Introduction

Land cover change is extensive, with over 40% of global terrestrial land converted to human-dominated uses (Foley et al. 2005; Ellis and Ramankutty 2008). Mammalian carnivore populations are in decline worldwide (Ripple et al. 2014), and agricultural land use is highlighted as a significant driver of extinction events (Dobrovolski et al. 2013). As a guild, carnivores are considered particularly sensitive to landscape changes due to traits such as long lifespans, slow reproduction rates, high trophic positions, large home ranges, and low metapopulation densities (Keinath et al. 2017). The full extent of carnivore responses to human modified landscapes is, however, dependent on the ecology of the focal species (Ryall and Fahrig 2006). Within the guild, generalists display elevated plasticity to land use modifications compared to specialists (Gehring and Swihart 2002), with a range of species adapting to life across gradients of human disturbance (e.g. Bouyer et al. 2014; Jerosch et al. 2017; Suraci et al. 2019).

The ecological consequences of land use changes are a function of the interplay between a species’ adaptability and the configuration of altered landscapes. Following human modification, the size, shape, composition, and connectivity of remaining natural habitats affect carnivores by either limiting or facilitating animal movement (Crooks et al. 2011). As an example, agricultural land flanked by or containing remnant habitat patches displays enhanced carnivore occupancy (Santos et al. 2016). Persistence can be further facilitated by the degree of functional impedance posed by the agricultural land use itself; if the matrix is permeable to, or can be successfully utilized by a species, a population’s survival probability further increases (Ferreira et al. 2018). Through the quantification of carnivore movements within heterogeneous landscapes, scientists, resource managers, and conservationists gain explicit and mechanistic insights into the habitat requirements of species persisting alongside agriculture (Nathan et al. 2008; van Moorter et al. 2016). These insights, in turn, inform more effective management actions, such as establishment of protected areas (Desmet 2018), implementation of land-sharing schemes within agriculture itself (Perfecto and Vandermeer 2008), or creation of functionally optimized wildlife corridors (Chetkiewicz and Boyce 2009).

Oil palm (Elaeis guineensis) agriculture is a significant driver of tropical forest loss and degradation (Phalan et al. 2013). Relative to highly diverse primary forests, the establishment of oil palm plantations results in marked declines in species’ abundances, assemblage complexity, and fundamental ecosystem functions (Dislich et al. 2017). Forest specialists with restricted dietary niches or behaviors cannot adapt to agricultural conversion (Foster et al. 2011). As such, species that do persist within the monoculture are often generalists (be it in dietary or habitat terms), pests, or invasive (Rembold et al. 2017). Given the forecasted increases in global agricultural degradation (Tilman et al. 2001), it is crucial to understand how generalists utilize these landscapes and what actions, if any, may be taken to facilitate biodiversity retention (Sodhi et al. 2010).

The island of Borneo is a biodiversity hotspot (Myers et al. 2000), yet is experiencing one of the highest deforestation rates due to oil palm agriculture (Gaveau et al. 2019). A total of 25 small- to medium-sized carnivore species are found in Borneo, representing one of the richest co-existing extant guilds in the world (Ross et al. 2017). The guild includes a spectrum of forest specialist and generalist species; the island has been highlighted as a region of carnivore conservation priority (Schipper et al. 2008). Effective management of the guild, however, is hindered by a paucity of knowledge regarding the ecology and behaviors of these species, particularly tolerance and fine-scale responses to human-modified landscapes (Mathai et al. 2016). Studies report limited carnivore flexibility to oil palm plantations, such as distinct avoidance by GPS collared Sunda clouded leopards (Neofelis diardi; Hearn et al. 2018). Most research thus concludes the guild will only be sustained with strict land-sparing conservation efforts (e.g. Yue et al. 2015; Wearn et al. 2017). However, documentation of several generalist species in close proximity to and within oil palm plantations suggests there are more nuanced processes facilitating carnivore persistence (Jennings et al. 2015; Evans et al. 2016a). One such species is the Malay civet (Viverra tangalunga), a 3–7 kg ground-dwelling, nocturnal, and predominantly solitary carnivore recorded in a range of habitat types, including within oil palm plantations (Duckworth et al. 2016). With advancements in animal tracking technologies, an unprecedented quantity and quality of spatiotemporal movement data can be directly collected from the species (Evans et al. 2016b). This opportunity facilitates the explicit study of carnivore functional responses to degraded and heterogeneous landscapes, which will provide insights into ecological mechanisms of carnivore persistence.

