Spatial heterogeneity facilitates coexistence between striped hyaenas and sympatric carnivores

Interspecific competition plays a key role in shaping carnivore communities. Top‐down effects can impact the coexistence of superordinate and subordinate competitors and prey across a shared landscape. Limited resources, their abundance and diminishing habitats can all exacerbate interspecific competition. Sometimes, behavioural mechanisms or adaptations are critical for subordinate carnivores to coexist with larger species. Here, we investigated the distribution pattern of the striped hyaena (Hyaena hyaena) population with respect to habitat features, sympatric carnivores and their prey species across a large landscape.


| INTRODUC TI ON
The ecological processes of carnivory and predation occupy important ecological niches.They can shape the structure of ecosystems and even facilitate the presence of other species evolving under different selective pressures (Dröge et al., 2017;Ripple et al., 2014).
As they speciated into many forms during rapid adaptive radiation, some carnivores became more specialized than others.Those species evolving similar morphologies, activity patterns and predation strategies over time may have exhibited higher degrees of interspecific competition, depending on the availability and use of prey resources, and how much they overlapped (Eaton, 1979;Harihar et al., 2011).The evolution of obligate 'hyper-carnivores', for example, through the adoptation in increase of body mass and dental specialization, allowed them to outcompete more massive, facultative predators for certain prey resources (Van Valkenburgh, 1999;Viranta, 1996).Over time, competition among them would either lead to some species displacing their similar sympatric competitors or eventually adopting different ecological niches (Dayan & Simberloff, 1996;Van Valkenburgh, 1999).
Today, interspecific competition still plays a deterministic role in shaping the composition of carnivore communities.This happens both through the active displacement of competitors, but also by shaping the adaptations of other species in the community (Dröge et al., 2017;Elbroch & Kusler, 2018).A synthesis by Prugh and Sivy (2020) suggested that about 30% of mortality of mesopredators is caused by larger predators, which in turn affects mesopredator distribution, habitat use, prey preference and abundance in a shared landscape.Conversely, even in resource-rich habitats, greater pressure from large carnivores may increase food acquisition costs for smaller carnivores, which can negatively impact their reproduction, growth and viability even more than interspecific predation (Creel, 2001;Creel & Christianson, 2008;Elbroch & Kusler, 2018;Gorman et al., 1998;Palomares et al., 2016).Successful coexistence between predators over time may be facilitated by species-specific behavioural traits, and/or scale-dependent and spatiotemporal partitioning and avoidance, which can balance the trade-offs between resource acquisition and predation/encounter risk (Heithaus, 2001;Vanak et al., 2013).Active avoidance at a fine scale, whereby subordinate predators are more aware of the movements of large predators, can also facilitate the successful occupation of the same habitats and ultimately coexistence (Sollmann et al., 2012;Vanak et al., 2013).In previous studies, for example, puma (Felis Concolor), cheetah (Acinonyx jubatus), African wild dog (Lycaon pictus), leopard (Panthera pardus) and dhole (Cuon alpinus) tended to avoid areas where large apex predators, like jaguar (Panthera onca), lion (Panthera leo), tiger (Panthera tigris) and spotted hyaena (Crocuta crocuta), are present in high numbers or were highly active (Creel & Creel, 1996;de Azevedo & Murray, 2007;Gorman et al., 1998;Hayward & Slotow, 2009;Rabinowitz & Nottingham, 1986).
Ecological attributes like habitat structure and cover, prey abundance, behaviour and seasonality may also influence coexistence by differentiating or partitioning resource use patterns over time.
In Africa, although painted dogs avoided areas with relatively high lion density, they still showed significant overlap with lions in the wet season; high vegetative cover and pronounced seasonality of wild dog movements, likely facilitated coexistence with lions by minimizing negative interactions (Vanak et al., 2013).Other behaviours used by smaller carnivores to facilitate coexistence with larger species in their guild include reducing or optimizing movements, use of fine-scale habitat 'refuges', and aggression or the deployment of chemical repellents (Allen et al., 2016;Dröge et al., 2017;Krebs et al., 2004;Ritchie & Johnson, 2009;Vanak et al., 2013).
