Study site
Simbalbara Wildlife Sanctuary (SWLS): SWLS is located in Paonta valley of Sirmour district in Himachal Pradesh. It covers an area of 27.88 km2 with altitude ranging from ~ 450 to 650 m asl and lies between latitude 30°24'21" N to 30°28'13" N and longitude 77°27'18" E to 77°31'26" E (Fig 1). The region's hilly terrain is representative of the lower Shiwaliks that further emerge into middle and upper Shiwaliks, and the Western boundary is adjoining the Kalesar National Park of Haryana Forest Division. The sanctuary has a subtropical climate with hot summers and severe winters. The summer temperatures touch as high as 46°C and drop to 6°C during winters, with a mean annual rainfall of 1,260 mm49. Flora and fauna of the regions show similarities with the Western Himalayas, Punjab Plain and Upper Gangetic Plains50 and represents the bio-geographic province 4A51. The area is covered by moist Sal bearing forest and northern dry mixed deciduous forests52, which is considered as the westernmost limit of Sal distribution in India53. The prominent mammal species are panthera pardus, Nemorhaedus goral, Rusa unicolor, Muntiacus muntjac, Sus scrofa, Canis aureus, Hystrix indica, Axis axis, Paguma larvata and Martes flavigula53.
Churdhar Wildlife Sanctuary (CWLS): CWLS is located in the Himachal Pradesh, Western Himalaya, India. It covers an area of 56 km², and a wide altitudinal range varying from 1900-3647 m asl and lies between 30°48'39"- 30°54'39" N latitude and 77°29'32"- 77°29'49" E longitude (Fig 1). The average annual rainfall in this area is about 1,200 mm, 75% of which is received from the southwest monsoon. The representative vegetation of the sanctuary is Quercus leucotricophora, Rhododendron arboreum, Neolitsea pallens, Lyonia ovalifolia, Pyrus pashia, Picea smithiana, Quercus semecarpifolia, Abies pindrow, etc.54. The mammalian diversity of the region includes species such as Panthera pardus, Ursus thibetanus, Nemorhaedus goral, Moschus chrysogaster, etc.30.
Field sampling design
We photographed mammals using camera traps in a 27.88 km2 area of SWLS and 56 km2 area of CWLS. We selected camera-trap locations based on accessibility, terrain features, animal trails, and nallahs (seasonal drainages) with carnivore signs55. We deployed a single Cuddeback X-Change™ colour model (Cuddeback, Green Bay, WI, USA) with motion sensors at each location. We set a time lag of 2s between animal detections and fastened cameras to trees at 30-45 cm above the ground for 29 and 58 days (average) for SWLS and CWLS, respectively. We deployed camera-traps (grid size: 1 km2) in SWLS during two sampling blocks: March 2021-April 2021 (n=17) and April 2021-May 2021 (n=14), and in CWLS, during a single sampling block: October 2021- December 2021 (n=31) (Fig 2).
Species richness and RAI1: We estimated species richness as the total number of species detected during the study period. With the assumption that the photodetection rate is influenced by animal abundance, the photographic trapping rate has been widely used to estimate relative abundance56. Furthermore, a significant correlation between trapping rates and independent density estimates supports its use as a relative abundance index57,58. We calculated RAI1 (Relative Abundance Index) as a total number of independent photographs for each species divided by total trap nights and multiplied by 10057. The criteria to determine a photographic event (a species occurrence) were (1) consecutive photographs of the same species within 0.5 hours (30 minutes) were counted as one species occurrence, (2) the stamped time of the first photograph of these consecutive photographs was taken as the species-occurrence time. After 30 minutes, additional photos of the same species were considered another occurrence event, and (3) different identifiable individuals were treated as a separate occurrence even though they appeared in the same photograph, or the photographs were taken within 30 minutes58. The analysis was carried out in a windows-based MS office excel worksheet using the data analysis tool.
Trap effort: We assessed RAI2 (the number of trap nights required to get a single photograph of the species) and RAI3 (the number of trap nights required to get a first photograph of the species) to understand the time required to detect mammals if they are present at a sampling location56. We calculated RAI2 as total trap nights divided by the number of independent photos of each species and RAI3 through frequency distribution of nights to the first detection of each photo-captured species. We analysed all analyses in a windows-based MS office excel worksheet using the data analysis tool.
Furthermore, to quantify the optimal number of camera stations and days (i.e., how many locations and days needed to be sampled to capture most mammals), we plotted mammal species detected against sample locations and days and fitted a hyperbola curve. We created this species accumulation curve (SAC) for all mammals pooled across camera stations and days to evaluate the sampling quality and survey effort needed for determining species richness. To eliminate the order in which data was recorded, we randomised the data 100 times using the vegan package59 with R software v. 3.5.2.
Density
Identifiable individuals
Camera-traps have been used to estimate the density of tigers, leopards, and other carnivores in various landscapes60,43,61. In both study areas, we distinguished leopards, leopard cats, and small Indian civets by their natural body and face markings with distinctive patterns, enabling the identification of individual animals. Since we deployed a single camera trap at each location, both flanks of all the captured individuals were not obtained. Therefore, we used the flank with the maximum number of photographs. Further, we created individual detection histories using binary format (detection or non-detection of the individual) and other trap-specific details (spatial coordinates of the trap, time, and date). We considered each camera station a proximity detector, allowing animal detection at multiple traps on any given occasion. We used the spatially explicit capture-recapture (SECR) method to estimate density using a maximum-likelihood-based approach9. This method eliminates the subjectivity of calculating an effective trap area to estimate density62. We considered a 10 km buffer width around the trapping grids to ensure that no individuals outside the buffered regions had any probability of being photographed by the camera trap63.
