Diversity of Beetles Across a Land-Use Gradient in Sabah, Malaysia

Diversity of beetles across a land-use gradient in Sabah, Malaysia

Adam Christian Sharp

Department of Life Sciences

Imperial College London

A thesis submitted for the degree of Doctor of Philosophy

June 2018

Declaration of originality

All work in this thesis is my own, and I have acknowledged instances where I have collaborated with other researchers. Where others are included as authors on individual chapters, those individuals have provided comments on that chapter. Two anonymous reviewers provided a great deal of insight into the analyses in Chapter 3 when that chapter was submitted for publication in the Journal of

Applied Ecology.

Supervisors

Dr Robert Ewers

Maxwell Barclay

The copyright of this thesis rests with the author and is made available under a Creative Commons

Attribution Non-Commercial No Derivatives licence. Researchers are free to copy, distribute or transmit the thesis on the condition that they attribute it, that they do not use it for commercial purposes and that they do not alter, transform or build upon it. For any reuse or redistribution, researchers must make clear to others the licence terms of this work.

Abstract

Borneo is one of the most biodiverse regions on the planet but half of Borneo’s forest has been logged and millions of hectares replaced by oil palm plantation. Decomposing biodiversity into its underlying component parts may reveal mechanisms by which diversity loss might be mitigated. I sampled beetles (order: Coleoptera) at the Stability of Altered Forest Ecosystems Project in Sabah to quantify shifts in biodiversity associated with logging and clearing of tropical lowland forest for oil palm plantation. While logged forest maintained high species richness, the beetle species which persisted there were of lower conservation value. Beta-diversity mitigated losses in alpha-diversity in heavily- logged forest to some extent, and in multiple beetle taxa beta-diversity was greatest in that habitat. In both logged and unlogged forest, high beta-diversity was maintained through balanced variation in community composition. Oil palm plantations supported very few beetle species and although beta- diversity remained high, that beta-diversity was attributed to fluxes in the abundances of common species. At small spatial scales, forest quality and topographic roughness were significant determinants of beta-diversity. In unlogged forest, analysis of beta-diversity revealed sites of distinct ecological communities relevant to some that appeared to be defined by microclimate. At larger scales, spatial turnover in community composition was the strongest contributor to diversity followed by habitat structure, microclimate and then topography. Oil palm plantations are of low value and large areas of unlogged forest are evidently of highest value for the conservation of biodiversity.

Degraded forest retains remarkable diversity through shifts in the spatial arrangement of ecological communities, and thus I recommend that small areas of even heavily-logged forest should be preserved where they are created in agricultural matrices. Future studies on land-use change in

Borneo should incorporate beta-diversity into their designs as the component reveals mechanisms by which diversity loss might be mitigated.

Acknowledgements

This work was funded by the Stability of Altered Forest Ecosystems (SAFE) Project, who are in turn funded by the Sime Darby Foundation.

I would like to thank the SAFE Project and Royal Society’s South East Asia Rainforest Research

Partnership for logistical support, and the Sabah Foundation, Maliau Basin Management Committee, the State Secretary, Sabah Chief Minister’s Departments, the Malaysian Economic Planning Unit and the Sabah Biodiversity Council for permission to conduct research. I would also like to thank the

Natural History Museum in London for allowing me access to facilities to carry out taxonomic work.

The ASTER Global Digital Elevation Model (GDEM) v2 data was retrieved from the online NASA

Earthdata Search portal at https://search.earthdata.nasa.gov/search. ASTER GDEM is a product of

NASA and METI.

Data in the field was collected largely by the SAFE Project team of research assistants – most notably the highly-proficient SAFE Project Invertebrate Team which includes Madani bin Samat (Opong),

Mainus bin Tausong, Melvin bin Teronggoh, Muhamed bin Juhanis (Mamat), Risman bin Ajang and

Sisoon bin Maunut (Siun). Dr Edgar Turner started the long-term insect sampling work at SAFE

Project in 2010 which formed the basis for the majority of this research. Ryan Gray and Min Sheng

Khoo were fundamental to keeping SAFE Project running on a day-to-day basis and making sure the team were able to complete their field work. Dr Arthur Chung of the Forest Research Centre in

Sepilok, Sabah, offered ongoing valuable advice and assisted in securing research permits to work in the state. There are also countless researchers, students and volunteers who I have crossed paths with who made my time at SAFE so enjoyable.

At the Natural History Museum, I would like to thank Max Barclay for his limitless knowledge and support when I was inundated with tens of thousands of beetles. Dr Richard Thomson was also incredibly helpful in aiding with the taxonomic identification of weevils. Alice Haughan, Kara Taylor,

Laura Healy and Noel Juvigny-Khenafou all used the SAFE Project insect samples as part of their respective MSc/MRes research projects and contributed to processing or identifying insects in some way. Surveys were facilitated by the Natural History Museum Learning Managers and Investigate

Team. At the Oxford Museum of Natural History, Darren Mann and Guillaume de Rougemont provided taxonomy of dung beetles and rove beetles respectively.

The Forest Ecology and Conservation group at Imperial College London provided insights into everything from statistical to ecological theory. Prof Rob Ewers contributed excellent supervision for the entirety of the PhD. Dr Lan Qie provided invaluable advice on scientific writing style. Dr Marion

Pfeifer and Dr Stephen Hardwick kindly allowed use of the habitat data they collected in the field.

Clare Wilkinson, Liv Daniel, Mike Boyle, Nichar Gregory, Phil Chapman, Ross Gray, Sarab Sethi and

Dr Terhi Riutta have all been constant help and inspiration.

Finally, thank you to all my friends and family who put up with me for the last four years.

Table of Contents

Declaration of originality ...... 3

Supervisors ...... 3

Abstract ...... 4

Acknowledgements ...... 5

Chapter 1 - Introduction ...... 11

1.1 Biodiversity and its importance ...... 11

1.2 Biodiversity in the tropics ...... 12

1.3 Borneo as a biodiversity hotspot ...... 13

1.4 Threats to the biodiversity of Borneo ...... 14

1.5 Past-documented impacts of land-use modification on biodiversity ...... 16

1.6 Spatial scale and conservation ...... 18

1.7 Habitat quality and conservation ...... 19

1.8 Selecting a suitable study taxon for researching biodiversity patterns ...... 20

1.9 The aims of this thesis ...... 21

1.10 Thesis structure ...... 22

Chapter 2 – High conservation value scarabs are hit hardest by tropical deforestation ...... 24

2.1 Abstract ...... 24

2.2 Introduction...... 24

2.3 Methods ...... 26

2.4 Results ...... 28

2.5 Discussion ...... 29

Table 2.1 ...... 31

Table 2.2 ...... 34

Table 2.3 ...... 35

Figure 2.1 ...... 36

Figure 2.2 ...... 37

Figure 2.3 ...... 38

Chapter 3 – Forest quality, forest area and the importance of beta-diversity for protecting Borneo’s beetle biodiversity ...... 39

3.1 Abstract ...... 39

3.2 Introduction...... 40

3.3 Methods ...... 42

3.3.1 Study area ...... 42

3.3.2 Insect sampling ...... 43

3.3.3 Calculating diversity metrics ...... 43

3.3.4 Relating insect trap data to forest distance ...... 44

3.3.5 Defining distance and forest quality variables ...... 44

3.3.6 Developing bootstrap models ...... 45

3.3.7 Creating nominal forest areas ...... 46

3.4 Results ...... 47

3.5 Discussion ...... 49

Table 3.1 ...... 53

Table 3.2 ...... 55

Figure 3.1 ...... 56

Figure 3.2 ...... 57

Figure 3.3 ...... 58

Figure 3.4 ...... 59

Figure 3.5 ...... 60

Chapter 4 – Tropical logging and deforestation impacts multiple scales of weevil beta-diversity ...... 61

4.1 Abstract ...... 61

4.2 Introduction...... 62

4.3 Methods ...... 64

4.4 Results ...... 69

4.5 Discussion ...... 70

Table 4.1 ...... 74

Figure 4.1 ...... 75

Figure 4.2 ...... 76

Figure 4.3 ...... 77

Figure 4.4 ...... 78

Chapter 5 – Congruence and Importance to Conservation of Beta-Diversity in Borneo Beetles ...... 79

5.1 Abstract ...... 79

5.2 Introduction...... 80

5.3 Methods ...... 83

5.3.1 Study area ...... 83

5.3.2 Insect sampling ...... 84

5.3.3 Habitat variables ...... 84

5.3.4 Comparing responses of alpha- and beta-diversity to disturbance ...... 85

5.3.5 Modelling large-scale beta-diversity patterns ...... 85

5.3.6 Identifying small-scale beta-diversity hotspots ...... 86

5.4 Results ...... 87 9

5.5 Discussion ...... 89

Table 5.1 ...... 94

Table 5.2 ...... 95

Figure 5.1 ...... 96

Figure 5.2 ...... 97

Figure 5.3 ...... 98

Figure 5.4 ...... 99

Figure 5.5 ...... 100

Chapter 6 – Conclusions and synthesis ...... 101

6.1 Overview ...... 101

6.2 Unlogged forest is irreplaceable for biodiversity ...... 101

6.3 Beta-diversity is a stronger force than alpha-diversity in heavily-modified landscapes ...... 102

6.4 Beta-diversity is mediated by environment ...... 104

6.5 Implications for conservation ...... 105

6.6 Concluding remarks ...... 106

References ...... 107

Chapter 1 – Introduction

In this thesis, I examine how the biodiversity of beetles (order: Coleoptera) is impacted by modification of natural habitat in Sabah, Malaysian Borneo. I sampled beetles at the Stability of

Altered Forest Ecosystems (SAFE) Project which comprises a land-use gradient extending from unlogged forest, through forest that has been logged to various extents, to oil palm plantation (Ewers et al. 2011) I aimed to assess if and how diversity changes are implemented through logging and clearing of forest. Biodiversity is immensely valuable to humans, both directly and indirectly (Foley et al. 2005; Graves et al 2017), and it is imperative to investigate the processes underlying diversity change if we are to potentially mitigate losses. This work will hopefully contribute to that field of science, which will in turn lead to better design of conservation areas in Borneo.

1.1 Biodiversity and its importance

Biodiversity is the variation among living organisms at every possible scale, and includes variation between individuals of the same species, between different species and in ecosystems (Magurran

2004). High diversity of species exists when those species occupy separate niche spaces deﬁned by resource requirement and availability (Hutchinson 1959) and is therefore reliant on environmental heterogeneity (Tews et al. 2004; Báldi 2008). Natural levels promote high rates of ecological processes, such as decomposition and seed dispersal, and the resilience of community structure against perturbations, including introduction of invasive species (Hooper et al. 2005; Worm et al.

2006; Thompson et al. 2014). As a result, biodiversity is imperative to ecosystems because it confers stability (Chapin et al. 2000; Ewers et al. 2015).

Biodiversity loss in naturally heterogeneous systems is likely to strongly reduce ecosystem processes

(Chan et al. 2006; Tylianakis et al. 2008) and several hypotheses provide explanations for this biodiversity-stability relationship. Environmental pressures are unlikely to remove all taxa performing any one functional role because change is unlikely to affect all those taxa negatively (the portfolio effect; Tilman et al. 1998; Elmqvist et al. 2003; Ewers et al. 2015). Lawton (1994) suggests that, in a community where several taxa perform similar functional roles, sequential removal of them 11

accompanies an exponential decay in ecosystem functioning where process rates start to become limited after the majority have been removed. Conversely, Naeem (1998) proposes that functional redundancy within a community allows natural processes to remain constant even if only one such taxon persists. In either case, ecological compensation for loss of taxa has been observed and can be either numerical in response, i.e. the remaining taxa increase in abundance, or per-capita, i.e. individuals of the remaining taxa each contribute at a greater rate (Ruesink & Srivastava 2001).

Our reliance on ecosystem processes renders biodiversity indirectly valuable (Foley et al. 2005).

These so-called ecosystem services include crop pollination, nutrient cycling, carbon storage, water cycling and pest control (Millennium Ecosystem Assessment 2005) and can be assigned a monetary value derived from the potential cost of artiﬁcial replacement. By this definition, many services can be valued globally at billions of U.S. dollars (Abson & Termansen 2011; Ninan & Inoue 2013).

Biodiversity is also largely considered to be directly valuable to mankind. The majority of people place high inherent value on the existence of many species and the differences between species (Graves et al. 2017), and that value translates even when those species might be foreign or never to be encountered by individuals (Christie et al. 2006). Even taxonomic groups which might be considered unlikely to attract public affection have had significant historical and cultural impact on cultures across the world (Leather et al. 2015). Because of the immeasurable direct and indirect values placed on biodiversity, it can easily be claimed that high biodiversity levels are of global importance.

1.2 Biodiversity in the tropics

Large-scale spatial patterns in biodiversity exist for a range of abiotic factors (Gaston 2000). Species richness increases non-linearly with area (Ackerman et al. 2007; Hannus & Von Numers 2008) as, in a heterogeneous environment, increasing search area is likely to sample further habitats to which specialist taxa are adapted (MacArthur & Wilson 1967). Richness also increases with both water availability (O’Brien 1993; Chappuis et al. 2012) and energy availability (Currie 1991; Kumschick et al.

2009), and due to latitudinal environmental gradients escalates towards the equator (Hawkins et al.

2003; Hillebrand 2004; Kreft & Jetz 2007). The tropics are extremely diverse because of their large

surface area and climatic conditions (Gaston 2000; Myers et al. 2000). The geographical zone is not only one of current high diversity, but a continuing source of rapid evolution which emanates taxa northwards and southwards (Jablonski et al. 2006).

1.3 Borneo as a biodiversity hotspot

Southeast Asia is an especially biodiverse region. Sundaland is one of three “hottest hotspots” recommended by Myers et al. (2000) for conservation priority alongside Madagascar and the

Philippines for its immense biodiversity. Approximately 5.0% of plant species and 2.8% of vertebrate species are endemic there (Myers et al. 2000). Historic climate oscillations have provided continual environmental heterogeneity and changes in island connectivity within the archipelago, resulting in a high degree of current endemism (Lohman et al. 2011; Wong 2011; Culmsee & Leuschner 2013) with forest refugia preserving a large portion of ancestral taxa through the Pleistocene (Woodruff 2010).

Borneo, being naturally dominated by forest (Ashton 2009), has acted as one such refugium and retained many taxa which were rendered locally extinct elsewhere (Wilting et al. 2012). High rates of immigration and internal diversiﬁcation have contributed to rapid rates of evolution on Borneo (de

Bruyn et al. 2014). Internal diversiﬁcation has been promoted by the large surface area of Borneo (the third-largest island in the world) and the accompanying range of habitats, in line with island biogeography theory (MacArthur & Wilson 1967). Because of these historical and geographical process, the island is one of the most biodiverse places in the world (Phillips et al. 1994; Kier et al.

2005).

Landscape-scale patterns in the diversity of Bornean plants exist and elevation is the strongest inﬂuence on those patterns (Grytnes & Beaman 2006; Slik et al. 2009; Culmsee & Leuschner 2013).

The interior of Borneo consists almost entirely of mountainous sedimentary shale supporting a diversity of vegetation, similar to the rest of Sundaland (Ashton 2009), as the high-elevation plants are likely to have speciated before the formation of the archipelago (Culmsee & Leuschner 2013). High plant diversity in the centre of Borneo is attributed to the mid-domain effect (Slik et al. 2003), whereby because of dispersal constraints, high diversity is likely in the middle of islands of habitat by spatial

probability alone (Colwell & Lees 2000). The geology of coastal Borneo is predominantly hilly sandstone deposited in areas of past epicontinental sea, and it is here that the majority of endemic plant species persist (Ashton 2009). Variation in the community composition of low-elevation vegetation is caused by gradients in precipitation, temperature seasonality, soil composition and past climatic conditions, as well as the dispersal barriers imposed by mountain ranges (Slik et al. 2003,

2009). Northeast Borneo, although not the most diverse region in terms of vegetation, exhibits a particularly distinct ﬂoral composition (Slik et al. 2003). It follows that Northeast Borneo possesses unique ecological communities of fauna reliant on that vegetation.

1.4 Threats to the biodiversity of Borneo

Humans have modiﬁed approximately half of the Earth’s surface (Millennium Ecosystem Assessment

2005) and natural landscapes of particularly high biodiversity, especially tropical forests, are increasingly threatened with replacement by agriculture and urbanisation (Fitzherbert et al. 2008;

Lewis et al. 2015; Taubert et al. 2018). Over one third of global tropical deforestation occurs in Asia

(Hansen et al. 2008). Within Southeast Asia, two-thirds of the deforestation between 2000 and 2010 occurred on islands (Stibig et al. 2014).

Borneo has lost a signiﬁcant proportion of its forest cover in recent decades. In 1973, 75.7% of

Borneo’s surface area was forested, and 30.2% of that area had been cleared by 2010. Deforestation rates varied across the island in this period due to major differences in management, with, as an example, 39.5% of forest cleared within the Malaysian state of Sabah but only 8.4% cleared within

Brunei (Gaveau et al. 2014). In fact, 80% of Malaysian Borneo was impacted by high-impact logging or clearing between 1990 and 2009 (Bryan et al. 2013). The mountainous centre of Borneo, dubbed the “Heart of Borneo”, maintains a single, large expanse of unlogged forest (Gaveau et al. 2014).

Most, 97% (Gaveau et al. 2014), of logging occurred in Borneo’s biodiverse peripheral lowland forest because of its accessibility (Bryan et al. 2013) and for the high abundance of hardwood trees such as dipterocarps (Slik et al. 2003; Ashton 2009). Selective logging for this valuable timber is commonly followed by clearing and conversion to agriculture (Gaveau et al. 2016; Tsujino et al. 2016), and by

2010 33% of total cleared land had been converted to plantation littered with small forest fragments.

At that point, 10% of Borneo’s surface area consisted either oil palm or timber plantation (Gaveau et al. 2014).

The oil palm (Elaeis guineensis) is native to West Africa and its cultivation as a crop in Malaysia began in 1917. It is only after the industrialisation of Malaysian agriculture in the 1970’s that signiﬁcant areas of forest were converted to oil palm plantation. At present, Malaysia is the largest producer of palm oil in the world, and the product is used not only as cooking oil but in various applications ranging from biofuel to cosmetics (Basiron 2007; Fitzherbert 2008). The Malaysian Palm

Oil Council describes farming of palm oil as “sustainable” (Basiron 2007), but such large-scale land- use modiﬁcation for agriculture has previously had negative implications for biodiversity (Tilman et al.

2001; Foley et al. 2005).

Land-use modification impacts biodiversity through both biotic and abiotic changes in habitat structure

(Báldi 2008). Replacement of natural forest representing hundreds of tree genera (Slik et al. 2003;

Nakagawa et al. 2013) with oil palm monoculture represents an inherent crash in tree taxonomic diversity, but also a decline in habitat heterogeneity for forest animals (Azhar et al. 2015). Mature oil palm plantations are largely characterised by low tree biomass, relatively open canopy and shallow, dry soils (Pfeifer at al. 2016) as well as high daytime ambient air temperatures and low daytime humidity (Hardwick et al. 2015). Despite low spatial heterogeneity, oil palm plantations are attributed with high temporal heterogeneity through the agricultural cycle (Luskin & Potts 2011). Conversely to clearing of forest for agriculture, logging results in forest which is highly spatially heterogeneous in structure due to uneven intensity (Burivalova et al. 2014). Logged forest areas differ significantly in attributes such as tree biomass, soil composition, ground-level vegetation structure and daily microclimatic variation (Hardwick et al. 2015; Pfeifer et al. 2016). Such widespread modification in habitat structure inevitably has consequences for biodiversity.

1.5 Past-documented impacts of land-use modification on biodiversity

Past studies have attempted to quantify the changes in biodiversity associated with logging and clearing of tropical lowland forest in Borneo. Recognising how biodiversity has previously been studied requires understanding of the three partitions of it; alpha-diversity is at a single point in space or time, beta-diversity is between points in space or time, and gamma-diversity is total diversity in a landscape or region (Whittaker 1972). Each of those three components can be quantified using a wide range of indices, and there therefore exists considerable disagreement in how best to quantify biodiversity (Magurran 2004). As such, previous biodiversity studies from Borneo have reported different diversity indices focussing on many taxa.

Declines in total number of species in oil palm plantation relative to adjacent natural habitats have been detected in multiple disparate taxa, including birds (Edwards et al. 2010; Azhar et al. 2011), mammals (Bernard et al. 2009; Jennings et al. 2015), amphibians (Gillespie et al. 2012; Faruk et al.

2013), ants (Brühl & Eltz 2009; Fayle et al. 2010) and beetles (Chung et al. 2000). These studies are based on gamma-diversity (comparison of the total number of species persisting in different habitats) and have been highly effective in demonstrating the conservation value of forest in comparison to plantation. However, this research has been limited in that they do not identify or control for spatial scale, and so cannot be related to one another or to applied topics such as selecting suitably size forest areas for conservation. Furthermore, taxa rarely coexist in equal numbers (Hubbell 2001), so relying solely on taxa richness to quantify diversity disregards community composition (measures of the identity and abundance of different taxa; Magurran 2004). Community composition can respond to habitat change independently of richness (Beck et al. 2006; Banks-Leite et al. 2012), and it is necessary to quantify both in order to examine the ecological processes which themselves underpin biodiversity (Whittaker 1965).

Similar observations have been made in comparison of logged forest with unlogged forest. Several studies have found little or no change in the number of species persisting in logged compared to nearby unlogged forest (Hamer et al. 2003; Cleary et al. 2007; Edwards et al. 2014a). Where differences have been observed between these two habitat types is in species identity (Hamer et al.

2003; Cleary et al. 2007). Cleary et al. (2009) noted fewer endemic species of butterflies in logged 16

forest compared to unlogged. It is therefore possible that while logged forest might maintain relatively high species richness, the conservation value of those species themselves may be lower than in natural habitat.

Beta-diversity is largely neglected from studies of logging and oil palm plantation impacts on biodiversity in Borneo yet is inherently linked to forest structure and heterogeneity (Whittaker 1972;

Tscharntke et al. 2012). Some previous work on beta-diversity does exist, and those pieces have varied both in the approaches the authors’ have taken to quantify diversity and the questions addressed. Hamer & Hill (2000) discovered that spatial scale was important in deciphering the responses of Lepidoptera to logging. Benedick et al. (2006) applied beta-diversity to identify stronger between-fragment variation in butterfly community structure compared with within-fragment. Kitching et al. (2013) identified large-scale spatial turnover in plant species in unlogged forest but not in logged forest and linked this finding to differences in moth community structure between habitat types. Wearn et al. (2016) have been the only authors to examine patterns of beta-diversity across the entire land- use gradient from unlogged forest, through logged forest to oil palm plantation. They found that increased beta-diversity in modified landscapes mitigated loss of mammal species and that that effect varied by spatial sampling grain. However, while these findings certainly have important implications for mammalian and tropical forest conservation, there remains significant scope for expansion.

Many previous studies have quantified beta-diversity using metrics that are either directly dependent on alpha-diversity or dissimilar in scale. This not only prohibits the comparison of the importance of alpha-diversity compared to beta-diversity, but also invalidates the calculation of beta over gradients in alpha (Jost 2007, 2010). As a result of mismatched diversity indices, as well as primary focus on species richness instead of community composition, there remains considerable debate on the responses of beta-diversity to logging and replacement of Borneo’s tropical forest with oil palm plantation.

Thankfully, there exists an array of statistical tools available to ecologists that can be utilised in aim of further contributing to this field of study. Jost (2007) has developed methods of quantifying each of alpha- and beta-diversity in independent forms that allow meaningful comparison. These measures of

“true” diversity provide the potential to standardise evaluations of biodiversity and allow comparisons 17

between alpha- and beta-diversity that would have previously been invalid. Baselga (2017) has separately devised methods of quantifying the underlying processes contributing to beta-diversity, namely true turnover in community composition and variation derived from gradients in taxa abundance. Further methods have been developed by Legendre & De Cáceres (2013) which allow the study of beta-diversity through partitioning of community variance. The variance-based methods provide scope to quantify spatial turnover at large scales (Legendre & Gallagher 2001) as well as identify sites at small scales of particular significance to local beta-diversity (Legendre & De Cáceres

2013). These tools have incredible potential for expanding on our knowledge on the impacts of land- use change on beta-diversity.

1.6 Spatial scale and conservation

Spatial scale is a signiﬁcant factor determining ecological responses to land-use modiﬁcation

(Magurran 2004; Berry et al. 2008; Wearn et al. 2016). It is also important in linking ecological studies on biodiversity to conservation. Accounting for relevant spatial scaling through concepts like beta- diversity has potential to suggest appropriate areas of natural habitat that should be preserved in order to best maintain natural levels of biodiversity (Socolar et al. 2016). This concept is particularly relevant considering the ongoing decline in tropical forest areas and increase in fragmentation (Lewis et al. 2015; Taubert et al. 2018).

