nature of the fragmentation and loss within my

study landscape or if my results represent general patterns in lemur responses to habitat loss and

fragmentation. By comparing my study site to another with a different landscape configuration

but a similar lemur community will help us to better understand how the pattern of habitat loss

and fragmentation influence lemur species richness and occurrence.

113

5.5 Implications for Lemur Conservation Lemurs are the most endangered mammal group in the world (Schwitzer et al. 2014). Habitat

loss and fragmentation are the greatest threats, followed by illegal hunting (Schwitzer et al.

2014). With a wealth of biodiversity, Madagascar is one of the worlds 36 Biodiversity Hotspots

(CEPF, 2016) and lemurs are endemic to the island nation (Schwitzer et al. 2013). Madagascar

is one of the poorest countries in the world with over 92% of its residents living on less than two

U.S dollars a day (World Bank, 2013). Although wealthy in natural resources, such as

gemstones and other minerals, forestry products, and oil and gas, Madagascar is under-

developed and is one of the most corrupt countries in the world. Since the 1950s an estimated

40% of the forest has been converted to non-forest habitats (Harper et al. 2007) while poverty

has increased. My research can provide a useful conservation framework to inform lemur

conservation actions and plan future directions for conserving lemurs.

The largest threat to lemurs in Ankarafantsika National Park is from habitat loss and

fragmentation from fire (Bloesch, 1999). Other threats include degradation of habitat by

resource extraction (Steffens, personal observation), hunting pressure from humans (García &

Goodman, 2003), and predation, especially from introduced carnivores (Steffens, personal

observation). Many of the threats are likely exacerbated by habitat loss and fragmentation.

The conservation implications of my study are complex. On the one hand, I found the two

smaller bodied Microcebus species to be quite tolerant of even extreme fragmentation and forest

loss. Albeit likely for different reasons, these two species are capable of surviving in small

fragments. One area of concern for these two species is that they may be forming a source-sink

metapopulation where migrants from the continuous forest or largest fragments move into the

smallest fragments into untenable populations (for example single individuals). For M. murinus,

the situation is not as problematic because they have a large distribution across western

Madagascar (Mittermeier et al. 2010) and are listed as least concern by the IUCN (2016).

However, M. ravelobensis has a very limited distribution, is only protected in Ankarafantsika,

and is considered endangered due to recent habitat loss and a suspected population crash (IUCN,

2016). Although I suggest M. ravelobensis is edge tolerant, there are many factors that are

compounded by habitat loss and fragmentation that may impact their population in a fragmented

landscape, such as increased predation pressure from introduced predators (Farris et al. 2014).

Therefore, for Microcebus species in fragmented landscapes, I conservatively recommend

114

protecting large tracts of forest, and further research into the possibility of source-sink dynamics

and potential increase in predation risk related to edge effects within fragments.

For C. medius the situation is of moderate conservation concern. C. medius is classified as least

concern and is widely distributed, and occurs in many protected areas (IUCN, 2016). However,

the results of my study suggest that within fragmented landscapes C. medius needs medium to

large fragments that have low isolation. Therefore, I recommend protecting larger fragments and

improving connectivity among fragments.

For the large bodied lemurs in my study, I found two species (E. mongoz and A. occidentalis)

were likely extirpated from my study site, two species (P. coquereli and L. edwardsi) had too

few occurrence points to conduct meaningful statistical assessment of their distribution, and one

species (E. fulvus) formed a stable metapopulation but was less sensitive to habitat isolation. All

the larger bodied species except E. fulvus are considered endangered by the IUCN (2016). The

conservation implications for E. fulvus are similar for C. medius. From a species richness

perspective my work on species-area relationships demonstrates the immediate need to protect

as many large tracts of habitat as possible. However, this may not be enough to stem the local

extinction of L. edwardsi, and especially P. coquereli; the latter occurring only in the largest

fragments and not even within fragments nearest to the continuous forest. For these species and

the two (E. mongoz and A. occidentalis) extirpated from my site, I recommend four important

conservation measures: First, protection from fire, which is the greatest threat to habitat in my

study site; Second, increased education regarding hunting, as all of these species are

preferentially hunted (García & Goodman, 2003); Third, re-establishment of corridors and

increasing the size of habitat by planting; Finally, considering possible reintroduction of these

species when large tracts of forest have been re-grown, protected, and connected to other forest

through corridors. However, further study is needed to determine the efficacy of such measures.

5.6 Significance Understanding how primates and specifically lemurs respond to habitat loss and fragmentation

is of critical importance for primate conservation biogeography. Madagascar is facing one of the

greatest threats to biodiversity due to habitat loss, and lemurs are on the brink of losing

important remaining habitat. The goal of my study was to determine how lemur species, at the

community level and individually, respond to habitat loss and fragmentation at both the patch-

115

and landscape-level in a fragmented landscape. My study continues to show the overwhelming

influence of area on primate species richness and occurrence at a patch-level but not necessarily

at a landscape-level. Therefore, it is important to consider primate biogeography at both a patch-

and landscape-level when developing a conservation action plan. A patch-level is important

because patches or habitat fragments are common units of management for conservation. Land

managers need to know what size of patches/fragments to protect in order to retain the

maximum number of species, or to protect particular species of interest. However, a landscape

approach, in addition to a patch-level approach, will yield important suggestions on how to

protect primate species. What we learn to protect lemur species can be applied to other primates

and arboreal mammals in other parts of the world where habitat loss and fragmentation are

rampant.

116

References Alonso, L.E., Schulenburg, T.S., Radiobe, S. & Missa, O. (2002). A biological assessment of the

Reserve Naturelle Integrale d’Ankarafantsika, Madagascar. RAP Bulletin of Biological Assessment No. 23. Washington, D.C.: Conservation International. Retrieved from https://library.conservation.org/Published%20Documents/2009/RAP%2023_Madagascar.pdf#page=73

Anzures-Dadda, A., & Manson, R. H. (2007). Patch- and landscape-scale effects on howler monkey distribution and abundance in rainforest fragments. Animal Conservation, 10(1), 69–76.

Arrhenius, O. (1921). Species and Area. Journal of Ecology, 9(1): 95-99. Arroyo-Rodríguez, V., Mandujano, S., & Benítez-Malvido, J. (2008). Landscape attributes

affecting patch occupancy by howler monkeys (Alouatta palliata mexicana) at Los Tuxtlas, Mexico. American Journal of Primatology, 70(1), 69–77

Arroyo-Rodríguez, V., & Dias, P. A. (2010). Effects of habitat fragmentation and disturbance on howler monkeys: a review. American Journal of Primatology, 72(1), 1–16.

Arroyo-Rodríguez, V., González-Perez, I. M., Garmendia, A., Solà, M., & Estrada, A. (2013a). The relative impact of forest patch and landscape attributes on black howler monkey populations in the fragmented Lacandona rainforest, Mexico. Landscape ecology, 28(9), 1717-1727.

Arroyo-Rodríguez, V., Moral, E. C. D., Mandujano, S., Chapman, C. A., Reyna-Hurtado, R., & Fahrig, L. (2013b). Assessing habitat fragmentation effects on primates: The importance of evaluating questions at the correct scale. In L. K. Marsh, & C.A. Chapman (Eds.), Primates in Fragments. (pp. 13–28). New York: Springer.

Arroyo-Rodríguez, V., & Fahrig, L. (2014). Why is a landscape perspective important in studies of primates? American Journal of Primatology, 76(10), 901-909.

Baguette, M. (2004). The classical metapopulation theory and the real, natural world: a critical appraisal. Basic and Applied Ecology. 5(3), 213-224.

Bannar-Martin, K. H. (2014). Primate and nonprimate mammal community assembly: The influence of biogeographic barriers and spatial scale. International Journal of Primatology, 35(6), 1122-1142.

Beaudrot, L. H., & Marshall, A. J. (2011). Primate communities are structured more by dispersal limitation than by niches. Journal of Animal Ecology, 80(2), 332-341.

Bellier, E., Monestiez, P., Durbec, J. P., & Candau, J. N. (2007). Identifying spatial relationships at multiple scales: principal coordinates of neighbour matrices (PCNM) and geostatistical approaches. Ecography, 30(3), 385-399.

Benchimol, M., & Venticinque, E. M. (2014). Responses of primates to landscape change in Amazonian land-bridge islands—A multi-scale analysis. Biotropica, 46(4), 470-478.

Bivand, R. S., Hauke, J., & Kossowski, T. (2013). Computing the Jacobian in Gaussian spatial autoregressive models: An illustrated comparison of available methods. Geographical Analysis, 45(2), 150-179.

117

Bivand, R.S., Piras, G. (2015). Comparing implementations of estimation methods for spatial econometrics. Journal of Statistical Software, 63(18), 1-36. Retrieved from http://www.jstatsoft.org/v63/i18/.

Bloesch, U. (1999). Fire as a tool in the management of a savanna/dry forest reserve in Madagascar. Applied Vegetation Science, 2(1), 117-124.

Borcard, D., & Legendre, P. (2002). All-scale spatial analysis of ecological data by means of principal coordinates of neighbour matrices. Ecological Modelling, 153(1), 51-68.

Borcard, D., Legendre, P., Avois-Jacquet, C., & Tuomisto, H. (2004). Dissecting the spatial structure of ecological data at multiple scales. Ecology, 85(7), 1826-1832.

Bowman, J., Jaeger, J. A. G., & Fahrig, L. (2002). Dispersal distance of mammals is proportional to home range size. Ecology, 83(7), 2049-2055

Boyle, S. A., & Smith, A. T. (2010). Can landscape and species characteristics predict primate presence in forest fragments in the Brazilian Amazon? Biological Conservation, 143(5), 1134-1143.

Boyle, S. A., Lenz, B. B., Gilbert, K. A., Sprionello, W. R., Gómez, M. S., Setz, E. Z., Marajó dos Reis, A., Ferreira da Silva, O., Keuroghlian, A., & Pinto, F. (2013). Primates of the biological dynamics of forest fragments project: A history. In L. K. Marsh, & C.A. Chapman (Eds.), Primates in Fragments (pp. 57-74). Springer, New York: Springer.

Bradter, U., Kunin, W. E., Altringham, J. D., Thom, T. J., & Benton, T. G. (2013). Identifying appropriate spatial scales of predictors in species distribution models with the random forest algorithm. Methods in Ecology and Evolution, 4(2), 167-174.

Braune, P., Schmidt, S., & Zimmerman, E. (2005) Spacing and group coordination in a nocturnal primate, the golden brown mouse lemur (Microcebus ravelobensis): the role of olfactory and acoustic signals. Behavioral Ecology and Sociobiology, 58(6), 587-596.

Brooks, T.M., Mittermeier, R.A., Mittermeier, C.G., Da Fonseca, G.A., Rylands, A.B., Konstant, W.R., Flick, P., Pilgrim, J., Oldfield, S., Magin, G. and Hilton-Taylor, C., (2002). Habitat loss and extinction in the hotspots of biodiversity. Conservation biology, 16(4), 909-923.

