Predicting Connectivity, Population Size and Genetic Diversity of Sunda Clouded Leopards Across Sabah, Borneo

Landscape Ecol (2019) 34:275–290

https://doi.org/10.1007/s10980-018-0758-1 (0123456789().,-volV)(0123456789().,-volV)

RESEARCH ARTICLE

Predicting connectivity, population size and genetic diversity of Sunda clouded leopards across Sabah, Borneo

Andrew J. Hearn . Samuel A. Cushman . Benoit Goossens . Joanna Ross . Ewan A. Macdonald . Luke T. B. Hunter . David W. Macdonald

Received: 23 April 2018 / Accepted: 9 November 2018 / Published online: 16 January 2019 Ó The Author(s) 2019

Abstract quantify the differences in connectivity metrics from Context The Sunda clouded leopard is vulnerable to an empirically optimized model of landscape resis- forest loss and fragmentation. Conservation of this tance with one based on expert opinion. species requires spatially explicit evaluations of the Methods We investigated connectivity metrics for effects of landscape patterns on genetic diversity, Sunda clouded leopards across Sabah, based on an population size and landscape connectivity. empirically optimised, movement based model, and an Objectives We sought to develop predictions of expert-opinion derived model. We used simulation Sunda clouded leopard population density, genetic modelling to predict and compare the patterns and diversity and population connectivity across the state causes of differences in the local neighbourhood of Sabah, Malaysian Borneo. We also wished to population density, distribution, and genetic diversity

A. J. Hearn (&) Á J. Ross Á E. A. Macdonald Á B. Goossens D. W. Macdonald Sustainable Places Research Institute, Cardiff University, Wildlife Conservation Research Unit (WildCRU), 33 Park Place, Cardiff CF10 3BA, UK Department of Zoology, University of Oxford, Oxford, UK L. T. B. Hunter e-mail: [email protected] Panthera, New York, NY, USA

S. A. Cushman US Forest Service, Rocky Mountain Research Station, 2500 S Pine Knoll Dr, Flagstaff, AZ 86001, USA

B. Goossens Danau Girang Field Centre, c/o Sabah Wildlife Department, Wisma Muis, 88100 Kota Kinabalu, Sabah, Malaysia

B. Goossens Sabah Wildlife Department, Wisma Muis, 88100 Kota Kinabalu, Sabah, Malaysia

B. Goossens Organisms and Environment Division, School of Biosciences, Cardiff University, Sir Martin Evans Building, Museum Avenue, Cardiff CF10 3AX, UK

