Habitat Connectivity for the Conservation of Small Ungulates in A Human-Dominated Landscape

Abstract

Conserving landscape connections among favorable habitats is a widely used strategy to maintain populations in an increasingly fragmented world. A species can then exist as a metapopulation consisting of several subpopulations connected by dispersal. Our study focuses on the importance of human–wildlife coexistence areas in maintaining connectivity among primary habitats of small ungulates within and outside protected areas in a large landscape in central India. We used geospatial information and species presence data to model the suitable habitats, core habitats, and connectivity corridors for four antelope species in an ~89,000 km² landscape. We found that about 63% of the core habitats, integrated across the four species, lie outside the protected areas. We then measured connectivity in two scenarios: the present setting, and a hypothetical future setting—where habitats outside protected areas are lost. We also modelled the areas with a high risk of human-influenced antelope mortality using eco-geographical variables and wildlife mortality records. Overall, we found that the habitats in multiple-use forests play a central role in maintaining the connectivity network for antelopes. Sizable expanses of privately held farmlands and plantations also contribute to the essential movement corridors. Some perilous patches with greater mortality risk for species require mitigation measures such as underpasses, overpasses, and fences. Greater conservation efforts are needed in the spaces of human–wildlife coexistence to conserve the habitat network of small ungulates.

1. Introduction

Dispersal is an essential process which allows for gene flow between populations leading to a higher genetic diversity and hence, greater resilience toward demographic stochasticity and environmental fluctuations [1,2,3]. Many studies have emphasized the need for landscape-scale conservation planning focused on facilitating dispersal among subpopulations to ensure their long-term survival as a metapopulation [4,5,6,7]. A set of small, isolated habitats would be less favorable for species persistence than if these habitats were connected by dispersal corridors to create a resilient metapopulation. Therefore, it is important to consider the outlying areas in addition to the designated wildlife areas (protected areas) for developing conservation plans. Conservation outside the protected areas is a more significant challenge due to increased exposure to human-induced disturbance. In these areas, wild animals not only face an increased risk of poaching [8] but also suffer due to habitat degradation [9] and environmental pollution [10]. Proximity to humans elevates the risk of disease [11,12,13] and induces physiological stress [14,15,16]. Human infrastructure like roads, railways, power lines, and also attacks by stray dogs cause wild animal deaths every year [17,18,19,20,21,22,23,24,25]. Therefore, specific conservation management policies for core habitats and connectivity corridors must be formulated to create a functional metapopulation. Another challenge is that a single management approach may not work for all species of different habitat preferences and dispersal abilities. Facilitating dispersal among distant populations of slow dispersers is a significant concern [26]. Slow dispersing species are at a greater risk of extirpation [27] as their rates of movement to suitable areas may be insufficient to respond to changes. Furthermore, maintaining habitat contiguity for habitat specialists may be more challenging than for habitat generalists. Therefore, multi-species landscape connectivity studies are needed to develop inclusive conservation plans that cover a wide range of species attributes [28]. Here, we present a study on four antelope species of central India and examine the importance of multiple-use forests for maintaining connectivity in a large landscape. Our study would complement previous studies on habitat connectivity for carnivores in this landscape [29,30,31,32], and can help in comprehensive land-use planning.

Numerous studies across the world have focused on mapping dispersal corridors, including for antelope species [33,34,35,36,37]. These studies typically involve the use of primary ground data on vegetation, species presence and absence, and sometimes even genetic data [38,39], in combination with remotely sensed data on vegetation and landcover types, and multiple layers of geospatial information such as landscape disturbance and habitat contours. The resulting framework of geoinformatics and geoinformation are now widely used as effective tools in biodiversity and conservation assessments and management. Our study is the first investigation on landscape connectivity for antelope species in the Indian subcontinent. In our study, we use spatial models to identify areas that facilitate or act as barriers to dispersal. An investigation on the habitat configuration and corridor structure of antelopes could also inform conservation decisions for other ungulate species in the long term [40]. The loss of herbivores could have a cascading effect on carnivore populations, but also have significant impact on vegetation [41].