The protected are network in Sabah, Malaysian Borneo, already comprises forest fragments of a wide range of sizes (Reynolds et al. 2011). Government-managed forest reserves are usually larger in area, for example Maliau Basin Conservation Area is 58,840 ha and Danum Valley Conservation Area is 43,800 ha. Privately-managed fragments, which are usually positioned within agricultural matrices, are considerably smaller and are commonly less than 100 ha in area (Tawatao et al. 2014; Lucey et al. 2017). Identifying the spatial scales which are relevant to natural patterns in beta-diversity has the potential to provide evidence on minimum thresholds of forest area suitable for protecting biodiversity

(Socolar et al. 2016). In specific, beta-diversity research may provide evidence on the potential

efficacy of small, privately-managed forest fragments in conserving similar biodiversity to larger forest reserves.

1.7 Habitat quality and conservation

Forest modification is a continuous process that generates high variation in the amount of physical damage inflicted on the forest (Burivalova et al. 2014; Pfeifer et al. 2016). Despite this fact, the majority of biodiversity studies from Borneo have focussed on discrete land-use categories, ie unlogged forest, logged forest and oil palm plantation, or even disregarded differences between unlogged and logged forest. Broad categorisation of habitat types may be sufficient in highlighting the need to focus conservation policy on particular regions or land use types but is unhelpful in linking with ecological processes that do not necessarily respond in a discrete fashion (Edwards et al. 2011).

Continuous measures of forest quality should be employed in the study of logged forest (Edwards et al. 2011). Large areas of logged forest are being added to the protected area network in Borneo

(Reynolds et al. 2011), yet there remains significant disagreement in value of logged forest to the conservation of biodiversity and logged forest is globally considered to be of lower biodiversity value compared to unlogged (Gibson et al. 2011). While further examination of logged forest is likely to confirm that unlogged forest support greatest overall levels of biodiversity, dissecting the responses of alpha- and beta-diversity to logging will provide insight into how best to manage logged forest

(Socolar et al. 2016). This could be through identifying ways in which to minimise biodiversity losses through logging, eg through identification of critical thresholds in forest quality for biodiversity, or by identifying environmental features which can be utilised in order to increase biodiversity in modified landscapes. This research is essential in informing the most effective ways of utilising limited conservation resources (Reynolds et al. 2011).

1.8 Selecting a suitable study taxon for researching biodiversity patterns

Diversity of large, charismatic animals are of limited relevance to the total biodiversity in Borneo, as their diversity patterns do not align well with those of other taxa (Meijaard & Nijman 2003). Insects are ideal subjects for studying spatial diversity patterns because of their high abundance and functional importance in tropical forests (Ewers et al. 2015), their co-evolution alongside other ecologically significant taxa (Farrell et al. 2001; McKenna et al. 2009; Ahrens et al. 2014) and their cultural relevance to people around the world (Leather et al. 2015).

Beetles (order: Coleoptera) are especially suitable because of their inherently high diversity, with over

400,000 species described worldwide (Hammond 1992) and an estimated 1.5 million species total

(Stork et al. 2015). The continuing radiation of beetles (Hunt et al. 2007) has led to a within-order functional diversity that is suitable for most ecological investigations, with beetles containing species linked to most functional groups, from detritivores through to parasitoids. Their proven sensitivity to environmental gradients (Chung et al. 2000; Ewers & Didham 2008; Ewers et al. 2015) renders them especially useful for quantifying impacts of habitat modiﬁcation, and they have been used as a focal taxon for the measurement of diversity in many study systems around the world (Weibull et al. 2003;

Fattorini 2006; García-López et al. 2012). In Borneo a high proportion of coleopteran families are represented (Chung et al. 2000) and beetles are proven as a strong indicator of other invertebrate and vertebrate groups (Edwards et al. 2014a).

Because of the immense diversity of beetles (Hammond 1992; Stork et al. 2015), it would be logistically impossible to study the diversity of the entire order. Instead, it makes sense to select subsets for analysis. Scarab beetles (superfamily: Scarbaeoidea) are one suitable subset. They are incredibly speciose in themselves, including taxa that are dung-feeding, carrion-feeding, vegetation- feeding, and wood-feeding among others. Scarabs have evolved alongside many species of plant and many species of mammals (Ahrens et al. 2014). As a result of their co-evolution with many other important forest species, they are excellent candidates for indicators of surrounding ecological communities. Previous studies have focussed solely on true dung beetles (family: Scarabaeidae; subfamily: Scarabaeinae) as indicators for the biodiversity value of logged forest and oil palm plantation (Edwards et al. 2014a) and riparian buffer strips (Gray et al. 2014) but have omitted other 20

members of the superfamily. Scarab beetles are relatively charismatic, are sought after by insect collectors and recognised by audiences around the world through appearances in wildlife documentaries and other science media (Leather et al. 2015). These traits render scarabs an ideal study group for quantifying the impacts of deforestation on biodiversity that is inherently valuable to people.

Rove beetles (family: Staphylinidae) are a second subset of beetles suitable for biodiversity analysis, being one of the most speciose taxonomic families in existence (Hammond 1992). The majority of rove beetles in tropical forests are active in the upper layers of soil, within leaf litter and around pieces of dead wood (Barton et al. 2011) and, unlike scarabs which fly frequently, are predominantly ground- dwelling. They occupy a similarly diverse range of niches (Barton et al. 2011). Rove beetles are highly abundant in tropical forests and sensitive to environmental change (Bohac 1999), and therefore lend themselves to diversity analyses considering community composition. Currently, there exists only taxonomic studies on the diversity of rove beetles in Borneo and these do not consider ecology or land-use.

Weevils (superfamily: Curculionoidea) are a group of beetles best suited for diversity analyses relating specifically to vegetation composition and structure. This is because weevils are mostly herbivorous, and their great taxonomic diversity (Oberprieler et al. 2007) is resultant of their co-evolution alongside flowering angiosperms (Farrell et al. 2001; McKenna et al. 2009). As such, they are likely to be strong indicators of changes in vegetation structure relating to land-use modification. However, there exists limited published material on the responses of this group, and previous studies focus on the natural history of weevils and their associations with dipterocarps (Lyal & Curran 2000; Iku et al. 2017). There is therefore significant in potential in the use of this group to the spatial responses of biodiversity to logging and clearing of forest.

1.9 The aims of this thesis

In this thesis, I aim to use a hyper-diverse study group, the beetles, to examine the impacts of logging and clearing of Borneo’s tropical lowland forest for oil palm plantations on biodiversity. I focus

primarily on quantifying patterns in beta-diversity with the aim of utilising those patterns to inform mitigation of biodiversity loss through land management. In particular, I examine the potential conservation value of logged forest, taking into account spatial scale, continuous measures of forest quality and community composition of my study group.

1.10 Thesis structure

The structure of this thesis is as follows:

• Chapter 2. High conservation value scarabs are hit hardest by tropical deforestation.

In the first of my data chapters, I use a highly-diverse and charismatic group of beetles, the

scarabs (superfamily Scarabaeoidea), to question whether species richness declines as a

result of logging or of replacement of forest with oil palm plantation. I expand on my findings

by examining whether shifts in community composition are damaging to the conservation

value of scarab species in each of unlogged forest, logged forest and oil palm plantations. I

define “high conservation value” as being one of endemic to Borneo or of high interest to the

general public. Scarabs lend themselves well to such valuation, having at least some

distribution data available to most species (Schoolmeesters 2018) and being of significant

cultural importance (Leather et al. 2015). I discuss the implications of shifts in conservation

value of species for protection of forest area.

• Chapter 3. Forest quality, forest area, and the importance of beta-diversity for

protecting Borneo’s beetle biodiversity.

I expand on Chapter 2 by disregarding the conservation value of individual species and

instead incorporating measures of community composition into a diversity analysis focussed

on rove beetles (family: Staphylinidae). Rove beetles are a taxonomically and functionally

diverse family (Hammond 1992; Barton et al. 2011) that are highly abundant in tropical

rainforests (Bohac 1999) and therefore broadly representative of the ecosystem as a whole.

Using Jost’s (2007) equations, I calculate alpha-, beta- and gamma-diversity of rove beetles

across the land-use gradient extending from unlogged forest to heavily-logged forest. I relate

these indices to continuous measures of forest quality and forest area in order to link spatial

biodiversity change with disturbance.

• Chapter 4. Tropical deforestation impacts multiple scales of weevil beta-diversity.

I use a primarily-herbivorous group of beetles (Farrell et al. 2001; McKenna et al. 2009), the

weevils (superfamily: Curculionoidea), to identify how topography and vegetation structure

interact to influence trends in beta-diversity at three spatial scales. I decompose community

beta-diversity into balanced variation in composition and variation attributable to abundance

gradients using the methods by Baselga (2017) and compare between forest and oil palm

plantation. I then compare those trends to temporal patterns in abundance of the most

common weevil species.

• Chapter 5. Congruence and Importance to Conservation of Beta-Diversity in Borneo’s

beetles.

My final data chapter brings together data on rove beetles, scarabs and weevils to quantify

the relative importance of beta-diversity compared to alpha-diversity and suggest ways in

which beta-diversity can be applied in the planning of protected areas. I use variance

partitioning (Legendre & De Cáceres 2013) to attribute large-scale variation in community

structure to spatial turnover in community, vegetation structure, microclimate and topography.

I then examine how identifying sites of high local contribution to beta-diversity might be

incorporated into the assignment of High Conservation Value forest areas in Sabah.

• Chapter 6. Conclusions and Synthesis.

Lastly, I bring together all my findings on the relative biodiversity value of unlogged forest, the

importance of beta-diversity in degraded landscapes, the effects of environment on beta-

diversity, and the potential roles of beta-diversity in land management.

Chapter 2 - High conservation value scarab beetles are hit hardest by tropical deforestation

Adam C. Sharp1, Maxwell V. L. Barclay2, Arthur Y. C. Chung3 & Robert M. Ewers1

1Department of Life Sciences, Imperial College London, Silwood Park Campus, Buckhurst Road,

Ascot SL5 7PY, UK. 2Department of Life Sciences, Natural History Museum, Cromwell Road, London

SW7 5BD, UK. 3Forest Research Centre (Sepilok), Sabah Forestry Department, PO Box 1407,

Sandakan, Sabah 90715, Malaysia.

2.1 Abstract

Half of Borneo’s forest has been logged, and logging is often followed by conversion to oil palm plantation. While it has been reported that plantations support few species, there remains considerable debate around the value of logged forest to biodiversity. We examine changes in species richness and community structure of scarab beetles (superfamily: Scarabaeoidea) between unlogged forest, logged forest and oil palm plantation in Sabah, Malaysia. While logged forest retains similar richness of scarabs to unlogged, we detected significant declines in the proportions of some high conservation value species – those which are either endemic or which generated the greatest public interest. Oil palm plantations supported few species of scarab beetle, and those which persisted there were of lowest conservation and cultural ecosystem service values. Our findings highlight the benefits of conserving unlogged forest over logged forest and warn against relying solely on species richness to quantify diversity across land-use gradients.

2.2 Introduction

Borneo’s tropical forest biodiversity is threatened by rapid expansion of industrial logging and subsequent conversion of forest to oil palm plantation (Fitzherbert et al. 2008; Lewis et al. 2015;

Tsujino et al. 2016). Plantations support few species and are regarded as being of little conservation value (Fitzherbert et al. 2008; Edwards et al. 2010; Fayle et al. 2010). Logging without clearing can alter community composition without reducing richness (Hamer et al. 2003; Cleary et al. 2007;

Edwards et al. 2014b), and so logged forest often maintains ecological stability (Ewers et al. 2015) and many ecosystem services (Edwards et al. 2014b). Large expanses of logged forest are added to the protected area network in Borneo (Reynolds et al. 2011), but without assessing the conservation value of the individual species that persist in logged forest it is impossible to make a conclusion on the value of this habitat.

Endemic species are considered of high conservation value (Myers et al. 2000). Fewer endemic butterflies have been observed in logged forest compared to pristine forest (Cleary et al. 2009), but these findings are yet to be expanded to other taxa. Cultural ecosystem services (CES) are often neglected from biodiversity studies (Daniel et al. 2012). Both the existence of many different species and the aesthetic value of those species add to CES (Graves et al. 2017). Species can be important to people even if they are unfamiliar or foreign (Christie et al. 2006). Assessing the CES provided by modified landscapes requires recognition of species identity.

We examined whether logging and conversion of forest to oil palm plantation in Borneo is detrimental to endemism and CES in scarab beetles (superfamily: Scarabaeoidea). Scarabs were an ideal study group for their large taxonomic and functional diversity (Ahrens et al. 2014), distribution data for many species (although unlikely complete; Schoolmeesters 2018), cultural importance (Leather 2015) and proven utility as an indicator group (Edwards et al. 2014a). They frequently appear in science communication and provoke interest from audiences ranging from young children to insect collectors.

We hypothesised that there would be fewer scarab species in plantation compared to forest. Endemic species often have narrower habitat niches (Essl at al. 2009; Theuerkauf et al. 2017) so we predicted that any modification of complex unlogged forest would reduce the diversity of endemic species. We

also predicted that scarab species contributing strongly to CES would be those with charismatic morphological adaptations that may be selected against through forest disturbance.

2.3 Methods

Scarabs were collected at the Stability of Altered Forest Ecosystems (SAFE) Project in Sabah,

Malaysian Borneo (Ewers et al. 2011). Sampling occurred in February 2011, November/December

2011, and June/July 2012 at sites positioned within the SAFE Project hierarchical design of 17 sampling blocks (Marsh & Ewers 2013). There were three “control” blocks in each of unlogged forest, logged forest and oil palm plantation. Control blocks had the same arrangement of sampling sites, with nine in each block, and covered approximately 0.2 km2 in area. Unlogged control blocks were within Maliau Basin Conservation Area (MBCA). Logged forest control sites were logged extensively in the 1970’s (removing 113 m3/ha) and again in the early 2000’s (removing a further 66 m3/ha,

Struebig et al. 2013). Plantation sites comprised closed or nearly-closed canopy habitat at the time of sampling. The remaining eight blocks comprised similar logged forest to the logged control block but differed in arrangement of sampling sites (a further 112 in total). All sites were separated by over 175 m, and all forest sites were connected within an expanse of continuous forest of over one million hectares in area.

Three insect traps were set surrounding every site during each trapping period for three days. The traps were positioned in a triangle with sides of approximately 56 m. Traps comprised a malaise-pitfall combination and were designed to capture beetles of various morphology and behaviour. The pitfall section comprised a plastic funnel of approximately 25 cm diameter dug flush into the ground over an ethanol-filled collection bottle. Protruding vertically from the funnel was a perspex flight-intercept trap of around 1 m2 area, channelling flying insects either downwards into the pitfall or upwards into a malaise net with a second ethanol-filled collection bottle. The traps were not baited.

Beetles were identified to species where possible and morphospecies within genera where not. From the nine control blocks, total species richness was extrapolated using five estimators: bootstrap

(Smith & van Belle 1984), Chao-2 (Chao 1987), Incidence-based Coverage Estimator (ICE, Lee &

Chao 1994) and first-order and second-order jackknife (Heltshe & Forrester 1983). Only the control blocks were used to calculate richness because they accounted for the same spatial turnover in species, but the other eight blocks were used in all further analysis. Incidence-based estimators were used because of the differences in detectability between scarab species; they accounted to some extent for some species existing solitarily and others in large numbers or groups. Those estimators also partially accounted for possible within-taxon differences in behaviour between land-uses, perhaps due to altered habitat complexity or reduced canopy height. For each estimator, species richness differences between land-uses were tested using a linear mixed-effect model with sampling period included as a random intercept.

We derived species endemism using range distributions retrieved from the online Catalogue of Life

(Schoolmeesters 2018). Species were categorized as endemic to Sabah, Borneo, Sundaland (land masses of the Sunda shelf) or nowhere. Proportional odds models were employed to identify shifts in the ratios of endemic species trapped between sites in unlogged forest, logged forest and plantation.

Species’ contribution to cultural ecosystem services were assigned using survey data from 100 children aged four to 16 visiting the Natural History Museum in London. Children were ideal participants for their lack of prior knowledge on ecology or entomology. A glass display case was prepared every day for five days with one randomly-selected beetle specimen representing each observed Scarabaeoidea genus. Specimens were mounted on pins and displayed in identical white trays. Specimens were shuffled in the case between surveys. Children were informed about the ecology of scarab beetles, allowed to ask their own questions, and then asked to choose the three beetles they found “most interesting”. Where multiple children were present, votes were cast by just one per presentation to ensure independent responses. Votes for each genus were allocated to each species and morphospecies in that genus, and then equally-sized “low”, “medium” and “high” interest categories were assigned. Shifts in interest category were identified using proportional odds models.

Differences in richness, endemism and public interest levels between land-uses were related to habitat structure using redundancy analysis (RDA). Community data was treated as presence/absence. Habitat variables were collected by Pfeifer et al. (2016) between 2010-2011 and were leaf litter depth (mm), mass of dead wood (kg), vegetation ground cover (%), above-ground 27

biomass of trees (AGB, Mg/ha), soil depth (mm), slope (°), canopy vine cover (%), and ambient temperature (°C) and humidity (%). Variables were added via forward selection to the RDA with significance determined via permutation of residuals.

2.4 Results

A total of 683 scarab beetles of 115 species or morphospecies were caught over the three periods: 95 from the Scarabaeidae, 13 from the Hybosoridae, five from the Lucanidae, one from the Geotrupidae and one from the Passalidae (Table 2.1, full dataset available online at www.safeproject.net). We identified 87 percent of taxa to species. The morphospecies belonged to the genera Maladera,

Panelus or Haroldius (all Scarabaeidae).

Habitat type explained a significant amount of variance in the estimated total richness of scarabs per control block (P < 0.001, Fig. 2.1, Table 2.2). All five estimators concluded no significant difference in total species richness between unlogged and logged forest control sites (P > 0.050) and significantly fewer species in oil palm plantation control sites relative to logged forest (P < 0.001). The second- order jackknife estimates were closest to mean values over all estimators and were 20.1 species

(95% CI: 10.4-38.4) in unlogged forest, 17.0 species (95% CI: 9.7-30.1) in logged forest, and 2.4 species (95% CI: 1.3-4.4) in oil palm plantation.

There were significant declines in both endemism and interest category both between unlogged and logged forest (P < 0.050) and logged forest and plantation (P < 0.001, Fig. 2.2, Table 2.3). Endemism and interest categories were not significantly correlated (Spearman’s ρ = -0.03, P > 0.050). When combined with second-order jackknife estimates of total scarab species richness, logged forest preserves just 69 percent of the Borneo-endemic species richness from unlogged forest, and just 66 percent of high-interest species. In contrast, logged forest harbours a greater number, 122 percent, of the unlogged richness of non-endemic species and 135 percent of the richness of low-interest species. We did not observe any Borneo-endemic scarabs in oil palm plantation, nor high-interest species.

The RDA explained a significant amount of variance in community structure (F3,13 = 1.69, P < 0.010,

2 R = 0.28, Fig. 2.3). Leaf litter depth (F1,13 = 1.81, P < 0.050), AGB (F1,13 = 1.66, P < 0.050) and

ambient temperature (F1,13 = 1.59, P < 0.050) were each selected for inclusion. The first RDA axis explained a significant amount of variance (F1,13 = 2.61, P < 0.050), but the second (F1,13 = 1.53, P >

0.050) and third did not (F1,13 = 0.93, P > 0.050). Land-use categories were separated across axis 1

(F2,14 = 5.71, P < 0.050) by AGB in the positive direction and litter depth in the negative. Categories were also separated across axis 2 (F2,14 = 28.88, P < 0.001) by AGB in the positive direction and temperature in the negative.

2.5 Discussion

Our results indicate that live woody biomass is important for maintaining natural scarab community structure. Species feeding on vegetation appeared to be most impacted by logging. Of the four

Borneo-endemic species observed only in unlogged forest, two, Dasyvalgus pusio and

Pseudohomonyx borneensis, were flower-feeding. Similarly, of the six high-interest species observed only in unlogged forest, four were either flower- or foliage-feeding, eg Ixordia pseudoregia and

Callistethus maculatus. We hypothesise that endemic vegetation-feeders may lose their specific food resource during logging. We also hypothesise that high-interest vegetation-feeders (mostly large or colourful in morphology) require frequent flights between food resources and may be more conspicuous to predators after the removal of woody biomass.

Species that were present in logged forest but not observed in unlogged forest were predominantly associated with vertebrates. Of the 15 species observed only in logged forest with geographic ranges extending beyond Sundaland, ten were dung-feeding and included six species of Onthophagus.

Furthermore, a single species of carrion-feeder, Phaeochroops giletti, was observed in unlogged forest, while three more species were caught in logged forest. All the carrion-feeding species were of low public interest. It can be inferred that changes to abiotic habitat structure (eg increased leaf litter) coupled with increased abundance in several vertebrate taxa (Ewers et al. 2015) allows the introduction of wide-ranging and generalist species with limited value to CES. We do note, however, that those species may contribute strongly to other important ecosystem services, including carrion removal.

In line with past studies (Edwards et al. 2010; Fayle et al. 2010), we found that oil palm plantations supported few species and were of little conservation value. We attribute this to harsh microclimate and comparatively low biomass of vegetation.

While these broad conclusions remain relevant, there exists a certain amount of uncertainty in our findings because of the noticeable under-sampling in our data. Many scarab species were detected only once or a few times (Table 2.1), and we identified a high number of species (n = 115) from a modest number of individuals (n = 683). As a result, it is impossible to determine true absences of individual species from habitat categories. However, we emphasise that our inferences at ecosystem- level remain valid because of the high number of species included in this study.

Due to the high proportions of endemics and taxa that were ranked most interesting to the public, we conclude that unlogged forest is irreplaceable for high levels CES and high numbers of endemic species. While the high scarab richness in logged forest supports previous findings that these modified landscapes are ecologically stable (Ewers et al. 2015), at least some unlogged forest must remain to protect species which are endemic or highest CES value. Future land-use change studies must consider more than taxa richness alone if we are to successfully conserve all aspects of biodiversity and ecosystem services.

Table 2.1. Summary of scarab taxa caught over the three sampling periods. Genera with a plus symbol contained morphospecies. Interest category was not derived for the genus Haroldius, as the two specimens we collected were unavailable at the time of the survey.

Endemism (N species) Interest Genus

Not Sundaland Borneo Sabah Votes Category

Geotrupidae:

Bolbochromus 1 0 0 0 2 Low

Hybosoridae:

Cyphopisthes 0 1 0 0 10 High

Madrasostes 0 3 0 3 3 Medium

Microphaeochroops 0 0 1 0 2 Low

Phaeochridius 0 0 1 0 0 Low

Phaeochroops 2 0 0 0 0 Low

Phaeochrous 1 0 0 0 2 Low

Pterorthochaetes 0 1 0 0 4 Medium

Lucanidae:

Aegus 1 2 2 0 9 High

Passalidae:

Pelopides 1 0 0 0 29 High

Scarabaeidae:

Adoretus 0 0 0 1 1 Low

Apogonia 0 2 7 0 4 Medium

Callistethus 0 1 0 0 21 High

Catharsius 0 0 1 1 13 High

Clinterocera 0 0 1 0 1 Low

Copris 1 0 0 0 4 Medium

Dasyvalgus 0 1 1 0 1 Low

Haroldius+ 0 0 0 0 - -

Ixorida 0 1 0 0 18 High

Lepadoretus 1 0 0 0 1 Low

Maladera+ 1 0 1 0 0 Low

Mericserica 1 0 0 0 1 Low

Microcopris 1 0 0 1 0 Low

Microserica 0 0 1 0 2 Low

Mimela 0 1 0 0 26 High

Nematophylla 0 0 0 1 4 Medium

Ochicanthon 0 0 3 1 4 Medium

Octoplasia 0 1 0 0 18 High

Oniticellus 1 0 0 0 0 Low

Onthophagus 12 10 7 4 9 High

Panelus+ 0 0 1 0 24 High

Paragymnopleurus 1 2 0 0 8 Medium

Proagoderus 0 0 0 1 7 Medium

Pseudohomonyx 0 0 2 0 6 Medium

Rhyparus 0 0 2 0 10 High

Saprosites 1 0 0 0 2 Low

Sisyphus 1 0 0 0 2 Low

Termitodiellus 0 0 1 0 3 Medium

Trichogomphus 0 1 0 0 49 High

Table 2.2. Summaries of linear mixed models predicting log10-transformed estimators from each land-

use category with sampling period as a random intercept. Model significance is derived from chi-

square test of variance explained against a null model with only random intercept for sampling period.

Unlogged forest and oil palm plantation estimates are presented relative to the logged forest estimate,

so P-values for these rows indicate difference from logged forest. The degrees of freedom (DF) used

in calculating significance of parameters are estimated by Satterthwaite's method.