123 276 Landscape Ecol (2019) 34:275–290 across the two different resistance maps, at two populations, however, there is substantial uncertainty dispersal distances. about all of these parameters. Results The empirical model produced higher esti- In the vast majority of applications expert-opinion mates of population size, population density, genetic has been used to parameterize resistance surfaces diversity and overall connectivity than the expert- (Spear et al. 2010; Zeller et al. 2012). This has opinion model. The overall pattern of predicted potentially serious limitations given that expert-opin- connectivity was similar between models. Both mod- ion is of unknown quality and may often fail to reflect els identified a large patch of core habitat with high accurately the resistance experienced by animals when predicted connectivity in Sabah’s central forest region, moving across the landscape (e.g., Shirk et al. 2010; and agreed on the location and extent of the main Wasserman et al. 2010; Shirk et al. 2015). This is isolated habitat fragments. particularly true of many threatened species, for which Conclusions We identified clear relationships even a basic understanding of their ecology is often between landscape composition and configuration lacking. A number of methods have been developed to and predicted distribution, density, genetic diversity estimate landscape resistance empirically using and connectivity of Sunda clouded leopard popula- genetic (e.g., Cushman et al. 2006; Shirk et al. 2010; tions. Core areas are comprised of large and unfrag- Castillo et al. 2014) and movement (e.g., Blazquez- mented forest blocks, and areas of reduced forest cover Cabrera et al. 2016; Cushman et al. 2016; Zeller et al. comprise barriers among patches of predicted remain- 2017, 2018) data. These approaches have the advan- ing habitat. tage that they are directly estimated using data from the key processes of interest. Indeed, when compared Keywords Clouded leopard Á Connectivity Á with expert-opinion or habitat suitability based mea- UNICOR Á CDPOP Á Fragmentation Á Habitat loss sures, resistance surfaces directly estimated from movement and genetic data have shown superior performance (e.g., Shirk et al. 2010; Wasserman et al. 2010; Mateo Sańchez et al. 2014, 2015; Zeller et al. Introduction 2018). Movement and genetic data are often lacking for many threatened species, however, and are typi- In the face of accelerating global habitat loss and cally very costly to acquire. In the absence of such fragmentation there is an increasing need to predict empirical data, expert opinion based estimates of accurately how changes to landscape structure affect landscape resistance may therefore provide a useful the population connectivity of threatened species initial prediction of population connectivity, particu- (Spear et al. 2010; Zeller et al. 2012; Cushman et al. larly for those species for which a basic understanding 2013a). Such insights can provide a foundation upon of habitat associations is available (e.g., Riordan et al. which to develop effective conservation action 2015; Moqanaki and Cushman 2016). (Chetkiewicz et al. 2006). At its core, population The forests of Borneo host one of the richest connectivity is the product of the movement of biological assemblages on Earth, yet the island is also individual organisms across a landscape, the surface a global hotspot of forest loss and degradation of which varies in its resistive qualities. Such move- (Gaveau et al. 2014; Cushman et al. 2017). These ments are shaped by the compounding influences of anthropogenic driven changes to Borneo’s forests are the composition and structure of the landscape (Zeller exemplified by the Malaysian state of Sabah, which et al. 2013), the distribution and density of the occupies the northern part of the island. In 2010, forest population (Cushman 2006), and the specific dispersal accounted for 47.5% of the state’s land area traits of the species (e.g., Abrahms et al. 2017). Of (35,006 km2), following a rapid decline from 78.6% these, the complex interplay between a species’ in 1973, representing the highest deforestation rate of dispersal characteristics and landscape features is all the political units on Borneo during this period arguably the most important factor mediating land- (Gaveau et al. 2014). Selective logging activities have scape resistance and subsequent population connec- been the primary driver of forest degradation through- tivity (Spear et al. 2010; Zeller et al. 2012). In most out the state, and the subsequent conversion of these degraded forests to mono-culture plantations, chiefly 123 Landscape Ecol (2019) 34:275–290 277 that of oil palm (McMorrow and Talip 2001), remains connectivity across the entire island of Borneo. They the principal driver of forest loss (Gaveau et al. 2014). estimated that between 2000 and 2010 the proportion In 2015 oil palm plantations accounted for around of landscape connected by dispersal had fallen by 21% of land area (15,442 km2) in 2015 (Malaysian approximately 24% and the largest patch size had Palm Oil Board 2016). Understanding the impact of declined by around 30%, leading to a 13% decline in such changes to species of conservation concern clouded leopard numbers. Macdonald et al.’s (2018) remains a research priority. analysis, however, was based on an expert-opinion Individual species responses to logging regimes derived model of Sunda clouded leopard resistance to vary, but research is increasingly showing that selec- movement, and so warrants empirical testing. In tively logged Bornean forests can retain considerable addition, conservation is conducted at the regional levels of pre-disturbance biodiversity (e.g., Meijaard scale by state and provincial governments and thus et al. 2005; Costantini et al. 2016), as well as the effective planning of such action requires the devel- capacity to serve as corridors for less disturbance opment of connectivity predictions at these spatial tolerant species moving between intact forest frag- scales. ments. The establishment of industrial scale planta- In this paper we had two main objectives. First, we tions of oil palm Elaeis guineensis, however, can lead sought to extrapolate the Hearn et al. (2018) empirical to dramatic declines in species richness (e.g., Fitzher- resistance model to predict population density, genetic bert et al. 2008) and greatly inhibit connectivity of diversity and population connectivity for Sunda forest dependent species (e.g., Hearn et al. 2018). clouded leopards across the full extent of Sabah. Thus, for species of conservation concern on Borneo Second, we wished to quantify the differences in there is an urgent need for connectivity modelling to predicted population density, genetic diversity and assess impacts of landscape change to inform the population connectivity obtained from the Hearn et al. development of effective conservation strategies. (2018) empirically optimized and the Macdonald et al. The Sunda clouded leopard Neofelis diardi is the (2018) expert-opinion resistance surfaces at the full apex carnivore on the Sundaic islands of Borneo and Sabah extent. We hypothesised that (H1) the empirical Sumatra, where it is threatened with extinction (Hearn resistance model would produce higher estimates of et al. 2015). This felid is charismatic (Macdonald et al. population size, population density, genetic diversity 2015), wide-ranging (Hearn et al. 2013), and closely and overall connectivity than the expert-opinion associated with forest (Hearn et al. 2016, 2017), and derived model, but that (H2) the overall pattern of thus serves as a potential flagship species for Bornean predicted connectivity would be the same in the two wildlife and a useful model with which to develop analyses, identifying the same major core areas and predictions of connectivity. In the first study to explore main areas of connectivity between them. patterns of connectivity for the Sunda clouded leopard, Brodie et al. (2015) used hierarchical modelling of camera-trap data to assess and identify potential Methods dispersal and corridor routes within a transboundary network of protected areas in Borneo. Hearn et al. Study area (2018) analysed movement data within a path-selection framework to develop the first multi-scale, The Malaysian state of Sabah occupies an area of empirical connectivity predictions for a population 73,631 km2 in the northernmost portion of Borneo of Sunda clouded leopards in eastern Sabah, and (Fig. 1). Akin with the rest of the island, Sabah is showed that movement is facilitated by forest canopy characterised by a rugged topography, particularly in cover and resisted by non-forest vegetation, particu- central and western areas, which give way to coastal larly plantation areas with low canopy closure. In the alluvial plains. only large-scale analysis of Sunda clouded leopard Considerable areas of highly disturbed, regenerat- connectivity, Macdonald et al. (2018) used spatially ing forests, characterised by areas of scrub and synoptic modelling, combining resistant kernel and grassland are also present in the state. A number of factorial least cost path analysis (Cushman et al. relatively small (280–1399 km2) patches of protected 2013a, 2014), to predict patterns and changes in primary forest remain in the state, including the 123 278 Landscape Ecol (2019) 34:275–290

Fig. 1 Map of the Malaysian state of Sabah, northern Borneo, Reserve. Commercial Forest Reserves are outlined in dashed showing land use in 2010 (Gaveau et al. 2014). Fully protected black lines; key areas include (8) Ulu Kalumpang, (9) Sapulut, forest areas (National Parks, Wildlife Reserves and Conserva- (10) Trus Madi, (11) Tankulap-Piningah, (12) Deramakot and tion Areas) are outlined in solid black lines and include: (1) (13) Segaliud Lokan Forest Reserves. The Yayasan Sabah Danum Valley and (2) Maliau Basin Conservation Areas, (3) Forest Management Area is outlined in dark red. Polygons Crocker Range, (4) Kinabalu and (5) Tawau Hills Parks, (6) represent the state owned, Permanent Forest Reserve system. Lower Kinabatangan Wildlife Sanctuary and (7) Tabin Wildlife (Color ﬁgure online)

Danum Valley, Maliau Basin and Imbak Canyon Tawau Hills Park, are no longer contiguous with the Conservation Areas, and the Crocker Range, Kinabalu core forest regions. and Tawau Hills Parks (Fig. 1), but the vast majority of remaining forest has undergone one or more rounds Development of resistance layers of selective logging (Reynolds et al. 2011). The state- owned Permanent Forest Reserve, which includes We investigated connectivity metrics for Sunda State Parks, Wildlife Reserves as well as commercial clouded leopards across the full extent of Sabah, Forest Reserves, now accounts for the majority of based on two resistant surfaces, each developed using remaining forest (Reynolds et al. 2011). The rugged different methodological approaches. The two resis- areas along the Crocker and Trusmadi mountain tance models consisted of Hearn et al’s (2018) ranges retain large areas of forest in the western and empirically optimised, movement based model, and southwestern parts of the state, and sizeable areas in Macdonald et al.’s (2018) expert-opinion derived the central areas of the state remain forested, most model, hereafter referred to as the Empirical model notably the 10,000 km2 Yayasan Sabah Forest Man- and Expert model, respectively. The Empirical model agement area (YSFMA; Reynolds et al. 2011). resistance layer was developed by Hearn et al. (2018) Deforestation rates have been signiﬁcantly higher in using conditional logistic regression in a path-selec- the eastern side of Sabah, however, and a number of tion context (e.g., Cushman et al. 2010; Cushman and protected forest areas, including the Tabin Wildlife Lewis 2010), applied to movement data derived from Reserve, Lower Kinabatangan Wildlife Sanctuary and GPS tagged Sunda clouded leopards residing in an approximately 4000 km2 study extent in eastern