Here we emphasize a geoinformatics template for analyzing the landscape conservation plans focused on key taxa of herbivores that can contribute to the overall biodiversity conservation in a landscape. We bring together ground data on species presence collected at large scales, remotely sensed data on vegetation and landcover, and geospatial information systems on roads, infrastructure, human settlements, and topography to derive landscape features that can contribute to maintaining dispersal processes and species metapopulations. We believe that the geoinformatics template that we present here can be more widely applied to address conservation imperatives in human-dominated landscapes. Specifically, we show how the spatial configuration of the suitable habitats, dispersal corridors, ecological barriers, and the degree of their overlap with the current protected area network and human use forests can be analyzed to create a metapopulation conservation framework.

2. Study Area

Our study focuses on an 89,398 km² multiuse landscape in central India (Figure 1). The topography includes escarpments, gorges, streams, plateaus, and broad valleys. The dominant vegetation type is tropical dry deciduous forest [42], and the landscape includes eight high-priority wildlife conservation areas (known as protected areas), multiple-use forests, agricultural fields, and significant area under human settlements (villages, towns, and cities) [43]. In addition to protected areas, numerous other forest patches, scrublands, and natural grasslands are administered as territorial forests with multiple-use forest management (MFM) practices [44,45]. Territorial forests constitute important spaces for coexistence as they are administered for both biodiversity conservation and sustainable livelihoods based on various forest resources excluding timber [46]. The major part of the land area is, however, under non-forest land use like farmlands, villages, and towns. The protected areas, territorial forests, and non-forest lands account for 9.31%, 30.3%, and 60.39% of the total landscape, respectively (Figure 1).

3. Focal Species

We focus on four species of wild antelopes (i) Blackbuck (Antilope cervicapra), (ii) Chinkara (Gazella bennettii), also known as Indian Gazelle, (iii) Nilgai (Boselaphus tragocamelus), also known as Blue Bull, and (iv) Four-horned antelope (Tetracerus quadricornis), also known as Chousingha. These species are morphologically and behaviorally heterogeneous, and have different habitat and foraging preferences. Chinkara and four-horned antelope (FHA), are shy and prefer natural forests and scrublands [47]. Nilgai is a habitat generalist [48] and is widely distributed in the region. Nilgai and blackbuck are comparatively more common around human habitations and can be frequently seen in agricultural fields [49,50,51,52]. Blackbuck, chinkara, and four-horned antelope are small ungulates that on an average weigh less than 50 kg while the nilgai weighs well over 200 kg [53].

4. Materials and Methods

We modeled the distribution of core habitats and the corridors that facilitate dispersal among them (Figure 2). Having identified core habitats and corridors, we quantified the level of connectivity between different core habitats. We also mapped the high-risk areas, which are either barriers or sites where accidental animal deaths may be high. The core habitats, corridors, and high-risk areas represent the essential sites that are needed to create functional metapopulations.

4.1. Ungulate Species Presence Data

Species presence records over such a large area can only be obtained by the state government agencies who possess the necessary logistics and infrastructure. The state department monitors not only species presences using surveys carried out by their staff but also records data on accidental deaths of wild animals. We were able to obtain this data set from the state forest department for use in our study.

Since species presence data are, in general, more reliable than species absence, our analysis was based entirely on species presence data. However, sites where species are currently absent are also known. The presence data is available at the scale of “forest compartments,” which are administrative units of area less than 10 km². We used a total of 83, 762, 209, and 1638 presence locations, along with randomly generated background points of species absences to model the suitable habitats for blackbuck, chinkara, four-horned antelope, and nilgai, respectively (Figure 3). The recommended number of background points is based on the type of algorithm used to model habitat suitability [54], with ~10,000 for regression-based approaches and equal numbers (compared to presence) for classification and machine-learning techniques. Since we used multiple algorithms, we reckoned that selecting background points equal to the number of presence points would be an optimal choice. We placed a 2 × 2 km² grid on the landscape, which yielded 23,051 cells, and species presence was recoded for each grid cell (Figure 3). The background points do not overlap with the presence grid cells and tended to match sites of known absences based on our field knowledge. The presence data therefore establishes the conditions in which a species is more likely to be present [55]. The final presence dataset comprises standardized survey data from 65% of the landscape’s forest areas (including both protected areas and territorial forest areas) supplemented with human-wildlife conflict records [56] from 91.78% of the landscape. We considered the conflict records in order to include data from the non-forest areas frequented by these antelope species (See Supplementary Information S1 for more details on data preparation). We finally checked for spatial autocorrelation in species presences using Moran’s I.