Burke, R. J., & Lehman, S. M. (2015). Edge effects on morphometrics and body mass in two sympatric species of mouse lemurs in Madagascar. Folia Primatologica, 85(5), 277-291.

Burnham, K. P., & Anderson, D. R. (2002). Model selection and multi-model inference: a practical information-theoretic approach. New York: Springer-Verlag.

Burnham, K. P., Anderson, D. R., & Huyvaert, K. P. (2011). AIC model selection and multimodel inference in behavioral ecology: Some background, observations, and comparisons. Behavioral Ecology and Sociobiology, 65(1), 23–35.

Butchart, S. H., Walpole, M., Collen, B., Van Strien, A., Scharlemann, J. P., Almond, R. E., Baillie, J.E., Bomhard, B., Brown, C., Bruno, J. & Carpenter, K. E. (2010). Global biodiversity: indicators of recent declines. Science, 328(5982), 1164-1168.

Cameron, A. C., & Trivedi, P. K. (2001). Essentials of count data regression. In B. H. Baltagi (Ed.) A companion to theoretical econometrics (pp. 331-348). Oxford: Blackwell Publishing Ltd.

118

Cant, J. G. H. (1986). Locomotion and feeding postures of spider and howling monkeys: field study and evolutionary interpretation. Folia Primatologica, 46(1), 1-14.

Cardillo, M., Mace, G. M., Gittleman, J. L., Jones, K. E., Bielby, J., & Purvis, A. (2008). The predictability of extinction: biological and external correlates of decline in mammals. Proceedings of the Royal Society of London B: Biological Sciences, 275(1641), 1441-1448.

Ceballos, G., Ehrlich, P. R., Barnosky, A. D., García, A., Pringle, R. M., & Palmer, T. M. (2015). Accelerated modern human–induced species losses: Entering the sixth mass extinction. Science Advances, 1(e1400253), 1-5.

Chapman, C. A., Chapman, L. J., & Wrangham, R. W. (1995). Ecological constraints on group size: an analysis of spider monkey and chimpanzee subgroups. Behavioral Ecology and Sociobiology, 36(1), 59-70.

Chapman, C. A., M. J. Lawes, L. Naughton-Treves, & Gillespie, T. R. (2003). Primate survival in community-owned forest fragments: are metapopulation models useful amidst intensive use? In L.K. Marsh (Ed.), Primates in fragments: Ecology and conservation (pp. 63–78). New York: Kluwer Academic/Plenum Publishers.

Chapman, C. A., Naughton-Treves, L., Lawes, M. J., Wasserman, M. D., & Gillespie, T. R. (2007). Population declines of colobus in western Uganda and conservation value of forest fragments. International Journal of Primatology, 28(3), 513-528.

Chapman, C. A., & Peres, C. A. (2001). Primate conservation in the new millennium: The role of scientists. Evolutionary Anthropology, 10(1), 16-33.

Clutton-Brock, T. H., & Harvey, P. H. (1977). Primate ecology and social organization. Journal of Zoology, 183(1), 1-39.

Coleman, B. D. (1981). On random placement and species-area relations. Mathematical Biosciences, 54(3-4), 191-215.

Connor, E. & McCoy, E. (1979). The statistics and biology of the species-area relationship. The American Naturalist, 113(6), 791-833.

Corbin, G. D., & Schmid, J. (1995). Insect secretions determine habitat use patterns by a female lesser mouse lemur (Microcebus murinus). American Journal of Primatology, 37(4), 317-324.

Cowlishaw, G. (1999). Predicting the Pattern of Decline of African Primate Diversity: An Extinction Debt from Historical Deforestation. Conservation Biology, 13(5), 1183-1193.

Cowlishaw, G., & Dunbar, R. I. (2000). Primate conservation biology. Chicago: University of Chicago Press.

Craule, M., Chikhi, L., Sousa, V., Olivieri, G. L., Rabesandratana, A., Zimmerman, E., & Radespiel, U. (2009). Influence of forest fragmentation on an endangered large-bodied lemur in northwestern Madagascar. Biological Conservation, 142(12), 2862-2871.

Cristobal-Azkarate, J., Vea, J. J., Asensio, N., & Rodríguez-Luna, E. (2005). Biogeographical and floristic predictors of the presence and abundance of mantled howlers (Alouatta palliata mexicana) in rainforest fragments at Los Tuxtlas, Mexico. American Journal of Primatology, 67(2), 209–222.

119

Crowley, B. E., Blanco, M. B., Arrigo-Nelson, S. J., & Irwin, M. T. (2013). Stable isotopes document resource partitioning and effects of forest disturbance on sympatric cheirogaleid lemurs. Naturwissenschaften, 100(10), 943-956.

Crowley, B., McGoogan, K. C., & Lehman, S. M. (2012). Edge effects on the foliar stable isotope values in a Madagascan tropical dry forest. Plos One, 7(9), e44538.

Crowley, B., Thorén, S., Rasoazanabary, E., Vogel, E. R., Meredith, B. A., Zohdy, S., Blanco M. B., McGoogan, K. C., Arrigo-Nelson, S., Irwin, M. I., & Wright, P. C., Godfrey, L. R., Koch, P. L., & Dominy, N. J. (2011). Explaining geographic variation in the isotope composition of mouse lemurs (Microcebus). Journal of Biogeography, 38(11), 1-16.

Curits, D. J., Zaramody, A., & Martin, R. D. (1999). Cathemerality in the mongoose lemur, Eulemur mongoz. American Journal of Primatology, 47(4), 279-298.

Dausmann, K. H., Glos, J., Ganzhorn, J. U., & Heldmaier, G. (2005). Hibernation in the tropics: lessons from a primate. Journal of Comparative Physiology B: Biochemical, Systemic, and Environmental Physiology, 175(3), 147-155.

De Knegt, H.J., van Langevelde, F.V., Coughenour, M.B., Skidmore, A.K., De Boer, W.F., Heitkönig, I.M.A., Knox, N.M., Slotow, R., Van der Waal, C. & Prins, H.H.T. (2010). Spatial autocorrelation and the scaling of species-environment relationships. Ecology, 91(8), 2455-2465.

De Knegt, H.J., van Langevelde, F., Skidmore, A.K., Delsink, A., Slotow, R., Henley, S., Bucini, G., De Boer, W.F., Coughenour, M.B., Grant, C.C. & Heitkönig, I. (2011). The spatial scaling of habitat selection by African elephants. Journal of Animal Ecology, 80(1), 270-281.

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(1), 19-26.

Di Fiore, A. & Campbell, C. J. (2007). The Atelines: variation in ecology, behavior, and social organization. In C. J. Campbell, A. Fuentes, K. C. MacKinnon, M. Panger & S. K. Bearder (Eds.), Primates in Perspective, (pp. 155–185). New York: Oxford University Press.

Diamond, J. M. (1975). The island dilemma: lessons of modern biogeographic studies for the design of natural reserves. Biological conservation, 7(2), 129-146.

Dray, S., Legendre, P., & Peres-Neto, P. R. (2006). Spatial modelling: a comprehensive framework for principal coordinate analysis of neighbour matrices (PCNM). Ecological Modelling, 196(3), 483-493.

Eigenbrod, F., Hecnar, S. J., & Fahrig, L. (2011). Sub-optimal study design has major impacts on landscape-scale inference. Biological Conservation, 144(1), 298-305.

Estrada, A., Garber, P. A., Pavelka, M., & Luecke, L. G. (2006). Overview of the Mesoamerican primate fauna, primate studies and conservation concerns: An introduction. In A. Estrada, P. A. Garber, M. S. M. Pavelka, & L. Luecke (Eds.), New perspectives in the study of Mesoamerican primates: Distribution, ecology, behavior and conservation, (pp. 1–22), Berlin: Springer.

120

Estrada, A., Luecke, L., Van Belle, S., Barrueta, E., & Meda, M. R. (2004). Survey of black howler (Alouatta pigra) and spider (Ateles geoffroyi) monkeys in the Mayan sites of Calakmul and Yaxchilán, Mexico and Tikal, Guatemala. Primates, 45(1), 33-39.

Fahrig, L. (2001). How much habitat is enough? Biological conservation, 100(1), 65-74. Fahrig, L. (2003). Effects of habitat fragmentation on biodiversity. Annual Review of Ecology,

Evolution, and Systematics, 34, 487–515. Fahrig, L. (2005). When is a landscape perspective important. Issues and perspectives in

landscape ecology. Retrieved from http://www.glel.carleton.ca/PEOPLE/PROFILES/lenoreFahrig/When%20is%20a%20Landscape%20Perspective%20Important.pdf

Farris, Z., Karpanty, S., Ratelolahy, F., & Kelly, M. (2014). Predator–primate distribution, activity, and co-occurrence in relation to habitat and human activity across fragmented and contiguous forests in Northeastern Madagascar. International Journal of Primatology, 5(35), 859-880.

Fietz, J. (1999). Monogamy as a rule rather than exception in nocturnal lemurs: The case of the fat-tailed dwarf lemur, Cheirogaleus medius. Ethology, 105(3), 255-272.

Fietz, J., & Ganzhorn, J. U. (1999). Feeding ecology of the hibernating primate Cheirogaleus medius: how does it get so fat? Oecologia, 121(2), 157-164.

Ford, S. M., & Davis, L. C. (1992). Systematics and body size: implications for feeding adaptations in New World monkeys. American Journal of Physical Anthropology, 88(4), 415-468.

Fox, B. J. & Fox, M. D. (2000). Factors determining mammal species richness on habitat islands and isolates: habitat diversity, disturbance, species interactions and guild assembly rules. Global Ecology and Biogeography, 9(1) ,19-37.

Frey, J. K., Bogan, M. A., & Yates, T. L. (2007). Mountaintop island age determines species richness of boreal mammals in the American Southwest. Ecography, 30(2), 231-240.

Gade, D. W. (1996). Deforestation and Its Effects in Highland Madagascar. Mountain Research and Development, 16(2), 101-116.

Ganzhorn, J. U. (1989). Niche separation of seven lemur species in the eastern rainforest of Madagascar. Oecologia, 79(2), 279-286.

Ganzhorn, J. U. (1997). Test of Fox's assembly rule for functional groups in lemur communities in Madagascar. Journal of Zoology, 241(3), 533-542.

Ganzhorn, J.U., Malcomber, S., Andrinantoanina, O., & Goodman, S.M., 1997. Habitat characteristics and lemur species richness in Madagascar. Biotropica 29(3), 331– 343.

Ganzhorn, J. U. (2003). Effects of introduced Rattus rattus on endemic small mammals in dry deciduous forest fragments of western Madagascar. Animal Conservation, 6(2), 147–157.