123 Landscape Ecol (2019) 34:275–290 279

Sabah, the Lower Kinabatangan Wildlife Sanctuary (Cushman et al. 2014). We seeded each of the two and surrounding oil palm matrix. Hearn et al. (2018) resistance layers with source points using the method used a multiscale approach and evaluated three scales described in Macdonald et al. (2018), in which of spatial shift to investigate relationships between clouded leopard habitat suitability is considered to clouded leopard movement paths and a range of be directly proportional to the inverse of resistance to landscape variables. To enable extrapolation of the movement. Specifically, we generated a raster of local-scale model to the entire state of Sabah, we identical pixel size and extent as the two resistance recomputed the path selection function based on layers, but with values randomly varying between 0 reclassified land cover data that was available at and 1. We then multiplied this uniform random map Sabah scale. We assumed that all kinds of upland with each of the two resistance layers, and selected all forest were equivalent, reclassifying two classes of pixels with value less than a constant chosen to montane forest to the same value as lowland tropical produce a set of 4540 points for the Empirical model forest which was present in the Hearn et al. (2018) resistance layer, which also produced a set of 3811 study area. points for the Expert resistance layer, since it had The expert model was developed by Macdonald higher overall resistance across Sabah and thus lower et al. (2018) who conducted a survey of the leading average suitability for clouded leopard occurrence. experts on Sunda clouded leopard ecology to obtain The true clouded leopard population size in Sabah is estimates of resistance values to be assigned to likely 1/4 of that of our source point population (Hearn different land cover types. The land cover classes et al. 2017). We chose to use a higher density of source were derived from those developed in a 250 m points to provide more spatial precision in the resolution 2010 land cover map of insular Southeast estimates of connectivity, given spatial uncertainty Asia, by Miettinen et al. (2012). Lowland forest and in the actual locations of clouded leopard home ranges lower montane forest were ranked as the highest and the fact that they are mobile animals and may quality habitat with mean scores of 4.50 and 4.25 out utilize multiple locations in their lifetimes (e.g., of 5, respectively, and urban, water and large-scale Moqanaki and Cushman 2016). plantation were given the lowest quality scores of 1.0, 1.05 and 1.11, respectively. Lowland Open and Resistant Kernel connectivity modelling Montane Open were also given very low-quality scores. These habitat suitability index scores were We used the least-cost resistant kernel approach translated into relative resistances by inverting and (Compton et al. 2007; Cushman et al. 2010) imple- scaling from a minimum of 1 to a maximum of 100, mented in UNICOR v2.0 (Landguth et al. 2012)to and subsequently applied to the Miettinen et al. (2012) predict the extent of the landscape connected by map to produce a resistance surface, within the Sabah dispersal across a 250,000 cost unit kernel width. This extent. kernel width was chosen since it is approximately the upper bound of the expected dispersal ability of Source points for connectivity modelling clouded leopards (Macdonald et al. 2018). The implementation of the cumulative resistant kernel For each of the two resistance layers we developed a used in this paper works by computing the cost- set of source points for use in connectivity modelling. distance kernel from each source location across the The resistance layers describe the local cost of moving resistance map and summing all such kernels to create through any given pixel which is the foundation for a cumulative resistant kernel surface that reflects the connectivity modelling; however, resistance surfaces incidence function of relative frequency of movement themselves are insufficient indicators of connectivity of the species through each location. Thus, the model (Cushman et al. 2009). Connectivity is a function of calculates the expected relative density of each species the resistance surface and the density, distribution and in each pixel around the source, given the dispersal dispersal ability of the dispersing population (Cush- ability of the species, the nature of the dispersal man et al. 2010). Thus, source points that reflect a function, and the resistance of the landscape (Compton realistic distribution and density of the population are et al. 2007; Cushman et al. 2010 ). The resistant kernel critical to reliable predictions of connectivity method of modelling landscape provides a 123 280 Landscape Ecol (2019) 34:275–290 comprehensive assessment of connectivity from the Predicted population size and genetic diversity source locations to all locations (i.e., many to all) and is computationally efficient, allowing implementation We used simulation modelling (e.g. Shirk et al. 2012; at broad scales and across multiple scenarios (e.g., Wasserman et al. 2012) with an individual-based, Cushman et al. 2012a, 2013a). spatially explicit population dynamics and genetics We compared the predicted connectivity obtained program (CDPOP version 1.0; Landguth and Cushman from the Empirical and Expert derived resistance 2010) to predict and compare the patterns and causes surfaces in several ways. First, we visually interpreted of differences in the local neighbourhood population the patterns of high and low resistance in each layer, density, distribution, and genetic diversity across the noting the major differences among them. Second, we two different resistance maps, at two dispersal visually interpreted the patterns of cumulative kernel distances. CDPOP simulates the birth, death, mating connectivity value in each layer, noting the major and dispersal of individuals in heterogeneous land- differences in predicted movement rates across the full scapes as probabilistic functions of the cost of extent of Sabah. Third, we computed FRAGSTATS movement through them. For each of the two (McGarigal et al. 2012) metrics on the mosaic of landscape resistance maps, we used the source cells patches predicted to be connected by dispersal in the used in the resistant kernel analysis as locations of cumulative kernel results for each resistance surface, simulated individual clouded leopards. We used across a range of connectivity thresholds (e.g., standard simulation parameters widely used in land- Wasserman et al. 2012). We chose two metrics to scape genetics simulation modelling (e.g., Cushman compute, the percentage of the landscape (PLAND) and Landguth 2010) and stipulated the population to and the correlation length (GYRATE_AM) predicted have 30 loci, with 10 alleles per locus, initially to be connected by dispersal at a given connectivity randomly assigned among individuals, and a mutation threshold value. The percentage of the landscape is the rate of 0.0005. We used an inverse square mating and most basic metric of landscape composition, yet dispersal probability function, with maximum disper- provides a useful quantification of the area predicted sal cost-weighted distances of 125,000 m and to be connected by dispersal. The correlation length 250,000 m, which reflect the estimated upper and measures the distance an organism can travel when lower range of expected clouded leopard dispersal placed at a random location in connected habitat and ability (Macdonald et al. 2018). Reproduction was assigned to move in a random direction before sexual with non-overlapping generations, and the reaching the edge of connected habitat (McGarigal number of offspring was based on a Poisson proba- et al. 2012), and has been shown to be a strong bility draw, with mean of 2. We ran 10 Monte Carlo predictor of functional connectivity (Cushman et al. runs in CDPOP for each of the two landscape 2012b). We chose connectivity thresholds at a range of resistance maps to assess stochastic variability. We cumulative resistant kernel values, including: 0, 10, simulated gene flow for 200 non-overlapping gener- 20, 40, 80, 160, 320 and 640, which span the range ations. Past studies have shown that this is sufficient from including all areas with any level of predicted time to ensure spatial genetic equilibrium (Landguth connectivity among them (0) to only those areas with et al. 2010a, b). We extracted several global measures exceptionally high predicted rates of clouded leopard of population genetic structure for the full study area at incidence and movement (640). Finally, we computed generation 200, including total population size, num- the intersection of the cumulative resistant kernel ber of alleles in the population, and observed and maps for the Empirical and Expert resistance models expected heterozygosity (Macdonald et al. 2018). We across a range of kernel density thresholds (Cushman analysed the differences in these global measures of et al. 2013b). We calculated the extent of each of the genetic structure between the two dispersal abilities three intersection components (connected in Empiri- using standard single factor analysis of variance cal only, Expert only, and connected in both. (ANOVA) implemented in the aov function of BASE package of R.

