Study area
The present study was conducted in Uttarkashi district, Uttarakhand, located between 38° 28′ to 31°28′ N latitude and 77°49′ to 79°25′ E longitude with an area of about 8016 km2, covering primarily hilly terrain with an altitudinal range of 715–6717 m (Fig. 3). The terrain is mountainous, consisting of undulating hill ranges and narrow valleys with temperate climatic conditions. The district lies in the upper catchment of two major rivers of India, viz., the Ganges (Bhagirathi towards upstream) and the Yamuna. The major vegetation types of the study area are Himalayan moist temperate forest, sub-alpine forest and alpine scrub59. The Uttarkashi district forests are managed under three Forest Divisions viz., (i) Uttarkashi Forest Division (ii) Upper Yamuna Badkot Forest Division and (iii) Tons Forest Division) with two Protected Areas (PAs) (i) Gangotri National Park and (ii) Govind Pashu Vihar National Park. The forested habitats of the study landscape are home to top conservation priority species, including Asiatic Black bear (Ursus thibetanus), Musk deer (Moschus spp.), Common leopard (Panthera pardus), Himalayan brown bear (Ursus arctos isabellinus) and Western Tragopan (Tragopan melanocephalus), Himalayan monal (Lophophorus impejanus). The study was conducted after a study permit issued by the Chief Wildlife Warden, Forest Department, Uttarakhand government, vide letter no. 848/5-6 dated 31/08/2019, we have not handled the species for doing research. Instead, remote camera traps have been used for collecting the data with the permission of the Chief Wildlife Warden, Government of Uttarakhand. Further, informed consent was taken before interviewing the local communities. The data was collected according to the institutional guidelines and approved by the Research Advisory and Monitoring Committee of the Zoological Survey of India.
Map of the study area Uttarkashi, Uttarakhand. ArcGIS 10.6 (ESRI, Redlands, CA) was used to create the map. (Map created using ArcGIS 10.6; http://www.esri.com).
Sampling protocol
The basic sampling protocol and assumptions for multi-species occupancy modelling are identical to the single-species case7. Briefly, a set of 62 intensive sites, were randomly selected, and each site i was surveyed j times. During each survey, detection/non-detection of S focal species was recorded. Additionally, direct or indirect evidences of species presence from the different areas were also recorded.
Data collection
The complete study area was divided into 10 × 10 km grids, consisting of n = 60 grids. Based on the reconnaissance survey, out of these 60 grids, we selected 25 girds that were accessible to conduct the survey and have the species presence. Further, these grids were divided into 2 × 2 km grids to maximize our effort so that all logistically accessible grids could be covered, and we conducted intensive sampling in N = 62 grids after excluding the grids with human settlements. T The field surveys were conducted during 2018–2019, and a team of researchers systematically visited selected grids to collect data on the detection/non-detection of these ungulates. A total of 62 camera traps were deployed in selected grids, and 650 km were traversed, accounting for N = 54 trails in these sampled grids. These camera traps were visited once in every fifteen days for replacing the batteries as well as documenting the presence of the species through the sign surveys. The ultra-compact SPYPOINT FORCE-11D trail camera (SPYPOINT, GG Telecom, Canada, QC) and Browning trail camera (Defender 850, 20 MP, Prometheus Group, LLC Birmingham, Alabama, https://browningtrailcameras.com) camera traps were used to detect the presence/absence of ungulate species. The cameras were mounted 40–60 cm above ground on natural trails without lures.
Data exploration
While deploying camera traps, we also noted habitat variables through on-site observation such as distance to the village and human disturbance. We tested site covariates for collinearity and discarded one of a pair if the Pearson’s correlation was greater than 0.760. Hence, we assumed each of the site covariates could influence the occupancy and detectability of these ungulates.
Covariates
We hypothesized that habitat variables may influence these ungulates’ occupancy and detection probability. A total of 21 variables were extracted either from the field or using the ArcGIS v. 10.6 software (ESRI, Redlands, CA), and only 14 were retained after collinearity testing60 (Table 3). These covariates were classified into the following categories (Topographic variables, Habitat variables and anthropogenic variables). The topographic variables (elevation, slope and aspect) were generated using 30× resolution SRTM (Shuttle Radar Topography Mission) image downloaded from EarthExplorer (https://earthexplorer.usgs.gov/). The habitat/ land cover classification was carried out using Landsat 8 satellite imagery (Spatial resolution = 30 m) downloaded from Global Land Cover Facility by following the methodology suggested by61 using the ArcGIS v. 10.6 software (ESRI, Redlands, CA). The study area was classified into nine Land use/land cover (LULC) classes viz., West Himalayan Sub-alpine birch/fir Forest (FT 188), West Himalayan upper oak/fir forest (FT 162), West Himalayan Dry juniper forest (FT 180), Ban oak forest (FT 152), Moist Deodar Forest (FT 155), Western mixed coniferous forest (FT 156), Moist temperate Deciduous Forest (FT 157) which were used for further analysis considering their importance to species ecology and behavior60. The values for all the covariates were extracted at 30 m resolution, and a single value per site was obtained by averaging all the pixel values within each sampling site (camera trap locations).