Ganzhorn, J. U., & Kappeler, P. M. (1996). Lemurs of the Kirindy forest. Primate report, 46(1), 257-274.

Ganzhorn, J. U., & Schmid, J. (1998). Different population dynamics of Microcebus murinus in primary and secondary deciduous dry forests of Madagascar. International Journal of Primatology, 19(5), 785–796.

121

Ganzhorn, J. U., Fietz, J., Rakotovao, E., Schwab, D., & Zinner, D. (1999). Lemurs and the regeneration of dry deciduous forest in Madagascar. Conservation Biology, 13(4), 794-804.

Garbutt, N. (2007). Mammals of Madagascar: A complete guide. Yale University Press. García, G. & Goodman S. M. (2003). Hunting of protected animals in the Parc National

d'Ankarafantsika, northwestern Madagascar. Oryx, 37(1), 115-118. Gillespie, T. & Chapman, C.A. (2006). Prediction of parasite infection dynamics in primate

metapopulations based on attributes of forest fragmentation. Conservation Biology, 20(2), 441-448.

Gilroy, J. J., Medina Uribe, C. A., Haugaasen, T., & Edwards, D. P. (2015). Effect of scale on trait predictors of species responses to agriculture. Conservation Biology, 29(2), 463-472.

Gleason, H. A. (1922). On the relation between species and area. Ecology, 3(2), 158-162.

Goodman, S. M., & Benstead, J. P. (2005). Updated estimates of biotic diversity and endemism for Madagascar. Oryx, 39(01), 73-77.

Gray, T. N. E., Phan, C., & Long, B. (2010). Modelling species distribution at multiple spatial scales: gibbon habitat preferences in a fragmented landscape. Animal Conservation, 13(3), 324-332.

Griffith, D. A., & Peres-Neto, P. R. (2006). Spatial modeling in ecology: the flexibility of eigenfunction spatial analyses. Ecology, 87(10), 2603-2613.

Groom, M. J., Meffe, G. K., & Carroll, C. R. (2006). Principles of conservation biology. Sinauer Associates Inc. Sutherland.

Gurd, D. B., Nudds, T. D., & Rivard, D. H. (2001). Conservation of mammals in eastern North American wildlife reserves: how small is too small? Conservation Biology, 15(5), 1355-1363.

Hanski, I. (1994a). Patch-occupancy dynamics in fragmented landscapes. Trends in Ecology & Evolution, 9(4), 131–135.

Hanski, I. (1994b). A practical model of metapopulation dynamics. Journal of Animal Ecology, 63(1), 151–162.

Hanski, I. (1994c). Spatial scale, patchiness and population dynamics on land. Philosophical Transactions of the Royal Society of Biological Sciences. 343(1303): 19-25

Hanski, I. (1997). Metapopulation Dynamics: From concepts and observations to predictive models. In I. Hanski, & M. E. Gilpin (Eds.), Metapopulation biology: Ecology, genetics, and evolution, (pp. 69-91). Academic Press.

Hanski, I. (1998). Metapopulation dynamics. Nature, 396(6706), 41-49

Hanski, I. (1999). Metapopulation ecology. Oxford University Press. Hanski, I. (2005). The shrinking world: ecological consequences of habitat loss (Vol. 14).

Oldendorf/Luhe: International Ecology Institute. Hanski, I. (2010). The theories of island biogeography and metapopulation dynamics: Science

marches forward, but the legacy of good ideas lasts a long time. In J. B. Losos, R. E. &

122

Ricklefs (Eds.), The theory of island biogeography revisited, (pp. 186-213). Princeton: Princeton University Press.

Hanski, I., & Gilpin, M. E. (1991). Metapopulation dynamics: brief history and conceptual domain. Biological Journal of the Linnean Society. 42(1-2), 3-16

Hanski, I., & Ovaskainen, O. (2003). Metapopulation theory for fragmented landscapes. Theoretical Population Biology, 64(1), 119–127.

Hanya, G., & Aiba, S. I. (2010). Fruit fall in tropical and temperate forests: Implications for frugivore diversity. Ecological research, 25(6), 1081-1090.

Harcourt, A. H., & Doherty, D. A. (2005). Species-area relationships of primates in tropical forest fragments: a global analysis. Journal of Applied Ecology, 42(4), 630–637.

Harper, G. J., Steininger, M. K., Tucker, C. J., Juhn, D., & Hawkins, F. (2007). Fifty years of deforestation and forest fragmentation in Madagascar. Environmental Conservation, 34(04), 1–9.

Harrison, S. (1991). Local extinction in a metapopulation context: an empirical evaluation. Biological Journal of the Linnean Society. 42(2), 73-88.

Harrison, S., & Taylor, A. D. (1997). Empirical evidence for metapopulation dynamics. In I. Hanski, & M. E. Gilpin (Eds.). Metapopulation biology: Ecology, genetics, and evolution, (pp. 27-42). Academic Press.

Holland, J. D., Bert, D. G., & Fahrig, L. (2004). Determining the spatial scale of species' response to habitat. BioScience, 54(3), 227-233.

Holland, J. D., Fahrig, L., & Cappuccino, N. (2005). Body size affects the spatial scale of habitat–beetle interactions. Oikos, 110(1), 101-108.

Hutchinson, G. E. (1965). The ecological theater and the evolutionary play. Yale University Press.

Irwin, M. T., Raharison, J. L., & Wright, P. C. (2009). Spatial and temporal variability in predation on rainforest primates: do forest fragmentation and predation act synergistically? Animal Conservation, 12(3), 220-230.

Isbell, L. A., & Van Vuren, D. (1996). Differential costs of locational and social dispersal and their consequences for female group-living primates. Behaviour, 133(1), 1-36.

IUCN 2016, IUCN Red List of Threatened Species. Version 2015.4. In IUCN Red List of Threatened Species. Retrieved 1 March, 2016 from www.iucnredlist.org.

Jackson, H. B., & Fahrig, L. (2012). What size is a biologically relevant landscape? Landscape Ecology, 27(7), 929-941.

Jackson, H. B., & Fahrig, L. (2014). Are ecologists conducting research at the optimal scale? Global Ecology and Biogeography, 24(1), 52-63.

Jaquiéry, J., Guélat, J., Broquet, T., Berset-Brändii, L., Pellegrini, E., Moresi, R., Hirzel, A. H., & Perrin, N. (2008). Habitat quality effects on metapopulation dynamics in greater white-toothed shrews, Crocidura russula. Ecology, 89(10), 2777-2785

Jones, J. H. (2011). Primates and the evolution of long, slow life histories. Current Biology, 21(18), R708-R717.

123

Kamilar, J. M. (2009). Environmental and geographic correlates of the taxonomic structure of primate communities. American Journal of Physical Anthropology, 139(3), 382-393.

Kenyon, M. (2007). The ecology of the golden-cheeked gibbon (Nomascus gabriellae) in Cat Tien National Park, Vietnam. (Doctoral dissertation). Cambridge, UK: University of Cambridge.

Kun-Rodrigues, C., Salmona, J., Besolo, A., Rasolondraibe, E. Ravarivola, C., Marques, T. A., & Chikhi, L. (2014). New density estimates of a threatened sifaka species (Propithecus coquereli) in Ankarafantsika National Park. American Journal of Primatology, 76(6), 515-528.

Kundu, D. (1993). Asymptotic theory of least squares estimator of a particular nonlinear regression model. Statistics & probability letters, 18(1), 13-17.

Lahann, P. (2007). Feeding ecology and seed dispersal of sympatric cheirogaleid lemurs (Microcebus murinus, Cheirogaleus medius, Cheirogaleus major) in the littoral rainforest of south-east Madagascar. Journal of Zoology, 271(1), 88–98.

Lahann, P., & Dausmann, K. H. (2011). Live fast, die young: flexibility of life history traits in the fat-tailed dwarf lemur (Cheirogaleus medius). Behavioral ecology and sociobiology, 65(2), 381-390.

Lambeck, R. J. (1997). Focal species: A multi-species umbrella for nature conservation. Conservation biology, 11(4), 849-856.

Lawes, M. J., Mealin, P. E., & Piper, S. E. (2000). Patch occupancy and potential metapopulation dynamics of three forest mammals in fragmented Afromontane forest in South Africa. Conservation Biology, 14(4), 1088-1098.

Lehman, S. M. (2004). Distribution and diversity of primates in Guyana: Species-area relationships and riverine barriers. International journal of Primatology, 25(1), 73-95.

Lehman, S. M., Mayor, M., & Wright, P. C. (2005). Ecogeographic size variations in sifakas: A test of the resource seasonality and resource quality hypotheses. American journal of physical anthropology, 126(3), 318-328.

Lehman, S.M.L, Rajaonson, A. & Day, S. (2006a). Edge effects and their influence on lemur density and distribution in Southeast Madagascar. American Journal of Physical Anthropology, 129(2), 232-241.

Lehman, S.M.L, Rajaonson, A. & Day, S. (2006b). Lemur responses to edge effects in the Vohibola III classified forest, Madagascar. American Journal of Primatology, 68(3), 292-299.

Lehman, S. M. (2014). Effects of altitude on the conservation biogeography of lemurs in southeast Madagascar. In N. B. Grow, S. Gursky-Doyen, & A. Krzton (Eds.). High Altitude Primates (pp. 3-22). Springer New York.

Levins, R. (1969), Some demographic and genetic consequences of environmental heterogeneity for biological control, Bulletin of the Entomological Society of America, 15, 237–240

Levins, R. 1970. Extinction: Some mathematical problems in biology. In M. Gesternhaber (Ed.) American Mathematical Society (pp. 77–107), Providence.

Lomolino, M. (2000). Ecology's most general, yet protean pattern: The species-area relationship. Journal of Biogeography, 27(1), 17-26.

124

Lomolino, M. V. (1982). Species-area and species-distance relationships of terrestrial mammals in the Thousand Island Region. Oecologia, 54(1), 72-75.

Mac Nally, R. (1996). Hierarchical partitioning as an interpretative tool in multivariate inference. Australian Journal of Ecology, 21(2), 224-228.

Mac Nally, R. (2000). Regression and model-building in conservation biology, biogeography and ecology: the distinction between–and reconciliation of–‘predictive’ and ‘explanatory’ models. Biodiversity & Conservation, 9(5), 655-671.

MacArthur, R., & Wilson, E. (1967). The theory of biogeography, Princeton: Princeton University Press.

Mandujano, S., & Escobedo-Morales, L. (2008). Population viability analysis of howler monkeys (Alouatta palliata mexicana) in a highly fragmented landscape in Los Tuxtlas, Mexico. Tropical Conservation Science, 1(1), 43–62.

Manly, B. F. J. (2004). Multivariate statistical methods: A primer, Chapman and Hall/CRC.