Reproductive behaviour, including the timing of births, as well as the selective location and security of dens, may be effective strategies to reduce intraguild predation of young which is an important cause of mortality for many smaller African carnivores (Mills, 1993;Mills & Mills, 2014).Despite the potential obvious importance of den-site selection to carnivore coexistence, this topic has rarely been explored empirically as a strategy by one or more species for mitigating competition, and ultimately facilitating persistence among larger carnivores.
Hyaenids are among those carnivore species that nearly always occur with other larger or similarly sized carnivores.They first evolved during the Oligocene from feliform ancestors during a time of great competition for large prey (Werdelin & Solounias, 1991;Westbury et al., 2021), subsequently radiating into a diversity of species and specialized ectomorphs.Optimized digestive and enhanced immune systems help hyaenids leverage pathogen-rich food sources otherwise not exploited by other carnivores, an advantage that may have allowed them to become the most diverse carnivore group in the late Miocene (Benbow et al., 2015;Westbury et al., 2021).
Of the four extant hyaenids, the striped hyaena has the widest distribution.Unlike spotted hyaenas, which are effective predators, striped hyaenas, which range from North Africa to the Indian subcontinent, are predominantly scavengers (AbiSaid & Dloniak, 2015;Alam & Khan, 2015;Mills & Hofer, 1998).Across their range, striped hyaenas co-occur with many other large predators, including tigers, leopards, dholes, lions and wolves (Canis lupus), frequently occurring near human settlements (Arivazhagan et al., 2007;Singh et al., 2014).Both co-predators and humans provide optimum scavenging opportunities in the form of wild prey carcasses and anthropogenic waste respectively.In turn, scavengers like hyaenas can disrupt the energy balance of their sympatric predators through kleptoparasitism which forces predators to hunt more frequently (Mattisson et al., 2016; Mellard camera-trapping, interspecific competition, meso-carnivores, random forest, sign surveys, Southern India, spatial incumbency, top-down effect et al., 2021).Land use and landcover changes driven by several anthropogenic factors have already limited resource availability in many ecosystems and may also exacerbate interspecific competition.In the long run, this can result in more frequent direct interactions between hyaenas with their dominant counterparts and thus may limit the coexistence of hyaenas in many of their shared landscapes.
To date, few have studied striped hyaenas and little is understood about the impact of large carnivores on the distribution pattern of the species.Some prior studies across five Indian tiger reserves have indicated that striped hyaena occupancy changes in response to the presence of other large carnivores (Alam et al., 2009;Gupta et al., 2009;Harihar et al., 2010;Sankar et al., 2010;Jhala et al., 2011).In Sariska and Achanakmar Tiger Reserves, where tiger populations are low relative to historical densities, the local striped hyaena population density appeared high; conversely, hyaenas were found at lower population densities in reserves, where more tigers were present.Despite these apparent top-down pressures, declines in hyaena populations have thus far only been ascribed to anthropogenic factors, like habitat destruction (Alam et al., 2009).Moreover, large predators also facilitate striped hyaena persistence and coexisting through the provisioning of prey carcasses.
In this study, we investigated the effects of habitat characteristics and other predators on the presence of striped hyaenas as well as predicted potential habitat in the Western & Eastern Ghats Part of Tamil Nadu (WEGPTN).Based on available literature, we also underscored the potential role of den sites as part of strategies to facilitate coexistence with southern India's other large carnivores, including tigers, leopards and dholes, with which hyaenas occur.We hypothesized that in our study area, fine-scale space use by hyaenas, including the use of habitat and denning refugia, will help facilitate the trade-offs between risk and opportunity related to the predators that striped hyaenas likely depend on for their survival.We note that this region encompasses the only major breeding population of striped hyaenas in Tamil Nadu, one that would serve as a critical source for the potential recovery of the species more widely across the southern subcontinent.