We quantified the spatial movements and habitat use of Malay civets, a generalist carnivore persisting within the degraded lowland forest fragments and oil palm plantations of northeastern Borneo. We combined animal movement data from GPS collars with high-resolution remotely sensed landscape characteristics to investigate the patterns and underlying landscape drivers of civet space use. Specifically, we sought to identify if oil palm plantations were permeable to civet movements, and if so, to determine the extent of usage. We then related patterns of core and total home range sizes to an individual’s use of oil palm agriculture to determine habitat suitability. We explored how habitat utilization of civets varied in response to proximity to agricultural and natural edges, terrain ruggedness, forest structure, and forest functional diversity. We sought to provide insights into effective conservation management strategies to facilitate carnivore persistence across agriculturally modified landscapes.

Methods

Study area

The research was conducted in the Lower Kinabatangan Floodplain in eastern Sabah, Malaysian Borneo (approximate range: 5° 18′ N to 5° 42′ N and 117° 54′ E to 118° 33′ E; Fig. 1). The Kinabatangan River is Sabah’s largest river, stretching 560 km and encompassing a total catchment area of 1.68 million ha (Harun et al. 2015). Selective logging for economically viable trees began in the 1950s and marked the beginning of the region’s large-scale forest degradation (McMorrow and Talip 2001). The wet tropical climate and alluvial soils are highly suitable for oil palm cultivation (Pirker et al. 2016), and the establishment of large-scale plantations began in the early 1980s. The majority of the forests in the Floodplain have been converted into oil palm plantations (Francis 2017), and the Kinabatangan estates account for 28% of Sabah’s total cultivated area of oil palm crop (Abram et al. 2014). The remaining patches of natural land cover flanking the river comprise a mixture of dry lowland, semi-inundated, swamp, and riparian forests interspersed with small grasslands. As of 2014, over 98% of the remaining forested areas in the Floodplain were within 1 km of a forest edge (Francis 2017).

Fig. 1
figure 1

Map of the Lower Kinabatangan Floodplain showing the protected areas and field centre (Danau Girang Field Centre) in Sabah, Malaysian Borneo. The gray areas in the main map denote non-forested land classes dominated by oil palm plantations and small settlements; specifically labeled lots indicate the focal study area

Full size image

Within this mosaic landscape is the 27,000 ha Lower Kinabatangan Wildlife Sanctuary. The Sanctuary is composed of 10 forest lots of varying degrees of isolation and disturbance history. The protected area network includes seven Forest Reserves, and in combination with the Sanctuary, sums to approximately 45,000 ha along 150 km of the lower reaches of the river (Fig. 1). An estimated 30,000 ha of additional forest exist outside protected areas on private or state lands (Abram et al. 2014). This research occurred within the Sanctuary lots 5, 6, and 7 and the surrounding oil palm estates. The Lower Kinabatangan Floodplain thus represents an ideal landscape to assess biodiversity persistence to forest fragmentation and degradation by oil palm plantations.