Marsh, L. K. (2003). The Nature of Fragmentation. In L. K. Marsh (Ed.), Primates in fragments: Ecology and conservation (pp1-10). New York: Kluwer Academic/Plenum Publishers.

Marshall, A. R., Jørgensbye, H. I. O., Rovero, F., Platts, P. J., White, P. C. L., & Lovett, J. C. (2010). The species-area relationship and confounding variables in a threatened monkey community. American Journal of Primatology, 72(4), 325–336.

Martin, T. (1981). Species-area slopes and coefficients: A caution on their interpretation. The American Naturalist, 118(6), 823-837.

Matthysen, E. (2012). Multicausality of dispersal: a review. In J. Clobert, M. Baguette, T. G. Benton, & J. M. Bullock, (Eds.), Dispersal ecology and evolution, (pp. 3-18). Oxford: Oxford University Press.

McGarigal, K., & Cushman, S. A. (2002). Comparative evaluation of experimental approaches to the study of habitat fragmentation effects. Ecological Applications, 12(2), 335-345.

McGarigal, K., & McComb, W. C. (1995). Relationships between landscape structure and breeding birds in the Oregon Coast Range. Ecological Monographs, 65(3), 235-260.

McGoogan, K. C., (2011). Edge effect on the behaviour and ecology of Propithecus coquereli in Northwest Madagascar. (Doctoral dissertation). Retrieved from T-Space, http://hdl.handle.net/1807/31861

McGuinness, K. A. (1984). Species–area curves. Biological Reviews, 59(3), 423-440.

Milton, K. & May, M. L. (1976). Body weight, diet and home range area in primates. Nature, 259(5543), 459-462.

Mittermeier, R. A., Louis, E. E., Richardson, M., Schwitzer, C., Langrand, O., Rylands, A. B., Hawkins, F., Rajaobelina, S., Ratsimbazafy, J., Rasoloarison, R., Roos, C., Kappeler, P.M., MacKinnon, J. (2010). Lemurs of Madagascar. Washington: Conservation International.

Moilanen, A., & Hanski, I. (1998). Metapopulation dynamics: effects of habitat quality and landscape structure. Ecology, 79(7), 2503-2515

125

Müller, P., Velo, A., Raheliarisoa, E. O., Zaramody, A., & Curtis, D. J. (2000). Surveys of sympatric lemurs at Anjamena, north-west Madagascar. African Journal of Ecology, 38(3), 248-257.

Murcia, C. (1995). Edge effects in fragmented forests: Implications for conservation. Trends in ecology & evolution, 10(2), 58-62.

Nilsson, S. G., Bengtsson, J., & As, S. (1988). Habitat diversity or area per se? Species richness of woody plants, carabid beetles and land snails on islands. The Journal of Animal Ecology, 57I2), 685-704.

Norscia, I. (2008). Pilot survey of Avahi population (woolly lemurs) in littoral forest fragments of southeast Madagascar. Primates, 49(1), 85-88.

O’Hara, R. B., & Kotze, D. J. (2010). Do not log-transform count data. Methods in Ecology and Evolution, 1(2), 118-122.

Oksanen, J. (2004). Incidence Function Model in R. 1-21. Retrieved June 1st 2013 from http://cc.oulu.fi/~jarioksa/opetus/openmeta/metafit.pdf

Olivieri, G. L., Sousa, V., Chikhi, L., & Radespiel, U. (2008). From genetic diversity and structure to conservation: genetic signature of recent population declines in three mouse lemur species (Microcebus spp.). Biological Conservation, 141(5), 1257-1271.

Ordóñez-Gómez, J. D., Arroyo-Rodríguez, V., Nicasio-Arzeta, S., & Cristóbal-Azkarate, J. (2014). Which is the appropriate scale to assess the impact of landscape spatial configuration on the diet and behavior of spider monkeys? American journal of primatology, 77(1), 56-65.

Patterson, J. E., & Malcolm, J. R. (2010). Landscape structure and local habitat characteristics as correlates of Glaucomys sabrinus and Tamiasciurus hudsonicus occurrence. Journal of Mammalogy, 91(3), 642-653.

Pimm, S. L., Russell, G. J., Gittleman, J. L., & Brooks, T. M. (1995). The future of biodiversity. Science, New Series, 269(5222), 347-350.

Preston, F. (1962). The canonical distribution of commonness and rarity: Part I. Ecology, 43(2), 185-215.

Pyritz, L. W., Büntge, A. B., Herzog, S. K., & Kessler, M. (2010). Effects of habitat structure and fragmentation on diversity and abundance of primates in tropical deciduous forests in Bolivia. International Journal of Primatology, 31(5), 796-812.

Quinn, G. P., & Keough, M. J. (2002). Experimental design and data analysis for biologists. Cambridge University Press.

R Core Team (2015). R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria. URL https://ArcGIS.R-project.org/.

Raboy, B. E., Neves, L. G., Zeigler, S., Saraiva, N. A., Cardoso, N., dos Santos, G. R., Ballou, J. D., & Leimgruber, P. (2010). Strength of habitat and landscape metrics in predicting golden-headed lion tamarin presence or absence in forest patches in southern Bahia, Brazil. Biotropica, 42(3), 388-397.

Radespiel, U. (2000). Sociality in the gray mouse lemur (Microcebus murinus) in northwestern Madagascar. American Journal of Primatology, 51(1), 21-40.

126

Radespiel, U., Ehresmann, P., & Zimmermann, E. (2003). Species-specific usage of sleeping sites in two sympatric mouse lemur species (Microcebus murinus and M. ravelobensis) in northwestern Madagascar. American Journal of Primatology, 59(4), 139-151.

Radespiel, U., Jurić, M., & Zimmermann, E. (2009). Sociogenetic structures, dispersal and the risk of inbreeding in a small nocturnal lemur, the golden–brown mouse lemur (Microcebus ravelobensis). Behaviour, 146(4), 607-628.

Radespiel, U., Lutermann, H., Schmelting, B., Bruford, M. W., & Zimmermann, E. (2003). Patterns and dynamics of sex-biased dispersal in a nocturnal primate, the grey mouse lemur, Microcebus murinus. Animal Behaviour, 65(4), 709–719.

Radespiel, U., Rakotondravony, R., & Chikhi, L. (2008). Natural and anthropogenic determinants of genetic structure in the largest remaining population of the endangered golden-brown mouse lemur, Microcebus ravelobensis. American Journal of Primatology, 70(9), 860-870.

Radespiel, U., Reimann, W., Rahelinirina, M., & Zimmermann, E. (2006). Feeding ecology of sympatric mouse lemur species in Northwestern Madagascar. International Journal of Primatology, 27(1), 311–321.

Radespiel, U., Sarikaya, Z., Zimmermann, E., & Bruford, M. W. (2001). Sociogenetic structure in a free-living nocturnal primate population: sex-specific differences in the grey mouse lemur (Microcebus murinus). Behavioral Ecology and Sociobiology, 50(6), 493-502.

Rakotondranary, S. J., & Ganzhorn, J. U. (2012). Habitat separation of sympatric Microcebus spp. in the dry spiny forest of south-eastern Madagascar. Folia Primatologica, 82(4-5), 212-223.

Rakotondravony, R., & Radespiel, U. (2009). Varying patterns of coexistence of two mouse lemur species (Microcebus ravelobensis and M. murinus) in a heterogeneous landscape. American Journal of Primatology, 71(11), 928–938.

Randrianambinina, B., Rakotondravony, D., Radespiel, U., & Zimmermann, E. (2003). Seasonal changes in general activity, body mass and reproduction of two small nocturnal primates: a comparison of the golden brown mouse lemur (Microcebus ravelobensis) in Northwestern Madagascar and the brown mouse lemur (Microcebus rufus) in Eastern Madagascar. Primates, 44(4), 321-331.

Rasmussen, M. A. (1999). Ecological influences on activity cycle in two cathemeral primates, Eulemur mongoz (mongoose lemur) and Eulemur fulvus fulvus (common brown lemur). Unpublished PhD thesis, Duke University, Durham, NC.

Rasolooarison, R. M., Goodman, S. M., & Ganzhorn, J. U. (2000). Taxonomic revision of mouse lemurs (Microcebus) in the western portions of Madagascar. International Journal of Primatology, 21(6), 963-1019.

Ratkowsky, D. A., & Giles, D. E. (1990). Handbook of nonlinear regression models. New York: M. Dekker.

Rawson, B. M., Clements, T., & Hor, N. M. (2009). Status and conservation of yellow-cheeked crested gibbons (Nomascus gabriellae) in the Seima Biodiversity Conservation Area, Mondulkiri Province, Cambodia. In S. Lappan, & D. Whittaker (Eds.), The gibbons: New perspectives on small ape socioecology and population biology (pp. 387-408). New York: Springer.

127

Rawson, B., Traeholt, C. (2011) Nomascus, gabriellae. In All the World's Primates. N. Rowe & M. Myers, (Eds.), Retrieved from http://alltheworldsprimates.org/Home.aspx

Razafy Fara L. (2003). Rapport sur l’actualisation de la carte de vegetation du Parc National d` Ankarafantsika. ANGAP, GFA. 26p

Reed, K. E. & Fleagle, J. G. (1995). Geographic and climatic control of primate diversity. Proceedings of the National Academy of Sciences of the United States of America, 92(17), 7874-7876.

Rendigs, A., Radespiel, U., Wrogemann, D., & Zimmermann, E. (2003). Relationship between microhabitat structure and distribution of mouse lemurs (Microcebus spp.) in northwestern Madagascar. International Journal of Primatology, 24(1), 47-64.

Richard, A. F. (1985). Primates in nature (Vol. 4). New York: WH Freeman.

Ricketts, T. H. (2001). The matrix matters: effective isolation in fragmented landscapes. The American Naturalist, 158(1), 87-99.

Ricklefs, R. E., & Lovette, I. J. (2001). The roles of island area per se and habitat diversity in the species–area relationships of four Lesser Antillean faunal groups. Journal of Animal Ecology, 68(6), 1142-1160.

Rodriguez-Toledo, E. M., Mandujano, S., & García-Orduna, F. (2003). Relationships between forest fragments and howler monkeys (Alouatta palliata mexicana) in southern Veracruz, Mexico. In L. K. Marsh (Ed.) Primates in fragments: Ecology and conservation (pp. 79–97). New York: Kluwer Academic/Plenum Publishers.

Rosenzweig, M. L. (1995). Species diversity in space and time. Cambridge University Press.

Sales, L. P., Hayward, M. W., & Passamani, M. (2016). Local vs landscape drivers of primate occupancy in a Brazilian fragmented region. Mammal Research, 61(1), 73-82.

Sato, H. (2012). Frugivory and seed dispersal by brown lemurs in a Malagasy tropical dry forest. Biotropica, 44(4), 479-488.