Estimator Model Parameter Estimate SE DF t P

Bootstrap Intercept (Logged Forest) 1.23 0.13 3.24 9.48 < 0.010

χ2 = 27.04, DF = 2 Unlogged Forest 0.07 0.15 17.87 0.48 > 0.050

P < 0.001 Oil Palm Plantation -0.84 0.14 17.15 -6.10 < 0.001

Chao-2 Intercept (Logged Forest) 1.39 0.14 4.18 10.12 < 0.001

χ2 = 29.72, DF = 2 Unlogged Forest 0.34 0.19 18.33 1.85 > 0.050

P < 0.001 Oil Palm Plantation -1.01 0.17 17.23 -5.80 < 0.001

ICE Intercept (Logged Forest) 1.11 0.11 3.46 9.78 < 0.01

χ2 = 28.63, DF = 2 Unlogged Forest 0.19 0.14 18.01 1.37 > 0.050

P < 0.001 Oil Palm Plantation -0.77 0.13 17.12 -5.91 < 0.001

First-Order Jackknife Intercept (Logged Forest) 1.19 0.12 3.33 9.69 < 0.01

χ2 = 27.49, DF = 2 Unlogged Forest 0.10 0.14 17.92 0.68 > 0.050

P < 0.001 Oil Palm Plantation -0.81 0.13 17.14 -6.07 < 0.001

Second-Order Jackknife Intercept (Logged Forest) 1.23 0.12 3.45 9.90 < 0.010

χ2 = 27.35, DF = 2 Unlogged Forest 0.07 0.15 17.98 0.49 > 0.050

P < 0.001 Oil Palm Plantation -0.85 0.14 17.17 -6.13 < 0.001

Table 2.3. Model summaries for proportional odds models predicting proportions of endemic and high-

interest scarab species in each land-use category. Probit link-functions were chosen over others for

lowest AIC value. Threshold parameters describe the transition point in probit-scale at which there is

a switch from one category to the next in sequence for logged forest. Unlogged forest and oil palm

plantation parameters describe the extent each of those cut-offs are shifted by (equally) given a land-

use category different to logged forest. Overall model significance was calculated by likelihood-ratio

test against a null model with threshold parameters which do not shift in response to land-use

category. As there were differing numbers of species observed across the land-use gradient,

weighting was assigned to grant equal leverage to each land-use category in models.

Model Parameter Estimate SE Z P

Endemism Threshold: Not|Sundaland -0.54 0.11

LR = 114.38, DF = 2 Threshold: Sundaland|Borneo 0.01 0.11

P < 0.001 Threshold: Borneo|Sabah 0.97 0.11

Unlogged Forest 0.28 0.14 2.06 < 0.050

Oil Palm Plantation -1.30 0.16 -8.18 < 0.001

Interest Threshold: Low|Medium -0.66 0.11

LR = 223.99, DF = 2 Threshold: Medium|High 0.09 0.10

P < 0.001 Unlogged Forest 0.34 0.14 2.36 < 0.050

Oil Palm Plantation -2.09 0.19 -11.06 < 0.001

Figure 2.1. Estimated total richness of scarab beetles within each control block (unlogged forest, UF, logged forest, LF, and oil palm plantation, OP) according to each of the five estimators. The 95 percent confidence intervals presented were derived via 10,000 bootstrap samples from the relevant mixed model.

Figure 2.2. Estimated richness (R) and proportions of endemic (E) and high-interest (I) scarabs in each of unlogged forest (UF), logged forest (LF) and oil palm plantation (OP). Total richness is estimated via second-order jackknife. Within each bar, portions of endemism and interest categories derived from proportional odds models are overlaid. Significant (P < 0.050) reduction in total richness is detected only between logged forest and plantation, while significant reduction in endemism and interest is detected between unlogged and logged forest, and logged forest and plantation.

Figure 2.3. RDA describing differences in community structure between unlogged forest (UF), logged forest (LF), and oil palm plantation (OP). The ellipse for each land-use category represents 95 percent confidence around the true mean location and is derived from standard error.

Chapter 3 - Forest quality, forest area and the importance of beta-diversity for protecting Borneo’s beetle biodiversity

Adam C. Sharp1, Maxwell V. L. Barclay2, Arthur Y. C. Chung3, Guillaume de Rougemont4, Edgar C.

Turner5 & Robert M. Ewers1.

1Department of Life Sciences, Imperial College London, Silwood Park Campus, Buckhurst Road,

Ascot SL5 7PY, UK. 2Department of Life Sciences, Natural History Museum, Cromwell Road, London

SW7 5BD, UK. 3Forest Research Centre (Sepilok), Sabah Forestry Department, PO Box 1407,

Sandakan, Sabah 90715, Malaysia. 4Oxford University Museum of Natural History, University of

Oxford, Parks Road, Oxford OX1 3PW, UK. 5Department of Zoology, University of Cambridge,

Downing Street, Cambridge CB2 3EJ, UK.

3.1 Abstract

The lowland forest of Borneo is threatened by rapid logging for timber export and clearing for the expansion of timber and oil palm plantations. This combination of processes leaves behind landscapes dotted with small, often heavily-degraded forest fragments. The biodiversity value of such fragments, which are easily dismissed as worthless, is uncertain. We collected 187 taxa of rove beetles across a land-use gradient in Sabah, Malaysia, spanning pristine tropical lowland forest to heavily-degraded forest. Using these data, we identified shifts in alpha-, beta-, and gamma-diversity in response to forest quality and distance, then applied our findings from continuous expanses of forest to make predictions on hypothetical forest areas. We found that maintaining high forest quality is more important than forest area for conserving rare taxa (those important for conserving biodiversity per se), and that very small areas (10 ha) are likely to harbour the same richness of rove beetles as larger areas (100 ha) of equal forest quality. We estimate a decline in richness of 36% following heavy logging (removal of 95% of the vegetation biomass) from a forest area of 100 ha or less. Maintaining

large forest area as well as high forest quality is important for conserving community composition, likely to be more important for conserving ecosystem functioning. We predict a decline of 35% in community diversity in conversion of a 100 ha area of unlogged forest to a 10 ha area of heavily- logged forest. Despite significant declines in alpha-diversity, beta-diversity within small rainforest areas may partially mitigate the loss of gamma-diversity, reinforcing the concept that beta-diversity is a dominant force determining the conservation of species in fragmented landscapes. In contrast to previous findings on larger animals, our results suggest that even small fragments of degraded forest might be important reservoirs of invertebrate biodiversity in tropical agriculture landscapes. These fragments, especially of lightly-logged forest, should be conserved where they occur and form an integral part of management for more sustainable agriculture in tropical landscapes.

3.2 Introduction

Borneo’s forest is amongst the most biodiverse on the planet (Kier et al. 2005; de Bruyn et al. 2014), but that forest is threatened by rapid logging and clearing for the expansion of plantations (Fitzherbert et al. 2008; Lewis et al. 2015; Tsujino et al. 2016). Plantations support little biodiversity in comparison to adjacent natural habitats (Chung et al. 2000; Edwards et al. 2010; Fayle et al. 2010). While most studies agree that unlogged forest supports the greatest amount of biodiversity (Gibson et al. 2011), the ongoing global expansion of logging and fragmentation (Taubert et al. 2018) requires us to also consider the value of degraded forest to conservation. This is especially important in the Malaysian

Borneo state of Sabah, where large amounts of logged forest have been progressively added to the protected area network (Reynolds et al. 2011).

Degraded forest can protect a large proportion of natural biodiversity (Edwards et al. 2011; Wearn et al. 2017). Modified forest is often highly variable in disturbance and structure (Burivalova et al. 2014;

Pfeifer et al. 2016), yet many studies on Borneo’s biodiversity have been unable to account for that variation and thus treat logged and unlogged as discrete forest types. Habitat quality needs to be considered on a continuous scale to understand the subtle responses of biodiversity to disturbance

(Edwards et al. 2011; Burivalova et al. 2014). The same is true for fragmentation - remnant forest fragments support a greater level of biodiversity than surrounding plantation (Edwards et al. 2010),

and this biodiversity value increases with fragment size (Struebig et al. 2011). It is difficult, however, to directly quantify independent effects of forest quality and fragment size because the two are often interlinked (Tawatao et al. 2014; Taubert et al. 2018). Understanding how forest quality and area together impact biodiversity might highlight low-effort strategies for maintaining greater levels of natural diversity in agriculture-dominated landscapes.

To add further complexity to this knowledge-gap, diversity itself is comprised of three components that could all be impacted individually by forest degradation. Diversity at any one point is alpha-diversity and diversity between points is beta-diversity. Total diversity of an area is gamma-diversity (Whittaker

1972; Magurran 2004). Beta-diversity is often neglected from studies of logging and oil palm plantation impacts on biodiversity (but see Hamer & Hill 2000), yet is particularly relevant for a spatial problem like forest fragmentation (Tscharntke et al. 2012) and in highly diverse tropical forests where beta-diversity can be the largest contributor to gamma-diversity (Beck et al. 2012).

Previous work has examined the responses of beta-diversity to logging, and found that logged forest is highly beta-diverse (Hamer & Hill 2000; Pfeiffer & Mezger 2012; Kitching et al. 2013; Wearn et al.

2016). However, where beta-diversity has been quantified, it is often in metrics that are either directly dependent on alpha-diversity (Jost 2007) or dissimilar in scale (Benedick et al. 2006; Lucey et al.

2014). This not only prohibits the comparison of alpha-diversity with beta-diversity, but also invalidates calculation of beta over gradients in alpha (Jost 2007). We remain, therefore, largely uninformed regarding the independent response of beta-diversity to conversion of Borneo’s forest, despite the fact that beta-diversity is probably the largest contributor to forest-level gamma-diversity

(Beck et al. 2012).

We aimed to quantify the independent impacts of forest quality and distance between sample sites on each component of biodiversity, paying close attention to how alpha- and beta-diversity shape gamma-diversity. To achieve this, we chose to study rove beetles (order: Coleoptera, family:

Staphylinidae). Rove beetles were ideal for their high taxonomic and functional diversity (Barton et al.

2011), high abundance in tropical rainforests, and environmental sensitivity (Bohac 1999). While it is never possible to quantify biodiversity change in every taxon present, focussing an analysis on the rove beetles alone should provide conclusions of broad relevance to the wider ecosystem. We 41

hypothesised that beta-diversity would make a larger contribution to the gamma-diversity of forest areas than alpha-diversity, and that beta-diversity would increase with distance and disturbance. We use our findings to predict diversity reductions that would result from different scenarios of forest degradation and conversion in Borneo. From these predictions, we generate rules of thumb that can be applied to inform best management practice in preserving tropical biodiversity within predominantly agricultural landscapes.

3.3 Methods

3.3.1 Study area

This study was conducted at the Stability of Altered Forest Ecosystems (SAFE) Project (Ewers et al.

2011) in Sabah, Malaysian Borneo (Fig. 3.1). The project utilises a fractal sampling pattern that was designed specifically to study diversity at multiple spatial scales (Ewers et al. 2011; Marsh & Ewers

2013). The SAFE Project comprises 7,200 ha of forest that was unevenly logged once in the 1970’s

(removing 113 m3 ha-1) and again between 2000 and 2008 (removing 66 m3 ha-1, Struebig et al.

2013), resulting in a logged forest area with widely varying forest quality (Pfeifer et al. 2016). Sample sites were spread across 10 discrete blocks (Ewers et al. 2011). One sampling block was in primary forest at Maliau Basin Conservation Area (MBCA), with two-thirds of sites in forest that has never been modified and one-third of sites in forest that has undergone light logging to obtain timber for construction of the adjacent field centre. A second sampling block was in a continuous expanse of logged forest. A third block was in a Virgin Jungle Reserve that had experience illegal logging around the periphery, and a further seven blocks were in or adjacent to the logged forest of the SAFE area

(Fig. 3.1). Within each block, sample sites were clustered at three spatial scales. Three sites were separated at second-order scale by approximately 102.25 m, with those clusters of sites separated at third-order scale by 102.75 m and again at fourth-order scale by 103.25 m. At the time of sampling, all locations were connected as part of a large, continuously forested area that extends across more than one million hectares (Reynolds et al. 2011).

3.3.2 Insect sampling

We sampled insects at up to 166 second-order sites twice in 2011, once in February and again in

November/December. Three insect traps were positioned around each site and were separated by

101.75 m (first-order scale). Traps were a design combining pitfall (diameter approximately 25 cm), flight-interception (surface area approximately 1 m2) and malaise traps and left active for three days.

Insects were directed either upwards or downwards into collection bottles filled partially with 70% ethanol, and trap samples were combined at each site. Rove beetles were identified to the lowest taxonomic level possible, but the abundant subfamily Aleocharinae was removed from analysis as it proved impossible to identify them to any meaningful level.

3.3.3 Calculating diversity metrics

Most studies on forest conversion in Borneo employ species richness to examine diversity changes and are useful in assessing the vulnerability of threatened or charismatic taxa. However, taxa rarely coexist in equal numbers (Hubbell 2001) so relying solely on taxa richness to quantify diversity disregards community composition (measures of the identity and abundance of different taxa,

Magurran 2004) which can change independently to species richness (Beck et al. 2006; Banks-Leite et al. 2012). It is important to consider both in biodiversity management to protect both rare or vulnerable taxa alongside the common taxa that likely contribute the majority of ecosystem function

(Slade et al. 2011; Winfree et al. 2015). In recognition of this, we chose to calculate diversity using the equations derived by Jost (2007). These metrics generate independent measures of alpha- and beta- diversity, and allow the examination of trends in both taxa richness and community composition.

Each of alpha-, beta- and gamma-diversity were calculated in three ways: weighting in favour of rare taxa, weighting all taxa equally, and weighting in favour of common taxa, corresponding to q = 0, 1 and 2 respectively in Jost’s diversity equations. When q = 0, alpha-diversity is equivalent to mean taxa richness across a set of sample sites, and gamma-diversity is equivalent to total taxa richness for all sample sites combined. When q = 1, diversity measurements are equivalent to Shannon indices and, when q = 2, diversity measurements are equivalent to Simpson’s indices. These three weightings

encompass a gradient of arguments for the conservation of biodiversity. A weighting of q = 0 treats all species equally, regardless of abundance, and is applicable to arguments based on conserving species richness and biodiversity per se. By contrast, q = 2 weights in favour of common species – those most likely to be dominating the ecology of the ecosystem (Slade et al. 2011; Winfree et al.

2015) – and is applicable to arguments about conserving ecosystem function. Only sites which caught at least one rove beetle were used, as some diversity metrics are undefined when gamma-diversity is

0 (Jost 2007).

3.3.4 Relating insect trap data to forest distance

Examining the rove beetles caught at any one site would have limited our study to alpha-diversity alone. We chose to randomly group multiple sites together in order to develop estimates of beta- diversity and gamma-diversity. Distance between sites within a group was used as a proxy for forest area and related to each of beta- and gamma-diversity. Combinations of three sites were used as the simplest method of generating 2-dimensional shapes from second-order sites. All combinations were nested within blocks and sampling periods.

3.3.5 Defining distance and forest quality variables

Each three-site combination was assigned continuous forest quality and between-site distance values. Quality was quantified by Pfeifer et al. (2016), who calculated estimates of above ground biomass (AGB) from 25 x 25 m (0.0625 ha) vegetation plots at each of the 166 second-order sites.

High levels of disturbance were characterised by low AGB, while primary forest sites had the highest

AGB. We estimated, from Pfeifer et al. (2016), that the mean unlogged above-ground biomass of vegetation was 524 ± 54 Mg/ha. We exploited the hierarchical nature of the fractal sampling design to assign a forest quality value to each three-site combination that was relevant to spatial scale.

Combinations where the three second-order points were clustered around the same third-order point were given the mean AGB value of all second-order points clustered around that third-order point; the same approach was extended to the fourth-order. Three-site combinations over greater than fourth- 44

order scale were given the mean AGB value of the entire sampling block. Distance between sites was calculated as the total centroid distance formed by the three chosen points (Fig. 3.2), and ranged from

400 m to 2,200 m.

3.3.6 Developing bootstrap models

To accurately model the response of rove beetle diversity to AGB and centroid distance, we had to group sites into combinations of three many times rather than just once. Within a classical linear model, this process would require resampling a single site to generate multiple data points, and therefore introduce a high level of pseudoreplication. We combatted this problem by randomly selecting 40 three-site groups in iteration, and fitting linear models predicting each of alpha-, beta- and gamma-diversity using q = 0, q = 1 and q = 2 within each iteration (nine models per iteration). All models except gamma-diversity where q = 0 were linear mixed models with log-transformed diversity as the response, AGB and distance and their interaction terms as fixed effects, and sampling period as a random intercept. The gamma-diversity/q = 0 models were generalized linear models with poisson error family, log-link function and an observation-level random intercept to account for overdispersion (Harrison 2014). All explanatory variables were log-transformed. Within any one iteration, no site was used more than once to avoid pseudoreplication, and we were able to sample from 6,684 possible three-site combinations. Overall model parameter estimates were taken as the mean of respective estimates from each iteration and were therefore effectively bootstrapped.

There was a weak positive correlation between sampling block area and AGB at the SAFE Project which manifested itself in false relationships between alpha-diversity and centroid distance. Alpha- diversity must be independent of distance or area (Whittaker 1972), and so we forced this independence on our bootstrapped dataset. This was achieved by randomly generating 100,000 iterations of 40 groups prior to analysis, fitting models of alpha-diversity within each of those iterations, and then selecting for analysis the 10,000 of those 100,000 iterations where the sum of squares of distance parameter estimates on alpha-diversity was minimised.

For each of the 10,000 iterations, model selection was achieved using backward selection of the terms AGB, AGB2, distance, and their interactions according to the rule of marginality. Applying Z- tests to determine whether mean parameter estimates differed significantly from 0 resulted in selection of all possible terms for beta- and gamma-diversity because of the large number of iterations we were able to perform (high number of iterations = low standard error and high Z-value). To increase the rigour of our model selection, we added the condition that effect sizes of each parameter estimate must be greater than 0.2 (a small effect according to Cohen, 1969, p.23).

3.3.7 Creating nominal forest areas

To translate our results - generated from samples collected in continuous forest areas - to recommendations about the potential impact of forest fragmentation in agricultural landscapes, we extrapolated forest area from the three-site combinations (Fig. 3.2). We considered the area inside each three-site combination as being equivalent to the core of a forest fragment; i.e. the part of a forest fragment that is not influenced by the edge between forest and its surrounding matrix habitat.

Work elsewhere has demonstrated that 100 m is the approximate extent to which forest edges influence diversity and abundance of tropical leaf-litter invertebrates (Laurance et al. 2002), and 80% of beetle species have edge effect extents of 100 m or less (Ewers & Didham 2008). We therefore added a 100 m buffer around the periphery of each three-site combination and refer to this total area as nominal forest area. Our three-site combinations ranged in nominal forest areas from approximately 10 ha to 100 ha. This represents an appropriate range for study because more than

90% of remnant fragment fragments following clearing of tropical forests are smaller than 100 ha

(Ranta et al. 1998; Ribeiro et al. 2009). The most recent trends in global deforestation show that forest fragments are ever-decreasing in area (Taubert et al. 2018).

We quantified the shape of each nominal forest area using the Shape Index (SI) derived by Patton

(1975) and adapted for metric units (Didham & Ewers 2012). A value of SI = 1.0 represents a perfect circle, and greater values are progressively more elongate. Because the effects of forest fragment shape are likely dependent on the surrounding matrix, and because we were unable to simulate the

complex shapes of fragments observed in real-world landscapes (Ewers & Didham 2008) using simple triangles, we chose to control for shape instead of including this variable in our models and so limited our selected forest areas to those with SI < 1.4.

Three further bootstrap models (one for each value of q) were fitted to predict gamma-diversity in the nominal forest areas from their size and forest quality. As before we generated 100,000 random iterations of 40 three-site combinations (from 4,593 possible combinations). We assigned each combination a value of AGB as before, but calculated the nominal forest area for each three-site combination instead of centroid distance. From these we selected the 10,000 iterations which minimised parameter estimates of nominal forest area on alpha-diversity, and fitted bootstrap models of gamma-diversity alone using the same methods to which we fitted our centroid distance models.

We used these final gamma-diversity models to produce estimates of proportional change in total rove beetle diversity given proportional change in forest quality or nominal forest area.

3.4 Results

A total of 11,352 rove beetles were caught over the two sampling periods, of which 2,905 belonged to subfamilies other than Aleocharinae. There were 1,138 beetles belonging to the Staphylininae, 1,083 to the Oxytelinae, 534 to the Paederinae, 87 to the Osoriinae, 37 to the Tachyporinae, 14 to the

Euaesthetinae, six to the Omaliinae and six to the Steninae. From those eight subfamilies, there were

187 reproducible taxonomic units, of which 40 were named species accounting for 618 individuals. A further 143 unnamed taxa (n = 1,674) were known to be separate species of non-aleocharine rove beetle. The remaining 613 individuals were identifiable to one out of the genera Anotylus (n = 315),

Mitomorphus (n = 82) or Thinocharis (n = 71) or the tribe Xantholini (n = 145, full dataset available online at www.safeproject.net). A total of 90 sites caught non-aleocharine rove beetles in the first sampling period and 144 in the second. The number of non-aleocharine rove beetles caught per site increased with AGB (Fig. 3.3).

Where q = 0 (taxa richness, Table 3.1), alpha-diversity at the lowest measured AGB was approximately 30% of that at primary forest AGB (524 Mg/ha). There was a peak in alpha-diversity at

around 200 Mg/ha; equivalent to the highest values of AGB measured in logged forest (Fig. 3.4a).

The bootstrap beta-diversity model for q = 0 (pseudo-R2 = 0.44) explained more variation than alpha- diversity (pseudo-R2 = 0.09). Beta-diversity was highest at low AGB and was greater than alpha- diversity in heavily-logged forest (AGB < 57 Mg/ha). There was a small interaction effect between

AGB and centroid distance whereby beta-diversity increased with distance at highest forest quality

(Fig. 3.4d). Th effect of distance was small enough that it was not selected in our models of gamma- diversity where q = 0, and only AGB influenced gamma-diversity (Fig. 3.4g). As with alpha-diversity, gamma-diversity was greatest at around 200 Mg/ha.

AGB had a similar effect size on alpha-diversity where q = 1 (individual-weighted diversity, Table 3.1) compared to q = 0, and peaked around 200 Mg/ha (Fig. 3.4b). Alpha-diversity for q = 1 at lowest AGB was around 41% of that of primary forest AGB. Our beta-diversity model for q =1 explained approximately the same amount of variation as the alpha-diversity model (pseudo-R2 = 0.09 and 0.11 respectively). Beta-diversity increased with centroid distance across all values of AGB (albeit a small effect) and was highest at low AGB and high distance (Fig. 3.4e). Gamma-diversity where q = 1 increased with both AGB and centroid distance, with AGB having a far stronger effect, and was highest again at intermediate forest quality (Fig. 3.4h).

The models of diversity where q = 2 explained very little variation in the data (Table 3.1). We detected very small shifts in alpha-diversity, which was highest in logged forest (Fig. 3.4c), and beta-diversity, which increased slowly with centroid distance (Fig. 3.4f). As a result, gamma-diversity for q = 2 peaked in logged forest and increased with centroid distance (Fig. 3.4i), but the effect of AGB was stronger than that of distance.

We estimated proportional loss of gamma-diversity for each of q = 0, q = 1 and q = 2 in response to proportional losses in vegetation biomass and nominal forest area (Table 3.2). Where q = 0 (taxa richness), gamma-diversity after heavy logging (95% AGB removed) was reduced by 36% compared to gamma-diversity in an unlogged 100 ha forest area (Fig. 3.5a). Gamma-diversity for taxa richness also peaked in lightly-logged forest (66% AGB removed) at 164%. Nominal forest area had no effect on total taxa richness. Forest area did affect gamma-diversity where abundance was considered (q =

1 and 2), but this effect was smaller than that of AGB (Fig. 3.5b-c). Reducing a 100 ha nominal forest 48

area to a 10 ha area while maintaining forest quality would result in a loss of 10% of the q = 1 gamma- diversity and 9% of the q = 2 gamma-diversity. Combining this reduction in forest area with heavy logging (removal of 95% of AGB/ha) would result in a 35% loss of gamma-diversity where q = 1 and

31% loss of gamma-diversity where q = 2.

3.5 Discussion

Preserving both high forest quality and protecting as much forest area as possible are both important to the conservation of biodiversity in Borneo. Our data for rove beetles – an abundant and ecologically important group of invertebrates (Bohac 1999) – demonstrate strongly contrasting responses to forest quality and forest area at the spatial scales relevant to heavily fragmented landscapes, with clear implications for forest management in the region. Our data indicate that forest quality is more important than area in propagating biodiversity at these small scales, but even the smallest, most- degraded forest areas may maintain a relatively large amount of gamma-diversity through shifts in beta-diversity.