123 Landscape Ecol (2019) 34:275–290 281

Results maps is related to how they treat areas, particularly in the state’s western half, that are classiﬁed by Miettinen Landscape resistance and resistant kernel et al. (2012) as Plantation/Regrowth and Lowland connectivity models Mosaic, and by Gaveau et al. (2014) as Agroforest/ Forest regrowth. The Expert model predicted these There were striking differences between the Empirical areas to be relatively high resistance, while the and Expert resistance layers in predicted resistance to Empirical model predicted these areas to be relatively clouded leopard movement across the full extent of low resistance (mid blue). Some areas classiﬁed as Sabah (Fig. 2—top row). Both maps predict low Severely degraded and logged forest by Gaveau et al. resistance (dark blue) in areas of primary and selec- (2014), but as Plantation/Regrowth by Miettinen et al. tively logged forest, and both predict high resistance (2012), such as parts of south eastern Tabin Wildlife (red-yellow) in areas of non-forest, which, in the Reserve, Ulu Kalumpang Forest Reserve (contiguous eastern half of the state, are characterised primarily by with Tawau Hills Park) and the Bukit Pithon Forest oil palm plantations. The main difference between the Reserve (north east of the YSFMA) were predicted as

Fig. 2 Landscape resistance (top row) and cumulative resistant dark blue to high (100) in red; Cumulative resistant kernels kernel (bottom row) maps for Sunda clouded leopard movement developed using a 250,000 cost unit dispersal threshold; Red applied to the full extent of the State of Sabah, Malaysia. areas are predicted to have high density/frequency of utilization, Landscape resistance model based on multi-scale optimization blue areas low, and black areas are predicted to not be utilized by of a path-selection function; Resistance ranges from low (1) in clouded leopards. (Color ﬁgure online) 123 282 Landscape Ecol (2019) 34:275–290 relatively low resistance in the Empirical map, but as connected at the very highest connectivity values (less relatively high (yellow-orange) in the Expert map. than 10% at cumulative kernel values [ 640). The cumulative resistant kernel prediction of Correlation length of connected habitat was similar connectivity across Sabah differed substantially between the Expert and Empirical models for all areas between the two resistance models (Fig. 2, bottom connected with cumulative kernel value [ 0 (Fig. 3b). row). Both cumulative resistant kernel maps showed a There was a clear threshold at cumulative resistant major core area of connectivity, which encompasses kernel value of approximately 20, above which the the YSFMA and contiguous commercial Forest two connectivity surfaces departed in correlation Reserves of Deramakot, Tankulap-Piningah and length, with the Empirical kernel surface remaining Segaliud Lokan, to the north, and Sapulut to the highly connected while the correlation length of the southwest. Both approaches predicted losses of func- Expert surface declined dramatically (decrease of tional connectivity between the YSFMA and the 29.60 km correlation length, 37% less than the Lower Kinabatangan Wildlife Sanctuary, Tabin Wild- correlation length of the Empirical model at a life Reserve and Tawau Hills Park. The two models’ connectivity threshold of [ 40 cumulative resistant main differences were in the western half of the State, kernel value). This relative difference in correlation where the Expert resistance surface predicted much length of connected habitat remained the same up to less connectivity than the Empirical resistance surface. cumulative resistant kernel threshold of [ 320, and This is a result of the differing resistance assigned to then declined substantially at the highest connectivity areas classiﬁed as Plantation Regrowth and Lowland threshold value ([ 640). Mosaic by Miettinen et al. (2012), and as primarily Agroforest/Forest regrowth Plantation Regrowth by Intersection analysis Gaveau et al. (2014), as noted above. Despite these differences, both models predicted that these western We computed the intersection of the cumulative and northern regions, which include the regionally resistant kernel maps for the Empirical and Expert important Crocker Range and Kinabalu Parks, resistance maps, across a range of kernel density remained functionally connected to the core areas of thresholds (Fig. 4). We calculated the extent of each of connectivity in and around the YSFMA and adjacent the three intersection components [connected in the commercial Forest Reserves. Empirical map only, connected in the Expert map We computed FRAGSTATS metrics on the mosaic only, and connected in both maps (Fig. 5)]. At low of patches predicted to be connected by dispersal in the cumulative kernel thresholds, the vast majority of the cumulative resistant kernel results for the two resis- area predicted to be connected in either map is tance surfaces, across a range of connectivity thresh- connected in both maps (Fig. 4a, 5). The proportion of olds. The percentage of Sabah predicted to be intersection declines as the cumulative resistant kernel connected by dispersal across cumulative kernel threshold increases, such that at high levels of density thresholds was substantially different between connectivity there is much less overlap between the the two resistance layers (Fig. 3a). At all dispersal areas predicted to be connected in the two approaches thresholds, cumulative resistant kernel surface (Fig. 4e–g). At the very highest level of connectivity obtained from the Empirical model had higher exten- value there is much less total area predicted to be siveness than that obtained from the Expert model. connected in either analysis, and only at this very high The relative difference in the percentage of Sabah level of connectivity value does the Expert analysis connected by dispersal increased as the connectivity predict connectivity in areas not also predicted to be threshold increased. At the most liberal threshold connected in the Empirical analysis. (cumulative kernel value greater than 0) the Empirical model cumulative resistant kernel surface had a 16% Predicted population size and genetic diversity greater area than the Expert model resistant kernel surface. This difference increased at the higher levels We used the program CDPOP to predict and compare of cumulative kernel value, reaching a difference of differences in population size, distribution, and 20% at cumulative kernel values greater than 320 genetic diversity across the two different resistance (Fig. 3a). Both maps had low extents predicted to be maps. There was considerable local variation in 123 Landscape Ecol (2019) 34:275–290 283