Civet GPS collaring

Live trapping began in October 2013 and was completed in December 2019. We captured civets using either custom-built box traps or commercial spring-loaded traps nocturnally set to avoid capture of non-target diurnal wildlife. Civets were anesthetized by a qualified veterinarian and fitted with GPS collars (e-obs 1A, e-obs GmbH, Gruenwald, Germany) set to collect hourly positions from 18:00 through 06:00. Successful trapping occurred across the study area, and collars were deployed within remnant forest ‘blocks’ (i.e. Sanctuary lots 6 and 7), narrow forest corridors (i.e. Sanctuary lot 5), and directly along plantation boundaries; all animals were captured within 3 km of an oil palm plantation. All collared individuals were male, with the exception of one experimental deployment on a female, due to morphometric constraints (head:neck circumference ratios were sexually dimorphic, and it was difficult to safely collar females; Supplementary Table 1). For further capture methodology, anaesthesia protocols, and tag specifications, see details provided in Evans et al. (2016b). All animal handling protocols were approved by the Sabah Wildlife Department and the Sabah Biodiversity Centre (license ref.no: JKM/MBS.10000–2/2 JLD.6[8]). Capture techniques and GPS collar deployment followed the guidelines set by the American Society of Mammalogists (Sikes et al. 2016).

Table 1 Resources included as environmental covariates in resource utilization functions evaluating space use of Malay civets (Viverra tangalunga) at the home range scale
Full size table

Civet space use

Civet space use was assessed using the biased random bridges (BRB) model, a movement-based home range estimator that accounts for the spatial and temporal information of GPS collar data (Benhamou and Cornélis 2010; Benhamou 2011). This estimator produces a 3-dimensional probabilistic map of animal space use, called a utilization distribution (UD), which quantifies the relative intensity of animal space use within a landscape. The method develops UDs from an animal’s movement paths rather than discrete point clouds, producing highly informative models of space use well suited to assessments of habitat utilization. The selection and optimization of three parameters (maximum time threshold Tmax, minimum step length Lmin, and minimum smoothing parameter hmin) within the modeling framework accounted for the serial autocorrelation in civet tracking data and the locational error of the GPS units. The same values were applied to all individuals in the study. Briefly, Tmax was set to 7800 s (2 h plus 10 min of tolerance to account for variable GPS fix times). Based on the results of collar accuracy testing (Evans et al. 2016b), Lmin was set to 17 m. Following the protocols presented by Benhamou (2004) and Benhamou and Cornélis (2010), hmin was set to 40 m. Civet UDs were constructed at 10 m resolution, and total and core home ranges designated as the area encompassed by the 90% and 50% contours, respectively, as recommended per Börger et al. (2006). All 18:00 h fixes were removed prior to analyses to account for the spatial pseudo-replication of diurnal sleeping sites within the data structure; thus, models represent ‘active’ nocturnal civet ranges. For those individuals that were collared twice, data from the longest running unit deployment were used in BRB analyses. Home range models were developed with R v3.5.0 and the adehabitat packages (Calenge 2006).

Differences in the mean home range sizes of civets that entered oil palm plantations, and those that did not, were evaluated with Welch’s two sample t-tests. The relationship between home range size and the proportion of the home range contour containing oil palm was assessed using a general linear model (GLM). Models were fitted with a Gaussian error distribution and identity link function, and modeling assumptions assessed and met by optimizing standardized residual plots.

Habitat resource mapping

We mapped the landscape of the Lower Kinabatangan Floodplain in 2016 using the Global Airborne Observatory (GAO; formerly the Carnegie Airborne Observatory; Asner et al. 2012). The GAO is a mapping system utilizing laser-guided image spectroscopy through the simultaneous aerial deployment of a light detection and ranging (LiDAR) scanner and a visible-to-shortwave infrared (VSWIR) imaging spectrometer. The observatory contains a Global Positioning System-Inertial Measurement Unit to facilitate high levels of locational precision and accuracy in the data collection process. The LiDAR system produces high resolution, three-dimensional information of terrain and vegetation structure (Lefsky et al. 2002). Measures of vegetative radiance collected by the VSWIR provide insights into forest canopy functional traits and landscape beta diversity (Asner et al. 2017).