Occupancy modelling framework
We used multi-species occupancy modelling62 of barking deer, goral and sambar to estimate the probability of the species (s) occurred within the area (i) sampled during our survey period (j), for accounting the imperfect detection of the species8. Distinguishing the true presence/absence of a species from detection/non-detection (i.e., species present and captured or species present but not captured) requires spatially or temporally replicated data. We used camera stations to record the presence/absence of species along with sign survey in all the studied grids. The camera traps were placed along the trail/transects in the studied grids hence each grid needs to be visited once in every fifteen days to check the camera traps as well as to document the presence of the studied species. Therefore, we treated 15 trap nights as one sampling occasion at a particular camera station resulting in ~ 7 sampling occasions per camera station.Our aim was to record the presence/ absence of the species at a particular gird hence we incorporated sign survey data if the species was not detected in camera station but recorded through sign survey. We pooled the presence/absence data in a single sheet of each species following6 and fitted occupancy and detectability models using programme Mark63,64. We model the species (s) presence (ysij = 1) and absence (ysij = 0) at site i during survey j, and the sampling protocol was identical to single species case65, where the Bernoulli random variable was conditional on the presence of species s (Zs = 1) following6
$${text{y}}_{sij} sim {text{ Bernoulli}}left( {{text{p}}_{sij} {text{z}}_{si} } right),$$
where Psij represents the probability of detecting species S during replicate survey j at site i and Zsi = presence or absence of species s at site i.
Furthermore, we model the latent occupancy state of species s at site i as a multivariate Bernoulli random variable:
$${text{Z}}_{i} sim {text{MVB}}left( {uppsi _{i} } right)$$
where Zi = {Z1i, Z2i….., ZSi} is an S-dimensional vector of 1’s and 0’s denoting the latent occupancy state of all S species and (ψi) is a 2S-dimensional vector denoting the probability of all possible sequences of 1’s and 0’s Zi can attain such that ∑ ψi = 1 with corresponding probability mass function (PMF) adopted from6,64.
$$fleft( {{text{Z}}_{i} } right) = {text{ exp}}left( {left( {{text{Z}}_{i} {text{log}}(uppsi_{{text{i}}} {1}/uppsi_{{text{i}}} 0} right) , + {text{ log}}left( {uppsi_{{text{i}}} 0} right)} right).$$
The quantity f = log (ψi1/ψi0), is the log odds species S occupies a site often referred to as a ‘natural parameter’.
Since we are modeling three ungulate species (S = 3), 2S = 23 the possible encounter histories included in the dataset were eight, if neither of the two species were detected the value of ‘00’ was assigned; similarly ‘01’ indicates detection of species 1; ‘02’ indicates detection of species 2; ‘03’ indicates detection of both the species; ‘04’ indicates detection of species 3; ‘05’ indicates detection of species 1 and species 3; ‘06’ indicates detection of species 2 and species 3 and ‘07’ indicates detection of all the three species. We modelled constant occupancy and detection probability for each of the three species. Hence, we specified 6 f and p parameters, an intercept (β) for each of one-way f parameter and detection parameter p following64.
$$f_{{1}}=upbeta_{{{1},}} ;;{text{p}}=upbeta_{{4}}$$
$$f_{{2}} = upbeta_{{{2},}} ;{text{p }} = , upbeta 5$$
$$f_{{3}} = , upbeta_{{{3},}}; {text{p }} = , upbeta_{{6}}$$
We fit a set of models including the detection probability as a constant, p(.), and variable function to occupancy ψ(covariate) for site-specific covariates and models include occupancy as constant ψ(.) and variable function of the detection p(covariates) for the respective site covariates.