Schad, J., Sommer, S., & Ganzhorn, J. U. (2004). MHC variability of a small lemur in the littoral forest fragments of southeastern Madagascar. Conservation Genetics, 5(3), 299-309.

Schäffler, L., & Kappeler, P. M. (2014). Distribution and abundance of three cheirogaleid species in Menabe Central, Western Madagascar. Lemur News, 18, 38-43.

Scheiner, S. M. (2003). Six types of species-area curves. Global Ecology and Biogeography, 12(6), 441-447.

Schliehe-Diecks, S., Eberle, M., & Kappeler, P. M. (2012). Walk the line-dispersal movements of gray mouse lemurs (Microcebus murinus). Behavioral Ecology and Sociobiology, 66(8), 1175–1185.

Schmid, J., & Rasoloarison, R. M. (2002). Lemurs of the Réserve Naturelle Intégrale d’Ankarafantsika, Madagascar. In L. Alonso, T. Schulenberg, S. Radilofe, & O. Missa (Eds.). A biological assessment of the Reserve Naturelle Integrale d’Ankarafantsika, Madagascar. RAP Bulletin of Biological Assessment No. 23. (pp. 73-82). Conservation International. Washington, DC. Retrieved from https://library.conservation.org/Published%20Documents/2009/RAP%2023_Madagascar.pdf#page=73

128

Schoener, T. W. (1976). The species-area relation within archipelagos: models and evidence from island land birds. In Proceedings of the XVI International Ornithological Congress (Vol. 6, pp. 629-642). Australian Academy of Science: Canberra, ACT.

Schwitzer, N., Randriatahina, G. H., Kaumanns, W., Hoffmeister, D., & Schwitzer, C. (2007). Habitat Utilization of Blue-Eyed Black Lemurs, Eulemur macaco flavifrons (Gray, 1867), in Primary and Altered Forest Fragments. Primate Conservation, 22(1), 79–87. doi:10.1896/052.022.0106

Schwitzer, C., Mittermeier, R. A., Davies, N., Johnson, S., Ratsimbazafy, J., Razafindramanana, J., Louis Jr., E. E. & Rajaobelina, S. (2013). Lemurs of Madagascar: A strategy for their conservation 2013–2016. IUCN SSC Primate Specialist Group, Bristol Conservation and Science Foundation, and Conservation International, Bristol, UK.

Schwitzer, C., Mittermeier, R. A., Johnson, S. E., Donati, G., Irwin, M., Peacock, H., et al. (2014). Conservation. Averting lemur extinctions amid Madagascar's political crisis. Science, 343(6173), 842–843

Shapiro, S. & Wilk, M. (1965). An analysis of variance test for normality (complete samples). Biometrika, 52(3-4), 591.

Shrum, M. (2008). A preliminary study of the lemur diversity in the Mahavavy-Kinkony Wetland Complex and Ankarafantsika National Park in Northwest Madagascar: with emphasis on the reassessment of the status of the mongoose lemur (Eulemur mongoz) and its habitat. Master’s Thesis. Oxford Brookes University, UK.

Steffens, T. & Lehman S.M. (2013). Lemur species-area relationship in a fragmented landscape: a test of the habitat heterogeneity hypothesis. Frontiers of Biogeography. 4(suppl. 1), 175.

Steffens, T. S. & Lehman, S. M. (2016). Factors determining Microcebus abundance in a fragmented landscape in Ankarafantsika National Park, Madagascar. In S. M. Lehman, U. Radespiel, & E. Zimmermann, (Eds.), The Dwarf and Mouse Lemurs of Madagascar: Biology, Behavior and Conservation Biogeography of the Cheirogaleidae (pp. 477-497). Cambridge: Cambridge University Press.

Sutherland, G. D., Harestad, A. S., Price, K., & Lertzman, K. P. (2000). Scaling of natal dispersal distances in terrestrial birds and mammals. Conservation Ecology. 4(1)16.

Swart, J., & Lawes, M. J. (1996). The effect of habitat patch connectivity on samango monkey (Cercopithecus mitis) metapopulation persistence. Ecological Modelling, 93(1), 57–74.

Tabachnick, B. G. & Fidell, L. S. (2007). Using multivariate statistics. Boston: Pearson.

Tews, J., Brose, U., Grimm, V., Tielbörger, K., Wichmann, M. C., Schwager, M., & Jeltsch, F. (2004). Animal species diversity driven by habitat heterogeneity/diversity: the importance of keystone structures. Journal of Biogeography, 31(1), 79-92.

Thorén, S., Quietzsch, F., & Radespiel, U. (2010). Leaf nest use and construction in the golden-brown mouse lemur (Microcebus ravelobensis) in the Ankarafantsika National Park. American Journal of Primatology, 72(1), 48-55.

Thorén, S., Quietzsch, F., Schwochow, D., Sehen, L., Meusel, C., Meares, K., & Radespiel, U. (2011). Seasonal changes in feeding ecology and activity patterns of two sympatric mouse lemur species, the gray mouse lemur (Microcebus murinus) and the golden-brown

129

mouse lemur (M. ravelobensis), in Northwestern Madagascar. International Journal of Primatology, 32(3), 566–586.

Thornton, D. H., & Fletcher, R. J. (2014). Body size and spatial scales in avian response to landscapes: a meta-analysis. Ecography, 37(5), 454-463.

Thornton, D. H., Branch, L. C., & Sunquist, M. E. (2011). The relative influence of habitat loss and fragmentation: Do tropical mammals meet the temperate paradigm? Ecological Applications, 21(6), 2324-2333.

Tjørve, E. (2003). Shapes and functions of species–area curves: a review of possible models. Journal of Biogeography, 30(6), 827-835.

Tjørve, E. (2009). Shapes and functions of species-area curves (II): a review of new models and parameterizations. Journal of Biogeography, 36(8), 1435-1445.

Tjørve, E., & Turner, W. R. (2009). The importance of samples and isolates for species–area relationships. Ecography, 32(3), 391-400.

Traeholt, C., Bunthoen, R., Rawson, B. M., Samuth, M., Vrak, C., & Vuthin, S. (2006). Status review of pileated gibbon, Hylobates pileatus and yellow-cheeked crested gibbon, Nomascus gabriellae in Cambodia. Fauna & Flora International. Retrieved from https://www.researchgate.net/profile/Benjamin_Rawson/publication/271210923_Status_review_of_Pileated_Gibbon_Hylobates_pileatus_and_yellow-cheeked_crested_gibbon_Nomascus_gabriellae_in_Cambodia/links/54c23b200cf219bbe4e64ef0.pdf

Triantis, K. A., Guilhaumon, F., & Whittaker, R. J. (2012). The island species–area relationship: biology and statistics. Journal of Biogeography, 39(2), 215-231.

Turner, M. G., Gardner, R. H., & O'neill, R. V. (2001). Landscape ecology in theory and practice (Vol. 401). New York: Springer.

Turner, M. G. (2005). Landscape ecology: What is the state of the science? Annual Review of Ecology, Evolution, and Systematics, 36(1), 319-344.

Ulrich, W., & Buszko, J. (2003). Self-similarity and the species–area relation of Polish butterflies. Basic and Applied Ecology, 4(3), 263-270.

Ulrich, W., & Buszko, J. (2004). Habitat reduction and patterns of species loss. Basic and Applied Ecology, 5(3), 231-240.

Usherwood, J. R., Larson, S. G., & Bertram, J. E. (2003). Mechanisms of force and power production in unsteady ricochetal brachiation. American journal of physical anthropology, 120(4), 364-372.

Valenta, K. (2014). Primate-plant interactions in the tropical dry forests of Northwestern Madagascar: Seed dispersal and sensory ecology of Eulemur fulvus (Doctoral dissertation). Retrieved from T-Space, https://tspace.library.utoronto.ca/handle/1807/68424

Vance, M. D., Fahrig, L., & Flather, C. H. (2003). Effect of reproductive rate on minimum habitat requirements of forest-breeding birds. Ecology, 84(10), 2643-2653.

Warren, R. D., & Crompton, R. H. (1997). A comparative study of the ranging behaviour, activity rhythms and sociality of Lepilemur edwardsi (Primates, Lepilemuridae) and

130

Avahi occidentalis (Primates, Indriidae) at Ampijoroa, Madagascar. Journal of Zoology, 243(2), 397-415.

Watson, H.C. (1859) Cybele Britannica, Vol. 4. Longman, London.

Weidt, A., Hagenah, N., Randrianambinina, B., Radespiel, U., & Zimmermann, E. (2004). Social organization of the golden brown mouse lemur (Microcebus ravelobensis). American Journal of Physical Anthropology, 123(1), 40-51.

Wells, N. A. (2003) Some hypotheses on the Mesozoic and Cenozoic paleoenvironmental history of Madagascar. In S. M Goodman, & J. P. Benstead (Eds.) The natural history of Madagascar, (pp. 16-34). Chicago: University of Chicago Press

Whitmee, S., & Orme, C. D. L. (2013). Predicting dispersal distance in mammals: a trait-based approach. Journal of Animal Ecology, 82(1), 211-221.

Whittaker, R. J., & Triantis, K. A. (2012). The species–area relationship: an exploration of that ‘most general, yet protean pattern’. Journal of Biogeography, 39(4), 623-626.

Wiens, J. A. (1989). Spatial scaling in ecology. Functional Ecology, 3(4), 385-397.

Williams, C. B. (1964). Patterns in the balance of nature and related problems of quantitative ecology. London: Academic Press.

Williams, S. E., Marsh, H., & Winter, J. (2002). Spatial scale, species diversity, and habitat structure: small mammals in Australian tropical rain forest. Ecology, 83(5), 1317-1329.

World Bank. (2014). Population, total. Retrieved 1 March, 2016 from http://data.worldbank.org/indicator/SP.POP.TOTL.

Young, J. W., Patel, B. A., & Stevens, N. J. (2007). Body mass distribution and gait mechanics in fat-tailed dwarf lemurs (Cheirogaleus medius) and patas monkeys (Erythrocebus patas). Journal of Human Evolution, 53(1), 26-40.

Zimmermann, E., Cepok, S., Rakotoarison, N., Zietemann, V., & Radespiel, U. (1998). Sympatric Mouse Lemurs in North-West Madagascar: A New Rufous Mouse Lemur Species (Microcebus ravelobensis). Folia Primatologica, 69(2), 106–114.

Zeigler, S. L., De Vleeschouwer, K. M., & Raboy, B. E. (2013). Assessing Extinction Risk in Small Metapopulations of Golden-headed Lion Tamarins (Leontopithecus chrysomelas) in Bahia State, Brazil. Biotropica, 45(4), 528-535.

Zuckerberg, B., Desrochers, A., Hochachka, W. M., Fink, D., Koenig, W. D., & Dickinson, J. L. (2012). Overlapping landscapes: a persistent, but misdirected concern when collecting and analyzing ecological data. The Journal of Wildlife Management, 76(5), 1072-1080.