Data were collected at 3600 m above ground level at a velocity of 150 knots. Details of the LiDAR data acquisition, specifications, and processing can be found in Asner et al. (2018). The VSWIR spectrometer collected radiance signatures across 428 continuous spectral channels within the 350–2510 nm wavelength range (Asner et al. 2012). The sensor held a 34° field-of-view and provided radiance measures at 4 m resolution. Details regarding the processing of VSWIR image spectra and field calibration and validation protocols are described in Martin et al. (2018). In that work, the spectra were converted to verifiable estimates of canopy foliar chemical properties. Using the same methods of Asner et al. (2017), forest functional diversity classes were mapped by Brodrick et al. (in review) using K-means clustering of canopy foliar chemical traits. These analyses yielded a Sabah-wide map of 32 unique forest functional classes reported at a mapping resolution of 30 m.

Resource utilization functions

The landscape features influencing civet space use were assessed through the development of resource utilization functions (RUFs) calculated at the total (90%) home range extent (thus representing third-order selection, Johnson 1980). RUFs relate variations in an animal’s continuous and probabilistic utilization distribution (UD) surface to the underlying, spatially explicit environmental covariates (Marzluff et al. 2004; Millspaugh et al. 2006). Notably, this approach robustly minimizes habitat-induced fix biases and location uncertainties inherent in all GPS telemetry studies (Millspaugh et al. 2006; Hooten et al. 2013). Individual RUFs were calculated for each civet in a multiple regression framework as described by Marzluff et al. (2004) using the ruf package (Handcock 2015). This package allows models to account for the spatial autocorrelation in use values of adjacent UD pixels through the application of a Matern correlation function; following recommendations of Marzluff et al. (2004), the spatial dependence was set at 40 m and the smoothness of each surface at 1.5 for all animals. Each pixel value within an individual’s range was rescaled to a use metric between 1 (lowest use) and 99 (highest use) based on the relative volume of the UD (Kertson and Marzluff 2010). We log-transformed these values to meet assumptions of normality prior to analysis (Hooten et al. 2013).

We initially selected eight landscape characteristics (hereafter ‘resources’) as measures we expected to influence civet habitat utilization based on local expert opinion and literature of carnivore behaviour (Table 1). Each of these assessed parameters may influence civet space use by providing food, mitigating predation risks, or acting as cover from environmental conditions. Resources were described by one of four categories: landscape proximity metrics, landscape topography measures, forest canopy attributes, and forest beta diversity class. Civet space-use models and remote sensing data products were processed in R software and QGIS v3.10. Shapefile layers of oil palm plantation boundaries and waterways were created from Google Earth Pro satellite imagery to derive distances from the center of each UD pixel to the closest accessible feature. These Euclidean distances accounted for accessibility by taking into consideration that collared civets did not cross the main river.

We interpolated our selected metric of landscape topography [the vector ruggedness measure, VRM (Sappington et al. 2007)] from the GAO digital terrain model, and measures of forest canopy structure (average canopy height, heterogeneity of canopy height, and canopy cover) from the top of canopy height raster. A circular analysis window with a 10 m radius was centered on each UD cell to calculate the mean and standard deviation of canopy height values. Canopy cover was measured as the percent of canopy height pixels within this window less than 10 m in height. We extracted forest beta diversity values from Brodrick et al.’s (in review) forest functional class map for each UD pixel.

Collinearity between explanatory resources was assessed with Pearson correlations at the level of each individual’s dataset. Canopy cover was excluded from RUF analyses due to collinearity (|r|> 0.7) with mean canopy height across all civet datasets. Resource parameters were transformed to meet normality assumptions. Specifically, distance to water and plantations, as well as the mean and standard deviations of canopy height, underwent a square root transformation. Vector ruggedness measure was log transformed. The same model structure and transformations were used for each civet’s RUF.