4.2. Modelling the Suitable Habitats

We followed the ODMAP protocol for modeling habitat suitability [57]. Habitat suitability is then used to derive the resistance surface and the core habitat patches. We used an ensemble approach with seven algorithms—generalized linear model (GLM), generalized additive model (GAM), random forest (RF), support vector machine (SVM), multivariate adaptive regression splines (MARS), flexible discriminant analysis (FDA) and gradient boosting machine (GBM). GLM and GAM are classical regression models, while others are based on machine learning methods [55,58]. To evaluate model accuracy, we used a split-plot approach and measured the AUC (area under the receiver operating characteristics curve) value for each model. Although AUC is best applied when both presence and absence data are used, we used AUC because we could “eyeball” the randomly derived background points of absences based on our field knowledge. We prepared the final consensus map using the weighted average of model outputs, which had AUC greater than 0.7 [59]. The analyses were implemented in R package “sdm” [58] (see Supplementary Information S1 for more details).

We used seven predictor variables to model the suitable habitats. These include distances (Euclidean) from water sources (dwater), roads and railways (drora), and human settlements (dhustmts), slope position (slope), proportion of forest cover (forestp), pre-monsoon NDVI (ndvi_s), and post-monsoon NDVI (ndvi_w). We derived the vector layers of major roads, railways, and water sources from the online Open Street Map Database [60] and derived the dwater and drora layers using the spatial analyst tool of ArcMap 10.2 [61] at 30 m spatial resolution. We estimated the slope using the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) digital elevation model (DEM) [62]. through the DEM surface tool, a plugin extension tool [63] in ArcMap 10.2.

We digitized all human settlements using high-resolution Google Earth imagery by onscreen visual interpretation. We created 5062 polygons which represent the constructed portions in the landscape. We developed the dhustmts layer using the digitized polygons through the spatial analyst tool of ArcMap 10.2. We used the 30 m spatial resolution Landsat-8 imageries from the USGS earth explorer [64] for the pre-monsoon and the post-monsoon periods to derive the ndvi_s and ndvi_w layers.

We tested multicollinearity among the predictor variables using variance inflation factor analysis, considering a threshold of 0.9. The grain size of all environmental layers was rescaled to 2 × 2 km² to match the resolution of the response data.

4.3. Map Binarization and Morphological Spatial Pattern Analysis (MSPA)

We applied the specificity-sensitivity threshold using the R package PresenceAbsence [65] and converted the continuous gradient maps into binary distributions, creating two classes, suitable and unsuitable. Subsequently, we performed the Morphological Spatial Pattern Analysis (MSPA) on the suitable areas using the Guidos Toolbox [66,67] to identify the core habitat patches. MSPA, is “a customized sequence of mathematical morphological operators to describe the geometry and connectivity of the components of an image” [68] and has been used to assess the fragmentation of habitats and to map the corridors in a landscape [69]. In MSPA, the image is classified, based on the shape and physical appearance of its features, into seven categories: core, islet, loop, bridge, perforation, edge, and branch [66]. MSPA: While this approach is useful for mapping the core habitats, it is not ideal to map the corridors as it does not consider the dispersal capabilities (the maximum dispersal distance) of the species. A different approach is therefore needed to design the corridors that takes note of the dispersal abilities.

4.4. Modeling Dispersal Corridors

We used the randomized shortest path algorithm (RSPA) to model the dispersal corridors that connect the core habitats. Here we used the circuit theory approach (by setting Ɵ equal to 0 in RSPA) as it is considered the most appropriate method for modeling dispersal corridors when the starting data is species presence locations [70] (Ɵ equal to 1 would simulate the least-cost path). The algorithm then estimates density values for each grid cell in the landscape, with higher values representing a greater probability of a random walker passing through the cell [71]. These analyses were implemented in the R package “gdistance” [72]. The data needed for this analysis are, the core habitat patches, the resistance surface, and the maximum dispersal distances of the focal species. We created the resistance surface using a negative exponential transformation of the habitat suitability gradient maps [73] using the following equation:

$$100-99\times \left(\frac{\left(1-exp\left(-c\times h\right)\right)}{\left(1-exp\left(-c\right)\right)}\right)$$