Fig. 3 a Percentage of the landscape predicted to be connected and b Correlation length of connected habitat across a range of cumulative resistant kernel surface thresholds, for the Hearn et al. (2018) Empirical model and Macdonald et al. (2018) Expert-opinion model

simulated local neighbourhood population density of Segaliud Lokan Forest Reserves. The Empirical model clouded leopards across Sabah among the four scenarios also predicted relatively high population CDPOP scenarios (Fig. 6). The general pattern was densities in the highland areas of the Crocker and Trus for broader distribution and larger populations in the Madi mountain ranges and along the border with Empirical resistance map simulations than the Expert Kalimantan and Sarawak, in south-western Sabah, but resistance map simulations, and in the long versus these areas were predicted as only moderate density in short dispersal ability simulations. We found that the the Expert model scenarios. All four simulations show effect of resistance layer on simulated population size populations persisting in the relatively isolated forest across the full extent of Sabah was statistically patches of the Crocker Range and Kinabalu Parks, and signiﬁcant, whereas the effect of dispersal ability Tabin Wildlife Reserve. In contrast, only the simula- was not (Table 1; Fig. 7a). All four simulations tions on the Empirical resistance maps show the predicted that population density was highest within population persisting in Tawau Hills Park, and Sunda a core central area which encompassed the YSFMA clouded leopards were predicted to persist in the and adjacent Deramakot, Tankulap-Piningah and 123 284 Landscape Ecol (2019) 34:275–290

Fig. 4 Intersections between the cumulative resistant kernel both maps, across eight different cumulative kernel connectivity surfaces produced on the Expert model (Macdonald et al. 2018) values (shown in top left of each map) map only, Empirical model (Hearn et al. 2018) map only, and

Lower Kinabatangan Wildlife Sanctuary only in the long dispersal distance than the short dispersal Empirical simulation with the high dispersal ability. distance. There were significant differences in the number of alleles in the Sabah-wide clouded leopard population among resistance maps, but not dispersal abilities or Discussion their interaction (Table 1, Fig. 7b), with the simulations on the Empirical resistance map producing Our goals in this study were to predict the local significantly higher allelic richness than simulations neighbourhood population density, genetic diversity on the Expert resistance map. There were significant and map patterns of population connectivity for Sunda differences in the heterozygosity of the Sabah-wide clouded leopards across Sabah and compare the clouded leopard population among resistance maps differences in predictions obtained from expert-opin- and dispersal abilities, but not their interactions ion and empirically derived resistance maps. Consis- (Table 1, Fig. 7c). Specifically, heterozygosity was tent with our first hypothesis, we found that the significantly higher in the CDPOP simulations on the Empirical resistance model produced higher estimates Empirical resistance map than on the Expert resistance of population size, population density, genetic diver- map, and significantly higher in simulations with the sity and overall connectivity than the Expert model.

123 Landscape Ecol (2019) 34:275–290 285

Fig. 5 Intersection analysis of areas predicted to be connected by across a range of cumulative resistant kernel surface thresholds: dispersal on the Empirical model (Hearn et al. 2018) map only, ([ 0, [ 10, [ 20, [ 40, [ 80, [ 160, [ 320, [ 640) Expert model (Macdonald et al. 2018) map only, and both maps,

Fig. 6 Mean predicted local neighbourhood density of clouded dispersal ability (250,000 cost unit dispersal threshold); bottom leopards across Sabah under four simulation scenarios: First row: Long dispersal ability (125,000 cost unit dispersal column: Macdonald et al. (2018) Expert-opinion model; second threshold) column: Hearn et al. (2018) Empirical model; top row: Short

123 286 Landscape Ecol (2019) 34:275–290

Table 1 Results of Two-way Analysis of Variance of differences between simulated populations, alleles, and heterozygosity as function of dispersal ability and the resistance map used Effect Analysis of variance P-values Population size Alleles Heterozygosity

Dispersal ability 0.533 0.191 0.021 Resistance map 0.010 0.025 0.035 Interaction: dispersal ability—resistance map 0.323 0.369 0.730

Fig. 7 Boxplots of simulated mean a Sunda clouded population (Macdonald et al. 2018) resistance maps, at 250,000 cost units size across Sabah, b number of alleles in the Sabah-wide (High) and 125,000 cost units (Low) dispersal abilities. Error population of Sunda clouded leopards, c heterozygosity for the bars represent 95% Conﬁdence intervals and boxes represent Sabah-wide population of Sunda clouded leopards, for the interquartile ranges (25–75%) Empirical model (Hearn et al. 2018) and Expert-opinion model

The Expert based resistance layer produced by Mac- propagated to large differences in predicted popula- donald et al. (2018) was based on relative suitability tion density, genetic diversity and population connec- estimates for different landcover types, and several tivity, with the Empirical model predicting higher landcover types that are common in western Sabah density, genetic diversity and connectivity in these were absent in the Kinabatangan study area; thus the parts of Sabah. It is impossible to determine based on two resistance layers and connectivity predictions existing data which resistance parameterization is obtained from them may differ in areas where these more accurate. Further work will be required to cover types are common. In addition, the Expert document patterns of density, genetic diversity and resistance layer was based solely on the expert opinion movement and/or genetic differentiation as a function weights given to land cover classes, while the of landscape features in this region of Sabah where the Empirical model also included additional variables predictions described here differ. related to canopy cover. Thus, the Empirical model Our second hypothesis proposed that, despite their predicted relatively low resistance in some areas with differences, the overall pattern of predicted connec- moderately high canopy cover, even if they were in tivity would be similar between the two models, cover types given relatively high resistance in the identifying the same major core areas and main areas Expert model. For example, the Expert model pre- of connectivity between them. Additionally, we dicted much higher resistance in a large region of expected that there would be higher similarity between northern and western Sabah, specifically in areas the connectivity predictions than between the resis- classified as Plantation/Regrowth and Lowland tance surfaces themselves, since connectivity is a Mosaic by Miettinen et al. (2012) and as Agroforest/ spatially contagious spread process that smooths local Forest regrowth by Gaveau et al. (2014). The differ- differences in resistance. Our results largely support ences in inferred resistance value in these areas this hypothesis. Both analyses identified a large patch