In unlogged forest, high taxa richness (q = 0) is a result of there being many taxa at any given point

(high alpha-diversity). In a heterogeneous, unlogged landscape, increasing sample area is likely to encompass a greater array of habitat types and their associated specialist taxa (Báldi 2008), explaining the small increase in richness beta-diversity with distance in high-quality forest. This did not, however, translate into an increase in gamma-diversity with nominal forest area. Other studies on forest fragments in SE Asia have found that larger fragments harboured greater invertebrate richness

(Benedick et al. 2006; Lucey et al. 2014), but these analyses all contained data extending across a much greater range of fragment areas (over 123,000 ha and 500 ha respectively) than our largest nominal forest are (100 ha). It is possible that the spatial scale of forest areas we examined was not large enough to detect area-mediated shifts in gamma-diversity where q = 0. Reconciling previous results with ours would suggest that fragments greater in size than our maximum nominal forest area

(100 ha) have large increases in within-fragment beta-diversity that go beyond the area-related increases we have observed in relatively small areas. However, our results suggest that in forest

areas of < 100 ha, a size range that encompasses more than 90% of tropical forest fragments (Ranta et al. 1998; Ribeiro et al. 2009), habitat quality is a far stronger predictor of taxa richness than area.

Our results indicate that in highly-disturbed forest, beta-diversity in taxa richness (Fig. 3.4d) mitigates substantial losses in alpha-diversity (Fig.3.4a) to maintain moderately high total richness (Fig. 3.4g).

Beta-diversity is high for even small, heavily-logged areas, as we hypothesised. There are greater point-to-point differences in the taxa comprising rove beetle communities in such forest, not because the community is more spatially heterogeneous, but because naturally-common taxa have become less abundant (Fig. 3.3) and are therefore observed less frequently. The importance of beta-diversity to maintenance of high species richness in disturbed landscapes has been demonstrated previously in Borneo. For example, Wearn et al. (2016) showed that high beta-diversity contributes to the high richness of small and large mammals in Bornean logged forest, and Wang & Foster (2015) confirmed that high spatial beta-diversity can serve to maintain gamma-diversity of ants in plantation compared to natural forest. Our results therefore support past findings that beta-diversity plays a central role in mitigating biodiversity losses in human-modified landscapes (Tscharntke et al. 2012).

When all individuals of all species were equally weighted (q = 1), we found the highest alpha-diversity in unlogged and lightly-logged forest (Fig. 3.4b), indicating that all taxa in the community have relatively equal abundance at any given point. By contrast, lower alpha-diversity in highly-disturbed forest indicates that the community is dominated by a smaller number of taxa that make up a greater proportion of individuals at any one point. The high beta-diversity accompanying low alpha-diversity

(Fig. 3.4e) suggests that the dominant taxa varies among sample points. The increase in beta- diversity with distance indicates that for any level of forest quality, large areas should be preserved to protect a community in which no single taxon dominates the entire area. This effect of this relationship on gamma-diversity was small at the spatial scales we examined, but may become more important at larger spatial scales (Benedick et al. 2006; Lucey et al. 2014).

Our models explained very little of the variation in the data where we gave higher weighting to the most common taxa (q = 2). As a result, we conclude that in comparison to rare taxa, there is little change in the diversity of common rove beetle taxa in response to forest quality and forest area. The apparent resilience of these common taxa to forest modification suggests that they, even in unlogged 50

forest, are relatively generalist and are remarkably tolerant of any changes that might have accompanied the forest disturbance. It is these taxa which are most likely responsible for a large proportion of the ecosystem functioning in the landscape (Slade et al. 2011; Winfree et al. 2015), and our results therefore support previous reports that ecological processes have strong resilience to logging (Gray et al. 2014; Ewers et al. 2015).

All our models of gamma-diversity predicted peaks in overall diversity in lightly-logged forest (Fig. 3.5) due to high alpha-diversity (Fig. 3.4a-c). While unlogged forest is largely accepted to be more diverse than logged forest at greater scales (Gibson et al. 2011), our results suggest that this may not be true at these very small spatial scales. From an ecological point of view, these findings warn against studying “logged” forest as a discrete habitat category instead of a highly-variable array of mix of forest types (Burivalova et al. 2014; Pfeifer et al. 2016). From an applied perspective, these findings suggest that preserving small areas of lightly logged forest may be a relatively low-effort, highly- effective method of boosting biodiversity in agriculture-dominated landscapes.

Our approach of simulating nominal forest areas has several advantages over using data from existing forest fragments, but is not perfect. Small fragments tend to be more heavily disturbed

(Tawatao et al. 2014), which would have prevented us from separating the independent and interactive effects of fragment area and forest quality in real-world fragments. The real-world size and shape of forest fragments also confounds the explicit examination of beta-diversity. Moreover, using sub-samples from a large dataset collected in continuous habitats allowed us to make predictions for far more fragments than could possibly be sampled in fragmented landscapes. However, because our data did not come from existing fragments, our predictions about the exact effect of forest area on diversity retain a degree of uncertainty. We were not able to examine habitat shape because our sampling design was limited to very simple triangles, and so are unable to make generalisations to the convoluted shapes of real-world forest fragments (Ranta et al. 1998; Ewers & Didham 2008). Our design was also not able to identify changes in the invertebrate community which are linked to habitat edges (Ewers & Didham 2008; Terraube et al. 2016), a key contributor to shape effects. Similarly, our design did not account for relaxation in ecological community structure over time. As a result, our results possibly underestimate the diversity of staphylinid beetles able to persist in heavily-degraded

forest. Additionally, we were unable to examine whether specialist or endemic taxa are disproportionately impacted by forest degradation. If this were the case, then our findings based on simple diversity values may be overestimating the value of small, degraded forest areas to the conservation of biodiversity. Further work is therefore required to explicitly determine whether or not the shifts in beta-diversity we observed are entirely beneficial for the conservation of biodiversity in heavily-logged forest, or instead represent the replacement of rare and ecologically-important species with the opposite.

Despite these limitations, our predicted losses in taxa richness were roughly equivalent to those observed in studies of real-world fragmentation scenarios in the same geographic region. Tawatao et al. (2014) found that small, heavily-degraded forest fragments had a 50% reduction in leaf-litter ant richness (fragment gamma-diversity) compared to larger, high-quality forest fragments. This percentage is close to our estimated 36% decrease in rove beetle taxa richness under the same conditions. Similarly, Didham et al. (1998) observed a loss of 19% of common beetle species comparing Central Amazonian forest fragments across the same range of areas we were able to examine. This finding appears to concur with our conclusion that the common species are, although still sensitive to some extent to forest quality and area, largely resilient to forest disturbance. We argue that despite the short-falls in our method of modelling nominal forest areas, our approach represents a pragmatic solution to informing managers in how best to mitigate biodiversity losses in heavily-fragmented landscapes.

We found that conserving rare taxa of non-aleocharine rove beetles will require high quality forest to be maintained (Fig. 3.5a). Simultaneously, conserving common taxa, and by extension safeguarding the functioning of degraded forest ecosystems, will require retention of large forest fragments (Fig.

3.5b-c). We detected relatively minor declines in diversity in even the most degraded scenarios (Fig.

3.5), with evidence that increasing beta-diversity plays an important role in offsetting declines in alpha-diversity (Fig. 3.4d-e). Overall, our results indicate the balance between changes to alpha- and beta-diversity results in a remarkable biodiversity value for what, at first glance, might be considered near-worthless habitat: heavily-degraded forest.

Table 3.1. Parameter estimates for bootstrap models predicting each of alpha-, beta- and gamma-

diversity from AGB and centroid distance. Only final models after model selection are shown. Pseudo-

R2 values are the mean of McFadden’s (1974) pseudo-R2 when the selected model was applied to

each of 10,000 iterations of data combinations. Mean, Standard Deviation (SD) and Standard Error

(SE) are those of the respective parameter from 10,000 fitted mixed models. The effect size, d, is the

absolute value of Mean/S.D. (Cohen 1969).

Weighting Component Parameter Mean SD d SE Z P

Alpha Intercept -9.95 3.45 2.88 0.03 -288.14 < 0.001

Pseudo-R2 = 0.09 AGB 9.94 3.09 3.21 0.03 321.20 < 0.001

AGB2 -2.13 0.68 3.14 0.01 -314.26 < 0.001

Beta Intercept -2.65 42.23 0.06 0.42 -6.28 < 0.001

Pseudo-R2 = 0.44 AGB 5.55 37.70 0.15 0.38 14.73 < 0.001

AGB2 -1.51 8.27 0.18 0.08 -18.27 < 0.001 q = 0

Distance 2.59 13.99 0.19 0.14 18.51 < 0.001

AGB x Distance -2.50 12.48 0.20 0.12 -20.03 < 0.001

AGB2 x Distance 0.60 2.74 0.22 0.03 22.07 < 0.001

Gamma Intercept -7.06 3.07 2.30 0.03 -230.22 < 0.001

Pseudo-R2 = 0.05 AGB 8.44 2.76 3.06 0.03 306.15 < 0.001

AGB2 -1.87 0.61 3.08 0.01 -307.67 < 0.001

Alpha Intercept -7.22 2.48 2.91 0.02 -290.52 < 0.001

q = 1 Pseudo-R2 = 0.11 AGB 7.35 2.25 3.27 0.02 327.05 < 0.001

AGB2 -1.59 0.50 3.20 0.00 -320.27 < 0.001

Beta Intercept 4.54 3.16 1.43 0.03 143.48 < 0.001

Pseudo-R2 = 0.09 AGB -2.41 2.75 0.88 0.03 -87.62 < 0.001

AGB2 0.43 0.60 0.71 0.01 70.77 < 0.001

Distance 0.20 0.36 0.56 0.00 56.13 < 0.001

Gamma Intercept -5.74 3.01 1.91 0.03 -191.01 < 0.001

Pseudo-R2 = 0.04 AGB 6.49 2.58 2.51 0.03 250.99 < 0.001

AGB2 -1.45 0.57 2.54 0.01 -254.04 < 0.001

Distance 0.12 0.35 0.36 0.00 35.61 < 0.001

Alpha Intercept -0.29 3.26 0.09 0.03 -9.03 < 0.001

Pseudo-R2 = 0.01 AGB 1.42 2.91 0.49 0.03 48.74 < 0.001

AGB2 -0.35 0.64 0.55 0.01 -54.83 < 0.001

Beta Intercept 1.04 1.31 0.79 0.01 79.47 < 0.001 q = 2 Pseudo-R2 = 0.01 Distance 0.16 0.43 0.37 0.00 37.23 < 0.001

Gamma Intercept -4.52 2.79 1.62 0.03 -161.94 < 0.001

Pseudo-R2 = 0.03 AGB 5.09 2.44 2.08 0.02 208.30 < 0.001

AGB2 -1.13 0.54 2.10 0.01 -210.02 < 0.001

Distance 0.14 0.34 0.40 0.00 39.73 < 0.001

Table 3.2. Parameter estimates for bootstrap models predicting gamma-diversity from AGB and nominal forest area. Only final models after model selection are shown. Values are as in Table 1.

Weighting Parameter Mean SD d SE Z P

q = 0 Intercept -8.12 3.19 2.54 0.03 -254.45 < 0.001

Pseudo-R2 = 0.06 AGB 9.42 2.89 3.27 0.03 326.50 < 0.001

AGB2 -2.09 0.64 3.27 0.01 -327.30 < 0.001

q = 1 Intercept -6.66 3.01 2.21 0.03 -221.04 < 0.001

Pseudo-R2 = 0.06 AGB 7.57 2.72 2.78 0.03 277.98 < 0.001

AGB2 -1.69 0.60 2.81 0.01 -281.09 < 0.001

Area 0.08 0.28 0.28 0.00 27.99 < 0.001

q = 2 Intercept -5.23 2.81 1.86 0.03 -185.92 < 0.001

Pseudo-R2 = 0.04 AGB 6.03 2.57 2.35 0.03 235.02 < 0.001

AGB2 -1.34 0.57 2.37 0.01 -236.96 < 0.001

Area 0.07 0.28 0.24 0.00 23.93 < 0.001

Figure 3.1. Arrangement of sampling blocks in Sabah, Malaysia. Sampling blocks outside of the SAFE

Project area and VJR are indicated in the top-left panel by +. The sampling block design at MBCA is shown in the bottom-left panel and is of equivalent design to the logged forest control block labelled as LF (top-left).

Figure 3.2. Random three-site combinations represented within a sampling block (dashed line).

Sampling sites comprising one of the three sites within a single combination are black circles, while sampling sites which are not in a combination are grey circles.

Figure 3.3. Abundance of non-aleocharine rove beetles caught at individual sites across the land-use gradient. Model fit is from a generalized mixed model with poisson error, log link function, random intercept for sampling period and second observation-level random intercept for modelling overdispersion (Harrison 2014). Grey area represents 95% confidence intervals, which were estimated via bootstrapping.

Figure 3.4. Responses of each of alpha-, beta- and gamma-diversity to AGB and centroid distance

(beta- and gamma-diversity only). Alpha-diversity is represented as a scatter plot with no relation to centroid distance. Model estimate for alpha-diversity is represented by a black line, with 95% confidence intervals represented in grey. Beta- and gamma-diversity are both represented as contour plots as they can respond to both AGB and centroid distance. All axes are log-transformed.

Figure 3.5. Proportional responses of gamma-diversity to percentage changes in forest quality and forest area (starting from 100 ha). Model estimates are represented as contours.

Chapter 4 - Tropical logging and deforestation impacts multiple scales of weevil beta-diversity

Adam C. Sharp1, Maxwell V. L. Barclay2, Arthur Y. C. Chung3 & Robert M. Ewers1

1Department of Life Sciences, Imperial College London, Silwood Park Campus, Buckhurst Road,

Ascot SL5 7PY, UK. 2Department of Life Sciences, Natural History Museum, Cromwell Road, London

SW7 5BD, UK. 3Forest Research Centre (Sepilok), Sabah Forestry Department, PO Box 1407,

Sandakan, Sabah 90715, Malaysia.

4.1 Abstract

Half of Borneo’s forest has been logged and oil palm plantations have replaced millions of hectares of forest since the 1970’s. While this extensive land-use change has been shown to reduce species richness across landscapes, there is limited current knowledge on how deforestation affects the spatial arrangement of ecological communities. Identifying responses of beta-diversity to land-use change may reveal processes which could mitigate total biodiversity loss. We sampled weevils

(superfamily: Curculionoidea) at multiple spatial scales across a land-use gradient at the Stability of

Altered Forest Ecosystems (SAFE) Project in Sabah, Malaysia, in 2011-2012. We caught 160 taxa of weevil and calculated the response of alpha-diversity (1-ha scale) and beta-diversity (10-, 100-, and

1,000-ha scales) to disturbance. Alpha-diversity of weevils was greatest in unlogged forest but landscape-level beta-diversity (100- and 1,000-ha scale) was maintained across logged and unlogged due to high rates of spatial turnover. Turnover at smallest spatial scales (10-ha) in unlogged forest was highest in rough, flat terrain but smooth, sloping terrain had highest turnover in logged forest.

Logging of flat terrain at small spatial scales has potential to decrease beta-diversity at greater scales.

Plantation beta-diversity at landscape-level remained high but was propagated by abundance shifts of few species instead of spatial turnover of many species. High temporal beta-diversity in unlogged

forest was evident through periodic fluxes in abundance of many weevil species. We conclude that unlogged forest is irreplaceable for high beetle biodiversity but that increased spatial turnover in some terrains may help conserve beetle communities in heavily-degraded landscapes.

4.2 Introduction

Tropical forests are increasingly damaged by human activity (Lewis et al. 2015; Taubert et al. 2018).

Borneo’s tropical forest, the largest remaining in SE Asia, has suffered extensive logging and clearing since the 1970’s (Gaveau et al. 2014), predominantly for high-quality timber and the expansion of industrial plantations (Fitzherbert et al. 2008; Tsujino et al. 2016). Unlogged forest is largely considered to be more biodiverse than degraded forest (Gibson et al. 2011), and large-scale land-use in Borneo change has caused significant declines in species richness of birds, mammals and invertebrates (Chung et al. 2000; Fitzherbert et al. 2008; Edwards et al. 2010; Wearn et al. 2016).

While broad declines in species have been well-documented, there has been little focus on how forest modification impacts the spatial and temporal processes which define landscape-level community structure. Identifying how communities disassemble, especially in degraded forest, will inform how best to use limited conservation resources in increasingly degraded and fragmented landscapes.

The concept of beta-diversity, change in diversity between points, helps to explain community structure (Whittaker 1972; Magurran 2004). In Borneo, beta-diversity has been shown to contribute greatly to total diversity in Borneo’s modified landscapes (Hamer & Hill 2000; Pfeiffer & Mezger 2012;

Kitching et al. 2013; Wearn et al. 2016). Beta-diversity is inherently linked to search area, and so selecting the optimal scales at which to study this concept is important in deriving valid conclusions

(Berry et al. 2008; Pfeiffer & Mezger 2012). Similarly, decomposing beta-diversity into multiple scales may highlight those scales which are important to ecological community structure (Veech & Crist

2007; Astorga et al. 2014). Such research is necessary in landscapes where remnant forest fragments are becoming ever-smaller (Taubert et al. 2018).

Habitat heterogeneity is largely accepted as being a key driver of high beta-diversity (Veech & Crist

2007), and forest modification can both generate or remove habitat heterogeneity. At small scales,

differences in logging intensity and methods can result in vast variation in the quality and structure of remaining forest (Burivalova et al. 2014; Pfeifer et al. 2016). Despite this, studies comparing biodiversity in logged forest to unlogged forest are often limited in that they consider the two as discrete habitat types. Continuous variation in forest structure should ideally be considered to identify the subtle drivers of diversity (Edwards et al. 2011; Burivalova et al. 2014). Conversely to logged forest, oil palm plantations are considered spatially homogeneous habitats (Azhar et al. 2015) although temporally heterogeneous (Luskin & Potts 2011). There is currently limited understanding of how disturbance and habitat heterogeneity influence landscape-level patterns in beta-diversity.

There has also been little previous work on the effect of heterogeneity in topography on spatial biodiversity patterns. Ridges, valleys and plateaus foster ecologically distinct communities of trees in

Borneo (Webb & Peart 2000), yet we remain uninformed about how these small-scale patterns manifest in landscape-level beta-diversity of other taxa. At large scales, flatter terrain is logged more extensively compared to rugged terrain (Bryan et al. 2013). Uncovering the interactions between topography-mediated vegetation heterogeneity and non-random logging activity on beta-diversity may highlight processes by which diversity loss is either mitigated or exacerbated in degraded landscapes.

This is because logging prevalence may be indirectly linked with areas of highest beta-diversity through topography or vice versa.

Beta-diversity can be applied to differences in diversity between points in time as well as in space

(Magurran 2004), and this concept has already been examined in Borneo’s forest. Temporal beta- diversity has been shown to contribute significantly alongside habitat structure to community structure in some taxa (Hamer et al. 2005; Beck & Vun Khen 2007). It is therefore evident that temporal patterns in biodiversity should not be overlooked in studies on land-use (Hamer et al. 2005), however information on temporal beta-diversity in plantations compared to forest is currently lacking.

We aimed to quantify spatial and temporal patterns in beta-diversity across a land-use gradient spanning unlogged forest, logged forest and oil palm plantation in a massively-diverse taxonomic group: the weevils (superfamily: Curculionoidea; Oberprieler et al. 2007). Weevils, a predominantly herbivorous clade, are an ideal focus for their co-evolution alongside flowering angiosperms (Farrell et

al. 2001; McKenna et al. 2009) and therefore direct dependence on tropical forest heterogeneity and structure.

We hypothesised that while weevil alpha-diversity may be highest in unlogged forest, weevil spatial beta-diversity would be greatest in heterogenous logged forest in rugged terrain, for the increased spatial variation in vegetation structure, and lowest in homogeneous oil palm habitat. Uncovering the spatial scaling of beta-diversity in degraded landscapes should provide insight into critical thresholds of forest quality and area which should be maintained to preserve natural community structure. We also hypothesised that temporal beta-diversity would be highest in heavily-degraded habitat because of increased heterogeneity through time, which would emphasise mechanisms by which biodiversity loss from those habitats might be mitigated to some extent.

4.3 Methods

We collected weevils over three separate sampling periods (February 2011, November/December

2011 and June/July 2012) at the Stability of Altered Forest Ecosystems (SAFE) Project in Sabah,

Malaysian Borneo. The SAFE Project comprises 11 blocks of sampling points across a land-use gradient that extends from unlogged forest, through logged forest to oil palm plantation (Fig. 4.1a;

Ewers et al. 2011). Of those 11 blocks, there were three “control” blocks (Fig. 4.1b), six “experimental” blocks (Fig. 4.1c) and two “linear” blocks (Fig. 4.1d). The first control block was in Maliau Basin

Conservation Area (MBCA), where two-thirds of sampling sites had never been logged and one-third of sites had been logged lightly for construction of the nearby field centre. The second control block was positioned in oil palm plantation (OP) that had been planted between 2000 and 2006 and therefore consisted of a closed or nearly-closed oil palm canopy. The third control block (LF) along with six experimental blocks and two linear blocks were dispersed around the experimental area at the SAFE Project (Fig. 4.1e) and comprised of forest that had initially been logged in the 1970’s

(removing 113 m3 ha-1) and logged again between 2000 and 2008 (removing a further 66 m3 ha-1,

Struebig et al. 2013). Logging intensity in this area was vastly uneven, so logged sites included forest

of varying quality and structure (Pfeifer et al. 2016). All forest sites were connected within a single expanse of forest extending over one million ha in area.

Sampling points within blocks were positioned in a fractal design devised specifically for the study of beta-diversity over multiple spatial scales (Marsh & Ewers 2012). While the sampling blocks contained differing arrangements of sampling points (Fig. 4.1b-d), all arrangements followed the same hierarchical structure whereby first-order points (n = 579) were clustered around second-order points

(n = 193), which were in turn clustered around third-order (n = 83) and then fourth-order points (n =

37). Distance between sites of increasing order increased exponentially: first-order points were separated by 101.75 m (56 m), second-order points by 102.25 m (178 m), third-order points by 102.75 m

(562 m) and fourth-order points by 103.25 m (1,778 m, Fig. 1b-d).

Insect traps were set for three days per sampling period at each of the 579 first-order points in the

SAFE Project landscape. Traps were designed to target species of many different behaviours, and comprised a design combining pitfall (diameter 25 cm), flight-intercept (area 1 m2) and malaise traps.

Insects either fell into the bottom section of the trap (the pitfall) or were directed upwards into a collection bottle at the top of the malaise. Top and bottom samples were combined for analysis.

Weevils were separated from the rest of the captured insects and were sorted into morphospecies or, where possible, species.

In order to validly quantify trends in weevil beta-diversity, we had to identify the optimal grain of sampling points to avoid confusing the signals of alpha-diversity with beta-diversity (Jost 2007). This was achieved by testing for spatial autocorrelation between insect traps at the first-order, then grouping insect traps to second- and third-order and testing for spatial autocorrelation at those grouping levels. Spatial autocorrelation was calculated as the Spearman’s rank correlation coefficient of Bray-Curtis dissimilarity against geographic distance. All autocorrelations were calculated within sampling blocks nested within sampling period. Fourth-order points were not tested because there were insufficient points within a single block to calculate a meaningful correlation. We employed Z- tests to decipher which point-groupings were significantly positively spatially autocorrelated, and therefore represented a suitable grain for our analysis. There was significant positive autocorrelation between first-order points (Z = 2.75, Padj < 0.010) but not second-order (Z = 0.97, Padj > 0.050) or 65

third-order (Z = -0.17, Padj > 0.050, adjustment according to Holm, 1979). Second-order was therefore the smallest spatial scale before points became spatially autocorrelated and thus we chose this grouping of insect traps as the grain most suitable for our analysis.

We quantified alpha-diversity of weevils at second-order scale and beta-diversity at the full set of larger spatial scales. By grouping second-order points within higher hierarchical levels, we were able to explicitly examine beta-diversity at third-order, fourth-order and block-level scales. To convert these arrangements of points into sampling areas, we treated each sampling point as a circular area centred on that point (Fig. 4.1). We chose areas for these circles that increased in size exponentially with point order (as did number of insect traps grouped to respectively higher order) and that fitted into the SAFE Project sampling design with minimal overlap. We chose areas of 1-ha for second- order (relevant to alpha-diversity of weevils), 10-ha for third-order, 100-ha for fourth-order and 1,000- ha for block-level groupings of insect traps. There was an unavoidable degree of overlap in circular areas at each scale: < 1% at second-order, 1% at third-order, 13% at fourth-order and 8% at block- level.