| Study area
Our study area encompasses a vast multi-use landscape covering an area of 7319 sq km in Tamil Nadu of southern India.The WEGPTN includes a diversity of habitats, protected areas and mixed-use forests across a large and mostly contiguous forest landscape, including Mudumalai Tiger Reserve (MTR), Sathyamangalam Tiger Reserve (STR), Cauvery North Wildlife Sanctuary (CNWLS), as well as the Gudalur, Nilgiris, Coimbatore, Erode, Dharmapuri and Hosur Forest Divisions (Figure 1).The topography varies from 186 to 2530 m a.s.l., and is comprised of diverse tropical forest types like evergreen, moist deciduous, dry deciduous, mixed deciduous and open dry thorn (Champion & Seth, 1968).Annual mean temperature varies in the landscape from 12.7 to 27.92°C, and annual precipitation ranges from 608 to 3438 mm.Generally, STR Erode, Dharmapuri and Hosur Forest divisions are the driest parts of the region, while MTR, Gudalur, Nilgiris and the southern part of Coimbatore are usually marked by high rainfall.This region also hosts a great diversity of the subcontinent's large mammals.Among India's native large carnivores, tigers, leopards, dholes and striped hyaenas are all present, as are diverse ungulate species, including chital (Axis axis), sambar (Rusa unicolor), gaur (Bos gaurus), blackbuck (Antilope cervicapra), four-horned antelope (Tetracerus quadricornis) and barking deer (Muntiacus vaginalis).The Moyar River is the only perennial water source transecting the Moyar Valley of the Sigur Plateau, a unique part of our study area.It is characterized by a narrow strip of arid habitat stretched between two mountain chains running parallel to each other, and whose slopes serve as potential den sites for striped hyaenas (Arivazhagan et al., 2007).In addition, there are several human settlements in this mountainous Sigur Plateau landscape, which could lead to greater opportunities for hyaenas to scavenge livestock carcasses.

| Data collection and analysis
We collected data on the presence-absence of striped hyaenas, all three co-occurring obligate predators (i.e.tigers, leopards and dholes), and major prey species (Chital, Sambar and domestic cattle) from across our study area.We used both indirect sign (track/scat) and camera-trapping surveys to gather location and time/date data for these species.We sampled 248 trails of 2 km in length, including management roads from all representative habitats and forest divisions, and covering most forest ranges (Figure 1).Indirect evidence of different predator and prey species was confirmed based on distinct characteristics unique to each respective species, including track morphology, faecal material (scat) size and shape and scrape marks (Kolipaka, 2014).We installed camera traps at 194 sites across seven forest divisions from all representative habitats and accrued a total sampling effort of 4975 trap nights to complement our sign surveys with photographic evidence.We have used single-sided 50 camera traps on a rotational basis for data collection across 194 sites.Cameras relied on motion and infrared detection to record target species; they were installed at locations optimal for detecting the presence of passing mammals, including water holes, the junction of game trails, walking transects created by the forest department and area management roads.Sampling occurred throughout the two consecutive years (2021 and 2022) except in the rainy season (July-October) when visibility inside forests decreases due to rapidly growing vegetation in the forest undergrowth.