131

Appendices Appendix A: Cheirogaleus medius Occurrence Responses to Amount of Habitat within 10 Landscape Scales.

Model Scale (ha) Variable Coefficient Value

Standard Error Z-Value P-value

0.5 Intercept -4.47 1.81 -2.47 0.01 Amount of Forest 1.06e-03 4.01e-04 2.65 <0.01

1 Intercept -4.26 1.53 -2.78 <0.01 Amount of Forest 5.41e-04 1.81e-04 3.00 <0.01





32 Intercept -1.68 0.66 -2.55 0.01 Amount of Forest 1.54e-05 5.49e-06 2.81 <0.01

64 Intercept -1.64 0.69 -2.39 0.02 Amount of Forest 9.44e-06 3.67e-06 2.58 0.01

128 Intercept -1.44 0.67 -2.14 0.03 Amount of Forest 4.58e-06 1.94e-06 2.36 0.02

256 Intercept -0.81 9.96e-07 -1.21 0.22 Amount of Forest 1.34e-06 1.34e-06 1.34 0.18

Null Intercept 6.70e-17 0.30 0.00 1.00

ME=Moran’s Eigenvectors

Appendix B: Microcebus murinus Occurrence Responses to Amount of Habitat within 10 Landscape Scales.

Model Scale (ha) Variable Coefficient Value

Standard Error Z-Value P-value

0.5

Intercept -2.88 1.11 -2.60 <0.01 Amount of Forest 7.21E-04 2.18E-04 3.32 <0.01

Fitted ME 139 -129.30 8.80 -1.47 0.14 Fitted ME 23 1418.00 1155.00 1.23 0.22 Fitted ME 73 38.76 9.071 4.27 <0.01 Fitted ME 14 -29.20 11.24 -2.60 <0.01 Fitted ME 33 45.13 18.68 2.42 <0.01 Fitted ME 47 -78.61 23.76 -3.31 <0.01 Fitted ME 10 28.65 10.32 2.78 <0.01 Fitted ME 30 -244.30 212.60 -1.15 0.25 Fitted ME 141 -71.84 67.480 -1.07 0.28 Fitted ME 46 -17.92 7.70 -2.33 <0.01

132

Fitted ME 51 35.37 9.54 3.71 <0.01 Fitted ME 78 78.31 19.93 3.93 <0.01 Fitted ME 98 31.13 7.34 4.24 <0.01 Fitted ME 3 1606.00 1146.00 1.40 <0.01

Fitted ME 437 38.38 14.12 2.72 <0.01 Fitted ME 17 12.94 18.63 6.94 0.49 Fitted ME 436 26.06 9.13 2.85 <0.01 Fitted ME 435 -31.57 12.20 -2.59 <0.01 Fitted ME 26 -4622.00 3718.00 -1.24 0.21 Fitted ME 1 -69.48 44.93 -1.55 0.12 Fitted ME 18 359.50 252.40 1.42 0.15

1


Fitted ME 139 -120.70 80.74 -1.50 0.13 Fitted ME 23 117.40 79.43 1.48 0.14 Fitted ME 73 37.91 7.79 4.86 <0.01 Fitted ME 47 -25.61 19.28 -1.33 <0.01 Fitted ME 33 65.41 21.25 3.08 <0.01 Fitted ME 14 -44.48 14.04 -3.17 <0.01 Fitted ME 10 47.91 17.17 2.79 <0.01 Fitted ME 51 35.21 9.72 3.62 <0.01 Fitted ME 78 78.38 18.99 4.13 <0.01 Fitted ME 98 31.15 6.64 4.69 <0.01 Fitted ME 46 -99.58 35.73 -2.79 <0.01 Fitted ME 141 -74.85 73.68 -1.02 0.31 Fitted ME 28 -44.33 17.01 -2.61 <0.01 Fitted ME 437 40.43 13.22 3.06 <0.01 Fitted ME 26 -257.70 215.70 -1.20 0.23 Fitted ME 436 27.39 8.62 3.18 <0.01 Fitted ME 435 -31.86 10.87 -2.93 <0.01 Fitted ME 74 21.27 6.59 3.23 <0.01 Fitted ME 42 -58.30 22.51 -2.59 <0.01 Fitted ME 336 20.21 6.96 2.90 <0.01 Fitted ME 2 -10.16 9.83 -1.034 0.30

2


Fitted ME 139 -106.40 58.59 -1.82 0.07 Fitted ME 23 133.90 74.68 1.79 0.07 Fitted ME 73 38.38 8.42 4.56 <0.01 Fitted ME 33 55.94 18.42 3.04 <0.01 Fitted ME 47 -89.91 23.18 -3.88 <0.01 Fitted ME 14 -28.59 9.05 -3.16 <0.01 Fitted ME 10 28.53 10.26 2.78 0.01 Fitted ME 78 65.73 12.61 5.21 <0.01 Fitted ME 51 33.39 9.00 3.71 <0.01 Fitted ME 98 27.57 5.52 4.99 <0.01 Fitted ME 46 -16.98 7.23 -2.35 0.02

133

Fitted ME 141 -62.55 50.33 -1.24 0.21 Fitted ME 28 -48.94 13.60 -3.60 <0.01 Fitted ME 437 31.25 12.04 2.60 0.01 Fitted ME 26 -294.90 203.90 -1.45 0.15 Fitted ME 436 21.97 7.51 2.93 <0.01 Fitted ME 435 -27.30 10.26 -2.66 0.01

4


Fitted ME 139 -102.10 54.49 -1.87 0.06 Fitted ME 23 127.70 78.28 1.63 0.10 Fitted ME 33 47.70 16.70 2.86 <0.01 Fitted ME 47 -80.53 21.91 -3.68 <0.01 Fitted ME 73 28.68 5.68 5.05 <0.01 Fitted ME 78 53.12 9.81 5.42 <0.01 Fitted ME 51 25.41 6.63 3.83 <0.01 Fitted ME 98 21.84 4.67 4.68 <0.01 Fitted ME 14 -25.98 8.44 -3.08 <0.01 Fitted ME 10 28.50 10.05 2.84 <0.01 Fitted ME 46 -15.55 6.72 -2.31 0.02 Fitted ME 141 -60.69 45.66 -1.33 0.18 Fitted ME 28 -39.26 11.43 -3.43 <0.01 Fitted ME 437 24.58 5.98 4.11 <0.01 Fitted ME 26 -286.30 215.40 -1.33 0.18

8

Intercept -1.373 0.59 -2.328 0.02 Amount of Forest 8.78e-05 1.82E-05 4.824 <0.01

Fitted ME 139 -110.36 54.61 -2.021 0.04 Fitted ME 23 161.11 67.13 2.400 0.02 Fitted ME 33 96.04 28.02 3.427 <0.01 Fitted ME 47 -25.09 18.22 -1.377 <0.01 Fitted ME 73 69.84 13.49 5.175 <0.01 Fitted ME 78 140.61 28.35 4.960 <0.01 Fitted ME 51 66.76 20.52 3.253 <0.01 Fitted ME 98 61.19 14.82 4.130 <0.01 Fitted ME 14 -37.76 11.46 -3.294 <0.01 Fitted ME 10 32.30 11.50 2.809 <0.01 Fitted ME 141 -56.65 45.47 -1.246 0.21 Fitted ME 46 -150.25 43.89 -3.424 <0.01 Fitted ME 28 -82.66e 22.12 -3.737 <0.01 Fitted ME 437 35.96 13.85 2.598 <0.01 Fitted ME 26 -330.08 174.70 -1.889 <0.01 Fitted ME 436 43.00 9.81 4.384 <0.01 Fitted ME 435 -61.59 14.790 -4.166 <0.01 Fitted ME 217 -14.77 6.18 -2.391 0.02 Fitted ME 440 91.19 19.37 4.707 <0.01 Fitted ME 42 -92.63 28.18 -3.287 <0.01 Fitted ME 32 -36.31 21.42 -1.695 0.09

16 Intercept -1.16 0.82 -1.41 0.16

134

Amount of Forest 2.20E-05 5.64E-06 3.89 <0.01

Fitted ME 139 -1711.00 121900.0

0 -0.01 0.99 Fitted ME 23 20.79 6.43 3.23 <0.01 Fitted ME 33 52.56 16.24 3.24 <0.01 Fitted ME 73 40.79 8.32 4.90 <0.01 Fitted ME 78 72.70 15.69 4.63 <0.01 Fitted ME 47 -95.15 24.82 -3.83 <0.01 Fitted ME 51 33.67 7.90 4.26 <0.01 Fitted ME 98 29.43 6.02 4.89 <0.01 Fitted ME 18 34.51 17.03 2.03 0.04 Fitted ME 14 -24.81 8.36 -2.97 <0.01 Fitted ME 10 29.99 10.23 2.93 <0.01 Fitted ME 46 -1034.00 77100.00 -0.01 0.99 Fitted ME 28 -15.06 8.17 -1.84 0.07 Fitted ME 437 -50.99 22.08 -2.31 0.02 Fitted ME 436 36.96 12.80 2.89 <0.01 Fitted ME 435 24.12 8.00 3.02 <0.01 Fitted ME 217 -29.40 10.64 -2.76 0.01 Fitted ME 3 -4.78 5.38 -0.89 0.37 Fitted ME 1 262.70 193.30 1.36 0.17

Fitted ME 150 -24.22 16.88 -1.44 0.15

32

Intercept 0.18 0.77 0.24 0.81 Amount of Forest 6.15E-06 3.50E-06 1.76 0.08

Fitted ME 139 -121.60 73.48 -1.65 0.10 Fitted ME 23 238.50 263.50 0.91 0.37 Fitted ME 78 77.27 15.68 4.93 <0.01 Fitted ME 73 36.86 7.38 4.99 <0.01 Fitted ME 33 52.24 16.19 3.23 <0.01 Fitted ME 47 -106.80 26.73 -4.00 <0.01 Fitted ME 51 35.09 7.65 4.59 <0.01 Fitted ME 98 30.05 5.85 5.14 <0.01 Fitted ME 18 81.64 36.82 2.22 0.03 Fitted ME 14 -25.71 10.09 -2.55 0.01 Fitted ME 10 32.67 10.71 3.05 <0.01 Fitted ME 46 -61.57 41.46 -1.49 0.14 Fitted ME 46 -19.01 8.92 -2.13 0.03 Fitted ME 3 326.30 248.50 1.31 0.19 Fitted ME 17 16.17 21.15 0.76 0.44 Fitted ME 437 39.37 12.42 3.17 <0.01 Fitted ME 217 -6.29 5.03 -1.25 0.21 Fitted ME 436 23.94 7.48 3.20 <0.01 Fitted ME 435 -27.62 9.64 -2.87 <0.01 Fitted ME 1 -22.98 15.90 -1.45 0.15 Fitted ME 26 -772.10 772.20 -1.00 0.32