Each of total weevil count, number of species and Shannon diversity were calculated per second- order point, nested within sampling period. We calculated beta-diversity at all higher orders by using the abundance-based Bray-Curtis methods of Baselga (2017). These metrics are proportions and therefore insensitive to sampling effort, so two values can be validly compared where based on the same sampling grain. These equations were ideal because they provided scope to separate total beta-diversity into two components: balanced variation (equivalent to turnover in taxa richness), and abundance-gradient variation between sites (roughly equivalent to nestedness in richness). Balanced variation is the subset of beta-diversity which can be attributed to difference in species identity between sampling sites. Abundance-gradient variation is the subset of beta-diversity that is attributable to uneven numbers of individuals between sites. Bray-Curtis dissimilarity is equal to the sum of these two components.

We derived measures of forest quality and topographic variation that were relevant to each of the four spatial scales we had selected for analysis. These measures were calculated for forest sites only as plantation sites comprised mature oil palm and had been terraced. Forest quality was taken as above- 66

ground biomass of vegetation (AGB) and was measured at every second-order point in the landscape by Pfeifer et al. (2016) between 2010 and 2011. No forest modification occurred at SAFE Project between this period and our third and final insect collection in 2012. AGB was derived for third-order, fourth-order and block-level scales by calculating the mean of AGB values at second-order points below that higher point in the hierarchical design.

We calculated two complementary measures of topographic variation – topographic roughness

(Ascione et al.2008) and slope. Both were derived from ASTER satellite elevation data with a 30 m x

30 m resolution (NASA/METI/AIST/Japan Spacesystems, and U.S./Japan ASTER Science Team,

2009). Topographic roughness (TR) per pixel within each circular area was calculated as:

푀푒푎푛 − 푉푎푙푢푒 푇푅 = 푅푎푛푔푒 where “Mean” refers to the mean elevation of that pixel and it’s eight neighbours, “Value” refers to the elevation of that centre pixel, and “Range” refers to the difference between maximum and minimum elevation in the pixel and it’s eight neighbouring pixels. This index of roughness returned a value between negative one and one and described how each pixel is elevated or lowered in relation to surrounding pixels. We converted these values to a single value per circular area by calculating the mean of the absolute values. Slope was taken as the “Range” value and was calculated per circular area as the mean of per-pixel values. Roughness and slope in forest were negatively correlated at all spatial scales (second-order: ρ = -0.40, P < 0.001; third-order: ρ = -0.41, P < 0.001; fourth-order: ρ = -

0.54, P < 0.010; block-level: ρ = -0.81, P < 0.010). The strong negative correlation between topography variables suggested that our forest sites were positioned along a topographic gradient from high TR/low slope to low TR/high slope. Because of this strong relationship, we chose to focus on roughness in our analysis, we which believed to be more clearly relatable to vegetation structure

(Webb & Peart 2000).

All of weevil count, alpha-diversity (richness and Shannon index) and beta-diversity (total, balanced variation and abundance-gradient variation) were analysed for both oil palm plantation and forest

(categorical, ignoring forest quality) using generalized mixed models with sampling period as a random intercept. Within forest sampling blocks, forest quality and roughness were correlated at 67

some spatial scales (second-order: ρ = -0.07, P > 0.050; third-order: ρ = -0.28, P < 0.050; fourth- order: ρ = -0.33, P > 0.050; block-level: ρ = -0.19, P > 0.050). This multicollinearity prohibited the inclusion of both raw AGB and roughness within regular models and so we instead created models based on principal components (Dodge 2003). We log10-transformed AGB so that both variables were normally-distributed, scaled and centred variables on 0, transformed those variables into orthogonal principal components and then performed mixed models with those components as predictors.

Because principal component transformation preserves the linearity of input variables, we were able to back-transform to derive parameter estimates for the original (log10-transformed) variables.

Statistical significance of variables was estimated via permutation: variables were randomized 9,999 times in turn, transformed into principal components as before, used to fit new models by maximum likelihood, and the proportion of those randomized models which were more statistically likely than our single observed model was calculated as a P-value.

Final models were generated via backward-selection on the terms AGB, roughness and their interaction; terms were removed sequentially where P > 0.050. Models which were late in the model selection process and therefore univariate (after the interaction term and one main term had been dropped) did not incorporate principal components but were instead linear models with a single predictor. In this way, we predicted weevil count and species richness via generalized mixed models, and Shannon index and each measure of logit-transformed beta-diversity via linear mixed models. In all models, sampling period was included as a random intercept.

We expanded on our results on spatial beta-diversity by examining whether temporal variation in weevil community structure differed significantly across this land-use gradient. This was achieved by taking each sampling block in turn, selecting only the weevil species where we caught 10 or more individuals over the three sampling periods in that block, and calculating a χ2 statistic describing the unevenness in the abundances of that species caught between sampling periods. With those statistics, we predicted unevenness in abundance first between oil palm plantation and forest, and then used the principal component methods previously described to predict evenness from forest

AGB and roughness. As before, we employed backward-selection on the variables and their interactions but included species as a random intercept.

4.4 Results

We caught 3,447 beetles of the Curculionoidea superfamily: 1,119 in the first sampling period, 370 in the second and 1,958 in the third. We identified 160 species and morphospecies (referred to henceforth as “species”). Species of each of Curculionidae (141 species), Anthribidae (nine species),

Attelabidae (five species) and Brentidae (five species) were present. Most individuals, 3,220, belonged to the Scolytinae subfamily within the family Curculionidae. The most abundant species caught belonged to the genera Coccotrypes, Xylosandrus, Xyleborus, Ambrosiodomus,

Scolytoplatypus and Debus, all within the Scolytinae (full dataset available online at www.safeproject.net).

Significant trends were identified in alpha-diversity, both between forest and oil palm plantation and within forest disturbed to various extents. There was no significant difference (estimate: 2.38, Z =

1.83, P > 0.050) in the number of weevils caught per second-order point between oil palm plantation and forest (Fig. 4.2a). Despite the similarities in weevil abundance, we found significantly more

(estimate: 1.72, Z = 5.24, P < 0.001, Fig. 4.2b) weevil species at forest points compared to plantation and greater Shannon diversity (estimate: 0.57, Z = 5.94, P < 0.001, Fig. 4.2c). Within forest sites, there was a significant increase in number of weevils caught (slope: 0.50, Z = 4.97, P < 0.001, Fig.

4.2a), number of species caught (slope: 0.31, Z = 3.74, P < 0.001, Fig. 4.2b) and Shannon diversity with increasing AGB (slope: 0.22, Z = 4.50, P < 0.001, Fig. 4.2c). Roughness was not selected for inclusion in any models of count or alpha-diversity.

Trends in beta-diversity were predominantly driven by balanced variation in weevil diversity in forest and abundance-gradient variation in oil palm plantation. At 10-ha scales, balanced variation in weevil diversity was significantly greater in forest compared to plantation (estimate: 0.31, Z = 4.30, P <

0.001, Fig. 4.3a) but there was a significant decline in abundance-gradient variation (estimate: -0.85,

Z = -2.16, P < 0.050, Fig. 4.3d). Despite the decline in abundance-gradient variation, total beta- diversity was greatest in forest at 10-ha scale (estimate: 0.70, Z = 2.68, P < 0.010, Fig. 4.3g). Within forest, significant interactions were observed between AGB and roughness (Table 4.1). Balanced variation in diversity was greatest at high AGB in high roughness or at low AGB in low roughness (LR3

= 14.88, P < 0.010) and the same was true of total beta-diversity (LR3 = 25.32, P < 0.001). Neither 69

AGB (estimate: 0.00, Z = -0.32, P > 0.050) nor roughness (estimate: 5.93, Z = 1.01, P > 0.050) had a significant effect on abundance-gradient variation (Fig. 4.3d).

We found no significant effect (P > 0.050) of either AGB or roughness on forest beta-diversity at scales greater than 10-ha. There were, however, significant differences between oil palm plantation and forest. Total beta-diversity did not vary between the two land-use categories at 100-ha scale

(estimate: -0.20, Z = -0.52, P > 0.050, Fig. 4.3h) or 1,000-ha scale (estimate: -0.32, Z = -0.75, P >

0.050, Fig. 4.3i). However, abundance-gradient variation was greatest in oil palm plantation at both

100-ha (estimate: 1.28, Z = 2.62, P < 0.050, Fig. 4.3e) and 1,000-ha scales (estimate: 1.84, Z = 4.10,

P < 0.001, Fig. 4.3f), while balanced variation was lowest (100-ha estimate: -1.10, Z = -2.92, P <

0.010, Fig. 4.3b & 1,000-ha estimate: -1.65, Z = -3.56, P < 0.010, Fig. 4.3c). Balanced variation increased with spatial scale in forest and abundance-gradient variation increased in plantation.

Our results showed unevenness in weevil abundance over the three sampling periods. Temporal unevenness was greater in oil palm plantation than forest (estimate: 1.93, Z = 3.90, P < 0.001) and increased with AGB within forest (estimate: 0.90, Z = 2.32, P < 0.050, Fig. 4.4). In oil palm plantation

(a single sampling block), there were four mid-sized scolytids (Xylosandrus crassiusculus, Xyleborus monographus, Xyleborinus artestriatus and one morphospecies of Ambrosiodomus) which were caught in significantly uneven numbers across the sampling periods (χ2 > 5.99, P < 0.050). In unlogged forest (also a single sampling block; MBCA) there were eleven such species, including

Coccotrypes spp., Debus spp. and Ambrosiodomus spp., which were all members of the Scolytinae.

4.5 Discussion

Our results on weevil diversity add to extensive evidence that unlogged forest is unmatched for high levels of biodiversity. However, while of lesser conservation value than unlogged forest, logged forest likely retains stable community structure through shifts in beta-diversity. Small-scale trends in weevil beta-diversity highlight the importance of topography in determining the conservation value of modified forest and may indicate processes by which targeted industrial logging impacts biodiversity loss.

Weevil alpha-diversity was greatest in unlogged forest and any intensity of forest modification caused declines. Both high species richness and high Shannon index (Fig. 4.2b-c) indicates that many weevil species are present at any one point (1-ha area) and that they are represented in relatively even abundances. Such high alpha-diversity is likely a consequence of high tree species richness (Slik et al. 2003) and therefore large numbers of microhabitats available to large numbers of herbivorous insects (Fig. 4.2a). Selective removal of tree biomass (Pfeifer at al.2016) removes whole microhabitats and therefore causes declines in some species, particularly specialist wood-boring weevils (Farrell et al. 2001). Our results therefore support previous findings that unlogged forest supports maximum numbers of species (Gibson et al. 2011). Despite high numbers of weevils caught in oil palm plantation (Fig. 4.2a), both species richness and Shannon index were low (Fig. 4.2b-c), indicating that most plantation points we sampled were dominated by one or two species in high abundance. These species were presumably generalist pests of the high-density palm fruits and contribute very little to landscape biodiversity.

Despite the large differences in alpha-diversity across the land-use gradient, total landscape spatial beta-diversity (1,000 ha scale, Fig. 4.3i) remained constant. This is in concordance with previous work that found beta-diversity was a significant force in mitigating biodiversity loss in heavily-modified landscapes (Hamer & Hill 2000; Pfeiffer & Mezger 2012; Kitching et al. 2013; Wearn et al. 2016).

However, expanding on those studies, we found that beta-diversity in oil palm plantation is driven primarily by fluxes in the abundance of just a handful of species (Fig. 4.3d-f, Fig. 4.4), presumably in response to either pest management strategies or periods of particularly harsh climate in plantations

(Hardwick et al. 2015). Conversely, high beta-diversity in forest was driven by constant, even turnover in many weevil species (Fig. 4.3b-c). Landscape-level differences in forest beta-diversity manifested themselves in temporal patterns, whereby unlogged forest saw stronger fluxes in the abundance of several species (Fig. 4.4). Such patterns could be driven by the phenology of dipterocarp and other canopy trees (Iku et al. 2017) that are high-value and therefore removed first in logging.

Our results reveal a spatial hierarchy in forest beta-diversity patterns. In rough terrain at small scales

(10-ha), unlogged forest is highly beta-diverse because of balanced turnover in weevil species at small scales (Fig. 4.3a, g). This beta-diversity could be attributed to local differences in vegetation

species and structure accompanying flats, ridges and gullies (Webb & Peart 2000). With logging and removal of that vegetation, beta-diversity is reduced, caused in by part by the fact that differences in tree communities associated with topography do not extend strongly to seedlings and regrowth (Webb

& Peart 2000). In smooth terrain at the same scales, heavily-logged forest is more beta-diverse than unlogged forest (Fig. 4.3a, g). We hypothesis that in these conditions, with increased canopy openness (Pfeifer et al. 2016), there is a greater density and diversity of generalist seedlings and shrubs that provide heterogeneous habitat for herbivores. Because there is no difference between logged and unlogged forest beta-diversity at greater scales (Fig. 4.3h-i), it can be implied that landscape-level beta-diversity in logged forest is maintained through high beta-diversity on plateaus and low beta-diversity in rough terrain, while the opposite is true in unlogged forest. Therefore, while there is no difference between landscape-level beta-diversity in logged and unlogged forest, that apparent resilience to logging is itself a result of small-scale shifts in beta-diversity.

The trends in weevil beta-diversity, differing over spatial scales, can be applied to make broad recommendations for land management. Balanced variation in weevil species was high across all forest qualities (Fig. 4.3a-c) and so our results provide evidence that invertebrate community structure is maintained in even heavily-degraded forest. Maximum total beta-diversity was found at the largest spatial scales (Fig. 4.3i), although there was no difference between logged and unlogged forest at scales greater than 10-ha (Fig. 4.3g-i). We therefore suggest that 100-ha should be a considered a cut-off, below which local forest quality and topography become important to beta-diversity.

Worryingly, the relationships between roughness, slope and AGB indicate that non-random logging activity may have detrimental effects on weevil beta-diversity. Logging is often targeted on flat land

(Bryan et al. 2013), and this was evident at our study site where AGB remaining after logging correlated positively with slope at smallest (1-ha) scales (ρ = 0.21, P < 0.010). Because of the strong negative correlation between slope and roughness (ρ = -0.40, P < 0.001), it could be deduced that the effects of slope are opposite to the effects of roughness. Therefore, high beta-diversity in unlogged forest on rough terrain (Fig. 4.3g) suggests high beta-diversity in unlogged forest on flat areas.

Similarly, high beta-diversity in logged forest on smooth terrain suggests high beta-diversity in logged forest on large-scale slopes. Following this logic, areas of high beta-diversity in unlogged forest (on

rough flats) are at greatest risk of modification, while areas of high beta-diversity in logged forest (on smooth slopes) are unlikely to be created through modification. Such uneven logging activity at small scales has the potential to reduce beta-diversity at greater spatial scales. Our results on high alpha- diversity already add to extensive evidence that unlogged forest supports greatest levels of biodiversity, however these potential impacts on beta-diversity suggest further mechanisms by which biodiversity may be reduced in heavily-degraded landscapes.

We recommend that future research on Borneo’s biodiversity investigate multiple spatial scales and the action of topography on spatial patterns. Likewise, the neglected impacts of temporal beta- diversity may be responsible for long-term biodiversity losses in logged forest and require further examination. While logged forest supports fewer species than unlogged (Gibson et al. 2011), shifts in community structure maintain equal rates of some important ecosystem processes (Ewers et al.

2015), which is likely facilitated in part by high beta-diversity. Conversely, while oil palm plantations maintain similar beta-diversity to forest, that beta-diversity arises mostly from spatial and temporal patchiness in a low number of abundant species. In-depth study of beta-diversity evidently reveals mechanisms by which ecological communities alter following extensive land-use change.

Table 4.1. Model summaries for principal component models at 10-ha spatial scale. Only models

which included both AGB and roughness are shown. Predictor estimates were back-calculated after

fitting linear mixed models with principal components as predictors.

Mixed Model Statistics Back-Transformed Estimates Model Parameter Est. SE Z P Parameter Est. P

Balanced Intercept 0.36 0.23 1.57 > 0.050 AGB -2.19 > 0.050

Variation PC1 0.11 0.07 1.71 > 0.050 TR -60.18 < 0.050

PC2 0.22 0.07 2.91 < 0.010 AGB x TR 28.03 < 0.050

PC3 -2.46 1.03 -2.40 < 0.050

Bray-Curtis Intercept 0.92 0.08 11.59 < 0.001 AGB -3.04 < 0.010

Dissimilarity PC1 0.09 0.06 1.57 > 0.050 TR -79.29 < 0.001

PC2 0.24 0.07 3.59 < 0.001 AGB x TR 36.29 < 0.001

PC3 -3.24 0.89 -3.65 < 0.001

Figure 4.1. Maps of the sampling design, where (a) shows the location of the MBCA (unlogged forest) control block, LF (logged forest) control block, OP (oil palm plantation) control blocks and SAFE

Project area (eight further blocks). The arrangement of 1-ha (light grey area), 10-ha (white area), 100- ha (dark grey area) and 1,000-ha circular areas (dashed perimeter) are shown for each of the three designs of sampling block: (b) control blocks, (c) experimental blocks and (d) linear blocks. The spatial arrangement of the six experimental blocks and two linear blocks within and adjacent to the

SAFE Project area are shown in (e), where those eight sampling blocks are represented inside 1,000- ha circles and comprise forest logged to various extents.

Figure 4.2. The (a) total abundance, (b) species richness and (c) Shannon diversity of weevils caught at second-order points (1-ha scale). Within each panel, samples collected from oil palm plantation are presented on the left (crosses) and forest on the right (dots). The large cross in each plot represents mixed model predictions (with vertical 95% confidence intervals) for oil palm plantation. The plotted lines (with light grey 95% confidence intervals) represent mixed model predictions of the y-axis from

AGB for points in forest only.

Figure 4.3. Beta-diversity of weevils at three spatial scales: 10-ha (first column), 100-ha (second column) and 1,000-ha (3rd column). Beta-diversity is partitioned by rows, where “balanced variation” is beta-diversity attributable to turnover in community structure, “abundance-gradient variation” is beta- diversity attributable to differences in weevil abundance between sites, and “Bray-Curtis dissimilarity” is total beta-diversity. Within panels, samples collected from oil palm plantation is represented on the left (crosses) and forest on the right (dots). Large crosses (with vertical 95% confidence intervals) indicate model predictions for oil palm plantation. Plotted lines (with grey 95% confidence intervals) represent response of forest sites to AGB and topographic roughness. Roughness is a continuous measure in our analyses, so for visualisation purposes we set “Low” roughness as a value of 0.06 and

“High” roughness as a value of 0.14.

Figure 4.4. Temporal unevenness in the abundances of weevil species caught per sampling block.

Samples collected from oil palm plantation are represented on the left (crosses), and forest on the right (dots). The critical value of χ2 on two degrees of freedom for P = 0.050 is shown as a grey dashed line, so all points above that line represent one species that was caught in significantly uneven abundances across sampling periods.

Chapter 5 - Congruence and Importance to Conservation of Beta-

Diversity in Borneo Beetles

Adam C. Sharp1, Maxwell V. L. Barclay2, Arthur Y. C. Chung3 & Robert M. Ewers1

1Department of Life Sciences, Imperial College London, Silwood Park Campus, Buckhurst Road,

Ascot SL5 7PY, UK. 2Department of Life Sciences, Natural History Museum, Cromwell Road, London

SW7 5BD, UK. 3Forest Research Centre (Sepilok), Sabah Forestry Department, PO Box 1407,

Sandakan, Sabah 90715, Malaysia.

5.1 Abstract

Beta-diversity has potential in identifying sites of unique biodiversity and might be useful in assigning

High Conservation Value areas (HCVs) in modified landscapes. We assessed the importance of beta- diversity and the level of congruence in diversity patterns between three taxonomic groups at different spatial scales. We sampled rove beetles (family: Staphylinidae), scarab beetles (superfamily:

Scarabaeoidea) and weevils (Curculionoidea) over a land-use gradient extending from unlogged forest, through logged forest to oil palm plantation at the SAFE Project in Sabah, Malaysia. Each beetle group differed in their response of alpha-diversity to logging, but in all three groups beta- diversity was greatest in heavily-logged forest and had a greater contribution to gamma-diversity than did alpha-diversity in that habitat. At large scales (> 70,000 ha), redundancy analysis revealed that turnover explained over 70% of constrained variance in the community composition of each taxonomic group, followed by habitat structure and microclimate. At small scales (< 350 ha), sites of greatest local contribution to beta-diversity (LCBD) did not correlate well with sites of highest alpha- diversity. High-LCBD sites were strongly congruent between rove beetles and weevils (but not scarabs) in unlogged forest and that relationship deteriorated with land use intensity. Our results suggest that local beta-diversity “hotspots” for some taxa exist only in unlogged forest and that 79

logging uniformly boosts beta-diversity through increased habitat heterogeneity. While there is no alternative for biodiversity to protecting large expanses of tropical forest, trends in beta-diversity might better inform the placement of small protected areas in logged and modified forest landscapes.

5.2 Introduction

Beta-diversity, the spatial or temporal variation in biodiversity, is an important component of community structure in natural habitats (Magurran 2004; Beck et al. 2012). High levels are propagated by environmental heterogeneity at multiple spatial scales (Astorga et al. 2014; Morante-

Filho et al. 2016) and understanding beta-diversity is of great relevance to land-use modification

(Tscharntke et al. 2012; Socolar et al. 2016). Quantifying how beta-diversity responds to disturbance is necessary to predict future biodiversity losses. It may, however, also be possible to utilise beta- diversity to the advantage of conservation efforts by highlighting optimum sizes and spatial arrangements of protected areas (Socolar et al. 2016). Informed land management is imperative given limited conservation resources and rapid encroachment of industrial agriculture on tropical forests (Fitzherbert et al. 2008; Lewis et al. 2015; Taubert et al. 2018).

Borneo, the third largest island in the world, has been subject to widespread deforestation since the

1970’s (Gaveau et al. 2014). Logging for high-quality timber is often followed by conversion of tropical forest to palm oil plantation, of which oil palm is the most common (Fitzherbert et al. 2008; Tsujino et al. 2016). The impacts of such extensive land-use change on biodiversity tend to be either negative or neutral. Reduced diversity has been reported in oil palm plantations compared to adjacent natural habitat (Chung et al. 2000; Hamer et al. 2003; Edwards et al. 2010), however in many cases logged forest retains a high proportion of natural biodiversity levels (Edwards et al. 2011; Wearn et al. 2017).

Some previous studies of Borneo’s biodiversity have quantified beta-diversity. Beck et al. (2012) found that beta-diversity in moths was greater than expected by chance, while alpha-diversity, diversity at a given point (Magurran 2004), was lower. Similarly, Wearn et al. (2016) reported greater mammalian beta-diversity than expected by chance in each of unlogged forest, logged forest and oil palm plantation. Such work has been imperative in highlighting the importance of beta-diversity in

both natural and modified habitat. However, there exists scope for expansion on those ideas, particularly in separating the independent effect of alpha- and beta-diversity and accounting for community composition as well as species richness (Jost 2007).

Furthermore, there remains need to identify the relevant environmental drivers of beta-diversity. Beta- diversity is propagated by environmental heterogeneity (Veech & Crist 2007). Oil palm plantations are spatially homogenous habitats compared to natural forest (Azhar et al. 2015), but logged forest is highly heterogenous in comparison with unlogged forest (Burivalova et al. 2014; Pfeifer et al. 2016).

As a result, logged forests are highly beta-diverse (Hamer & Hill 2000; Pfeiffer & Mezger 2012;

Kitching et al. 2013; Wearn et al. 2016). Uncovering which aspects of environmental heterogeneity directly drive high-beta-diversity will be insightful in understanding how diversity responds to disturbance.

It may be possible to utilise beta-diversity in conservation management if multiple taxa show congruent responses of beta-diversity to land-use change (Socolar et al. 2016). Other forms of diversity are already widely used, notably in assigning areas of High Conservation Value forest (HCV areas or HCVFs) to be spared during the conversion of landscapes to oil palm plantation in Borneo

(RSPO 2013). Forest can be assigned “high” conservation value according to definition HCV1 in the

HCVF Toolkit available to private land managers (Jennings et al. 2003) if that forest “contains globally, regionally or nationally significant concentrations of biodiversity values”. However, despite this stipulation, HCV areas in Borneo are often of low biodiversity value compared to nearby natural habitat or publicly-owned forest fragments (Tawatao et al. 2014), and researchers agree that greater scientific input is required in their planning (Lucey et al. 2017). Given the current failings in HCV areas to conserve biodiversity effectively, creation of new HCV areas should rely more heavily on empirical evidence, which might be part-based on beta-diversity.