(1)

Here, h is the habitats suitability value. For c Keeley et al. (2016) [74] used eight different values (0.25, 0.5, 1, 2, 4, 8, 16, 32). The values 32 and 0.25 would mean a large deviation and a small deviation from the linear transformation, respectively. We assume that the c value would be located away from the extremities for slow dispersal species like antelopes. Therefore, we created resistance maps with the c values of 2, 4, and 8 and “eyeballed” to select the most realistic map (c = 4) based on our field experience. The negative exponential method provides a robust estimate of resistance when dispersing animals traverse through unsuitable sites [74]. Dispersal ability is an important attribute for designing corridors, but there are no available estimates of the maximum or average dispersal distances of Indian antelopes. Therefore, we approximated their maximum dispersal distances using their home range sizes and their body mass (see [75]). There are reports of the home range sizes of blackbuck, nilgai, and chinkara [76,77,78,79,80,81,82,83,84], but none for the four-horned antelope [85], so we had to rely on average body mass for analysis. The maximum dispersal distances used as cut-off values for blackbuck, chinkara, four-horned antelope, and nilgai are 46 km, 45 km, 36 km, and 80 km, respectively. While we believe that these values are appropriate for our study area, we do not expect them to be the absolute maximum dispersal distances for these species (see Supplementary Information S2 for more details). The resistance values of each grid cell in the landscape connecting the core habitats and the maximum dispersal ability of the species were used to derive the probabilistic dispersal surface, where each cell is assigned a corridor value (range 0 to 100). The corridor values along with the core habitats were used to create and integrated antelope conservation area that highlights high probability corridor regions and core habitats.

4.5. Quantifying Functional Connectivity in the Landscape

Identifying the key sites responsible for a disproportionate increase in connectivity is vital to designing effective dispersal corridors. We used Graphab version 2.4 [86] to measure a total of 19 landscape connectivity metrics to identify the most crucial areas. Graphab software is useful for modeling ecological networks using landscape graphs, and is often applied to quantify the importance of individual habitat patches, habitat clusters, and connections, using specific landscape measures based on habitat configuration and resistance. The connectivity metrics and their measurements are detailed in the Graphab 2.4 user manual [87], and we do not describe them here. The primary input data needed are the core habitats, the resistance surfaces, and the maximum dispersal costs (threshold) for each species. We determined the dispersal thresholds within the Graphab environment using the species’ estimated maximum dispersal distances.

The local connectivity metrics evaluate the ability of each habitat patch (or patch cluster) for facilitating dispersal and helps in prioritizing the crucial habitats. Global connectivity metrics on the other hand describe the level of connectedness or fragmentation in the whole landscape. We used the global metrics (

Table 1. Weighted global connectivity metrics used to measure connectivity in two scenarios.

Metric	Formula	Meaning
Flux (F)	$F={\displaystyle {\displaystyle \sum }_{i=1}^n{\displaystyle \sum }_{j=1}^n}{a}_j^{\beta }{e}^{-\alpha d_{ij}}$j≠i	Sum of potential dispersions from all patches
Equivalent Probability (EC)	$\begin{array}{l}EC\\ =\sqrt{{\displaystyle {\displaystyle \sum }_{i-1}^n{\displaystyle \sum }_{j=1}^n}{a}_i}{a}_j{e}^{-\alpha d_{ij}}\end{array}$	Square root of the sum of products of capacity of all pairs of patches weighted by their interaction probability.
Probability of Connectivity (PC)	$\begin{array}{l}PC\\ =\frac{1}{A^2}{\displaystyle {\displaystyle \sum }_{i-1}^n{\displaystyle \sum }_{j=1}^n}{a}_i{a}_j{e}^{-\alpha d_{ij}}\end{array}$	This indicates probability that two points randomly placed in the study area are connected.
Integral Index of Connectivity (IIC)	$\begin{array}{l}IIC\\ =\frac{1}{A^2}{\displaystyle {\displaystyle \sum }_{i-1}^n{\displaystyle \sum }_{j=1}^n}{\frac{a_i{a}_j}{1+n{l}_i}}_j\end{array}$	Product of patch capacities divided by the number of links between them, the sum is divided by the square of the area of the study zone.