64 Intercept 0.32 0.77 0.42 0.68 Amount of Forest 2.66E-06 2.10E-06 1.27 0.21

135

Fitted ME 139 -127.60 79.33 -1.61 0.11 Fitted ME 23 229.20 262.70 0.87 0.38 Fitted ME 78 77.60 15.57 4.98 <0.01 Fitted ME 73 35.15 7.01 5.02 <0.01 Fitted ME 47 -109.20 26.90 -4.06 <0.01 Fitted ME 51 34.74 7.61 4.56 <0.01 Fitted ME 33 50.80 15.83 3.21 <0.01 Fitted ME 18 29.91 5.76 5.19 <0.01 Fitted ME 10 83.43 36.72 2.27 0.02 Fitted ME 14 33.71 10.79 3.12 <0.01 Fitted ME 141 -25.94 10.05 -2.58 0.01 Fitted ME 46 -64.06 43.79 -1.46 0.14 Fitted ME 3 -19.89 9.14 -2.18 0.03 Fitted ME 17 336.10 247.80 1.36 0.18 Fitted ME 437 17.69 21.40 0.83 0.41 Fitted ME 217 39.55 12.19 3.25 <0.01 Fitted ME 436 -5.94 4.99 -1.19 0.23 Fitted ME 435 23.48 7.28 3.22 <0.01 Fitted ME 1 -26.08 9.18 -2.84 <0.01 Fitted ME 26 -23.15 15.69 -1.48 0.14

128

Intercept 0.33 0.71 0.47 0.64 Amount of Forest 4.00E-07 1.28E-06 0.31 0.76

Fitted ME 139 -128.10 78.06 -1.64 0.10 Fitted ME 23 121.40 138.60 0.88 0.38 Fitted ME 78 72.07 13.81 5.22 <0.01 Fitted ME 73 32.77 6.48 5.06 <0.01 Fitted ME 47 -98.75 24.61 -4.01 <0.01 Fitted ME 51 33.90 7.66 4.43 <0.01 Fitted ME 98 28.12 5.25 5.36 <0.01 Fitted ME 33 46.58 15.23 3.06 <0.01 Fitted ME 14 -58.58 20.67 -2.83 <0.01 Fitted ME 10 48.87 15.60 3.13 <0.01 Fitted ME 141 -67.49 44.74 -1.51 0.13 Fitted ME 46 -21.38 9.88 -2.16 0.03 Fitted ME 3 633.40 316.80 2.00 0.05 Fitted ME 17 510.30 180.30 2.83 <0.01 Fitted ME 437 36.70 11.09 3.31 <0.01 Fitted ME 26 -388.60 398.20 -0.98 0.33 Fitted ME 436 -31.73 16.48 -1.93 0.05 Fitted ME 435 21.12 6.59 3.20 <0.01 Fitted ME 6 -23.05 8.25 -2.80 0.01

256

Intercept 0.94 0.79 1.19 0.23 Amount of Forest -7.37E-07 8.24E-07 -0.90 0.37

Fitted ME 139 -130.50 79.97 -1.63 0.10 Fitted ME 23 167.10 238.90 0.70 0.48 Fitted ME 78 73.41 13.93 5.27 <0.01 Fitted ME 73 33.36 6.55 5.09 <0.01

136

Fitted ME 47 -102.90 25.80 -3.99 <0.01 Fitted ME 51 36.16 8.07 4.48 <0.01 Fitted ME 10 52.36 16.06 3.26 <0.01 Fitted ME 14 -57.64 20.10 -2.87 <0.01 Fitted ME 98 28.37 5.33 5.32 <0.01 Fitted ME 33 49.59 15.81 3.14 <0.01 Fitted ME 3 649.30 327.00 1.99 0.05 Fitted ME 17 -69.65 45.16 -1.54 0.12 Fitted ME 437 -24.45 10.93 -2.24 0.03 Fitted ME 1 524.30 185.50 2.83 <0.01 Fitted ME 26 37.00 11.20 3.30 <0.01 Fitted ME 436 -33.55 17.31 -1.94 0.05 Fitted ME 435 -520.20 683.70 -0.76 0.45 Fitted ME 6 21.17 6.65 3.18 <0.01

NULL

Intercept 0.46 0.58 0.80 0.42 Fitted ME 139 -129.66 79.59 -1.63 0.10 Fitted ME 23 125.61 154.95 0.81 0.42 Fitted ME 73 32.75 6.45 5.08 <0.01 Fitted ME 78 72.30 13.76 5.26 <0.01 Fitted ME 47 -99.48 24.70 -4.03 <0.01 Fitted ME 51 34.32 7.59 4.52 <0.01 Fitted ME 10 49.71 15.39 3.23 <0.01 Fitted ME 14 -58.85 20.55 -2.86 <0.01 Fitted ME 33 47.19 15.19 3.11 <0.01 Fitted ME 98 28.13 5.25 5.36 <0.01 Fitted ME 3 637.02 317.37 2.01 0.04 Fitted ME 46 -22.24 9.84 -2.26 0.02 Fitted ME 141 -68.34 45.02 -1.52 0.13 Fitted ME 1 513.66 180.11 2.85 <0.01 Fitted ME 17 36.78 11.05 3.33 <0.01 Fitted ME 436 -32.01 16.53 -1.94 0.05 Fitted ME 435 -401.26 444.52 -0.90 0.37 Fitted ME 6 21.09 6.57 3.21 <0.01

ME=Moran’s Eigenvectors

Appendix C: Microcebus ravelobensis Occurrence Responses to Amount of Habitat within 10 Landscape Scales.

Model Scale (ha) Variable Coefficient

Value Standard

Error Z-Value P-value

0.25

Intercept -1.69 0.57 -2.98 <0.01 Amount of Forest 7.23E-04 2.67E-04 2.71 0.01

Fitted ME 24 41.11 8.00 5.14 <0.01 Fitted ME 49 32.79 7.73 4.24 <0.01 Fitted ME 85 36.27 11.39 3.19 <0.01 Fitted ME 127 -56.99 15.35 -3.71 <0.01 Fitted ME 80 -27.05 6.61 -4.09 <0.01 Fitted ME 15 51.43 11.63 4.42 <0.01

137

Fitted ME 104 31.85 16.13 1.98 0.05 Fitted ME 81 28.69 8.62 3.33 <0.01 Fitted ME 105 15.31 6.94 2.21 0.03 Fitted ME 130 -56.46 15.53 -3.64 <0.01 Fitted ME 133 29.47 7.61 3.87 <0.01 Fitted ME 2 -335.50 16690.00 -0.02 0.98 Fitted ME 78 -24.05 7.06 -3.41 <0.01 Fitted ME 140 19.94 6.81 2.93 <0.01 Fitted ME 136 -14.37 8.96 -1.60 0.11 Fitted ME 138 14.88 4.12 3.61 <0.01 Fitted ME 120 18.01 14.26 1.26 0.21 Fitted ME 40 -12.99 5.67 -2.29 0.02 Fitted ME 116 19.51 10.61 1.84 0.07 Fitted ME 92 -31.42 11.39 -2.76 0.01 Fitted ME 17 5.02 3.55 1.41 0.16

0.5

Intercept -1.37 0.51 -2.70 0.01 Amount of Forest 2.97E-04 1.25E-04 2.37 0.02

Fitted ME 24 41.18 7.97 5.17 <0.01 Fitted ME 49 32.44 7.70 4.22 <0.01 Fitted ME 80 -27.28 6.64 -4.11 <0.01 Fitted ME 85 36.62 11.29 3.24 <0.01 Fitted ME 127 -56.86 15.28 -3.72 <0.01 Fitted ME 15 51.09 11.62 4.40 <0.01 Fitted ME 104 32.32 16.31 1.98 0.05 Fitted ME 105 15.35 6.95 2.21 0.03 Fitted ME 81 28.39 8.47 3.35 <0.01 Fitted ME 130 -56.78 15.44 -3.68 <0.01 Fitted ME 133 29.58 7.62 3.88 <0.01 Fitted ME 2 -333.10 16700.00 -0.02 0.98 Fitted ME 78 -24.29 7.03 -3.46 <0.01 Fitted ME 140 19.99 6.71 2.98 <0.01 Fitted ME 136 -14.19 8.86 -1.60 0.11 Fitted ME 138 15.08 4.12 3.66 <0.01 Fitted ME 120 18.21 14.44 1.26 0.21 Fitted ME 40 -13.00 5.65 -2.30 0.02 Fitted ME 116 19.04 10.22 1.86 0.06 Fitted ME 92 -31.78 11.22 -2.83 <0.01 Fitted ME 17 4.78 3.46 1.38 0.17

1

Intercept -1.34 0.45 -2.98 <0.01 Amount of Forest 1.29E-04 5.91E-05 2.18 0.03

Fitted ME 24 41.52 7.95 5.22 <0.01 Fitted ME 49 26.50 6.97 3.80 <0.01 Fitted ME 80 -21.17 4.49 -4.71 <0.01 Fitted ME 85 27.44 8.49 3.23 <0.01 Fitted ME 127 -66.78 17.02 -3.92 <0.01 Fitted ME 15 55.47 13.91 3.99 <0.01 Fitted ME 104 10.65 3.61 2.95 <0.01

138

Fitted ME 130 -66.40 20.69 -3.21 <0.01 Fitted ME 133 37.55 8.95 4.19 <0.01 Fitted ME 81 22.45 6.78 3.31 <0.01 Fitted ME 2 -333.80 16520.00 -0.02 0.98

Fitted ME 116 11.14 5.38 2.07 0.04 Fitted ME 78 -18.92 4.91 -3.85 <0.01 Fitted ME 140 18.84 5.24 3.60 <0.01 Fitted ME 134 -23.29 10.99 -2.12 0.03 Fitted ME 138 14.64 4.15 3.53 <0.01 Fitted ME 40 -15.24 7.93 -1.92 0.05 Fitted ME 129 -23.03 11.11 -2.07 0.04 Fitted ME 41 35.09 26.64 1.32 0.19 Fitted ME 203 9.74 6.44 1.51 0.13 Fitted ME 17 4.68 3.39 1.38 0.17 Fitted ME 75 -10.15 3.89 -2.61 0.01

2


Fitted ME 24 39.71 7.61 5.22 0.00 Fitted ME 49 30.07 7.32 4.11 <0.01 Fitted ME 80 -21.66 4.53 -4.78 <0.01 Fitted ME 85 28.99 8.57 3.38 <0.01 Fitted ME 15 52.13 12.31 4.24 <0.01 Fitted ME 127 -113.80 42.97 -2.65 0.01 Fitted ME 104 45.96 23.73 1.94 0.05 Fitted ME 130 -50.65 15.42 -3.28 <0.01 Fitted ME 133 36.16 10.07 3.59 <0.01 Fitted ME 81 23.32 6.68 3.49 <0.01 Fitted ME 2 -330.00 16440.00 -0.02 0.98