The standardized resource coefficients (β) from civet RUFs were used to produce a population-level assessment of Malay civet space use. We calculated the mean standardized coefficient (β*) and 95% confidence interval (CI) for each resource, such that CIs encompassed inter-animal variation (Marzluff et al. 2004). Resources with 95% CIs that did not include zero were considered significant predictors of intense use at the population-level (Marzluff et al. 2004). The magnitude of statistically significant mean standardized coefficients indicated the relative importance of the resource in determining intensity of use by Malay civets (Millspaugh et al. 2006).

Results

Space use by Malay civets

A total of 24 GPS collars were deployed on 21 Malay civets, resulting in the collection of 20,834 successful GPS fixes (Supplementary Table 1); three individuals were collared twice due to prematurely lost or malfunctioning units. Data collection spanned 1–246 nights (mean ± SE: 105 ± 15 nights). Based on the unit failures of Female 1 and Male 5, these datasets were omitted from further analyses; all results thus describe the space-use patterns of male Malay civets.

Total home ranges (90% contours) varied in size from 45.9 to 214.8 ha (mean ± SE: 113.4 ± 13.8 ha; Supplementary Table 1). Core home ranges (50% contours) measured from 12.2 to 75.3 ha (mean ± SE: 35.8 ± 5.20 ha). Eight civets did not enter oil palm plantations during the collaring period, while 11 individuals utilized both forested and plantation land cover (Fig. 2). The presence of oil palm within home ranges varied within the population (Fig. 3); across all individuals, the median percentage of total home range containing plantation was 15.3% (range 0–67.8%). The median percentage of a civet’s core home range comprising oil palm was 17.2% (range 0–79.8%).

Fig. 2
figure 2

Space-use models for Malay civet (Viverra tangalunga) males calculated from biased random bridges (BRB) approach within the Lower Kinabatangan Floodplain from 2013 to 2020. Maps show each individual’s utilization distribution (UD) surface clipped at 90%; each raster is overlaid with contour outlines delineating the boundaries of core (50%) and total (90%) home ranges

Full size image
Fig. 3
figure 3

Habitat composition of total (90%) and core (50%) biased random bridges (BRB) home ranges of 19 Malay civet (Viverra tangalunga) males in the Lower Kinabatangan Floodplain, Sabah, Malaysia

Full size image

Civets that accessed oil palm plantations held significantly larger total and core home ranges than those individuals that remained in the forests (t =  − 3.913, d.f. = 14.801, p < 0.005; t =  − 3.689, d.f. = 13.665, p < 0.005, respectively). Mean (± SE) total and core home range sizes for civets that accessed oil palm were 136.5 ± 52.9 ha and 43.2 ± 6.02 ha, respectively. In contrast, the mean (± SE) total and core home range sizes for civets that used only forests were 65.7 ± 8.52 ha and 18.8 ± 2.71 ha, respectively.

There was a significant positive relationship between the overall range size and the percentage of the total home range containing oil palm (F1,17 = 23.8, p < 0.0005, Adj. R2 = 0.559; Fig. 4a). A 10% increase in the percentage of oil palm within an individual’s total range resulted in an overall increase of range size by 9.7 ± 4.03 ha (t = 4.879, p < 0.0005). Similarly, there was a significant positive relationship between the core range area and the percentage of the core range containing oil palm (F1,17 = 13.68, p < 0.005; Adj. R2 = 0.413; Fig. 4b). A 10% increase in the percent of oil palm within a civet’s core range resulted in the core range increasing by 4.6 ± 1.2 ha in area (t = 3.699, p < 0.005).