One way in which creation of new protected areas might be informed is through local contribution to beta-diversity (LCBD; Legendre & De Cáceres 2013). Contribution to local beta-diversity is unlikely to be spatially even in heterogenous habitat (Legendre & De Cáceres 2013). Sites of high contribution

(ie, sites that are home to a distinct ecological community compared to their surroundings) may indicate “forest areas that are in or contain rare ecosystems”; a phrase relevant to the HCV3 criterion 81

in the HCVF Toolkit (Jennings et al. 2003). The response of beta-diversity to land-use modification depends on the scale at which it is examined (Berry et al. 2008; Wearn et al. 2016) and it is therefore important to study it at the appropriate spatial scale of the conservation initiative. HCV areas are commonly small (tens to hundreds of ha in area; Tawatao et al. 2014; Lucey et al. 2017), and so landscape-scale trends in beta-diversity rather than regional-scale trends would be most relevant to identifying those areas. Government-managed forest reserves are frequently tens of thousands of ha in area (eg in Sabah, Malaysian Borneo, Maliau Basin Conservation Area = 58,840 ha and Danum

Valley Conservation Area = 43,800 ha), and so larger-scale trends in beta-diversity would best describe the ecological communities which persist there. Because of this, identifying sites of highest

LCBD has greatest relevance to planning small protected areas such as HCVs, while identifying the importance of regional ecological processes such as spatial turnover has relevance instead to larger- scale conservation initiatives.

We quantified the beta-diversity responses of multiple taxonomic groups to logging and clearing of tropical forest. We chose three diverse beetle groups which could be sampled using the same methods, but that varied broadly in their life strategies to best represent whole ecosystems. These groups were rove beetles (family: Staphylinidae), scarab beetles (superfamily: Scarabaeoidea) and weevils (superfamily: Curculionoidea). Rove beetles, although diverse in diet and microhabitat preference, predominantly inhabit the upper layers of soil and leaf litter (Barton et al. 2011). Immature scarabs feed largely on dung, rotting wood or roots, while mature scarabs are commonly active fliers in search of dung, flowers, foliage or carrion (Ahrens et al. 2014). Weevils are primarily herbivorous or fungus-farming (Farrell et al. 2001; McKenna et al. 2009). Because of the diverse life strategies represented in these three taxonomic groups, they represent species exploiting many of the niches available in a tropical forest.

We aimed to (1) quantify the relative contribution of beta-diversity compared to alpha-diversity in to gamma-diversity in each beetle group, (2) quantify the relative contribution of spatial turnover compared with environmental variables to diversity at large spatial scales, and (3) identify the level of congruence in beta-diversity patterns at small scales relevant to HCV areas. In line with previous work

(Beck et al. 2012; Wearn et al. 2016), we hypothesised that beta-diversity is the dominant process

shaping ecological communities in response to land-use modification. We also hypothesised that spatial turnover would be a significant determinant of community at structure at large scales, and that congruent between-taxon trends in beta-diversity would reveal forest sites of unique biodiversity at small scales. Identifying environmental indicators of such sites has strong potential for assigning protected forest areas in the future.

5.3 Methods

5.3.1 Study area

We sampled beetles at the Stability of Altered Forest Ecosystems (SAFE) Project in Sabah,

Malaysian Borneo (Ewers et al. 2011). Sampling occurred over three discrete time periods to account for seasonality in biodiversity patterns: February 2011, November-December 2011 and June-July

2012. The sampling area spanned over 70,000 ha (area confined by perimeter around all sites) and comprised a total of 193 sampling sites divided into 11 blocks (Fig. 5.1). Of those 11 blocks, three were “control blocks” of 27 sites each, two were “linear blocks” of eight sites each and six blocks were

“experimental blocks” of 16 sites each. There was one control block in Maliau Basin Conservation

Area (MBCA), where two-thirds of sampling sites comprised unlogged forest and one-third of sites comprised forest which had been logged to a minor extent for timber for the adjacent field centre.

There were a second control block in oil palm plantation that comprised closed or nearly-closed canopy oil palm at the time of sampling. The remaining one control block, two linear blocks and six experimental blocks were positioned in or nearby to the SAFE Project experimental area of logged forest. That forest had been logged in the 1970’s (removing an average of 113 m3 ha-1) and later salvage logged between 2000 and 2008 (removing a further 66 m3 ha-1, Struebig et al. 2013). Logging varied in intensity, and as a result logged forest sites (n = 139) were diverse in forest quality and structure (Pfeifer et al. 2016). Both logged and unlogged forest sites were connected within a single expanse of forest which was over one million ha in area. Sampling sites were separated by

approximately 180 m; an appropriate scale in relation to the smallest HCV areas (Lucey et al. 2017;

Tawatao et al. 2014).

5.3.2 Insect sampling

At each of the 193 sites and within every sampling period, we set three insect traps centred on that site. Insect traps were positioned in a triangle with sides of approximately 56 m. The traps were a combination of pitfall (diameter 25 cm), flight intercept (1 m2) and malaise traps targeted at insects of various behaviours and taxonomic groups. Crawling insects fell into a collection bottle below the pitfall trap, while flying insects struck the flight intercept trap and either fell into the pitfall or were channelled upwards into a second collection bottle in the suspended malaise net. For each site, samples from the three insect traps were combined and all rove beetles, scarab beetles and weevils were identified to either species or morphospecies level. The highly abundant subfamily of rove beetle Aleocharinae was excluded from analysis as they proved impossible to taxonomically identify to any meaningful level.

5.3.3 Habitat variables

Environmental variables at each site were collected by Hardwick et al. (2015) and Pfeifer et al. (2016).

Hardwick et al. collected temperature and humidity data between 2011 and 2012, and for each site we calculated the mean midday air temperature (°C) and relative humidity (%) values per site from their data. Pfeifer et al. collected data on forest structure between 2010 and 2011, and from their data we used above-ground biomass of vegetation per hectare (AGB, Mg/ha), canopy openness (%), canopy vine cover (%), ground vegetation cover (%), soil depth (mm), leaf litter depth (mm), mass of dead wood (kg), average tree canopy area (m3), soil pH, soil temperature (°C), soil moisture (%), slope (°) and presence/absence of streams or rivers. We also derived site elevation (m) from ASTER satellite elevation data with a 30 m x 30 m resolution (NASA/METI/AIST/Japan Spacesystems, and

U.S./Japan ASTER Science Team, 2009).

5.3.4 Comparing responses of alpha- and beta-diversity to disturbance

We calculated alpha-diversity per site and beta-diversity per sampling block for each group of beetles to compare the relative importance of the two components of diversity. The alpha- and beta-diversity indices derived by Jost (2007) were used. These measures were ideal as they ensure alpha- and beta-diversity estimates are independent, and because the units of alpha- and beta-diversity are comparable. We used a value of q = 1 in Jost’s equations, which calculates diversity with equal weighting given to individuals rather than species (equivalent to Shannon index). Beta-diversity between sites within the same block were not significantly spatially autocorrelated in rove beetles (Z =

-0.22, p > 0.05), scarab beetles (Z = 1.10, p > 0.05) or weevils (Z = 1.46, p > 0.05), ensuring sites could be used as independent replicates for statistical analysis. For each taxonomic group of beetles, we predicted diversity from log-transformed AGB (a proxy for forest quality). Models of alpha-diversity were generalised linear mixed models with gamma-distributed error structure and sampling period included as a random intercept. Models of beta-diversity were linear mixed models also with sampling period included as a random intercept. In both cases we applied backwards selection via AICc on the terms AGB and AGB2.

5.3.5 Modelling large-scale beta-diversity patterns

We used redundancy analysis (RDA) to identify the environmental variables important to regional beta-diversity. Samples from the same site were pooled across multiple sampling periods for each of the three beetle groups, and the resulting community matrices were Hellinger-transformed as recommended by Legendre & Gallagher (2001). Because each community matrix included all the 193 sampling sites in our analysis, they represented changes in community composition over spatial scales > 70,000 ha. Environmental variables were transformed for normality; continuous variables without 0’s were log-transformed, continuous variables with 0’s were transformed by inverse hyperbolic sine, and percentage variables were converted to proportions and arcsine square root transformed. We also calculated distance-based Moran’s eigenvector maps (dbMEMs, Borcard &

Legendre 2002) from spatial coordinates of the sampling sites with the aim of quantifying the relative

importance of spatial turnover in determining community composition. From all dbMEMs, we selected only those which modelled negative spatial autocorrelation (ie spatial turnover in community). We used forward selection to identify which transformed environmental variables or negative dbMEMs explained a significant amount of variance in each of the three community matrices. Forward selection occurred via permutation of residuals under a null model. Variables were added sequentially where P

< 0.10.

From final RDAs for each of rove beetles, scarab beetles and weevils, we partitioned the amount of constrained variance into four categories: microclimate, spatial turnover, topography and vegetation structure variables. Microclimate variables included temperature, humidity, soil temperature, soil humidity and soil pH. Spatial turnover variables were negative dbMEMs. Topography variables were elevation, presence/absence of rivers and slope. Vegetation structure variables were AGB, canopy openness, vine cover, ground cover, soil depth, leaf litter depth, mass of dead wood and average tree canopy area. For each category in turn, we recalculated RDAs for each taxon but with variables within that category included as constraining variables and all others as conditional variables, thus isolating the effect of the target category. From those recalculated RDAs, we quantified the amount of variance explained by each of the four categories via anova (Legendre & De Cáceres 2013).

5.3.6 Identifying small-scale beta-diversity hotspots

After identifying the regional drivers of beta-diversity in each taxonomic group, we identified the sites of greatest local contribution to beta-diversity (LCBD) and tested whether those sites were also the sites of greatest alpha-diversity. We used all the sites within just the three control blocks as this resulted in equal sampling effort and arrangement of sampling points in unlogged forest, logged forest and oil palm plantation (Fig. 5.1). These areas represented approximately 350 ha of each habitat type. For each land-use category, we calculated the LCBD of each site according to Legendre & De

Cáceres (2013). From those LCBD values, we identified the relative beta-diversity hotspot sites for each taxonomic group, defined as sites contributing greater than mean beta-diversity. We also identified the alpha-diversity hotspot sites, defined in a similar manner. We calculated the correlation

between high alpha-diversity sites and high LCBD sites within each beetle group, and also calculated the correlation between high LCBD sites of different beetle groups. High/low LCBD and alpha- diversity (in binary form) were used in correlations in place of raw values because we anticipated that those values would be distributed such that there were few large values and many small values.

Therefore, it would not be intuitive for our correlations to incorporate the differences between the many similar, low values. Finally, we used partial correlation to determine which microclimate and vegetation structure variables correlated significantly with high LCBD sites in each of unlogged forest, logged forest and oil palm plantation.

5.4 Results

We caught 2,937 non-aleocharine rove beetles (family: Staphylinidae), 683 scarab beetles

(superfamily: Scarabaeoidea) and 3,365 weevils (superfamily: Curculionoidea) – a total of 6,985 beetles. We identified a total of 462 species or morphospecies (referred to as “species” from here on);

187 from the rove beetles, 115 from the scarabs and 160 from the weevils. The most abundant rove beetle species belonged to the genera Anotylus, Philonthus and Platystethus, while the most abundant scarab species belonged to the genera Onthophagus, Maladera and Phaeochroops. All of the most abundant weevil taxa belonged to the subfamily Scolytinae, and included the genera

Coccotrypes, Xylosandrus and Xyleborus (all data available online at www.safeproject.net).

Responses of alpha-diversity varied between taxonomic groups. Alpha-diversity was lower in oil palm plantation compared to forest in all of rove beetles (Fig. 5.2a; estimate = -1.01, Z = -7.63, P < 0.001), scarab beetles (Fig. 5.2b; estimate = -0.50, Z = -4.81, P < 0.001) and weevils (Fig. 5.2c; estimate = -

0.71, Z = -7.09, P < 0.001). However, within forest habitat, rove beetle alpha-diversity was highest at intermediate AGB equivalent to lightly-logged forest (AGB estimate = 4.77, Z = 2.55, P < 0.050; AGB2 estimate = -1.04, Z = -2.50, P < 0.050), while scarab beetle alpha-diversity was not affected by AGB

(AGB estimate = 0.00, Z = -0.06, P > 0.050) and weevil alpha-diversity peaked at highest AGB (AGB estimate = 0.82, Z = 36.06, P < 0.001; AGB2 estimate = -0.09, Z = -5.60, P < 0.001). Conversely, beta-diversity was greatest in heavily-logged forest in all beetle groups (Fig. 5.2d-f). Beta-diversity

decreased with AGB in rove beetles (AGB estimate = -1.98, Z = -2.82, P < 0.010), scarabs (AGB estimate = -44.38, Z = -3.69, P < 0.001; AGB2 estimate = 9.26, Z = 3.44, P < 0.010) and weevils (AGB estimate = -1.80, Z = -2.64, P < 0.050). Beta-diversity in oil palm plantation did not differ significantly to forest in rove beetles (estimate = 0.96, Z = 1.12, P > 0.050), but was lesser in scarabs (estimate = -

3.57, Z = -2.87, P < 0.010) and weevils (estimate = -2.50, Z = -3.81, P < 0.001). In all three taxonomic groups, beta-diversity was greater than alpha-diversity in oil palm plantation and also in heavily- logged forest. In unlogged forest, alpha-diversity was greater than beta-diversity in rove beetles and weevils but not scarabs.

RDA’s explained a significant amount of variance across the land-use gradient in rove beetles (F35 =

2.43, P < 0.001), scarab beetles (F49 = 1.71, P < 0.001) and weevils (F45 = 2.60, P < 0.001). Spatial turnover was the strongest overall driver of community composition and accounted for over 70% of constrained variance in each taxonomic group (Fig. 5.3). In all three groups, vegetation structure was the next strongest driver, followed by microclimate. Topography was important to weevil community structure only (a single topography variable, presence of rivers). Of the specific environmental variables, soil moisture, air temperature, AGB, canopy openness and mass of dead wood all explained a significant amount of variance in rove beetle community structure (P < 0.050, Table 5.1).

Soil moisture, soil temperature, AGB, average tree canopy area, ground cover and vine cover explained a significant amount of variance in scarab beetles. In weevils, variance was significantly attributed to air temperature, presence of rives, AGB, canopy openness and vine cover. AGB was the only variable to explain a significant amount of variance in the community structure of all three groups of beetles and was highly significant (P < 0.001) in each case.

Sites of high alpha-diversity rarely had high LCBD. In scarab beetles and weevils, high alpha-diversity correlated negatively with high LCBD across all land-uses (Fig. 5.4). In rove beetles, the same pattern held true only in unlogged forest. With increased disturbance (logging and then conversion to plantation), alpha-diversity of rove beetles became significantly positively correlated with LCBD.

Comparing sites of high LCBD between beetle groups, there was highly significant (P < 0.001) positive correlation between rove beetles and weevils in unlogged forest but little correlation (P >

0.050) in any other comparison (Fig. 5.5).

We observed significant correlations between environmental variables and LCBD in rove beetles and weevils, but not scarabs (Table 5.2). In unlogged forest, rove beetle LCBD was significantly positively correlated with air temperature, vegetation ground cover, leaf litter depth and soil depth and negatively correlated with soil moisture. Weevil LCBD was significantly positively correlated with air temperature, elevation and litter depth and negatively correlated with soil moisture and slope. In logged forest, there were no significant correlations between LCBD and environmental variables.

5.5 Discussion

Our results highlight the importance of beta-diversity in mitigating biodiversity loss in heavily-modified landscapes. Moreover, the presence of some congruence among taxa in beta-diversity patterns may have some utility for the planning of new HCV areas. Shifts in beta-diversity and informed land management at small-scales are, however, unlikely to protect natural biodiversity to the same extent as conserving large expanses of natural forest.

Responses of alpha-diversity to logging were not consistent between taxonomic groups. Rove beetle alpha-diversity was highest in lightly-logged forest (Fig. 5.2a), while scarab alpha-diversity was relatively low in forest of any quality (Fig. 5.2b) and weevil alpha-diversity was highest in unlogged forest (Fig. 5.2c). These findings are representative of previous literature that report differing responses of alpha-diversity to logging between taxa. Many studies have found greatest alpha- diversity in unlogged forest (Wearn et al. 2016; Gray et al. 2018) while others have found that logged forest harboured similar alpha-diversity to unlogged (Hamer et al 2003; Slade et al. 2011; Kimber &

Eggleton 2018). Some studies have found that beta-diversity increases in disturbed forest and mitigate losses from alpha-diversity (Woodcock et al. 2011; Wearn et al. 2016). In agreement with these beta-diversity findings, we detected highest beta-diversity in heavily-logged forest in all of rove beetles (Fig. 5.2d), scarab beetles (Fig. 5.2e) and weevils (Fig. 5.2f). It is evident that logging alters the spatial arrangement of ecological communities in ways which mitigate at least some loss of total biodiversity.

As has been previously demonstrated (Chung et al. 2000; Edwards et al. 2010; Hamer et al. 2003), oil palm plantations were low in beetle diversity. In scarabs (Fig. 5.2b, e) and weevils (Fig. 5.2c, f), plantations harboured both lower alpha- and beta-diversity than forest, suggesting that communities were dominated by a small number of taxa that varied little between sites. For rove beetles, only alpha-diversity was lower in oil palm plantation (Fig. 5.2a, d), showing that any given site is dominated by a small number of taxa but that there is some change in community composition between sites.

Beta-diversity had a stronger contribution to gamma-diversity than alpha-diversity in highly-disturbed habitat (Fig. 5.2). Between logged and unlogged forest, trends in beta-diversity were also the most congruent between taxonomic groups, indicating that study of how and why beta-diversity increases in modified habitat has great relevance in predicting and mitigating biodiversity losses from modified tropical forests (Socolar et al. 2016; Wearn et al. 2016). In rove beetles and weevils, alpha-diversity was greater than beta-diversity in unlogged forest and the opposite was true in logged forest and plantation. In scarab beetles, beta-diversity was greater than alpha- in all habitat. This suggest that, despite a lack of congruence in the responses of alpha-diversity to logging (Fig. 5.2a-c), natural forest communities are propagated mostly by alpha-diversity, while beta-diversity becomes the dominant force under disturbance.

At the largest scale we were able to study (> 70,000 ha in area), spatial turnover was the greatest driver by far of community structure in all three beetle groups (Fig. 5.3). Similar distance-related turnover has been observed in Borneo’s moths (Kitching et al. 2013) and trees (Berry et al. 2008).

Vegetation structure and microclimate were also important in all groups, and the variables which were selected for inclusion in RDAs (Table 5.1) can largely be traced back to the life histories in each taxonomic group. Soil variables (moisture and temperature) explained a significant amount of variance in community structures of rove and scarab beetles, whose larvae are soil-dwelling (Barton et al. 2011; Ahrens et al. 2014). The majority of weevils caught belonged to the subfamily Scolytinae, whose larvae are instead found in boreholes in dead or stressed trees (Farrell et al. 2001), and are therefore not directly affected by soil structure. Mature scarabs, largely comprising dung- and flower- feeders, are well adapted for frequent flight between food resources (Ahrens et al. 2014), and were unaffected by temperature variation or canopy openness. Conversely, rove beetles and weevils,

which are mostly small-bodied and do not rely on such long-distance flight (Farrell et al. 2001; Barton et al. 2011), were significantly affected by temperature and canopy openness and so were presumably influenced in community structure by tree fall gaps. Vine cover was significant to scarabs and weevils, both of which include flying foliage-feeding species (McKenna et al. 2009; Ahrens et al.

2014). AGB was the only environmental variable to explain a significant amount of variance in the community structure of all three beetle groups, indicating that many taxa have direct dependencies on the rainforest trees that are targeted by loggers. However, it seems that regardless of taxon-specific habitat associations and tree gap dynamics, protecting very large expanses of forest is the best method of preserving high community diversity across multiple taxa.

At smaller scales (< 350 ha), LCBD provided insight into spatial congruence between taxa. Sites of high LCBD did not correlate positively with sites of greatest alpha-diversity (Fig. 5.4), apart from in rove beetles in heavily-modified habitat. This suggests that landscape-scale beta-diversity is greatest at sites which are not necessarily high in diversity themselves. One possible explanation for such a relationship is that sites of high alpha-diversity may be rich in resources and subject to relatively gentle environmental conditions, allowing many species to persist there. By contrast, sites of high

LCBD may be low in resource and have harsher conditions and are therefore highly selective of which species can survive there. The correlations between LCBD and microclimate variables support this theory. In both rove beetles and weevils, sites of high LCBD correlated negatively with soil moisture and positively with temperature (Table 5.2). Such conditions would increase the risk of desiccation in many taxa, as well as alter the community composition of vegetation available to plant-feeding species. In rove beetles, the positive correlation between high alpha-diversity and high LCBD sites in plantation is perhaps because all sites are limited in resources and harsh in microclimate. In forest, both sites of high alpha-diversity and high LCBD are important in maintaining total diversity.

Sites of high LCBD were congruent only in unlogged forest in some taxa (Fig. 5.5). In both rove beetles and weevils, high LCBD was associated with harsh microclimate (low soil moisture and high temperature) and deep leaf litter, most likely providing habitat for specialist rove beetles hunting there and resting vegetation-feeding weevils. Separately, high LCBD in rove beetles correlated positively with ground cover of vegetation and soil depth, while high LCBD in weevils was also affected

significantly by topography. In logged forest, there was no significant correlation between high-LCBD sites in the two groups (Fig. 5.5) and no significant correlations with environmental variables (Table

5.2). Our results therefore suggest that despite lower landscape beta-diversity, “hotspots” of beta- diversity relevant to some taxa exist in unlogged forest.

No such hotspots were detected in logged forest, despite there being higher overall beta-diversity. We attribute this trend to environmental heterogeneity. Because logged forest is more spatially heterogeneous than unlogged forest (Pfeifer et al. 2016), all sites differ in community composition and there are therefore no distinct hotspots of beta-diversity. This logic does not apply to the scarab beetles, in which we did not detect any correlation with other taxonomic groups (Fig. 5.5) or environmental variables (Table 5.2). Scarabs appear to exist at such low alpha-diversity and at such high beta-diversity, perhaps because of their mobility at maturity, that small-scale variation in habitat has little or no effect.

Our results indicate that landscape-level patterns in beta-diversity were congruent in some, but not all, beetle taxa. It should be emphasised that the use of beta-diversity as a measure for the conservation value of forest is not straight-forward. At small spatial scales, high beta-diversity appears to represent some ecological response to perturbation (Fig 5.2d-f), rather than being directly indicative of natural biodiversity levels. Similarly, study of beta-diversity gives no indication of the rarity or specialism of species persisting in degraded habitats. Following on, we recommend that alpha- and beta-diversity as well as species identity are examined together in order to develop a suitable understanding of landscape-level biodiversity change.

This logic applies to our findings. We found that in unlogged forest, sites which contributed strongly to beta-diversity were highly correlated between rove beetles and weevils, and that correlation was associated largely with microclimate variation (Table 5.2). In those same beetle groups, community composition was sensitive to canopy openness at larger scales (Table 5.1), and so it is possible that small-scale microclimate variation is resultant of tree gap dynamics. If that is the case, then although

“hotspots” of beta-diversity might exist in unlogged forest at any single point in time (or relatively short period, as in our sampling design), those hotspots might be ephemeral in nature. Therefore, sites of locally-unusual microclimate variation in unlogged forest may not indicate permanent “forest areas 92

containing rare ecosystems” under HCV3 (Jennings et al. 2003). Instead, we suggest that forest which is particularly dynamic in structure with high tree turnover may be most suitable for conservation, although we stress that further explicit examination would be required to confirm such a hypothesis. Beyond studies on pure biodiversity measures, future work should also identify how shifts in beta-diversity relate to levels of endemism, specialism, or other aspects of “conservation value” of species.

Our findings have clear implications for the designation of HCVs in Borneo. It may be possible to conserve high levels of biodiversity within small areas of forest which are unlogged (associated with high alpha-diversity in some taxonomic groups) and are also diverse in microclimate (to envelope beta-diversity hotspots). Targeting such sites in logged forest would not be possible, and instead much larger areas of forest should be protected to capture as wide a range of environmental heterogeneity as possible. While beta-diversity gives great, albeit nuanced, insight into how best to designate small protected areas, the great importance of community turnover at large scales provides further evidence that natural levels of biodiversity can be maintained only in very large reserves of forest.

Beta-diversity is a major force shaping communities in tropical forests, and in our analysis, was more important and predictable than alpha-diversity in modified habitat. Although beta-diversity cannot be used alone to quantify biodiversity change across land-use gradients, understanding the relations between alpha- and beta-diversity has great relevance in planning conservation actions. Changes in the spatial arrangement of taxa provide some mitigating force against biodiversity loss and confer remarkable resilience to deforestation. Future research must incorporate spatial patterns in beta- diversity and environmental heterogeneity if we are to effectively address and mitigate the biodiversity impacts of habitat degradation.

Table 5.1. Variance explained by environmental variables in RDA’s on community composition of rove beetles, scarab beetles and weevils separately. Only variables included by forward selection (P <

0.100) are shown. Variables are split into categories: microclimate, topography (T.), and vegetation structure. Values are F-statistics derived via anova, and stars represent the significant levels P <

0.050 (*), P < 0.010 (**) or P < 0.001 (***).