) to bring out the importance of the core habitats in the multiple-use forests and non-forest lands for maintaining connectivity in the larger habitat network. We estimated the global connectivity metrics in two scenarios. In the first scenario, we measured the global metrics using all the core habitats. In the second scenario, we measured the global metrics using only the core habitats present inside protected areas. The latter is a hypothetical scenario in which habitats outside protected areas are lost. Using this, we evaluated the reduction in global connectivity when the habitats outside the protected areas are lost. Land degradation outside protected areas due to poor land administration may render core habitats unsuitable for wildlife species. Since future land use policies cannot be easily anticipated, we used a simple future scenario where the core habitats remain only in the protected areas. However, we use the same resistance surface in both the scenarios due to the lack of a better alternative. In reality, the resistance would increase if the landscape is unfavorably altered, but here we only aim to highlight the influence of outlying core habitats as virtual focal nodes in the overall habitat network.

4.6. Modeling Human-Influenced Antelope Mortality

To study human-influenced wildlife mortality, we obtained the wildlife mortality data of the past 20 years (2000–2019) from the Madhya Pradesh state wildlife mortality records [56]. In particular, we selected the incidents in which human activities or infrastructure had directly or indirectly resulted in the death of antelopes (Figure 4). This includes mortality incidents due to a wide variety of causes including poaching, road accidents, drowning in wells, accidental electrocution, and killing by stray dogs. We assumed that an area considered unsafe for one antelope species due to anthropogenic causes would also be unsafe for other antelopes. Therefore, we aggregated all the data into a single antelope-mortality dataset. The final dataset consisted of a total of 370 antelope mortality incidents from 318 locations (grid units). So, most of the grid cells had zero mortality incidents, inflating the zero counts in the data. Using these wildlife mortality records and a set of five eco-geographical variables, we developed a spatial risk map for anthropogenic antelope mortality. Excluding some of the peripheral areas for which wildlife mortality data were not available, the study area for this analysis was composed of 19,907 cells of a 2 × 2 km² grid.

The five predictor variables are: proportion of forest cover (forestp), distances (Euclidean) from roads and railways (drora), human settlements (dhustmts), forest boundary (distancefb), and human population density (population). We already described forestp, drora, and dhustmts layers in Section 4.2. We acquired human population data from NASA- Socioeconomic Data and Applications Center (SEDAC) [88] to prepare the population layer. We created the distancefb layer using the forest boundary shapefiles in the spatial analyst tool of ArcMap 10.2.

We first modeled the anthropogenic antelope mortality using a non-spatial Zero-Inflated Poisson (ZIP) model that incorporates covariate effects at the grid-cell scale and assessed spatial autocorrelation by examining the residuals of the regression. The ZIP model is a mixture of a probability mass distribution based at point zero and a Poisson distribution with mean μ_k. The probability of Y_k in the distribution based at structural zero is ω_k [89]. The parameters μ_k and ω_k are modeled by,

$$ln\left({\mu }_k\right)= X_k^T\beta +O_k+ {\psi }_k$$

(2)

and

$$ln \frac{{\omega }_k}{1-{\omega }_k} = v_k^T\delta +O_k^{\left(2\right)}$$

(3)

where, $X_k$ are explanatory spatial covariates associated with the mean process (μ_k) with $\beta$ oefficients, $O_k$ is a vector of offset for the Poisson mean and ${\psi }_k$ is a set of spatially autocorrelated random effects for grid cell k. The terms $v_k$, $O_k^{\left(2\right)}$ are respectively covariates and a second offset term (designated as (2)) to determine the probability that observation Y_k is in the point mass distribution, while $\delta$ is the corresponding vector of regression parameters. ZIP regressions were done using the “pscl” package in R [90]. We assessed spatial autocorrelation in the residuals using Moran’s I and a Monte Carlo significance test. A null distribution Moran’s I was computed after randomly permuting the antelope mortality data across the grid cells in the landscape 1000 times and computing Moran’s I for each permutation and comparing the observed Moran’s I against this null distribution. We then performed Bayesian spatial modelling of the number of antelope deaths per grid cell using the Conditional Autoregressive (CAR) approach and implemented this in the R package CARBayes [91]. For the ZIP response, the linear predictor was the same set of variables indicated above and was treated as random effects. To model autocorrelation in this Bayesian framework, we use the conditional autoregressive prior proposed in Leroux et al. (2000) [92] and implemented in the function S.CARleroux within the CARBayes package. This is an appropriate conditional autoregressive prior for the binomial, Gaussian, Poisson, or ZIP response variables [91].