Fig. 4
figure 4

Predicted relationship between the size of Malay civet (Viverra tangalunga) a total (90%) and b core (50%) biased random bridges (BRB) home ranges and percent of range containing oil palm plantation. Points are the recorded values from the 19 independent males. Solid lines indicate model estimates, and the shaded regions represent the 95% confidence intervals of the model

Full size image

Resource utilization functions

Nineteen resource utilization functions were calculated from civet collar data, and resource use was variable at the individual scale (Fig. 5). At the population-level, intensity of civet space use consistently increased with increasing proximity to oil palm plantation edges (89.5% of individuals) and with increasing height of forest canopy (63.2% of individuals; Fig. 5, Supplementary Table 2). Distance to plantation edge was the most important resource predicting intensity of space use by civets, followed by mean top of canopy height. No other landscape resources were consistently correlated with civet use intensity at the population level.

Fig. 5
figure 5

Standardized resource coefficients (β) from Malay civet (Viverra tangalunga) resource utilization function (RUF) models. Large points indicate the population-level mean β*, and error bars represent the 95% confidence intervals. Negative β values indicate civet use increased when continuous resource metrics were smaller, while positive β values indicate civet use increased when metrics were greater. Negative β values for forest functional classes indicate the class is used less than expected based on availability, while a positive β value indicates use of the class is greater than expected based on availability; the reference category is areas with no canopy cover. Asterisks indicate a statistically significant population-level effect. Water distance to water, Plant distance to oil palm edge, VRM vector ruggedness metric, mTCH mean top of canopy height, sdTCH standard deviation of top of canopy, F relates to categorical forest functional class numbers from Brodrick et al. (in review)

Full size image

Discussion

This study presents the first detailed insights into the influences of oil palm plantations and remnant protected forests on the spatial ecology of male Malay civets. Our merging of high spatiotemporal resolution GPS tracking and mapped habitat data allowed us to quantify patterns of generalist persistence in degraded landscapes. Plantations did not act as definitive barriers to movement; indeed, a majority of civet home ranges contained oil palm plantations. However, all animals in this study used protected remnant forests, emphasizing the likely value of these features for conservation. Proximity to plantation edges and forest canopy height were important features explaining intensity of space use by civets, highlighting the significance of proximities and local habitat attributes in shaping this species’ adaptability. Overall, our analyses indicate oil palm is a less suitable habitat for civets than remnant forest, but persistence within degraded landscapes may be met with a combination of land sharing and land sparing approaches.

Malay civet access of oil palm plantations confirms the species’ ability to forage within this agricultural matrix. Our work builds upon previous studies documenting civet presence within oil palm plantations (Jennings et al. 2010; Yue et al. 2015), and is the first to detail the extent to which this generalist uses human-modified lands. Plantation boundaries did not impede civet movements; in fact, individuals actively used oil palm agriculture at both the total and core home range scales. By quantifying the explicit space-use patterns of these animals across the heterogeneous landscape, our work provides preliminary mechanistic insights into the adaptability of this species to plantations, with individuals demonstrating a range of tolerance thresholds (Fig. 3). As Malay civets are considered dietary generalists (Colón and Sugau 2012), plantations likely provide supplemental food resources such as oil palm fruit or associated small mammals. Indeed, given their dietary flexibility, Malay civets may supply a degree of rodent-control service to oil palm plantations (e.g. Holzner et al. 2019), providing a conservation incentive to industry stakeholders.

Despite this behavioral flexibility, however, the positive correlations between home range size and extent of oil palm usage indicate plantations are of lower suitability than forested habitats. Carnivore home range size is negatively related to habitat suitability, a term encompassing resource quality, stability, and predictability at the scale of an individual’s perception (McLoughlin and Ferguson 2000). To obtain adequate resources within plantations, male civets cover larger areas, and thus expend more energy, than those individuals exclusively using forest fragments. Similar scaling patterns have been documented in other spatially adaptive carnivores using agriculture (e.g. Eurasian lynx Lynx lynx, Aronsson et al. 2016; feral cat Felis catus, Bengsen et al. 2016; red fox Vulpes vulpes, Walton et al. 2017), and represent the ecological costs of adaptation to modified landscapes. Mechanistically, risks not assessed in this study may offset civet foraging rewards obtained within oil palm plantations; bushmeat hunting, road accidents, conflict with domestic animals, and exposure to agrochemicals are greater within plantations (Azhar et al. 2013; Salim et al. 2014).