We reviewed key literature to understand den-site selection in carnivores and other mammals, and to identify the best approaches for this study (Bopanna, 2013;Boydston et al., 2006;Chourasia et al., 2020;Davies et al., 2016;Jackson et al., 2014;Khanal et al., 2017;Majumder et al., 2016;Mandal, 2018;Mukherjee et al., 2018;Nikunj et al., 2009;Singh et al., 2010;Singh et al., 2014;Vroom et al., 1980).Furthermore, data on the attributes of potential den sites for striped hyaenas were collected post-identification of critical microhabitat features.After identifying the core habitat of the striped hyaena based on our sign and camera trap surveys, we located hyaena dens by exploring potentially suitable sites like mountain slopes and rifts, the junction between mountain ranges and dry riverbeds and any natural foothold refuges, like modified porcupine (Hystrix indica) dens, as has been reported from the literature (Alam et al., 2014;Bopanna, 2013;Singh et al., 2014).We confirmed the hyaena den based on indirect signs of recent hyaena denning activity in the form of fresh tracks or scats outside of suspected dens, or the hoarding of bones inside or outside of the entrance, as is often characteristic of hyaena behaviour (Bopanna, 2013;Kruuk, 1976;Stewart et al., 2021).We also relied on the experiences of and sightings by local forest department staff and villagers to help us locate hyaena dens.In addition to occurrence surveys, camera-traps were also used to confirm activity at suspected hyaena den sites.
To determine the distribution patterns of striped hyaena across our study area, we collected information on 25 important explanatory variables that could potentially influence the distribution of hyaena prey (Appendix S1).Many of these variables were initially identified based on prior research works as relevant to striped hyaenas (Alam et al., 2009;Gupta et al., 2009;Harihar et al., 2010;Jhala et al., 2011;Sankar et al., 2010).These include topographic variables such as elevation, slope and terrain ruggedness.Initially identified variables also include all climatic variables from Worldclim (www.world clim.org/ current) and other landscape variables, such as land cover (Roy et al., 2015), tree cover (Sexton et al., 2013), distance from the nearest settlement and proximity to streams.We also downloaded a raster layer of variable tree cover from NASA Land Data Products (Townshend, 2016) at a 30-m spatial resolution, which allowed us to quantify the proportion of land in each pixel covered by woody vegetation >5 m in height.To account the effect of stream and settlement on the distribution of the species, we first procured their updated shape files for the year 2021 from the state forest department.Then, the shapefiles were rasterized using the Euclidean distances tool in ArcMap 10.8 (ESRI, 2014).
After procuring these datasets, all raster layers were projected into the WGS 1984 UTM projection system using the reproject tool in The map shows the boundary of the study area which is situated in the Western Eastern Ghats part of Tamil Nadu state of India, with the Geographic Coordinate System: WGS84 projection.The study area is a vast multi-use landscape covering an area of 7319 sq.km and stretched between 10.85° and 12. 86° in north and 76.24° and 78.74° in east.The study area encompasses seven forest divisions and two tiger reserves.The map also illustrates the locations of 194 camera traps and 248 sign survey trails used to collect presenceabsence data of striped hyaena, its co-predators and their prey species.
ArcMap 10.8.Afterwards, we resampled reprojected data at 1-km spatial resolution and clipped to our study area extend in ArcGIS.
After preprocessing of raster layers, we calculated terrain slope and ruggedness for our study area from the elevation layer downloaded from Earth Explorer (NASA/METI/AIST/Japan Spacesystems & U.S./ Japan ASTER Science Team, 2019).We also reclassified the original 28 discrete classes of the land use/land cover (LULC) layer into 16 broad categories (Appendices S2 and S3).All raster layers were then resampled at 1 sq km in ArcGIS.
To test all variables for potential pairwise correlation, we used a Pearson-product moment correlation test (Figure 2) and ultimately retained only those variables that were not strongly correlated (i.e.r < 0.6) for the final analyses (Table 1).These variables were also used to build the models for sympatric carnivores (tiger, leopard and dhole), and wild prey species (chital and sambar).Spatial incumbency of a species in a natural system is a very complex phenomenon and generally does not exhibit absolute linear relation in an uncontrol environment.Whereas linear models are parametric and hold several assumptions that rarely align with data collected from the free natural system.Therefore, to address the complexity of spatial incumbency of our target species, we used 'randomForest' package version 4.7-1.1 in the R program, a non-linear random forest classification algorithm (Breiman, 2001;R Development Core Team, 2018).Presence-absence points (GPS location) of each target species were gathered from our field data and plotted over 1 sq km grids in ArcGIS map after reprojecting those locations into UTM.