As we have assumed the independence among all three species, the model shows marginal occupancy probabilities of species 1, species 2 and species 3 varies as a function of environmental variables. We incorporated site-level characteristics affecting species-specific occurrence (f1: occupancy of species 1, f2: occupancy of species 2, & f3: occupancy of species 3) and detection probabilities using a generalized linear modelling approach42. This requires 9 parameters: an intercept (β1, β3, β5) and slope (β2, β4, β6) coefficient for each 1-way f parameter f1, f2, f3 and an intercept parameter for each detection parameter (β7, β8, β9). Below mentioned is the model for 1-way f parameters.
$$f_{{1}} = , upbeta_{{{1 } + }} upbeta_{{2}} left( {{text{Covariate}}} right),;;{text{ p }} = , upbeta_{{7}}$$
$$f_{{2}} = , upbeta_{{{3 } + }} upbeta_{{4}} left( {{text{Covariate}}} right),;;{text{ p}} = , upbeta_{{8}}$$
$$f_{{3}} = , upbeta_{{5}} + , upbeta_{{6}} left( {{text{Covariate}}} right),;;{text{ p }} = , upbeta_{{9}} .$$
All covariates were standardized before model fitting. We fitted the most complex model to each species and considered all possible combinations of covariates using the logit link function. Our rationale for including these variables in the occupancy and detectability component of the model was that we expected these variables to influence the occupancy and detectability of the study species.
Since multi-species occupancy simultaneously model environmental variables, & interspecific interactions. Further it also allows to understand the influence of environmental variables on one species occupancy, in the presence or absence of other sympatric species64. Hence, we also modeled two species occur together as a function of covariates. We examined how the variables of each camera site influenced the pair-wise interaction of the three ungulate species. This model assumes that the conditional probability of one species varies in the presence or absence of other species. We assumed f123: co-occurrence of species 1, species 2 & species 3 = 0, hence we did not include higher-order interactions in any of our models, we assumed the conditional probability of 3 species occurred together was purely a function of species-specific (f1, f2, f3) and pair-wise interaction (f12: co-occurrence of species1 & species 2, f13: co-occurrence of species 1 & species 3, f23: co-occurrence of species 2 & species 3) parameters. We modeled pair-wise interaction of species varies as a function of environmental variables keeping detection probability constant. Hence, we specified 15 f and p parameters, an intercept and slope coefficient for each of the one-way (f1, f2, f3) and the two-way f parameters (f12, f13, and f23); as well as an intercept parameter for each of the detection models. The model equation below implies for 2-way f parameters:
$$f_{{{12}}} = , upbeta_{{{7 } + }} upbeta_{{8}} left( {{text{Covariate}}} right),;;{text{ p }} = , upbeta_{{{13}}}$$
$$f_{{{13}}} = , upbeta_{{{9 } + }} upbeta_{{{1}0}} left( {{text{Covariate}}} right),;;{text{ p }} = , upbeta_{{{14}}}$$
$$f_{{{23}}} = , upbeta_{{{11 } + }} upbeta_{{{12}}} left( {{text{Covariate}}} right),;;{text{ p }} = , upbeta_{{{15}}} .$$
We also fitted models including co-occurrence and detection probability of a species varies as a function of environmental variables. Hence, we specified 18 f and p parameters, an intercept and slope coefficient for each of one-way (f1, f2, f3) and two-way f parameters (f12, f13, f23); and an intercept as well as the slope parameters for each of the detection models. The model equation below implies for 2-way f parameters:
$$f_{{{12}}} = , upbeta_{{{7 } + }} upbeta_{{8}} left( {{text{Covariate}}} right),{text{ p }} = , upbeta_{{{13 } + }} upbeta_{{{14}}} left( {{text{covariate}}} right)$$
$$f_{{{13}}} = , upbeta_{{{9 } + }} upbeta_{{{1}0}} left( {{text{Covariate}}} right),{text{ p }} = , upbeta_{{{15}}} + , upbeta_{{{16}}} left( {{text{covariate}}} right)$$
$$f_{{{23}}} = , upbeta_{{{11 } + }} upbeta_{{{12}}} left( {{text{Covariate}}} right),{text{ p }} = , upbeta_{{{17}}} + , upbeta_{{{18}}} left( {{text{covariate}}} right)$$
A total of 38 models were run to test the influence of environmental variables on occupancy and detection probability of species-specific (f1, f2, f3) and pair-wise interaction of the three ungulate species. The best-supported model was identified by selecting the model with the lowest AICc value and highest model weights66, where higher model weights indicate a better fit of the model to the data. Second-Order Information Criterion (AICc)67 values were used to rank the occupancy models, and all the models whose ΔAICc < 2 were considered as equivalent models. The Akaike weight represents the ratio of ΔAICc values for the whole set of candidate models, providing a strength of evidence for each model. The sign of the logistic coefficient of each variable (positive or negative) was used to determine the direction of influence of the variables on the occupancy and detection probability of the three ungulate species.
Activity pattern
We have compared activity patterns among species to see how overlapping patterns may relate to the competition using the package “Overlap” in R (R Development Core Team). The time and date printed on the photographs have been used to determine the daily activity pattern of individual species68. We used a Daily Activity Index (DAI) of half an hour duration to examine the daily activity. The coefficient of overlap is denoted by “Dhat1” values, ranging between zero (no overlap) and 1.0 (complete overlap).
Source: Ecology - nature.com