We obtained the posterior distribution of model parameters on 10,000 Markov Chain Monte Carlo (MCMC) samples generated from a single Markov chain, executed for 300,000 iterations with a burn-in of 150,000 and then thinned by 15, to reduce Markov chain autocorrelation. The results of the Bayesian Spatial Model were used to predict the areas with a high-risk of anthropogenic antelope mortality.

5. Results

5.1. Modelling the Suitable Habitats

We found no evidence of multicollinearity among the predictor variables. The weakest correlation was found between d_water and slope: 6.3 × 10⁻³, whereas the strongest correlation was between slope and forestp: 0.64. The variance inflation factors (VIF) of the predictor variables were all lower than typical cut-off value of 5, which indicate significant collinearity (forestp: 2.48, dhustmts: 1.39, ndvi_s: 1.50, ndvi_w: 1.96, drora: 1.18, slope: 1.84, and d_water: 1.09). Moran’s I values for spatial autocorrelation in species presence data for blackbuck, chinkara, four-horned antelope, and nilgai were 0.1742, 0.4855, 0.4241, and 0.4638 respectively. The seven algorithms that we used gave somewhat different accuracies within and across species. Using an accuracy cutoff of 0.7 to include models, we considered all algorithms in the final ensemble model for chinkara, four-horned antelope, and nilgai. For blackbuck alone, we used only the GAM and MARS for the final model as the other algorithms gave AUC values <0.7. We prepared the final maps by rescaling the suitability values from 0 to 100 (Figure 5) (ROC plots are in Supplementary Information S3).

5.2. Map Binarization and Morphological Spatial Pattern Analysis (MSPA)

The binarized habitat suitability maps (Figure 6) show the spatial distribution of suitable and core habitats. The proportions of suitable and core habitats as a function of total landscape area and among landuse categories varies across species (

Table 2. Percentage of suitable habitats and core habitats as a function of the total landscape (total) and the allocation of these areas among the three land-use categories, protected areas (PA), territorial forests (TF), and non-forest land (NF).

	Suitable Habitat				Core Habitat
	Total	PA	TF	NF	Total	PA	TF	NF
Blackbuck	40.51	12.61	38.65	48.74	5.24	17.36	52.72	29.90
Chinkara	25.25	29.87	54.98	15.15	5.34	35.51	59.04	5.44
FHA	20.21	23.99	59.19	16.81	3.04	43.67	48.04	8.28
Nilgai	27.71	24.09	57.73	18.18	8.07	36.30	56.99	6.71
Integrated Area	52.88	16.79	41.12	42.08	21.70	36.74	51.48	11.78

), and integrated area for suitable and core habitats were 52.88% and 21.70% respectively.

5.3. Modeling Dispersal Corridors

The corridor maps for individual species (Figure 7) and the integrated across species (Figure 8) show the regions that offer low resistance for dispersal in the landscape. The patterns are highly variable across species and the integrated area considers the minimal requirement for allowing dispersal across all four species.

5.4. Quantifying Potential Functional Connectivity

All global connectivity metrics assessed in two scenarios unequivocally demonstrate that there will be a much lower overall connectivity in the landscape if the habitats outside protected areas are not conserved (

Table 3. The percentage decrease in weighted global connectivity metrics in the future hypothetical scenarios when habitats outside protected areas lost, compared to the present.