Concurrently, each civet in this study utilized remnant forest patches, suggesting even this generalist species likely requires a degree of access to natural habitats. The conservation value of protected patches and riparian corridors such as the Lower Kinabatangan Wildlife Sanctuary has been reported in other anthropogenically-altered ecosystems (Wintle et al. 2019), but is frequently debated within the context of oil palm research (e.g. Yue et al. 2015; Wearn et al. 2017; Pardo et al. 2019). Our results provide further evidence that generalist carnivore persistence within oil palm landscapes may require small patches of even degraded forests outside of plantation boundaries (i.e. land-sparing); such features likely provide source habitats from which civets may ‘spillover’ to forage within oil palm plantations (Blitzer et al. 2012).

Moreover, with all other landscape variables held constant, areas with greater forest canopy height were more intensively used by civets. Top of canopy height is a proxy of forest quality, particularly in disturbed systems such as the Kinabatangan (Davies et al. 2017), suggesting this species positively responds to higher quality and less degraded forests whenever available. By principles of the environmental heterogeneity hypothesis (Simpson 1949), forests with taller canopies contain more ecological niches than comparatively simplified systems (Tews et al. 2004); wildlife diversity and species’ reproductive success have been positively related to elevated canopy height (Hinsley et al. 2006; Bergen et al. 2009). The interaction between 3-dimensional habitat metrics and animal movement ecology represents an expanding field (Davies and Asner 2014), and our study provides preliminary evidence of a ground-dwelling generalist functionally responding to forest structure within a degraded system. Taller fruiting trees, such as Ficus spp., may provide a large volume of reliable food for many lowland animals, civets included (Nakabayashi et al. 2017). Alternatively, these areas of higher quality forests may act as thermally optimized foraging grounds (Melin et al. 2014), provide adequate ground and mid-story vegetative cover, or concentrate female distributions around reproductive denning sites (Zhao et al. 2012); further research is warranted to uncover the specific structural drivers behind space-use patterns of Malay civets.

Intriguingly, there was no evidence of population-level consistency in civet responses to any of the assessed forest functional classes, suggesting forest structure may be more important than functional composition in determining civet use. This is likely borne from the high variability in civet responses and access to forest functional classes (Fig. 5). This large degree of individualism may exemplify the flexibility of this adaptable habitat generalist within the anthropogenically-altered Kinabatangan landscape; similar patterns have been recorded in both large and small carnivores [e.g. cougars Puma concolor, (Kertson and Marzluff 2010; Kertson et al. 2011), common genets Genetta genetta, (Santos et al. 2016)]. Alternatively, it is possible the vegetative diversity encompassed by this metric may not adequately capture resources used by civets. Taken all together, these findings may carry implications for landscape planning actions within degraded regions, such as protected area establishment or restoration efforts, by demonstrating this species is likely to respond positively to small increases in forest canopy, regardless of explicit vegetative beta diversity.