Fitted ME 116 27.52 12.28 2.24 0.03 Fitted ME 78 -19.42 4.93 -3.94 <0.01 Fitted ME 140 30.39 9.08 3.35 <0.01 Fitted ME 134 -13.28 8.92 -1.49 0.14 Fitted ME 138 25.71 7.65 3.36 <0.01 Fitted ME 40 -11.80 5.40 -2.19 0.03 Fitted ME 129 -35.12 15.37 -2.29 0.02 Fitted ME 75 -10.09 3.74 -2.70 0.01 Fitted ME 105 20.10 11.19 1.80 0.07 Fitted ME 203 9.70 6.50 1.49 0.14 Fitted ME 148 19.40 7.03 2.76 0.01

4


Fitted ME 24 39.91 7.62 5.24 <0.01 Fitted ME 49 30.00 7.32 4.10 <0.01 Fitted ME 80 -21.69 4.55 -4.77 <0.01 Fitted ME 85 29.08 8.54 3.41 <0.01 Fitted ME 15 52.18 12.32 4.24 <0.01 Fitted ME 127 -113.20 42.73 -2.65 0.01

139

Fitted ME 104 46.31 23.68 1.96 0.05 Fitted ME 130 -50.51 15.45 -3.27 <0.01 Fitted ME 133 36.12 10.05 3.60 <0.01 Fitted ME 81 23.29 6.66 3.50 <0.01 Fitted ME 2 -330.20 16430.00 -0.02 0.98

Fitted ME 116 27.41 12.25 2.24 0.03 Fitted ME 78 -19.38 4.93 -3.93 <0.01 Fitted ME 140 30.28 9.06 3.34 <0.01 Fitted ME 134 -13.16 8.92 -1.48 0.14 Fitted ME 138 25.66 7.64 3.36 <0.01 Fitted ME 40 -11.67 5.40 -2.16 0.03 Fitted ME 129 -34.86 15.34 -2.27 0.02 Fitted ME 75 -10.15 3.73 -2.72 0.01 Fitted ME 105 20.22 11.14 1.82 0.07 Fitted ME 203 9.72 6.54 1.49 0.14 Fitted ME 148 19.37 7.01 2.77 0.01

8


Fitted ME 24 40.10 7.65 5.24 <0.01 Fitted ME 49 30.32 7.37 4.12 <0.01 Fitted ME 80 -21.71 4.56 -4.77 <0.01 Fitted ME 85 29.38 8.53 3.44 <0.01 Fitted ME 127 -58.47 15.50 -3.77 <0.01 Fitted ME 15 51.83 12.22 4.24 <0.01 Fitted ME 104 33.93 17.84 1.90 0.06 Fitted ME 105 15.92 7.44 2.14 0.03 Fitted ME 130 -58.14 15.69 -3.71 <0.01 Fitted ME 133 30.72 7.93 3.87 <0.01 Fitted ME 81 23.25 6.63 3.51 <0.01 Fitted ME 2 -331.40 16530.00 -0.02 0.98 Fitted ME 78 -19.32 4.92 -3.93 <0.01 Fitted ME 140 20.47 6.56 3.12 <0.01 Fitted ME 136 -14.44 9.09 -1.59 0.11 Fitted ME 138 15.43 4.24 3.64 <0.01 Fitted ME 120 17.28 15.36 1.13 0.26 Fitted ME 40 -11.63 5.40 -2.15 0.03 Fitted ME 116 18.87 10.45 1.81 0.07 Fitted ME 75 -10.22 3.73 -2.74 0.01 Fitted ME 203 9.71 6.51 1.49 0.14

16


Fitted ME 24 40.15 7.66 5.24 <0.01 Fitted ME 49 30.41 7.37 4.12 <0.01 Fitted ME 80 -21.68 4.58 -4.73 <0.01 Fitted ME 127 -58.66 15.47 -3.79 <0.01 Fitted ME 85 29.47 8.46 3.48 <0.01 Fitted ME 15 52.10 12.27 4.25 <0.01

140

Fitted ME 104 34.08 17.73 1.92 0.05 Fitted ME 105 16.03 7.43 2.16 0.03 Fitted ME 2 -331.60 16530.00 -0.02 0.98

Fitted ME 130 -58.41 15.69 -3.72 <0.01 Fitted ME 133 30.79 7.93 3.88 <0.01 Fitted ME 81 23.25 6.57 3.54 <0.01 Fitted ME 78 -19.35 4.93 -3.92 <0.01 Fitted ME 140 20.52 6.54 3.14 <0.01 Fitted ME 136 -14.47 9.10 -1.59 0.11 Fitted ME 138 15.33 4.24 3.62 <0.01 Fitted ME 120 16.89 15.04 1.12 0.26 Fitted ME 40 -11.63 5.41 -2.15 0.03 Fitted ME 116 18.63 10.36 1.80 0.07 Fitted ME 75 -10.22 3.72 -2.75 0.01 Fitted ME 203 9.70 6.52 1.49 0.14

32


Fitted ME 24 40.21 7.63 5.27 <0.01 Fitted ME 49 30.11 7.32 4.11 <0.01 Fitted ME 85 29.47 8.41 3.51 <0.01 Fitted ME 127 -113.30 42.11 -2.69 0.01 Fitted ME 80 -21.77 4.61 -4.72 <0.01 Fitted ME 15 53.15 12.48 4.26 <0.01 Fitted ME 130 -50.82 15.65 -3.25 <0.01 Fitted ME 133 36.23 10.08 3.60 <0.01 Fitted ME 81 23.55 6.55 3.60 <0.01 Fitted ME 2 -330.40 16420.00 -0.02 0.98

Fitted ME 140 30.51 9.06 3.37 <0.01 Fitted ME 134 -12.88 8.63 -1.49 0.14 Fitted ME 78 -19.64 4.99 -3.94 <0.01 Fitted ME 116 27.39 12.09 2.27 0.02 Fitted ME 104 46.45 23.32 1.99 0.05 Fitted ME 138 25.55 7.59 3.37 <0.01 Fitted ME 129 -11.80 5.44 -2.17 0.03 Fitted ME 75 -34.96 15.26 -2.29 0.02 Fitted ME 105 -10.17 3.71 -2.74 0.01 Fitted ME 203 20.44 10.96 1.87 0.06 Fitted ME 148 9.70 6.58 1.48 0.14

64


Fitted ME 24 40.10 7.60 5.27 <0.01 Fitted ME 49 30.68 7.43 4.13 <0.01 Fitted ME 127 -3662.00 218800.00 -0.02 0.99 Fitted ME 85 30.63 8.64 3.54 <0.01 Fitted ME 130 -1656.00 106800.00 -0.02 0.99 Fitted ME 133 1242.00 72310.00 0.02 0.99 Fitted ME 80 -22.24 4.65 -4.78 <0.01

141

Fitted ME 15 50.02 11.32 4.42 <0.01 Fitted ME 81 24.29 6.63 3.66 <0.01 Fitted ME 2 -436.40 199500.00 0.00 1.00

Fitted ME 140 923.40 55900.00 0.02 0.99 Fitted ME 134 36.19 27100.00 0.00 1.00 Fitted ME 78 -20.11 5.01 -4.01 <0.01 Fitted ME 116 576.70 55830.00 0.01 0.99 Fitted ME 104 338.80 22620.00 0.02 0.99 Fitted ME 138 652.00 40000.00 0.02 0.99 Fitted ME 40 -11.73 5.31 -2.21 0.03 Fitted ME 129 -1258.00 78120.00 -0.02 0.99 Fitted ME 75 -10.21 3.68 -2.77 0.01 Fitted ME 148 161.00 47980.00 0.00 1.00 Fitted ME 252 641.80 38080.00 0.02 0.99

128


Fitted ME 24 40.59 7.65 5.31 <0.01 Fitted ME 49 30.08 7.27 4.14 <0.01 Fitted ME 127 -113.80 41.93 -2.71 0.01 Fitted ME 80 -21.77 4.60 -4.74 <0.01 Fitted ME 130 -50.99 15.61 -3.27 <0.01 Fitted ME 133 36.40 10.09 3.61 <0.01 Fitted ME 85 29.66 8.42 3.52 <0.01 Fitted ME 15 53.44 12.52 4.27 <0.01 Fitted ME 81 23.58 6.51 3.62 <0.01 Fitted ME 2 -330.10 16300.00 -0.02 0.98

Fitted ME 140 30.71 9.07 3.39 <0.01 Fitted ME 134 -12.87 8.50 -1.51 0.13 Fitted ME 78 -19.85 4.94 -4.02 0.00 Fitted ME 116 27.18 12.09 2.25 0.02 Fitted ME 104 46.39 23.29 1.99 0.05 Fitted ME 138 25.63 7.58 3.38 0.00 Fitted ME 40 -11.83 5.46 -2.17 0.03 Fitted ME 129 -35.31 15.22 -2.32 0.02 Fitted ME 75 -10.25 3.72 -2.75 0.01 Fitted ME 105 20.51 10.94 1.88 0.06 Fitted ME 203 9.84 6.73 1.46 0.14 Fitted ME 148 19.41 6.93 2.80 0.01

NULL

Intercept -0.29 0.21 -1.40 0.16 Fitted ME 24 40.10 7.59 5.28 <0.01 Fitted ME 80 -22.34 4.64 -4.81 <0.01 Fitted ME 127 -3665.00 218600.00 -0.02 0.99 Fitted ME 85 30.67 8.67 3.54 <0.01 Fitted ME 15 50.18 11.32 4.43 <0.01 Fitted ME 49 30.62 7.44 4.12 <0.01 Fitted ME 130 -1658.00 106800.00 -0.02 0.99 Fitted ME 133 1243.00 72290.00 0.02 0.99

142

Fitted ME 81 24.49 6.61 3.70 <0.01 Fitted ME 2 -435.30 198800.00 -2.00E-03 1.00

Fitted ME 140 924.20 55840.00 0.02 0.99 Fitted ME 134 36.64 26980.00 1.00E-03 1.00 Fitted ME 78 -20.26 5.01 -4.05 <0.01 Fitted ME 116 577.30 56080.00 0.01 0.99 Fitted ME 104 339.30 22940.00 0.02 0.99 Fitted ME 138 652.50 39920.00 0.02 0.99 Fitted ME 40 -11.83 5.30 -2.23 0.03 Fitted ME 129 -1258.00 78060.00 -0.02 0.99 Fitted ME 75 -10.18 3.68 -2.77 0.01 Fitted ME 105 161.40 48150.00 3.00E-03 1.00 Fitted ME 148 642.30 38030.00 0.02 0.99 Fitted ME 152 -832.10 48110.00 -0.02 0.99

ME=Moran’s Eigenvector