For analysis, we retained only one point per grid indicating only the presence or absence of species in a grid.Values of all predictor layers were extracted for those points and using a binary decision analytical framework, we assessed the differential presence-absence by integrating models with different numbers of forest 'trees'.
We continued this until predictions stabilized and yielded a minimum of 5000 (ntree) trees for each species.We determined the optimal number of variables to be randomly sampled (mtry) after running and comparing several models with different values.We found that the highest model accuracy was found at mtry = 4, and so this value was assigned for all final models; in addition, we retained the default value of '1' to tree depth from the tree node.All these optimizations were kept consistent for all four carnivores while for chital and sambar, we grew 6000 and 5000 trees respectively.The value of mtry was also optimized for both prey species based on model stabilization.The value of mtry was assigned 2 for chital while the model for sambar was stabilized at mtry = 3.
The accuracy of each model was checked using out-of-bag data, which consists of data points from our dataset that were not used in model calibration but left out for model validation (Breiman, 2001).
We applied true skill statistics to calculate the true-positive rate and false-positive rate and plotted ROC curves for model validation (Allouche et al., 2006).We assessed the overall variable importance for each species using two metrics: (1) the mean decrease in accuracy (the mean minimal depth of the tree node) and ( 2) the Gini index (Lerman & Yitzhaki, 1984).Where, a lower value of mean minimum depth corresponds to a relatively higher importance of a variable in the model for tree nodes, while a high Gini index implies greater relative importance for a given model.Furthermore, we checked how changes in the values of a predictor variable affected our calibrated model by plotting the likelihood of the species occurrence along its data range keeping other variables at a constant value (Cutler et al., 2007).

| Occurrence records from WEGPT
We found only 18 communal striped hyaena dens on two mountain slopes across our study area.Of these, 16 were 'clumped', relatively close together along one mountain slope, whereas the other two occurred on the second slope.Due to this unexpectedly low sample size, more complex analyses of den-site characteristics were unfortunately not possible.In addition, we recorded 72 distinct hyaena presence points.Out of these 19.4% (n = 14) were photographic evidence from 194 camera-trap stations, and 80.6% (n = 58) were indirect evidence of hyaena presence (e.g.scats, tracks, etc.) collected from 248 sign survey trails.For tigers, we recorded 128 distinct presence points.Among these points, 53.1% (n = 68) were photographic evidence and 46.9% (n = 60) were from indirect sign surveys.Whereas out of 165 distinct presence points recorded for leopard, 37% (n = 61) and 63% (n = 104) were photographic and indirect sign evidence respectively.Dholes were found to be rare in the landscape like hyaenas and were recorded from 101 distinct locations.Out of these, 57.4% (n = 58) were photographic evidence and

F I G U R E 2
The Pearson correlation test shows collinearity among the 25 potential environmental variables initially identified for analysis.Among these, only nine environmental variables such as Bio2, Bio3, Bio12, Bio14, Dem, slope, distance to stream, distance to settlement and tree density were found suitable for further analysis based on a set cut-off value of r < 0.6.the rest 42.6% (n = 43) were gathered on survey trails.Chital and sambar were more frequently recorded than their predators.Chital were recorded from 303 distinct locations out of which 30% (n = 19) and 70% (n = 212) were from camera trap stations and sign surveys respectively.Whereas we recorded 273 distinct presence locations of sambar among which 41.4% (n = 113) were photographic evidence and 58.6% (n = 160) were indirect signs.

| Distribution hyaena in WEGPT
Based on our sign and camera-trap surveys, we found that the striped hyaena population was mostly confined to the Bhavanisagar and Nilgiri eastern slope ranges of both the Sathyamangalam and Mudumalai Tiger Reserves, respectively, with unconfirmed reports of hyaenas originating from other adjoining forest ranges (Figure 3).