123 Landscape Ecol (2019) 34:275–290 287 of core habitat with high predicted cumulative resis- clouded leopard occurrence and dispersal. The best tant kernel connectivity value in the YSFMA and way to resolve the differences between the empirical contiguous commercial Forest Reserves. Importantly, and expert predictions would be to obtain additional however, the analysis based on the Empirical model data on clouded leopard occurrence patterns, genetic predicted this core area to extend west along the border structure and movement. Broad-scale monitoring of with Kalimantan and Sarawak, in south-western occurrence patterns would enable empirical estimates Sabah, and also westward, encompassing and linking of distribution and abundance that could be used to the mountainous areas of the Trus Madi Forest validate the two predictions presented here. In addi- Reserve and Crocker Range and Kinabalu Parks. In tion, further work with empirical modelling of resis- contrast, the Expert based resistance model predicted tance based on telemetry in other parts of Sabah, these areas to have low connectivity and low rates of ideally targeting the range of different age/sex classes, predicted movement of clouded leopards through and focusing on habitat types not included in the them. So while the models agreed in terms of the current model, would help to generalize the empirical location of the most important core area they differed model across the broader extent, enabling more robust substantially in the extent of that core population. comparison of the Empirical vs. Expert-opinion Despite their differences, both models agreed on models. This kind of meta-replicated study to gener- the location and extent of the main isolated fragments alise ecological relationships across broad scales has of internally connected habitat. Namely, they each been highly useful for other carnivore species (e.g., identified the Lower Kinabatangan Wildlife Sanctu- Short Bull et al. 2011; Shirk et al. 2014), and has been ary, Tabin Wildlife Reserve and Tawau Hills Park as identified as one of the keys to reliable inferences patches of habitat predicted to have extant clouded about pattern-process relationships at the landscape leopard populations, but predicted to be isolated from level (McGarigal and Cushman 2002). other populations. Efforts should be made to explore Future research should also focus on how different mechanisms to increase connectivity between these landscape management approaches affect the conser- areas and the main central forest, such as establish- vation of this felid. In this regard it would be valuable ment of riparian corridors, and identification and/or to develop a scenario-based analysis that includes creation of High Conservation Value forest areas likely development and conservation actions for the within plantations landscapes. region and use the modelling approach presented here to quantitatively measure their relative impacts on Scope and limitations population size, connectivity and genetic diversity. We hope that future work will help to close the gap in To apply the Hearn et al. (2018) resistance model to understanding through a combination of occurrence, the full extent of Sabah we assumed that all kinds of genetics and movement modelling for this species upland forest were equivalent, reclassifying two across its range. classes of montane forest to the same value as lowland Landscape connectivity predictions are the com- tropical forest (which was present in the Hearn et al. bined result of three main things: (1) the density and (2018)) study area. Whilst there are no empirical distribution of the source population, (2) landscape movement data to test these assumptions (e.g., the resistance and (3) dispersal ability. Our goal was to movement data used in the empirical model was from evaluate how the two different resistance maps, from a part of Sabah where these upland forest types are expert and empirical analysis, differ in their implica- absent), occurrence data from camera trap studies tions for connectivity. In our analysis the two resis- support the notion that clouded leopards are found in tance maps affect both (1) density and distribution of these forested uplands at relatively high densities source points and (2) landscape resistance. It would be (Hearn et al. 2017). However, in the absence of such good to isolate the effects of differential resistance by empirical movement data, and since the movement holding density and distribution of source points model was based on data from a small number of constant. However, source points must be located to individuals, we should view this model as preliminary. represent the density and distribution of the subject Further research should strive to improve ecolog- population. The two resistance maps suggest quite ical understanding of how landscape conditions affect different things about clouded leopard habitat and 123 288 Landscape Ecol (2019) 34:275–290 movement selection, such that it would be incorrect to accurately inform corridor conservation? J Appl Ecol use source points predicted by one to analyse connec- 54:412–422 Blazquez-Cabrera S, Gastoń A, Beier P, Garrote G, SimońMA´ , tivity of the other. The locations of the points needs to Saura S (2016) Influence of separating home range and be proportional to predicted suitability, and the dispersal movements on characterizing corridors and suitability is different in the expert and empirical effective distances. Landscape Ecol 31:2355–2366 analysis so the points are different and must be Brodie JF, Giordano AJ, Dickson B, Hebblewhite M, Bernard H, Mohd-Azlan J, Anderson J, Ambu L (2015) Evaluating different between the two analyses. As a result we multispecies landscape connectivity in a threatened tropi- cannot separate source points and landscape resis- cal mammal community. Conserv Biol 29(1):122–132 tance, and thus our analysis is of the differences in the Castillo JA, Epps CW, Davis AR, Cushman SA (2014) Land- predicted connectivity and population size implied by scape effects on gene flow for a climate-sensitive montane species, the American pika. Mol Ecol 23(4):843–856 the two maps in full, which includes both effects of Chetkiewicz C-LB, St. Clair CC, Boyce MS (2006) Corridors different source points and effects of different land- for conservation: integrating pattern and process. Annu scape resistance. Future work should assess the Rev Ecol Evol Syst 37:317–342 independent and interactive effects of source points Compton BW, McGarigal K, Cushman SA, Gamble LR (2007) A resistant-kernel model of connectivity for amphibians and resistance on connectivity. That is a more that breed in vernal pools. Conserv Biol 21(3):788–799 theoretical question than what this paper explored Costantini D, Edwards DP, Simons MJ (2016) Life after logging and would best be addressed in a simulation study in in tropical forests of Borneo: a meta-analysis. Biol Cons which neutral models (e.g., Cushman et al. 2013c; 196:182–188 Cushman SA (2006) Effects of habitat loss and fragmentation on Shirk et al. 2018) are used to generate landscapes of amphibians: a review and prospectus. Biol Conserv different patterns of resistance and different densities 128(2):231–240 and distributions of points, and then landscape Cushman SA, Landguth EL (2010) Spurious correlations and connectivity and landscape genetic analyses carried inference in landscape genetics. Mol Ecol 19:3592–3602 Cushman SA, Lewis JS (2010) Movement behavior explains out on combinations of resistance and source points, genetic differentiation in American black bears. Landscape enabling the separation of their effects. Ecol 25(10):1613–1625 Cushman SA, McKelvey KS, Hayden J, Schwartz MK (2006) Acknowledgements We thank Sabah Wildlife Department, Gene flow in complex landscapes: testing multiple Sabah Forestry Department and the Sabah Biodiversity Centre hypotheses with causal modeling. Am Nat 168(4):486–499 for permission to conduct research. This research was funded by Cushman SA, McKelvey KS, Schwartz MK (2009) Use of the Robertson Foundation, Sime Darby Foundation, Recanati- empirically derived source-destination models to map Kaplan Foundation, Clouded Leopard Project, Dr Holly Reed regional conservation corridors. Conserv Biol Wildlife Conservation Fund, Point Defiance Zoo and Aquarium, 23(2):368–376 Houston Zoo, and Panthera. Cushman SA, Chase M, Griffin C (2010) Mapping landscape resistance to identify corridors and barriers for elephant Open Access This article is distributed under the terms of the movement in southern Africa. In: Cushman SA, Huettmann Creative Commons Attribution 4.0 International License (http:// F (eds) Spatial complexity, informatics, and wildlife con- creativecommons.org/licenses/by/4.0/), which permits unre- servation. Springer, Tokyo, pp 349–367 stricted use, distribution, and reproduction in any medium, Cushman SA, Landguth EL, Flather CH (2012a) Evaluating the provided you give appropriate credit to the original sufficiency of protected lands for maintaining wildlife author(s) and the source, provide a link to the Creative Com- population connectivity in the US northern Rocky Moun- mons license, and indicate if changes were made. tains. Divers Distrib 18(9):873–884 Cushman SA, Shirk A, Landguth EL (2012b) Separating the effects of habitat area, fragmentation and matrix resistance on genetic differentiation in complex landscapes. Land- Data availability The datasets generated during and/or scape Ecol 27(3):369–380 analysed during the current study are not publicly available Cushman SA, Lewis JS, Landguth EL (2013a) Evaluating the due to the fact that they contain location data regarding a intersection of a regional wildlife connectivity network threatened species in a protected area, but are available from the with highways. Mov Ecol 1(1):12 corresponding author on reasonable request. Cushman SA, McRae B, Adriaensen F, Beier P, Shirley M, Zeller K (2013b) Biological corridors and connectivity. In: Macdonald DW, Willis KJ (eds) Key topics in conservation References biology 2. Wiley, Oxford, pp 384–404. https://doi.org/10. 1002/9781118520178.ch21 Abrahms B, Sawyer SC, Jordan NR, McNutt JW, Wilson AM, Cushman SA, Shirk AJ, Landguth EL (2013c) Landscape Brashares JS (2017) Does wildlife resource selection genetics and limiting factors. Conserv Genet 14(2):263–274 123 Landscape Ecol (2019) 34:275–290 289