Rove Scarab Variable Weevils Beetles Beetles

Humidity - - 1.51

Soil Moisture 1.86* 1.60* -

Soil Temperature - 1.56* - Microclim. Temperature 3.01*** - 2.06*

Rivers (P/A) - - 2.12* T. AGB 3.12*** 3.65*** 4.33***

Canopy Area 1.67 1.65* -

Canopy Openness 2.84** - 1.92*

Ground Cover - 1.54* 1.52

Veg.Structure Dead Wood 1.81* - -

Vine Cover - 1.87** 2.53**

Table 5.2. Partial correlations between high-LCBD sites and environmental variables in logged and unlogged forest. Correlation values which are significantly different from 0 are represented in bold.

Stars represent the significant levels P < 0.050 (*), P < 0.010 (**) or P < 0.001 (***).

Unlogged Forest Logged Forest

Variable Rove Scarab Rove Scarab Weevils Weevils Beetles Beetles Beetles Beetles

Humidity 0.36 -0.38 0.47 0.41 -0.25 0.03

Soil Moist. -0.62* 0.33 -0.73*** -0.10 -0.02 0.18

Soil pH -0.38 0.36 -0.40 -0.30 -0.15 0.43

Soil Temp. 0.23 -0.16 0.37 0.30 -0.21 0.06 Microclimate Temperature 0.52* -0.35 0.68** -0.41 0.25 -0.03

Elevation -0.02 -0.13 0.57* 0.11 0.04 0.18

Rivers -0.43 0.41 -0.39 -0.18 0.04 0.07

Top. Slope -0.34 -0.13 -0.60* 0.09 -0.22 0.34

AGB -0.39 0.37 -0.38 -0.30 -0.18 0.19

Canopy Ar. 0.03 -0.34 0.45 -0.24 0.02 -0.11

Canopy Op. -0.48 0.44 -0.29 -0.18 0.15 0.04

Ground Cover 0.61* -0.01 0.45 -0.22 -0.11 -0.17

Litter Depth 0.51* 0.00 0.67* 0.14 -0.18 -0.19

Dead Wood 0.38 -0.43 0.09 -0.26 0.10 0.05

VegetationStructure Soil Depth 0.51* -0.09 0.45 -0.17 -0.13 -0.10

Vine Cover 0.29 -0.37 0.44 -0.05 -0.10 0.17

Figure 5.1. Locations of sampling blocks in Sabah (far left panel) and designs of the three types of sampling blocks, where sampling sites are represented by black dots. There was one control block at each of the MBCA (unlogged forest), LF (logged forest) and OP (oil palm plantation) locations. Linear blocks and experimental blocks (all in logged forest) were all positioned at the SAFE location.

Figure 5.2. Responses of log-transformed alpha-diversity (a-c) and non-transformed beta-diversity (d- f) to logging and conversion to oil palm plantation (OP). Alpha-diversity points are one per-site, whereas beta-diversity points are one per-block. Oil palm plantation is represented in the left of each panel with crosses while forest is represented in the right of each panel with dots. The 95% confidence intervals (represented by vertical black bars for plantation and grey areas for forest) were extrapolated via bootstrapping.

Figure 5.3. Proportions of RDA constrained variance attributed to microclimate, spatial turnover, topography and vegetation structure. Significance of variance explained is represented by stars for P

< 0.050 (*), P < 0.010 (**) and P < 0.001 (***).

Figure 5.4. Correlations between high alpha-diversity sites and high LCBD sites in each of rove beetles, scarab beetles and weevils. “High” was defined as greater than the mean value of alpha- diversity per land-use, and the same was true for LCBD. Correlations were therefore on binary data.

Significance levels are represented at P < 0.050 (*), P < 0.010 (**) and P < 0.001 (***) levels.

Figure 5.5. Correlation between the high LCBD sites of each pair of beetle groups. “High” is defined as a greater than average contribution to beta-diversity compared to other sites in that land-use.

Correlations were therefore on binary data. Highly significant correlation (P < 0.001) is represented by

***. Only one correlation was significant (P < 0.050).

100

Chapter 6 – Conclusion and synthesis

6.1 Overview

In Borneo, natural tropical forest is threatened by logging for timber and clearing for the expansion of oil palm plantations (Gaveau et al. 2016; Tsujino et al. 2016). Plantations are largely considered to be of low biodiversity value (Fitzherbert et al. 2008), however logged forest has been observed as retaining a high level of biodiversity (Hamer et al. 2003; Cleary et al. 2007; Edwards et al. 2014). Both unlogged forest and logged forest areas are protected within the conservation network in Borneo

(Reynolds et al. 2011).

In this thesis, I quantified shifts in beetle (order: Coleoptera) diversity across the land-use gradient extending from unlogged forest, through logged forest to oil palm plantation at the Stability of Altered

Forest Ecosystems (SAFE) Project in Sabah, Malaysian Borneo. Expanding on previous work, I decomposed biodiversity into its component parts and identified many of the impacts of deforestation on beta-diversity. Those responses of beta-diversity can be applied to inform land management for the effective conservation of ecological communities (Socolar et al. 2016). In this final chapter, I will collate my findings, relate them to previous work, and discuss the implications for those findings for the conservation of biodiversity in Borneo.

6.2 Unlogged forest is irreplaceable for biodiversity

Unlogged tropical forest is largely considered to support highest levels of biodiversity and endemic species (Gibson et al. 2011) and my results support this view. My analysis showed that logged forest and oil palm plantations supported fewer species of “high conservation value” scarab beetle species, those which were either endemic or generated high interest with the general public, compared with unlogged forest (Chapter 2). While logged forest maintained high species richness of scarab beetles, oil palm plantations supported very low numbers of scarab species and that the species which persisted were widespread in distribution and of limited public interest.

101

My findings aligned with previous research that concluded oil palm plantations were of low value for the conservation of biodiversity (Fitzherbert et al. 2008) and added further evidence for their lack of contribution to cultural ecosystem services. Despite similar scarab species richness, logged forest differed in community composition from unlogged forest. The shift in composition appeared to be driven by removal of tree biomass and alteration of abiotic forest structure. Past studies have documented retainment of high species diversity in Borneo after logging (Hamer et al. 2003; Cleary et al. 2007; Edwards et al. 2014), but I added to that work by showing that fewer logged forest species were either endemic to Borneo or contributed strongly to ecosystem services. Therefore, while logged forests may at first appear equal in biodiversity value to unlogged forest, subtle shifts in community composition significantly reduce the value of that habitat.

In some taxa, alpha-diversity is significantly higher in unlogged forest compared with logged forest.

This was true in weevils (Chapter 4), and I would hypothesise that such a trend would apply mainly to plant-feeding taxa which are likely to be strongly associated with the large hardwood trees targeted during selective logging. Such declines in alpha-diversity between unlogged and logged forest were not evident in groups of beetles which were not primarily associated with vegetation (Chapter 5).

Chapters 3, 4 and 5 in this thesis highlighted how modifications in the spatial arrangement of species can confer, in some ways, a remarkable resilience of diversity to deforestation. However, my findings from scarab beetles and weevils in Borneo provide irrefutable evidence that unlogged forests are unmatched in their ability to propagate high levels of biodiversity.

6.3 Beta-diversity is a stronger force than alpha-diversity in heavily-modified landscapes

Beta-diversity has previously been shown to be an important contributor to total biodiversity in

Borneo’s heavily-degraded landscapes (Hamer & Hill 2000; Pfeiffer & Mezger 2012; Kitching et al.

2013; Wearn et al. 2016). However there has been little formal comparison of the effects of beta- diversity compared to alpha- across the land-use gradient from unlogged forest, through logged forest to oil palm plantation. This was, in part, because of differences in the units in which alpha- and beta- diversity are commonly calculated (Jost 2007) as well as the logistical difficulty in quantifying beta-

102

diversity. My results support previous work indicating the importance of beta-diversity in modified habitat, but expand on that work by direct comparison with alpha-diversity (Chapters 3 and 5).

On the land-use gradient from unlogged to heavily-logged forest, beta-diversity explained a greater amount of variation in rove beetle diversity than alpha-diversity but only where richness-based diversity measures were used (Chapter 3). Where diversity measures were based on community composition, alpha- and beta-diversity explained similar amounts of variation. In both cases, beta- diversity was greatest in heavily-logged forest and lowest in unlogged forest, and therefore mitigated losses in alpha-diversity with logging to some extent. It appears that some taxa become rarer in response to logging, owing to low alpha-diversity and high beta-diversity in the most degraded forest I sampled. However, because many of those taxa become rarer but are not removed from the local community, gamma-diversity remains relatively high in even low-quality forest (Chapter 3).

Responses of beta-diversity to logging of forest and replacement with oil palm plantation were relatively consistent between beetle taxa, being highest in heavily-logged forest in each of rove beetles, scarab beetles and weevils (Chapter 5). Conversely, responses of alpha-diversity to logging differed vastly between taxonomic groups of beetles, and there was very limited congruence between sites of highest alpha-diversity and sites of highest local contribution to beta-diversity. It therefore appears that despite indices of alpha-diversity being widely used to measure responses of taxonomic groups to logging and forest conversion, those indices are unlikely to be representative of the entire ecological community.

Beta-diversity might be more-often relevant outside of a studies focal taxonomic group, but its valid examination requires understanding of some caveats. This component of diversity can respond to land-use change differently according to spatial scale, and so should be measured at multiple scales

(Chapter 4; Hamer & Hill 2000). Also, beta-diversity can itself be decomposed into turnover/balanced variation and nested-ness/abundance gradient partitions (Baselga 2017) which separately provide further insight into the spatial arrangement of biodiversity (Chapter 4). However, beta-diversity alone cannot be used to quantify ecosystem degradation. Both alpha- and beta-diversity should be measured to achieve as complete an indication as possible of biodiversity change across land-use gradients. 103

6.4 Beta-diversity is mediated by environment

Beta-diversity is highest in highly-disturbed habitat (Chapters 3 and 5) but that response can be linked back to environmental variables. At large scales (> 70,000 ha), above ground biomass of vegetation was the only environmental variable that was a significant driver of beta-diversity in rove beetles, scarab beetles and weevils, indicating that removal of trees and natural variation in tree biomass significantly alters ecological communities (Chapter 5). From this it can be inferred that many taxa rely directly on those trees for some resource. Canopy vine cover was significant to flying herbivorous beetles, and because canopy openness, air temperature and soil moisture were all significant drivers of beta-diversity, it can also be implied that ephemeral tree gap dynamics are important in maintaining large-scale variation in community composition.

At small scales, impacts of environmental variables on beta-diversity varied between logged and unlogged forest (Chapter 5). In logged forest, no variables significantly affected beta-diversity in any beetle taxon. Conversely, some were significant in unlogged forest, but only in rove beetles and weevils. Similarly to at greater spatial scales, air temperature and soil moisture were revealed as significant drivers of beta-diversity, further highlighting the likely role of tree gap dynamics in propagating local beta-diversity.

Topography was a significant driver of beta-diversity but only in weevils. Presence of streams or rivers significantly altered weevil community composition (Chapter 5), as did topographic roughness; at spatial scales of around 10 ha, beta-diversity was highest in rough terrain in unlogged forest and smooth terrain in logged forest (Chapter 4). Those trends could be related back to spatial differences in vegetation structure (Webb & Peart 2000), which explains why only weevil beta-diversity (the only primarily-herbivorous taxonomic group I studied) was affected by topography.

Beetle taxa differed in which environmental variables affected their community composition and at which scales. Rove beetles and weevils were impacted by vegetation structure and microclimate at all spatial scales examined. Scarab beetles were only impacted by environmental variables at large spatial scales, perhaps because their relative mobility (Ahrens et al. 2014) allows them better ability to actively avoid unfavourable habitat. This provides further evidence that high beta-diversity is reliant on

104

habitat heterogeneity (Veech & Crist 2007). It appears that habitat heterogeneity at large spatial scales propagates high diversity in most taxa, but small-scale heterogeneity is important to only some

(Chapter 5). However, these results provide clear evidence that maintaining maximum ecosystem diversity, high habitat heterogeneity at all spatial scales in necessary.

6.5 Implications for conservation

With the ongoing deforestation of Borneo’s tropical lowland forest (Gaveau et al. 2014), this thesis has timely implications for land-use on the island. In keeping with the global consensus that unlogged forest is of most valuable biodiversity (Gibson et al. 2011), my results confirm that unlogged forest contains the greatest overall diversity of weevils (Chapter 4) and the highest levels of endemism and cultural ecosystem services in scarab beetles (Chapter 2). Similarly, high rates of spatial turnover in beetle community composition indicate that large forest areas are necessary to maintain natural levels of diversity at regional-scales (Chapter 5). It is clear that large reserves of unlogged forest are of immense conservation value.

Forest quality is a stronger driver of diversity change at small spatial scales (< 100 ha) compared with forest area (Chapter 3). My data on rove beetle diversity showed that forest quality alone is important to total species richness in small forest areas. When community composition was considered (taking into account abundances of each species), forest area had a small effect on total rove beetle diversity, but that effect remained lesser than forest quality. In both cases, beta-diversity was an important force increasing overall rove beetle diversity through mitigation of declines in alpha-diversity in heavily-logged forest. Despite the action of beta-diversity, reduced gamma-diversity was recorded in the most degraded habitat. Highest diversity was recorded in lightly-logged forest. From these results, I conclude that small forest fragments dispersed through agricultural matrices likely maintain high levels of biodiversity and should be preserved where they are created. In the real world, such fragments are usually formed from logged forest, and I recommended that lightly-logged forest fragments might present a relatively low-cost, high-return method of protecting relatively large amounts of biodiversity.

105

At large spatial scales (> 70,000 ha), spatial turnover in community structure was the strongest driver of beta-diversity by far in all groups, followed by vegetation structure and then microclimate.

Conversely, at small scales (< 350 ha) microclimate is most important. These findings suggest that at the small scales relevant to agriculture matrices, it is important to consider the positioning (relative to microclimate heterogeneity or topography) of protected areas so as to make optimum use of limited conservation resources. As has previously been noted by Lucey et al. (2017), this requires increased scientific input from ecologists. However, informed positioning of small protected areas such as HCVs is not a suitable replacement for maintaining very large forest areas, eg reserves. My results suggest that preserving such large areas may not require such extensive ecological advice on environmental heterogeneity or forest quality, and therefore present a more-certain although more-costly method of conserving biodiversity.

6.6 Concluding remarks

It remains obvious that unlogged forest is valuable for the conservation of high levels of biodiversity, and that the optimum method of preventing diversity loss would be to cease further logging and deforestation in Borneo. However, given the economic need to produce vast quantities of palm oil

(Basiron 2007; Fitzherbert 2008) and the rapid and extensive land-use modification that accompanies that process (Gaveau et al. 2014), we must assess the conservation value of even heavily-degraded forest habitat. Study of beta-diversity reveals ways in which the spatial and temporal arrangement of ecological communities responds to land-use change and, in many ways, mitigates biodiversity loss.

106

References

Abson, D. J., & Termansen, M. (2011). Valuing ecosystem services in terms of ecological risks and returns. Conservation Biology: The Journal of the Society for Conservation Biology 25(2), 250–258.

Ackerman, J. D., Trejo-Torres, J. C., & Crespo-Chuy, Y. (2007). Orchids of the West Indies: predictability of diversity and endemism. Journal of Biogeography 34(5), 779–786.

Ahrens, D., Schwarzer, J., & Vogler, A. P. (2014). The evolution of scarab beetles tracks the sequential rise of angiosperms and mammals. Proceedings of the Royal Society B: Biological

Sciences 281(1791), 20141470.

Ascione, A., Cinque, A., Miccadei, E., Villani, F., & Berti, C. (2008). The Plio-Quaternary uplift of the

Apennine chain: new data from the analysis of topography and river valleys in Central Italy.

Geomorphology 102(1), 105–118.

Ashton, P. S. (2009). Conservation of Borneo biodiversity: do small lowland parks have a role, or are big inland sanctuaries sufficient? Brunei as an example. Biodiversity and Conservation 19(2), 343–

356.

Astorga, A., Death, R., Death, F., Paavola, R., Chakraborty, M., & Muotka, T. (2014). Habitat heterogeneity drives the geographical distribution of beta diversity: the case of New Zealand stream invertebrates. Ecology and Evolution, 4(13), 2693–2702.

Azhar, B., Lindenmayer, D. B., Wood, J., Fischer, J., Manning, A., McElhinny, C., & Zakaria, M.

(2011). The conservation value of oil palm plantation estates, smallholdings and logged peat swamp forest for birds. Forest Ecology and Management 262(12), 2306–2315.

Azhar, B., Saadun, N., Puan, C. L., Kamarudin, N., Aziz, N., Nurhidayu, S., & Fischer, J. (2015).

Promoting landscape heterogeneity to improve the biodiversity benefits of certified palm oil production: evidence from Peninsular Malaysia. Global Ecology and Conservation 3, 553–561.

107

Báldi, A. (2008). Habitat heterogeneity overrides the species-area relationship. Journal of

Biogeography 35(4), 675–681.

Banks-Leite, C., Ewers, R. M. & Metzger, J. P. (2012). Unraveling the drivers of community dissimilarity and species extinction in fragmented landscapes. Ecology 93(12), 2560–2569.

Barton, P. S., Gibb, H., Manning, A. D., Lindenmayer, D. B., & Cunningham, S. A. (2011).

Morphological traits as predictors of diet and microhabitat use in a diverse beetle assemblage.

Biological Journal of the Linnean Society 102(2), 301–310.

Baselga, A. (2017). Partitioning abundance-based multiple-site dissimilarity into components: balanced variation in abundance and abundance gradients. Methods in Ecology and Evolution 8(7),

799–808.

Basiron, Y. (2007). Palm oil production through sustainable plantations. European Journal of Lipid

Science and Technology 109(4), 289–295.

Beck, J., Kitching, I. J. & Linsenmair, K. E. (2006). Effects of habitat disturbance can be subtle yet significant: biodiversity of hawkmoth-assemblages (Lepidoptera: Sphingidae) in Southeast-Asia.

Biodiversity and Conservation 15(1), 465–486.

Beck, J., & Vun Khen, C. (2007). Beta-diversity of geometrid moths from northern Borneo: effects of habitat, time and space. Journal of Animal Ecology 76(2), 230–237.

Beck, J., Holloway, J. D., Khen, C. V., & Kitching, I. J. (2012). Diversity partitioning confirms the importance of beta components in tropical rainforest Lepidoptera. The American Naturalist 180(3),

E64–E74.

Benedick, S., Hill, J. K., Mustaffa, N., Chey, V. K., Maryati, M., Searle, J. B., ... Hamer, K. C. (2006).

Impacts of rain forest fragmentation on butterflies in northern Borneo: species richness, turnover and the value of small fragments. Journal of Applied Ecology 43(5), 967–977.

108

Bernard, H., Fjeldså, J., & Mohamed, M. (2009). A case study on the effects of disturbance and conversion of tropical lowland rain forest on the non-volant small mammals in North Borneo: management implications. Mammal Study 34(2), 85–96.

Berry, N. J., Phillips, O. L., Ong, R. C., & Hamer, K. C. (2008). Impacts of selective logging on tree diversity across a rainforest landscape: the importance of spatial scale. Landscape Ecology 23(8),

915-929.

Bohac, J. (1999). Staphylinid beetles as bioindicators. Agriculture, Ecosystems & Environment 74(1-

3), 357–372.

Borcard, D., & Legendre, P. (2002). All-scale spatial analysis of ecological data by means of principal coordinates of neighbour matrices. Ecological Modelling 153(1-2), 51–68.

Bryan, J. E., Shearman, P. L., Asner, G. P., Knapp, D. E., Aoro, G., & Lokes, B. (2013). Extreme differences in forest degradation in Borneo: comparing practices in Sarawak, Sabah, and Brunei.

PLOS ONE, 8(7).

Burivalova, Z., Sekercioglu, C. H. & Koh, L. P. (2014). Thresholds of logging intensity to maintain tropical forest biodiversity. Current Biology 24(16), 1893-1898.

Chao, A. (1987). Estimating the population size for capture-recapture data with unequal catchability.

Biometrics 43(4), 783-791.

Chan, K. M. A., Shaw, M. R., Cameron, D. R., Underwood, E. C., & Daily, G. C. (2006). Conservation planning for ecosystem services. PLoS Biology 4(11), e379.

Chapin, F. S., Zavaleta, E. S., Eviner, V. T., Naylor, R. L., Vitousek, P. M., Reynolds, H. L., … Díaz,

S. (2000). Consequences of changing biodiversity. Nature 405(6783), 234–242.

Chappuis, E., Ballesteros, E., & Gacia, E. (2012). Distribution and richness of aquatic plants across

Europe and Mediterranean countries: patterns, environmental driving factors and comparison with total plant richness. Journal of Vegetation Science 23(5), 985–997.

109

Christie, M., Hanley, N., Warren, J., Murphy, K., Wright, R., & Hyde, T. (2006). Valuing the diversity of biodiversity. Ecological Economic 58(2), 304–317.

Chung, A., Eggleton, P., Speight, M., Hammond, P. & Chey, V. (2000). The diversity of beetle assemblages in different habitat types in Sabah, Malaysia. Bulletin of Entomological Research 90(6),

475–496.

Cleary, D. F. R., Boyle, T. J. B., Setyawati, T., Anggraeni, C. D., Loon, E. E. Van, & Menken, S. B. J.

(2007). Bird species and traits associated with logged and unlogged forest in Borneo. Ecological

Applications 17(4), 1184–1197.

Cleary, D. F. R., Genner, M. J., Koh, L. P., Boyle, T. J. B., Setyawati, T., de Jong, R., & Menken, S. B.

J. (2009). Butterfly species and traits associated with selectively logged forest in Borneo. Basic and

Applied Ecology 10(3), 237–245.

Cohen, J. (1969). Statistical Power Analysis for the Behavioural Sciences. New York: Academic

Press.

Colwell, R. K., & Lees, D. C. (2000). The mid-domain effect: geometric constraints on the geography of species richness. Trends in Ecology & Evolution 15(2), 70–76.

Culmsee, H., & Leuschner, C. (2013). Consistent patterns of elevational change in tree taxonomic and phylogenetic diversity across Malesian mountain forests. Journal of Biogeography 40(10), 1997–

2010.

Currie, D. J. (1991). Energy and large-scale patterns of animal- and plant-species richness. The

American Naturalist 137(1), 27.

Daniel, T. C., Muhar, A., Arnberger, A., Aznar, O., Boyd, J. W., Chan, K. M. A., … von der Dunk, A.

(2012). Contributions of cultural services to the ecosystem services agenda. Proceedings of the

National Academy of Sciences of the United States of America 109(23), 8812–8819.

110

de Bruyn, M., Stelbrink, B., Morley, R. J., Hall, R., Carvalho, G. R., Cannon, C. H., ... von Rintelen, T.

(2014). Borneo and Indochina are major evolutionary hotspots for Southeast Asian biodiversity.

Systematic Biology 63(6), 879–901.

Didham, R. K., Hammond, P. M., Lawton, J. H., Eggleton, P. & Stork, N. E. (1998). Beetle species responses to tropical forest fragmentation. Ecological Monographs 68(3), 295–323.

Didham, R. K. & Ewers, R. M. (2012). Predicting the impacts of edge effects in fragmented habitats:

Laurance and Yensen’s core area model revisited. Biological Conservation 155, 104–110.

Dodge, Y. (2003). The Oxford Dictionary of Statistical Terms. New York: Oxford University Press.

Edwards, D. P., Hodgson, J. A., Hamer, K. C., Mitchell, S. L., Ahmad, A. H., Cornell, S. J. & Wilcove,

D. S. (2010). Wildlife-friendly oil palm plantations fail to protect biodiversity effectively. Conservation

Letters 3(4), 236–242.

Edwards, D. P., Larsen, T. H., Docherty, T. D. S., Ansell, F. A., Hsu, W. W., Derhé, M. A., … Wilcove,

D. S. (2011) Degraded lands worth protecting: the biological importance of Southeast Asia’s repeatedly logged forests. Proceedings of the Royal Society B: Biological Sciences 278(1702), 82–90.

Edwards, D. P., Magrach, A., Woodcock, P., Ji, Y., Lim, N. T.-L., Edwards, F. A., … Yu, D. W.

(2014a). Selective-logging and oil palm: multitaxon impacts, biodiversity indicators, and trade-offs for conservation planning. Ecological Applications 24(8), 2029-2049.

Edwards, D. P., Tobias, J. A., Sheil, D., Meijaard, E., & Laurance, W. F. (2014b). Maintaining ecosystem function and services in logged tropical forests. Trends in Ecology and Evolution 29(9),

511–520.