At the population-level, civet landscape use was most strongly influenced by proximity to oil palm agricultural edges. Elevated use of edge habitats by carnivores has been documented in other human-modified systems (e.g. Pereira and Rodríguez 2010; Andersen et al. 2017), and suggests landscape proximity metrics may modulate civet persistence across the degraded Kinabatangan Floodplain. As this metric is non-directional relative to land cover (i.e. forest vs. plantation), this pattern may be explained by two processes. First, given civet use of both forests and plantations, repeated movements between protected areas and plantations are likely to elevate the relative importance of edge zones at the home range scale. This supports the camera trap findings of Jennings et al. (2015), whereby Malay civet occupancy within Indonesian plantations was positively related to the distance from native forest boundaries. Free roaming packs of domestic dogs within most Kinabatangan oil palm plantations may pose a predation threat to foraging civets; individuals may thus not forage too far from the protection provided by natural forest cover. In this case, civet persistence within oil palm plantations is directly facilitated by access to remnant forests, underlining the importance of landscape configuration in carnivore adaptability. Alternatively, civets within remnant forest fragments may positively respond to plantation-induced edge effects. Ordway and Asner (2020) documented significant alterations to the structure and functioning of Malaysian forests flanking oil palm plantations. Such changes may benefit generalists like Malay civets by providing localized increases in small mammal densities (Sálek et al. 2010) or fruit production by edge-preferring vines (Lovejoy et al. 1986). Given the positive relationship between canopy height and civet intensity of space use, however, other factors not explicitly assessed in the resource utilization function analyses may more eloquently explain these forest edge patterns.

Our results have several important caveats. Variables not explicitly accounted for in this study, such as the position of female civet ranges, conspecific densities, oil palm management strategies, temporal variations in forest phenology, or forest understory structure are likely to influence space-use patterns of male civets. Furthermore, within resource utilization functions (RUF), relationships between habitat resource proxies and intensity of space use are treated as linear functions, which may mask ecologically nuanced civet responses to continuously varying landscape characteristics and associated resources (e.g. Garabedian et al. 2017). Civet RUFs described habitat utilization processes driving nocturnal movements, which may differ from those driving other core behaviors such as diurnal rest or reproductive den site selection. Finally, the spatial extent of this study was limited to the Kinabatangan Floodplain, which underwent large-scale conversion to oil palm agriculture more than 30 years ago; civet home ranges and patterns of habitat use may not be consistent within pristine forests or in regions actively being converted to oil palm agriculture. Although the extensive degradation of the region embodies the future of many lowland ecosystems, the mechanisms driving civet persistence may be modulated by the time since agricultural conversion (Gibson et al. 2013).

Future research determining the persistence of this species within mosaic landscapes should seek to scale up patterns of individual movement strategies into long-term population viability assessments. Although the civets collared in this study were considered healthy by visual inspection, animals utilizing plantations may suffer from nutrient deficiencies, immunosuppression, or other sub-lethal physiological manifestations of poor health (Ellis et al. 2012). Genetic evaluation of civet population structure would provide more detailed insights into the dispersal patterns and broader landscape connectivity of the species within modified mosaic landscapes. Based on our results, it is unknown if individuals can disperse through or establish home ranges completely contained within oil palm plantations; indeed, the maximum distance a civet was recorded within a plantation was 980 m from a natural forest edge. Research by Brunke et al. (2019) quantified the genetic consequences of fragmentation in the Kinabatangan for small mammals; similar analyses of the civet population would generate minimum functional patch size and connectivity thresholds to even better inform landscape planning strategies.

Given the increasing scale and extent of land use change worldwide, there is an urgent need for effective planning in and around human-modified landscapes to mitigate biodiversity losses. Although the cessation of land conversion and expansion of protected areas are undeniably crucial to stabilizing biodiversity populations (Gibson et al. 2011), the conservation value of agriculturally degraded landscapes such as the Lower Kinabatangan Floodplain should not be overlooked. By evaluating the behaviors of a generalist carnivore within such a mosaic, we began to elucidate the patterns and possible price of biodiversity persistence. To adequately conserve carnivores, the protection status of the Lower Kinabatangan must be maintained, and efforts to expand the protected area network cultivated and enacted. Simultaneously, strategies to safely facilitate civet presence within plantations should be undertaken (e.g. Ashraf et al. 2018), as oil palm may pose an ecological trap to these behaviorally adaptive carnivores (Robertson and Blumstein 2019). The execution of such recommendations will provide benefits for other lowland wildlife species while simultaneously safeguarding critical ecosystem-level processes and functions.