Cushman SA, Lewis JS, Landguth EL (2014) Why did the bear charismatic cat: Felis felicis. Global Ecol Conserv cross the road? Comparing the performance of multiple 3:851–866 resistance surfaces and connectivity modelling methods. Macdonald EA, Cushman SA, Landguth EL, Hearn AJ, Malhi Diversity 6(4):844–854 Y, Macdonald DW (2018) Implications of rapid forest loss Cushman SA, Elliot NB, Macdonald DW, Loveridge AJ (2016) on population size, connectivity and genetic diversity of A multi-scale assessment of population connectivity in clouded leopards in Borneo. PLoS ONE 13(9):e0196974. African lions (Panthera leo) in response to landscape https://doi.org/10.1371/journal.pone.0196974 change. Landscape Ecol 31(6):1337–1353 Malaysian Palm Oil Board (2016) Economics and industry Cushman SA, Macdonald EA, Landguth EL, Malhi Y, Mac- development division. See http://bepi.mpob.gov.my/. donald DW (2017) Multiple-scale prediction of forest loss Accessed May 2016 risk across Borneo. Landscape Ecol 32(8):1581–1598 Mateo Sańchez MC, Cushman SA, Saura S (2014) Scale Fitzherbert EB, Struebig MJ, Morel A, Danielsen F, Bru¨hl CA, dependence in habitat selection: the case of the endangered Donald PF, Phalan B (2008) How will oil palm expansion brown bear (Ursus arctos) in the Cantabrian Range (NW affect biodiversity? Trends Ecol Evol 23(10):538–545 Spain). Int J Geogr Inf Sci 28(8):1531–1546 Gaveau DL, Sloan S, Molidena E, Yaen H, Sheil D, Abram NK, Mateo-Sańchez MC, Balkenhol N, Cushman SA, Pe´rez T, Ancrenaz M, Nasi R, Quinones M, Wielaard N, Meijaard E Domıńguez A, Saura S (2015) A comparative framework (2014) Four decades of forest persistence, clearance and to infer landscape effects on population genetic structure: logging on Borneo. PLoS ONE 9(7):e101654 are habitat suitability models effective in explaining gene Hearn AJ, Ross J, Pamin D, Bernard H, Hunter L, Macdonald flow? Landscape Ecol 30(8):1405–1420 DW (2013) Insights into the spatial and temporal ecology McGarigal K, Cushman SA (2002) Comparative evaluation of of the Sunda clouded leopard Neofelis diardi. Raffles Bull experimental approaches to the study of habitat fragmen- Zool 61(2):871–875 tation effects. Ecol Appl 12(2):335–345 Hearn AJ, Ross J, Brodie J, Cheyne S, Haidir IA, Loken B, McGarigal K, Cushman SA, Ene E (2012) FRAGSTATS v4: Mathai J, Wilting A and McCarthy J (2015) Neofelis dia- spatial pattern analysis program for categorical and con- rdi. The IUCN Red List of Threatened Species 2015 tinuous maps. Computer software program produced by the Hearn AJ, Ross J, Macdonald DW, Bolongon G, Cheyne SM, authors at the University of Massachusetts, Amherst. http:// Mohamed A, Samejima H, Brodie JF, Giordano A, Alfred www.umass.edu/landeco/research/fragstats/fragstats.html R, Boonratana R, Bernard H, Loken B, Augeri DM, Hey- McMorrow J, Talip MA (2001) Decline of forest area in Sabah, don M, Hon J, Mathai J, Marshall AJ, Pilgrim JD, Hall J, Malaysia: relationship to state policies, land code and land Breitenmoser-Wu¨rsten C, Kramer-Schadt S, Wilting A capability. Glob Environ Change 11(3):217–230 (2016) Predicted distribution of the Sunda Clouded leopard Meijaard E, Sheil D, Nasi R, Augeri D, Rosenbaum B, Iskandar Neofelis diardi (Mammalia: Carnivora: Felidae) on Bor- D, Setyawati T, Lammertink M, Rachmatika I, Wong A, neo. Raffles Bull Zool Suppl No. 33:165–172 Soehartono T, Stanley S, O’Brien T (2005) Life after Hearn AJ, Ross J, Bernard H, Bakar SA, Goossens B, Hunter logging: reconciling wildlife conservation and production LT, Macdonald DW (2017) Responses of Sunda clouded forestry in Indonesian Borneo. Centre for International leopard Neofelis diardi population density to anthro- Forestry Research, Bogor pogenic disturbance: refining estimates of its conservation Miettinen J, Shi C, Tan WJ, Liew SC (2012) 2010 land cover status in Sabah. Oryx. https://doi.org/10.1017/ map of insular Southeast Asia in 250-m spatial resolution. S0030605317001065 Remote Sens Lett 3(1):11–20 Hearn AJ, Cushman SA, Goossens B, Macdonald E, Ross J, Moqanaki EM, Cushman SA (2016) All roads lead to Iran: Hunter L, Abram NK, Macdonald DW (2018) Evaluating predicting landscape connectivity of the last stronghold for scenarios of landscape change for Sunda clouded leopard the critically endangered Asiatic cheetah. Anim Conserv connectivity in a human dominated landscape. Biol Con- 20(1):29–41 serv 222:232–240 Reynolds G, Payne J, Sinun W, Mosigil G, Walsh RPD (2011) Landguth EL, Cushman SA (2010) CDPOP: a spatially explicit Changes in forest land use and management in Sabah, cost distance population genetics program. Mol Ecol Malaysian Borneo, 1990-2010, with a focus on the Danum Resourc 10:156–161 Valley region. Philos Trans R Soc B 366:3168–3176 Landguth EL, Cushman SA, Murphy MA, Luikart G (2010a) Riordan P, Cushman SA, Mallon D, Shi K, Hughes J (2015) Relationships between migration rates and landscape Predicting global population connectivity and targeting resistance assessed using individual-based simulations. conservation action for snow leopard across its range. Mol Ecol Resourc 10(5):854–862 Ecography 39(5):419–426 Landguth EL, Cushman SA, Schwartz MK, McKelvey KS, Shirk, AJ, Wallin DO, Cushman SA, Rice CG, Warheit KI Murph M, Luikart G (2010b) Quantifying the lag time to (2010) Inferring landscape effects on gene flow: a new detect barriers in landscape genetics. Mol Ecol model selection framework. Mol Ecol 19:3603–3619 19(19):4179–4191 Shirk AJ, Cushman SA, Landguth EL (2012) Simulating pat- Landguth EL, Hand BK, Glassy J, Cushman SA, Sawaya MA tern-process relationships to validate landscape genetic (2012) UNICOR: a species connectivity and corridor net- models. International J Ecol. https://doi.org/10.1155/2012/ work simulator. Ecography 35:9–14 539109 Macdonald EA, Burnham D, Hinks AE, Dickman AJ, Malhi Y, Shirk AJ, Raphael MG, Cushman SA (2014) Spatiotemporal Macdonald DW (2015) Conservation inequality and the variation in resource selection: insights from the American marten (Martes americana). Ecol Appl 24(6):1434–1444 123 290 Landscape Ecol (2019) 34:275–290