Connectivity Metric	Chinkara	Nilgai	Blackbuck	Four-Horned Antelope
Sum Flux	59.78	18.01	11.65	25.33
Equivalent Connectivity	54.78	37.96	41.86	64.24
Probability of connectivity	30.01	14.41	17.53	41.26
Integral index of connectivity	17.49	10.41	7.78	23.99

). A visual representation of the nodes and their linkages clearly demonstrate that the connections between most patches will be destroyed if the territorial forest habitats cease to exist (Figure 9). The same message emerges from the local connectivity metrics, which demonstrate that the habitat patches in the territorial forests play a central role in the connectivity network (see Supplementary Information S4). An important observation here is that the habitat networks of all four species have a low degree of cohesion, i.e., the removal of a few nodes will disconnect the entire graph, making each node very valuable.

5.5. Modeling Human-Influenced Antelope Mortality

The areas with a high risk of anthropogenic antelope mortality were predicted using the CARBayes spatial model (Figure 10). The Moran’s I statistic obtained for the residuals of the non-spatial Zero-Inflated Poisson model was 0.04, which showed that spatial autocorrelation is not present in the residuals. “Distance from forest boundary” and “forest percentage” were found to be the most significant predictors associated with higher anthropogenic mortality risks, followed by the “distance from roads & rails.” The posterior median values for these three predictors were −0.5859, 0.3725, and −0.2505 respectively and the 95% credible intervals did not include zero. The effect of “human population density” and “distance from human settlements” were not statistically significant. The results are summarized in

Table A2. Posterior quantities and DIC of the CARBayes analysis.

	Median	2.50%	97.50%	n.Effective	Geweke.Diag
(Intercept)	−6.1251	−6.5779	−5.553	114.7	3.3
dhustmts	0.0996	−0.0845	0.2827	1802.7	−1.1
drora	−0.2505	−0.4468	−0.0644	1291.6	2.7
distancefb	−0.5859	−0.8221	−0.3579	2009.2	1.7
population	0.0102	−0.1534	0.1525	3645.7	−0.6
forestp	0.3725	0.2266	0.5195	3531	0
Omega—(Intercept)	−181.765	−634.328	−1.0285	234.6	140.1
tau2	10.3691	7.6972	13.742	179	−2.2
rho	0.9984	0.9939	0.9998	1221.1	0.4
DIC = 6310.09
p.d = 2141.987
LMPL = −3762.73

in the Appendix.

6. Discussion

This study is the first attempt to identify the core habitats and potential corridors for Indian antelopes. It is an important milestone toward a greater understanding of multi-species habitat distributions that are needed for more inclusive conservation planning. Our study highlights the importance of the different land uses and administrative classes in maintaining dispersal among favorable habitats for these antelope species.

Global connectivity metrics and corridor structure of the four focal species supports the view that some species are better connected than others within the same landscape. This may be a function of their home range size and dispersal ability, so while nilgai is widely distributed with habitats that are largely contiguous, chinkara, and four-horned antelope remain limited to certain pockets, and their core habitats are connected only when they are in proximity. Similarly, many of the blackbuck core habitats lack sufficient connectivity owing to the large non-habitat spaces between them. Therefore, some species are more prone to geographic isolation and population decline compared to others within the same landscape [26,27]. Management interventions must therefore focus on multiple species based on their traits that influence dispersal and geographical distributions. The maps show that most of the core habitats in the western part of the landscape are connected, while the core habitats in the eastern part are largely disconnected. Panna tiger reserve and Ranipur wildlife sanctuary are fairly connected for all antelope species. Similarly, Nauradehi wildlife sanctuary and Veerangana Durgavati wildlife sanctuary are sufficiently connected for all antelope species. Panna tiger reserve and Nauradehi sanctuary appear connected for nilgai, blackbuck, and chinkara, while they are disconnected for the four-horned antelope. All other major protected areas, namely Bandhavgarh tiger reserve, Sanjay Dubri tiger reserve, Guru Ghasidas sanctuary, and Kaimur wildlife sanctuary are disconnected from each other for all focal species.

A comparison of our study with the previous studies on dispersal of large carnivores in this landscape [29,31,32] suggests that the habitat network needed for small ungulates is more fragile than for top carnivores, as their core habitats must be located at close proximity. The removal of one or more nodal habitats can disrupt multiple connections. Furthermore, loss of habitats outside protected areas also leads to significant increase in habitat isolation. This is an important finding for conservation action as the population densities of some small ungulate species are already low.