In contrast, we found other large carnivore species to be well distributed across the entire study area.For example, chital, dhole and leopard were recorded from most of the areas we surveyed (Appendices S4, S12 and S16).In contrast, sambar and tiger were recorded largely from relatively less disturbed habitats (Appendices S8 and S20), although we also recorded their sporadic presence from Coimbatore, Dharmapuri and Hosur Forest Divisions, as well as from the Cauvery North Wildlife Sanctuary, which are areas with larger anthropogenic footprints.

| Predictive model of hyaena
Our best calibrated random forest model indicated that only approximately 9.2% (at a cut-off value of 0.7 of predicted probability) of the entire study area is suitable for striped hyaenas (Figure 3).We obtained an overall model error rate of 12.41% based on out of bag data (i.e.data that were not used in model construction by the algorithm, but rather kept aside for model validation); the false positivity rate (i.e. the % of 'absence' data misclassified as 'presence') was approximately 14%, and the false-negative rate (i.e. the % of 'presence' data misclassified as 'absence') was 11.11% (Figure 4).Based on the low error, false-positive and false-negative rates, we found our overall model accuracy to be high with an AUC value of 0.93.Both mean decrease in accuracy and the Gini index were congruent in their indications that elevation, Isothermality (Bio3), chital potential probability of presence, cattle density and dhole potential probability of presence appeared to be the most important variables, among those we included, that influenced the distribution of hyaenas in our study area (Figure 5).

TA B L E 1
The table shows the 16 variables that were selected to understand their impact on the distribution pattern of striped hyaena in the Western eastern ghats part of Tamil Nadu, India.These variables include non-correlated geoclimatic variables selected after the Pearson correlation test.These variables also include the occurrence probabilities of co-predators of hyaenas and prey species.The occurrence probabilities of prey species were calculated using geoclimatic variables through the random forest algorithm.Similarly, the occurrence probabilities of co-predators were calculated through the random forest algorithm but using both geoclimatic variables and predicted probabilities for prey species.We found that elevation appeared to be the most important variable overall, and negatively impacted striped hyaena occurrence, that is, as elevation increased, the probability of hyaena occurrence decreased (Figures 5 and 6).We found that isothermality (Bio3), a

| DISCUSS ION
The persistence of mammals, particularly carnivores, across landscapes and their 'spatial incumbency' are likely the result of many complex factors, including both bottom-up and top-down ecological mechanisms and processes (Wallach et al., 2015).For striped hyaenas in the Western and Eastern Ghats part of Tamil Nadu (WEGPTN), our calibrated model was consistent with our field observations in that 'suitable' habitat for hyaenas across our study area was sparse, constituting <1/10 of the entire area.The population was found to be majorly confined to the Moyar valley of WEGPTN which is spread in the Bhavanisagar and Nilgiri eastern slope range of Satyamangalam and Mudumalai tiger reserve respectively.
However, our field data also showed the intermittent distribution of the species from the Masinagudi and Sigur forest ranges of MTR.In consensus to field data, our model also predicted the Moyar valley and Sigur plateau as highly suitable for the species in WEGPTN.The predicted suitable habitat also includes some additional areas of STR and Erode Forest division.These areas are characterized by dry and arid climatic conditions as well as mostly dominated by dry thorn forests which provide optimal habitat for the species.In contrast to hyaena, the distribution of other predators living in this landscape was more widespread.Both hyaenas and the other carnivores relied extensively on lowland habitat, which was critically important in determining their spatial distribution across the landscape.