Shirk AJ, Schroeder MA, Robb LA, Cushman SA (2015) population connectivity of American marten (Martes Empirical validation of landscape resistance models: americana) in the northern Rocky Mountains, USA. insights from the Greater Sage-Grouse (Centrocercus Landscape Ecol 27(2):211–225 urophasianus). Landscape Ecol 30(10):1837–1850 Zeller KA, McGarigal K, Whiteley AR (2012) Estimating Shirk AJ, Landguth EL, Cushman SA (2018) A comparison of landscape resistance to movement: a review. Landscape regression methods for model selection in individual-based Ecol 27(6):777–797 landscape genetic analysis. Mol Ecol Resour 18(1):55–67 Zeller KA, Rabinowitz A, Salom-Perez R, Quigley H (2013) Short Bull RA, Cushman SA, Mace R, Chilton T, Kendall KC, The Jaguar corridor initiative: a range-wide conservation Landguth EL, Schwartz MK, McKelvey K, Allendorf FW, strategy. In: Ruiz-Garcia M, Shostell JM (eds) Molecular Luikart G (2011) Why replication is important in landscape population genetics, evolutionary biology and biological genetics: American black bear in the Rocky Mountains. conservation of neotropical carnivores. Nova Science Mol Ecol 20(6):1092–1107 Publishers, New York, pp 629–658 Spear SF, Balkenhol N, Fortin MJ, McRae BH, Scribner KIM Zeller KA, Vickers TW, Ernest HB, Boyce WM, Pollinger J, (2010) Use of resistance surfaces for landscape genetic Ernest H (2017) Multi-level, multi-scale resource selection studies: considerations for parameterization and analysis. functions and resistance surfaces for conservation plan- Mol Ecol 19(17):3576–3591 ning: Pumas as a case study. PLoS ONE 12:e0179570 Wasserman TN, Cushman SA, Schwartz MK, Wallin DO (2010) Zeller KA, Jennings MK, Vickers TW, Ernest HB, Cushman Spatial scaling and multi-model inference in landscape SA, Boyce WM (2018) Are all data types and connectivity genetics: Martes americana in northern Idaho. Landscape models created equal? Validating common connectivity Ecol 25(10):1601–1612 approaches with dispersal data. Divers Distrib 24:868–879 Wasserman TN, Cushman SA, Shirk AS, Landguth EL, Littell JS (2012) Simulating the effects of climate change on

123