Our study is also the first to model and map the areas where the antelopes are at a greater risk of dying due to anthropogenic influence. The transport network and human settlements, located within or adjacent to forest areas, emerge as sites where anthropogenic antelope mortality may be high. This pattern is also known for other wild herbivores in India and in other countries [17,18,93]. These unsafe areas act as potential barriers to effective dispersal. Most high-risk sites are present in the periphery of the protected areas and in the multiple-use forests. These areas may become death traps for the antelopes, either by poaching or accidental death. We find that most of the high-risk areas fall within potential corridor regions and may pose an elevated mortality risk for dispersing animals. This may explain why some of the antelope species are not actually found in many of the habitats that are suitable for the species. We suggest that the notion of mortality risk must be included in corridor studies so that more comprehensive habitat management policies can be formulated.

The distribution of the suitable habitats and the core habitats suggest that vast stretches of land outside protected areas act as primary habitats for the antelopes. The multiple-use forests provide crucial residential habitats that act as central nodes in the greater habitat network, connecting several protected areas. It is known that intrepid species like blackbuck and nilgai also make extensive use of the non-forest lands (like agricultural fields) for dispersal [49,50]. Wildlife management must therefore account for human activities in the multiple-use forests to minimize any disturbance to wild animals if dispersal is to be maintained.

Although we have not directly measured dispersal or the use of corridors by these four species, our findings show that a singular focus on protected areas misses or ignores a larger opportunity for more robust conservation practices. While strict protection or exclusionary measures of protection can neither be applied nor are desirable for habitats outside protected areas, large areas of suitable sites for wildlife and biodiversity conservation can be harnessed outside traditional protected areas to create resilient metapopulations. The genetic and demographic benefits of conserving species in metapopulations are well-known [6,94], but practical implementation will require careful analyses and planning for each site to create workable solutions. We have presented detailed analyses and high-resolution maps for an important conservation region in central India, which can be directly used by managers in planning conservation in the landscape. Moreover, future developmental activities in this region can use these maps to integrate conservation imperatives while planning agriculture, infrastructure, and urbanization.

Future studies in this field should focus on informing site-specific management interventions. Such issues can primarily be addressed using geoinformatics, where multiple imperatives can be integrated to bear in clear question in conservation and management planning. This will require a combination of high-resolution satellite data to detect large-scale changes in habitat and landcover, ground-based monitoring to detect and quantify fine-scale changes, and geospatial information on land use attributes, climate, and environment, all pulled together into a comprehensive geospatial information system. There is no substitute for quality ground monitoring of species distributions and abundance, which is needed to improve the strength and reliability of large-scale geospatial analyses. In human-dominated areas, other metrics such as habitat fragmentation and degradation, human-wildlife conflict, and dependence on forest resources should be systematically monitored and made available in public domain. The rapidly developing tools in geoinformatics can then be harnessed to address issues in conservation and management of ecologically rich landscapes.

The antelope core habitats and dispersal corridors mostly lie in the multiple-use forests, but sizeable expanses are also found in non-forest areas, mainly comprising privately owned farmlands and plantations. Since the multiple-use forests and non-forest areas are not primarily administered for wildlife conservation, these areas deserve careful attention. Our maps show that most of these forest patches lie adjacent to human settlements, and people use these areas as livestock grazing grounds and to collect minor forest produce. Evolving best practices to minimize human-wildlife interactions in these multiple-use areas would be the key for sustainable management. Some core habitats outside protected areas that are critical for the maintenance of connectivity for multiple species may warrant higher levels of protection. Specific land-use policies must be formulated to minimize human disturbance in the areas identified as antelope core habitats and corridors.

7. Conclusions

Wildlife policymakers and land-use planners must view multiple-use landscapes as unified conservation units. Metapopulations of small ungulates require core habitat patches at proximity as they do not have long-distance dispersal ability. Management interventions are needed to conserve these habitats from destruction and degradation, particularly when they fall outside the protected areas. Specific mitigation measures are needed to prevent poaching and accidental deaths in high-risk areas. It is vital to increase surveillance and reduce disturbance in the multiple-use forests containing the core habitats and dispersal corridors. Providing a safe passage for dispersal will be useful not only for the ungulates but also for the predators. We provide a widely applicable framework than can be widely used in conservation planning.