Our findings are also consistent with prior findings for the species (Sharma et al., 2011;Singh et al., 2010Singh et al., , 2014)) substantial rugged terrain bounded by parallel mountain chains and other plateaus in the area (May et al., 2008;Ritchie & Johnson, 2009;Singh et al., 2010).Such rugged terrain adds to landscape heterogeneity and may create 'competition refugia', including greater cover, more ways to avoid line-of-sight interactions and/or optimal but secretive or cryptic denning habitats.This may effectively mitigate interspecific competition and intraguild predation or other agonistic interactions, and further safeguard a species from worsening or otherwise encroaching anthropogenic threats (Finke & Denno, 2006;Karanth et al., 2017;Petren & Case, 1998;Ritchie & Johnson, 2009;Wallach et al., 2015).In the case of striped hyaenas, rugged terrain likely provides safer denning refugia, and thus opportunities to coexist with other predators where they overlap.This is at least consistent with our identification of all 18 hyaena den sites amidst the steep slopes of two mountain ranges in the Moyar.Furthermore, because steep slopes are an 'energy expensive terrain', mobility through such terrain may burden energetic expenditures relative to less steep terrain; larger predators therefore may be more inclined to avoid them, or use them less intensively (Dunford et al., 2020;Langman et al., 1995;Shepard et al., 2008).This could give striped hyaenas an advantage, allowing them to persist in the landscape over time.Such rough terrain not only provides competition refugia from larger guild members (Ashish et al., 2022) but may also allow hyaenas to establish a foothold among humans and guard dogs (Singh et al., 2010).
That striped hyaenas across our study area were consistently found in topographically and altitudinally diverse terrain is consistent with the findings of other studies for other carnivores.For example, leopards exploit structural complexity in the vertical strata of forests to coexist with larger predators (Karanth et al., 2017;Ritchie & Johnson, 2009).The persistence of more than one predator species in a landscape is also driven by other factors, like climatic attributes and prey abundance.Precipitation or lack thereof can also impact the availability of competition refugia for subordinate predators.Pumas that are sympatric with jaguars are known to utilize more open, drier habitats to avoid fatal interactions with the latter, more powerful felid (Sollmann et al., 2012).For our study area, we found that striped hyaena probability of occurrence was positively related to isothermality, aridity and annual mean temperature.Potential prey in the form of chital and sambar were also positively associated with hyaena occurrence, and the predictive value of increased domestic cattle density was likely reflective of increased scavenging opportunities for the species.
Finally, although striped hyaenas are known to co-occur with other predators across their range, including western India (Alam et al., 2014;Arivazhagan et al., 2007;Jhala et al., 2020;Singh et al., 2014), before this study, little information existed on the specific mechanisms that facilitate hyaena coexistence with other large predators.Despite small sample sizes, our evidence is consistent with hyaenas making fine-scale investments in the use of certain habitats and microhabitat features considered less suitable by other predators.We note that although striped hyaenas have a foothold in the Moyar Valley, a valley important to all large carnivores present in southern India, this population is very small, and thus potentially vulnerable to inbreeding depression and anthropogenic threats (Jhala et al., 2020).It is the only major breeding population of hyaenas in all of southern India, which makes it deserving of regional and possibly national management practices and/or conservation interventions (Jhala et al., 2020).We, therefore, recommend that broader surveys be conducted more widely across the region we identified as critical to striped hyaenas between the Western and Eastern Ghats, as these activities could identify additional habitats and individuals.We also call for the implementation of a long-term monitoring programme to map the trajectory and status of this hyaena population or meta-population, and a critical analysis of what if any management tools or approaches can effectively enhance suitable habitat.

ACK N O WLE D G E M ENTS
We are indebted to the Director, SACON for providing logistical the second author.We thank forest department staff and our research assistants for their help in the field.We also thank Ashish AP for his literary input and support.We are also grateful to S.P.E.C.I.E.S.
(USA) for providing equipment, and additional technical support, to conduct field surveys.