Unidentifiable individuals
Camera-trap distance sampling (CTDS): We computed the distance between the animal and the camera at snapshot moments during distance sampling with camera traps to ensure that animal movement does not bias the distribution of detection distances19. We thus defined a finite set of snapshot moments (2s apart) within the sampling period19 for a total number of thirty-one camera traps of SWLS and CWLS. We estimated the radial distance between each animal and the camera trap using a regression equation developed from the field calibration for each snapshot moment. We did this calibration for ten camera traps for distance sampling. The dependent variable in this equation was the ratio of an individual's actual height to its height in the photograph. The explanatory variable was the distance at which the individual was photo captured.
We determined actual heights for different species by comparing camera-trap photos of the species to the calibration pole height. We identified nine, eight, six, and five comparable images of adult males, adult females, sub-adults, and fawns for spotted deer Axis axis and sambar Rusa unicolor, respectively. If the population surveyed is not detectable during the data collection period selected for analysis, temporal sampling effort is overestimated, and thus density could be underestimated24. To avoid this bias, we included either the proportion of time when animals were available for detection as a parameter in the model or the defined sampling period when the entire population was available for detection19. We generated a temporal activity curve for each species, and the active period of each species was then considered the sampling period for the analysis.
We estimated density (D) following the equation for camera-trap point transects as
Where nk is the number of observations of animals at a point k (camera-trap location), ek is the temporal effort. Pk is the estimated probability of obtaining an image of an animal within θ degrees (angle covered by the camera's field of view), K is the total number of camera-trap locations and w (truncation distance) in front of the camera at a snapshot of the moment. We measured the effort at a point k as ek = θ Tk/2 πt where θ/2π describes the fraction of a circle covered by a camera, Tk is the period of camera deployment (in seconds), and t is the unit of time used to determine a finite set of snapshot moments within Tk (also in seconds). We defined the period of camera deployment as the time the target species was expected to be active during the sampling period. We used the distances ri to model the detection function and estimated Pk.
We censored distance data and modelled it with two different setups, 'user-manual' and 'empirical'. For the 'user-manual' setup, we assumed θ as 42° (0.733 radians). For the 'empirical' setup, we estimated θ by walking in front of the camera perpendicularly to the midline of the field of view and measured the distance from the operator to the midline that triggered the sensor while the camera was in setup mode. We repeated this procedure 3-4 times (walking five times from the left and five times from the right). We calculated the angle of view using basic trigonometric formulas and estimated realised θ. The value of θ by the empirical method was almost equal to that of the user-manual setup, so the value of θ used was 42° (0.733 radians).
We used the point transect distance sampling method in Distance20 for all analyses, where ek = θ Tk/2 πt is used to calculate the survey effort. For the analysis in distance software, we modelled the detection using the same functions as Howe et al.19: half normal with 0, 1 or 2 Hermite polynomial adjustment terms; hazard rate with 0, 1 or 2 cosine adjustments; uniform with 1 or 2 cosine adjustments. We constrained adjustment terms, where necessary, to ensure the detection function was monotonically decreasing. We selected candidate models of the detection function by comparing AIC values, acknowledging the potential for overfitting because many observations were not independent.
We also applied CTDS to compute the density of ungulates in both landscapes to see how densities vary for different grid sizes. We increased the size of sampling grids from 1 km2 to 1.5 km2 and 2 km2. Keeping the effort constant across grids, we repeated the CTDS analysis to estimate the densities of the ungulates. With an increase in the grid size, we randomly removed a few camera trap stations from the analysis. For SWLS, we analysed 11 camera traps in a grid size of 1.5 km2 and 9 cameras in a grid size of 2 km2, while for CWLS, 15 camera traps in a 1.5 km2 grid and 12 cameras in a 2 km2 grid.
Line transect distance sampling: We also used line-transects to determine the ungulate density40,64. We covered undulating terrain, dry riverbeds and mixed and Sal dominated forests and recorded every animal visually detected65. We laid eight transects, and each transect replicated six times, resulting in a total effort of 58 km.
We calculated the perpendicular distance (x) of an animal from the transect using a range finder (Inesis, Telemeter, 900) and a compass to determine the sighting angle (θ); and the radial distance via the equation x = r sin θ65. The method assumes that every animal on the transect path will be detected, and thus the animal detection probability is a declining function of perpendicular distance from the transect65. We fitted detection metric to the data to estimate the proportion of the population detected, which was used to estimate species population abundance with the standard estimator of the form:
Where N is abundance, A is the total survey area, n is the number of animals counted, w is the approximate distance view on each side of the transect, L is the length of the transect, and Pa is the detection probability of each animal. We used the statistical software package 'distance'20 to fit the models and estimate species abundance. It employs the models briefly described here and took size bias associated with the increased probability of detecting larger animal groups.