Elmqvist, T., Folke, C., Nystrom, M., Peterson, G., Bengtsson, J., Walker, B., & Norberg, J. (2003).

Response diversity, ecosystem change, and resilience. Frontiers in Ecology and the Environment

1(9), 488-494.

111

Essl, F., Staudinger, M., Stöhr, O., Schratt-Ehrendorfer, L., Rabitsch, W., & Niklfeld, H. (2009).

Distribution patterns, range size and niche breadth of Austrian endemic plants. Biological

Conservation 142(11), 2547–2558.

Ewers, R. M. & Didham, R. K. (2008). Pervasive impact of large-scale edge effects on a beetle community. Proceedings of the National Academy of Sciences of the United States of America

105(14), 5426–5429.

Ewers, R. M., Didham, R. K., Fahrig, L., Ferraz, G., Hector, A., Holt, R. D., ... Turner, E. C. (2011). A large-scale forest fragmentation experiment: the Stability of Altered Forest Ecosystems Project.

Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences 366(1582),

3292–3302.

Ewers, R. M., Boyle, M. J. W., Gleave, R. A., Plowman, N. S., Benedick, S., Bernard, H., … Turner, E.

C. (2015). Logging cuts the functional importance of invertebrates in tropical rainforest. Nature

Communications 6, 6836.

Farrell, B. D., Sequeira, A. S., O’Meara, B. C., Normark, B. B., Chung, J. H., & Jordal, B. H. (2001).

The evolution of agriculture in beetles (Curculionidae: Scolytinae and Platypodinae). Evolution 55(10),

2011-2027.

Fattorini, S. (2006). Detecting biodiversity hotspots by species-area relationships: a case study of mediterranean beetles. Conservation Biology 20(4), 1169–1180.

Fayle, T. M., Turner, E. C., Snaddon, J. L., Chey, V. K., Chung, A. Y. C., Eggleton, P. & Foster, W. A.

(2010). Oil palm expansion into rain forest greatly reduces ant biodiversity in canopy, epiphytes and leaf-litter. Basic and Applied Ecology 11(4), 337–345.

Fitzherbert, E. B., Struebig, M. J., Morel, A., Danielsen, F., Brühl, C. A., Donald, P. F., & Phalan, B.

(2008). How will oil palm expansion affect biodiversity? Trends in Ecology & Evolution, 23(10), 538–

545.

112

Foley, J. A., DeFries, R., Asner, G. P., Barford, C., Bonan, G., Carpenter, S. R., … Snyder, P. K.

(2005). Global consequences of land use. Science 309(5734), 570–574.

García-López, A., Micó, E., & Galante, E. (2012). From lowlands to highlands: searching for elevational patterns of species richness and distribution of scarab beetles in Costa Rica. Diversity and

Distributions 18(6), 543–553.

Gaston, K. J. (2000). Global patterns in biodiversity. Nature 405(6783), 220–227.

Gaveau, D. L. A., Sloan, S., Molidena, E., Yaen, H., Sheil, D., Abram, N. K., … Meijaard, E. (2014).

Four decades of forest persistence, clearance and logging on Borneo. PLOS ONE, 9(7).

Gaveau, D. L. A., Sheil, D., Husnayaen, Salim, M. A., Arjasakusuma, S., Ancrenaz, M., … Meijaard,

E. (2016). Rapid conversions and avoided deforestation: examining four decades of industrial plantation expansion in Borneo. Scientific Reports 6, 32017.

Gibson, L., Lee, T. M., Koh, L. P., Brook, B. W., Gardner, T. A., Barlow, J., … Sodhi, N. S. (2011).

Primary forests are irreplaceable for sustaining tropical biodiversity. Nature 478(7369), 378–381.

Graves, R. A., Pearson, S. M., & Turner, M. G. (2017). Species richness alone does not predict cultural ecosystem service value. Proceedings of the National Academy of Sciences of the United

States of America 114(14), 3774–3779.

Gray, C. L., Slade, E. M., Mann, D. J. & Lewis, O. T. (2014). Do riparian reserves support dung beetle biodiversity and ecosystem services in oil palm-dominated tropical landscapes? Ecology and

Evolution 4(7), 1049–1060.

Gray, R. E. J., Ewers, R. M., Boyle, M. J. W., Chung, A. Y. C., & Gill, R. J. (2018). Effect of tropical forest disturbance on the competitive interactions within a diverse ant community. Scientific Reports

8(1), 5131.

Grytnes, J. A., & Beaman, J. H. (2006). Elevational species richness patterns for vascular plants on

Mount Kinabalu, Borneo. Journal of Biogeography 33(10), 1838–1849.

113

Hamer, K. C., & Hill, J. K. (2000). Scale-dependent effects of habitat disturbance on species richness in tropical forests. Conservation Biology 14(5), 1435–1440.

Hamer, K. C., Hill, J. K., Benedick, S., Mustaffa, N., Sherratt, T. N., Maryati, M., & Chey, V. K. (2003).

Ecology of butterflies in natural and selectively logged forests of northern Borneo: the importance of habitat heterogeneity. Journal of Applied Ecology 40(1), 150–162.

Hamer, K. C., Hill, J. K., Mustaffa, N., Benedick, S., Sherratt, T. N., Chey, V. K., & Maryati, M. (2005).

Temporal variation in abundance and diversity of butterflies in Bornean rain forests: opposite impacts of logging recorded in different seasons. Journal of Tropical Ecology 21(4), 417–425.

Hammond, P. (1992). Species Inventory. In B. Groombridge (Ed.), Status of the Earth’s Living

Resources (1st ed., pp. 17–39). London: Chapman and Hall.

Hannus, J.-J., & Von Numers, M. (2008). Vascular plant species richness in relation to habitat diversity and island area in the Finnish Archipelago. Journal of Biogeography 35(6), 1077–1086.

Hardwick, S. R., Toumi, R., Pfeifer, M., Turner, E. C., Nilus, R., & Ewers, R. M. (2015). The relationship between leaf area index and microclimate in tropical forest and oil palm plantation: forest disturbance drives changes in microclimate. Agricultural and Forest Meteorology 201, 187–195.

Harrison, X. A. (2014). Using observation-level random effects to model overdispersion in count data in ecology and evolution. PeerJ 2, e616.

Hawkins, B. A., Field, R., Cornell, H. V., Currie, D. J., Guégan, J.-F., Kaufman, D. M., … Turner, J. R.

G. (2003). Energy, water, and broad-scale geographic patterns of species richness. Ecology 84(12),

3105–3117.

Heltshe, J. F., & Forrester, N. E. (1983). Estimating species richness using the jackknife procedure.

Biometrics 39(1), 1-11.

Hillebrand, H. (2004). On the generality of the latitudinal diversity gradient. The American Naturalist

163(2), 192–211.

114

Holm, S. (1979). A simple sequentially rejective multiple test procedure. Scandinavian Journal of

Statistics 6(2), 65–70.

Hooper, D. U., Chapin, F. S., Ewel, J. J., Hector, A., Inchausti, P., Lavorel, S., … Wardle, D. A.

(2005). Effects of biodiversity on ecosystem functioning: a consensus of current knowledge.

Ecological Monographs 75(1), 3–35.

Hubbell, S. P. (2001). The Unified Neutral Theory of Biodiversity and Biogeography. New Jersey:

Princeton University Press.

Hunt, T., Bergsten, J., Levkanicova, Z., Papadopoulou, A., John, O. S., Wild, R., … Vogler, A. P.

(2007). A comprehensive phylogeny of beetles reveals the evolutionary origins of a superradiation.

Science 318(5858), 1913–1916.

Hutchinson, G. E. (1959). Homage to Santa Rosalia, or Why are there so many kinds of animals? The

American Naturalist 93(870), 145–159.

Iku, A., Itioka, T., Kishimoto-Yamada, K., Shimizu-kaya, U., Mohammad, F. B., Hossman, M. Y., …

Meleng, P. (2017). Increased seed predation in the second fruiting event during an exceptionally long period of community-level masting in Borneo. Ecological Research 32(4), 537–545.

Jablonski, D., Roy, K., & Valentine, J. W. (2006). Out of the tropics: evolutionary dynamics of the latitudinal diversity gradient. Science 314(5796), 102–106.

Jennings, S., Nussbaum, R., Judd, N., Evans with, T., Azevedo, T., Brown, N., … Chunquan, Z.

(2003). The High Conservation Value Forest Toolkit. Retrieved from https://www.hcvnetwork.org/resources/global-hcv-toolkits (accessed 12 May 2018).

Jost, L. (2007). Partitioning diversity into independent alpha and beta components. Ecology 88(10),

2427–2439.

Kier, G., Mutke, J., Dinerstein, E., Ricketts, T. H., Küper, W., Kreft, H. & Barthlott, W. (2005). Global patterns of plant diversity and floristic knowledge. Journal of Biogeography 32(7), 1107–1116. 115

Kitching, R. L., Ashton, L. A., Nakamura, A., Whitaker, T., & Khen, C. V. (2013). Distance-driven species turnover in Bornean rainforests: homogeneity and heterogeneity in primary and post-logging forests. Ecography, 36(6), 675–682.

Kreft, H., & Jetz, W. (2007). Global patterns and determinants of vascular plant diversity. Proceedings of the National Academy of Sciences of the United States of America 104(14), 5925–5930.

Kumschick, S., Schmidt-Entling, M. H., Bacher, S., Hickler, T., Espadaler, X., & Nentwig, W. (2009).

Determinants of local ant (Hymenoptera: Formicidae) species richness and activity density across

Europe. Ecological Entomology 34(6), 748–754.

Laurance, W. F., Lovejoy, T. E., Vasconcelos, H. L., Bruna, E. M., Didham, R. K., Stouffer, P. C., ...

Sampaio, E. (2002). Ecosystem decay of Amazonian forest fragments: a 22-year investigation.

Conservation Biology 16(3), 605–618.

Lawton, J. H. (1994). What do species do in ecosystems? Oikos 71(3), 367-374.

Leather, S. R. (2015). Influential entomology: a short review of the scientific, societal, economic and educational services provided by entomology. Ecological Entomology 40(S1), 36–44.

Lee, S.-M., & Chao, A. (1994). Estimating population size via sample coverage for closed capture- recapture models. Biometrics 50(1), 88-97.

Legendre, P., & Gallagher, E. D. (2001). Ecologically meaningful transformations for ordination of species data. Oecologia 129(2), 271–280.

Legendre, P., & De Cáceres, M. (2013). Beta diversity as the variance of community data: dissimilarity coefficients and partitioning. Ecology Letters 16(8), 951–963.

Lewis, S. L., Edwards, D. P., & Galbraith, D. (2015). Increasing human dominance of tropical forests.

Science 349(6250), 827–832.

116

Lohman, D. J., de Bruyn, M., Page, T., von Rintelen, K., Hall, R., Ng, P. K. L., … von Rintelen, T.

(2011). Biogeography of the Indo-Australian Archipelago. Annual Review of Ecology, Evolution, and

Systematics 42(1), 205–226.

Lucey, J. M., Tawatao, N., Senior, M. J. M., Chey, V. K., Benedick, S., Hamer, K. C., ... Hill, J. K.

(2014). Tropical forest fragments contribute to species richness in adjacent oil palm plantations.

Biological Conservation 169, 268–276.

Lucey, J. M., Palmer, G., Yeong, K. L., Edwards, D. P., Senior, M. J. M., Scriven, S. A., … Hill, J. K.

(2017). Reframing the evidence base for policy-relevance to increase impact: a case study on forest fragmentation in the oil palm sector. Journal of Applied Ecology 54(3), 731–736.

Luskin, M. S., & Potts, M. D. (2011). Microclimate and habitat heterogeneity through the oil palm lifecycle. Basic and Applied Ecology 12(6), 540–551.

Lyal, C. H. C., & Curran, L. M. (2000). Seed-feeding beetles of the weevil tribe Mecysolobini (Insecta:

Coleoptera: Curculionidae) developing in seeds of trees in the Dipterocarpaceae. Journal of Natural

History 34(9), 1743–1847.

MacArthur, R. H., & Wilson, E. O. (1967). The Theory of Island Biogeography. New Jersey: Princeton

University Press.

Magurran, A. E. (2004). Measuring Biological Diversity. New Jersey: Wiley-Blackwell.

Marsh, C. J. & Ewers, R. M. (2013). A fractal-based sampling design for ecological surveys quantifying beta-diversity. Methods in Ecology and Evolution 4(1), 63–72.

McFadden, D. L. (1974) Conditional logit analysis of qualitative choice behaviour. In Frontiers in

Econometrics p. 105-142. New York: Wiley.

McKenna, D. D., Sequeira, A. S., Marvaldi, A. E., & Farrell, B. D. (2009). Temporal lags and overlap in the diversification of weevils and flowering plants. Proceedings of the National Academy of Sciences of the United States of America 106(17), 7083–7088.

117

Meijaard, E., & Nijman, V. (2003). Primate hotspots on Borneo: predictive value for general biodiversity and the effects of taxonomy. Conservation Biology 17(3), 725–732.

Millennium Ecosystem Assessment. (2005). Ecosystems and Human Well-Being: Synthesis.

Washington.

Morante-Filho, J. C., Arroyo-Rodríguez, V., & Faria, D. (2016). Patterns and predictors of β-diversity in the fragmented Brazilian Atlantic forest: a multiscale analysis of forest specialist and generalist birds. Journal of Animal Ecology 85(1), 240–250.

Myers, N., Mittermeier, R. A., Mittermeier, C. G., da Fonseca, G. A., & Kent, J. (2000). Biodiversity hotspots for conservation priorities. Nature 403(6772), 853–858.

Naeem, S. (1998). Species Redundancy and Ecosystem Reliability. Conservation Biology 12(1), 39–

45.

Nakagawa, M., Momose, K., Kishimoto-Yamada, K., Kamoi, T., Tanaka, H. O., Kaga, M., …

Nakashizuka, T. (2013). Tree community structure, dynamics, and diversity partitioning in a Bornean tropical forested landscape. Biodiversity and Conservation 22(1), 127–140.

NASA/METI/AIST/Japan Spacesystems, and U.S./Japan ASTER Science Team (2009). ASTER

Global Digital Elevation Model Terra ASTER v2. NASA EOSDIS Land Processes DAAC.

Ninan, K. N., & Inoue, M. (2013). Valuing forest ecosystem services: what we know and what we don’t. Ecological Economics 93, 137–149.

O’Brien, E. M. (1993). Climatic Gradients in Woody Plant Species Richness: Towards an explanation based on an analysis of Southern Africa’s woody flora. Journal of Biogeography 20(2), 181-198.

Oberprieler, R. G., Marvaldi, A. E., & Anderson, R. S. (2007). Weevils, weevils, weevils everywhere.

Zootaxa 520(1668), 491–520.

118

Patton, D. R. (1975). A diversity index for quantifying habitat edge. Wildlife Society Bulletin 3, 171–

173.

Pfeifer, M., Kor, L., Nilus, R., Turner, E., Cusack, J., Lysenko, I., … Ewers, R. (2016). Mapping the structure of Borneo’s tropical forests across a degradation gradient. Remote Sensing of Environment

176, 84–97.

Pfeiffer, M., & Mezger, D. (2012). Biodiversity Assessment in Incomplete Inventories: Leaf litter ant communities in several types of Bornean rain forest. PLOS ONE, 7(7), e40729.

Phillips, O. L., Hall, P., Gentry, A. H., Sawyer, S. A., & Vásquez, R. (1994). Dynamics and species richness of tropical rain forests. Proceedings of the National Academy of Sciences of the United

States of America 91(7), 2805–2809.

Ranta, P., Blom, T., Nimela, J., Joensuu, E. & Siitonen, M. (1998). The fragmented Atlantic rain forest of Brazil: size, shape and distribution of forest fragments. Biodiversity and Conservation 7(3), 385–

403.

Reynolds, G., Payne, J., Sinun, W., Mosigil, G. & Walsh, R. P. D. (2011). Changes in forest land use and management in Sabah, Malaysian Borneo, 1990-2010, with a focus on the Danum Valley region.

Philosophical Transactions of the Royal Society B: Biological Sciences 366(1582), 3168–3176.

Ribeiro, M. C., Metzger, J. P., Martensen, A. C., Ponzoni, F. J. & Hirota, M. M. (2009). The Brazilian

Atlantic Forest: How much is left, and how is the remaining forest distributed? Implications for conservation. Biological Conservation 142, 1141–1153.

RSPO (2013) Principles and criteria for the production of sustainable palm oil. Report submitted by the RSPO Executive Board for the Extraordinary General Assembly. Retrieved from http://www.rspo.org/key-documents/certification/rspo-principles-and-criteria (accessed 12 May 2018).

Ruesink, J. L., & Srivastava, D. S. (2001). Numerical and per capita responses to species loss: mechanisms maintaining ecosystem function in a community of stream insect detritivores. Oikos

93(2), 221–234.

119

Schoolmeesters P. (2018). Scarabs: World Scarabaeidae Database (version Jul 2017). In: Roskov Y.,

Abucay L., Orrell T., Nicolson D., Bailly N., Kirk P. M., … Penev L., eds. 2018. Species 2000 & ITIS

Catalogue of Life, 26th February 2018. Digital resource at www.catalogueoflife.org/col. Species 2000:

Naturalis, Leiden, the Netherlands.

Slade, E. M., Mann, D. J. & Lewis, O. T. (2011). Biodiversity and ecosystem function of tropical forest dung beetles under contrasting logging regimes. Biological Conservation 144, 166–174.

Slik, J. W. F., Poulsen, A. D., Ashton, P. S., Cannon, C. H., Eichhorn, K. A. O., Kartawinata, K., …

Wilkie, P. (2003). A floristic analysis of the lowland dipterocarp forests of Borneo. Journal of

Biogeography 30(10), 1517–1531.

Slik, J. W. F., Raes, N., Aiba, S.-I., Brearley, F. Q., Cannon, C. H., Meijaard, E., … Wulffraat, S.

(2009). Environmental correlates for tropical tree diversity and distribution patterns in Borneo.

Diversity and Distributions 15(3), 523–532.

Smith, E. P., & van Belle, G. (1984). Nonparametric estimation of species richness. Biometrics 40(1),

119-129.

Socolar, J. B., Gilroy, J. J., Kunin, W. E., & Edwards, D. P. (2016). How should beta-diversity inform biodiversity conservation? Trends in Ecology & Evolution 31(1), 67–80.

Stibig, H. J., Achard, F., Carboni, S., Rasi, R., & Miettinen, J. (2014). Change in tropical forest cover of Southeast Asia from 1990 to 2010. Biogeosciences 11(2), 247–258.

Stork, N. E., McBroom, J., Gely, C., & Hamilton, A. J. (2015). New approaches narrow global species estimates for beetles, insects, and terrestrial arthropods. Proceedings of the National Academy of

Sciences of the United States of America 112(24), 7519–7523.

Struebig, M. J., Kingston, T., Petit, E. J., Le Comber, S. C., Zubaid, A., Mohd-Adnan, A. & Rossiter, S.

J. (2011). Parallel declines in species and genetic diversity in tropical forest fragments. Ecology

Letters 14(6), 582–590.

120

Struebig, M. J., Turner, A., Giles, E., Lasmana, F., Tollington, S., Bernard, H., & Bell, D. (2013).

Quantifying the biodiversity value of repeatedly logged rainforests: gradient and comparative approaches from Borneo. In Advances in Ecological Research p. 183–224. New York: Academic

Press.

Taubert, F., Fischer, R., Groeneveld, J., Lehmann, S., Müller, M. S., Rödig, E., … Huth, A. (2018).

Global patterns of tropical forest fragmentation. Nature 554(7693), 519-522.

Tawatao, N., Lucey, J. M., Senior, M., Benedick, S., Vun Khen, C., Hill, J. K., & Hamer, K. C. (2014).

Biodiversity of leaf-litter ants in fragmented tropical rainforests of Borneo: the value of publically and privately managed forest fragments. Biodiversity and Conservation 23(12), 3113–3126.

Terraube, J., Archaux, F., Deconchat, M., van Halder, I., Jactel, H. & Barbaro, L. (2016). Forest edges have high conservation value for bird communities in mosaic landscapes. Ecology and Evolution

6(15), 5178–5189.

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.

Theuerkauf, J., Chartendrault, V., Desmoulins, F., Barré, N., & Gula, R. (2017). Positive range- abundance relationships in Indo-Pacific bird communities. Journal of Biogeography 44(9), 2161–2163.

Thompson, I. D., Okabe, K., Parrotta, J. A., Brockerhoff, E., Jactel, H., Forrester, D. I., & Taki, H.

(2014). Biodiversity and ecosystem services: lessons from nature to improve management of planted forests for REDD-plus. Biodiversity and Conservation 23(10), 2613–2635.

Tilman, D., Lehman, C. L., & Bristow, C. E. (1998). Diversity-stability relationships: statistical inevitability or ecological consequence? The American Naturalist 151(3), 277–282.

Tilman, D., Fargione, J., Wolff, B., D’Antonio, C., Dobson, A., Howarth, R., … Swackhamer, D. (2001).

Forecasting agriculturally driven global environmental change. Science 292(5515), 281–284.

121

Tscharntke, T., Tylianakis, J. M., Rand, T. A., Didham, R. K., Fahrig, L., Batáry, P., … Westphal, C.

(2012). Landscape moderation of biodiversity patterns and processes - eight hypotheses. Biological

Reviews 87(3), 661–685.

Tsujino, R., Yumoto, T., Kitamura, S., Djamaluddin, I. & Darnaedi, D. (2016). History of forest loss and degradation in Indonesia. Land Use Policy 57, 335–347.

Tylianakis, J. M., Rand, T. A., Kahmen, A., Klein, A.-M., Buchmann, N., Perner, J., & Tscharntke, T.

(2008). Resource heterogeneity moderates the biodiversity-function relationship in real world ecosystems. PLoS Biology 6(5), e122.

Veech, J. A., & Crist, T. O. (2007). Habitat and climate heterogeneity maintain beta-diversity of birds among landscapes within ecoregions. Global Ecology and Biogeography, 16(5), 650–656.

Wang, W. Y. & Foster, W. A. (2015). The effects of forest conversion to oil palm on ground-foraging ant communities depend on beta diversity and sampling grain. Ecology and Evolution 5(15), 3159–

3170.

Wearn, O. R., Carbone, C., Rowcliffe, J. M., Bernard, H. & Ewers, R. M. (2016). Grain-dependent responses of mammalian diversity to land use and the implications for conservation set-aside.

Ecological Applications 26(5), 1409–1420.

Wearn, O. R., Rowcliffe, J. M., Carbone, C., Pfeifer, M., Bernard, H., & Ewers, R. M. (2017).

Mammalian species abundance across a gradient of tropical land-use intensity: A hierarchical multi- species modelling approach. Biological Conservation 212, 162–171.

Webb, C. O., & Peart, D. R. (2000). Habitat associations of trees and seedlings in a Bornean rain forest. Journal of Ecology 88(3), 464–478.

Weibull, A. C., Östman, Ö., & Granqvist, Å. (2003). Species richness in agroecosystems: the effect of landscape, habitat and farm management. Biodiversity and Conservation 12(7), 1335–1355.

Whittaker, R. H. (1972) Evolution and measurement of species diversity. Taxon 21(2-3), 213–251.

122

Wilting, A., Sollmann, R., Meijaard, E., Helgen, K. M., & Fickel, J. (2012). Mentawai’s endemic, relictual fauna: is it evidence for Pleistocene extinctions on Sumatra? Journal of Biogeography 39(9),

1608–1620.

Winfree, R., W. Fox, J., Williams, N. M., Reilly, J. R. & Cariveau, D. P. (2015). Abundance of common species, not species richness, drives delivery of a real-world ecosystem service. Ecology Letters

18(7), 626–635.

Woodcock, P., Edwards, D. P., Fayle, T. M., Newton, R. J., Khen, C. V., Bottrell, S. H., & Hamer, K.

C. (2011). The conservation value of South East Asia’s highly degraded forests: evidence from leaf- litter ants. Philosophical Transactions of the Royal Society B: Biological Sciences 366(1582), 3256–

3264.

Woodruff, D. S. (2010). Biogeography and conservation in Southeast Asia: how 2.7 million years of repeated environmental fluctuations affect today’s patterns and the future of the remaining refugial- phase biodiversity. Biodiversity and Conservation 19(4), 919–941.

Wong, K. M. (2011). A Biogeographic History of Southeast Asian Rainforests. In R. Wickneswari & C.

Cannon (Eds.), Managing the Future of Southeast Asia’s Valuable Tropical Rainforests (Vol. 2, pp.

21–55). Dordrecht: Springer Netherlands.

Worm, B., Barbier, E. B., Beaumont, N., Duffy, J. E., Folke, C., Halpern, B. S., … Watson, R. (2006).

Impacts of biodiversity loss on ocean ecosystem services. Science 314(5800), 787–790.

123