CO N FLI C T O F I NTE R E S T S TATE M E NT
We have no conflict of interest.
climatic variable indicative of the range of temperature fluctuations within a calendar month and its relationship to annual mean temperature, was also important; specifically, our model suggested that areas subjected to fewer temperature fluctuations were more suitable for striped hyaenas.The occurrence probability of chital and precipitation of driest month (Bio14) showed a strong positive impact on the species distribution while precipitation of wettest quarter (Bio16) showed a strong negative relation.Hyaenas showed a weak positive relation with proximity to settlement but also a negative relation as the distance from settlement increased away from 10 km.The partial dependency graph shows that sambar occurrence probability has a very weak and positive effect on the hyaena occurrence.In LULC, only thorn forest was observed to have a strong and positive effect on the distribution of the species.The dry deciduous forest type showed a weak positive relation while the other classes had a negative relation with hyaena occurrence probability.Finally, hyaenas were more likely F I G U R E 3 The Western and Eastern Ghats part of Tamil Nadu shows the occurrence probability of striped hyaena at a spatial scale of 1 km 2 with the Geographic Coordinate System: WGS84 projection.The distribution probability of the species was predicted through random forest algorithm using predictors such as selected geoclimatic variables as well as modelled occurrence probabilities of co-predators of hyaena and their prey species.The areas represented by the red colour on the map indicate a low predicted probability of hyaena and the green colour represents a high occurrence probability of the species.The map also shows presence points of the hyaenas recorded during our field sampling.Projection of the map is EPSG:32643 -WGS 84/UTM zone 43 N. F I G U R E 4 The accuracy of the calibrated model was determined by plotting the receiver operating characteristics (ROC) curve which is a trade-off between the proportion of correctly identified (True positive rate) and wrongly identified (false-positive rate) occurrence of hyaena from the dataset.tooccur in areas with a lower to medium occurrence probability of leopards and tigers.Whereas hyaena exhibited a strong negative relationship with the probability of dhole occurrence.Likewise, annual precipitation (Bio12) also showed a strong negative relation with hyaena occurrence probability.
, which have underscored the importance of rugged terrain in shaping hyaena presence.The Moyar Valley of the Sigur plateau is geographically complex and has F I G U R E 5 This variable importance graph shows the relative importance of all explanatory variables in determining the occurrence probability of hyaena.Mean decrease in accuracy (depth of forest node) and mean decrease in the Gini index are two different measures of random forest used in assessing the importance of a variable in explaining the current distribution of the species.In both measures, variables Dem (elevation), Bio3, Chital, Cattle density and Dhole appeared to be important variables determining the distribution of hyaena in the landscape (Dem = Elevation, Bio3 = Isothermality, Chital = Occurrence probability of chital, Dhole = Occurrence probability of dhole, Bio14 = Precipitation of Driest Month, Bio12 = Annual Precipitation, Settlement = distance from settlement, Cattle density = No. of cattle per sq.km., Sambar = Occurrence probability of sambar, Leopard = Occurrence probability of leopard, Tiger = Occurrence probability of tiger, Bio2 = Mean Diurnal Range and Stream = distance from stream).

F
Partial dependency graphs of the nine most important variables show their marginal effect on hyaena occurrence.Here, the marginal effect refers to how changes in the value of a variable keeping others at a constant value affect the likelihood of occurrence of hyaena.These graphs cogently show changes in the likelihood of hyaena occurrence with values of important variables (Dem = Elevation, Bio3 = Isothermality, Chital = Occurrence probability of chital, Dhole = Occurrence probability of dhole, Bio14 = Precipitation of Driest Month, Bio12 = Annual Precipitation, Settlement = distance from settlement, Cattle density = No. of cattle per sq.km. and Sambar = Occurrence probability of sambar).
support and the Chief Wildlife Warden, Tamil Nadu, for granting permission to work across forest divisions.The project was funded by the Science and Engineering Research Board, a statutory body of the Department of Science & Technology, Government of India, under the Core Research Grant (Grant no.EMR/2017/001849), and Ramanujan Fellowship scheme (Grant no.SB/S2/RJN049/2016) to