Methodological approaches for estimating populations of the endangered dhole Cuon alpinus

Abstract

Large carnivores are important for maintaining ecosystem integrity and attract much research and conservation interest. For most carnivore species, estimating population density or abundance is challenging because they do not have unique markings for individual identification. This hinders status assessments for many threatened species, and calls for testing new methodological approaches. We examined past efforts to assess the population status of the endangered dhole (Cuon alpinus), and explored the application of a suite of recently developed models for estimating their populations using camera-trap data from India's Western Ghats. We compared the performance of Site-Based Abundance (SBA), Space-to-Event (STE), and Time-to-Event (TTE) models against current knowledge of their population size in the area. We also applied two of these models (TTE and STE) to the co-occurring leopard (Panthera pardus), for which density estimates were available from Spatially Explicit Capture-Recapture (SECR) models, so as to simultaneously validate the accuracy of estimates for one marked and one unmarked species. Our review of literature (n = 38) showed that most assessments of dhole populations involved crude indices (relative abundance index; RAI) or estimates of occupancy and area of suitable habitat; very few studies attempted to estimate populations. Based on empirical data from our field surveys, the TTE and SBA models overestimated dhole population size beyond ecologically plausible limits, but the STE model produced reliable estimates for both the species. Our findings suggest that it is difficult to estimate population sizes of unmarked species when model assumptions are not fully met and data are sparse, which are commonplace for most ecological surveys in the tropics. Based on our assessment, we propose that practitioners who have access to photo-encounter data on dholes across Asia test old and new analytical approaches to increase the overall knowledge-base on the species, and contribute towards conservation monitoring of this endangered carnivore.

Keywords: Camera trapping; Canids; Conservation monitoring; Model evaluation; Site-based abundance; Space-to-event; Time-to-event; Unmarked species.

Figure 1. Spatial locations of studies pertaining to population assessments of the dhole across the species’ distribution range.

Spatial locations of studies pertaining to population assessments of the dhole across the species’ distribution range (n = 38). The sizes of circles reflect the relative sizes of the study areas. All three panels include all study areas, but the point locations have been separated out based on the methodology used (RAI, occupancy/distribution, site-based abundance, or spatial capture–recapture) for ease of interpretation.

Figure 2. Study area map showing Radhanagari Wildlife Sanctuary.

Study area map showing Radhanagari Wildlife Sanctuary (351 km²), with the surrounding forest–non-forest habitat matrix, large water reservoirs, locations of camera traps and villages. Camera trap surveys were conducted in April–May 2019. Inset shows location of the study area within the larger Western Ghats landscape in India.

Figure 3. Estimates of density and abundance of leopards in Radhanagari Wildlife Sanctuary.

(A) Estimates of density (left) and abundance (right) of leopards in Radhanagari Wildlife Sanctuary (April–May 2019), based on Spatially Explicit Capture–Recapture models. Density calculated as individuals per 100 km². Error bars for the likelihood-based estimates indicate Standard Errors, and those for Bayesian estimates are Standard Deviation values. (B) Estimates of density (left) and abundance (right) of leopards in Radhanagari Wildlife Sanctuary (April–May 2019), based on space-to-event models (Moeller et al. 2011). Density calculated as individuals per 100 km². L, L+10% and L+20% are three scenarios, where viewshed areas were calculated as ‘largest’, ‘largest+10%’ and ‘largest+20%’. Error bars for density are 95% confidence intervals.

Figure 4. Estimates of density and abundance of dholes based on space-to-event ‘STE’ and Site-based Abundance (SBA) models in Radhanagari Wildlife Sanctuary.

(A) Estimates of density (left) and abundance (right) of dholes in Radhanagari Wildlife Sanctuary (April–May 2019), based on space-to-event ‘STE’ models. (B) Estimates of density (left) and abundance (right) of dholes in Radhanagari Wildlife Sanctuary (April–May 2019), based on site-based abundance models (beta-binomal/Poisson mixture models; Martin et al., 2011). Density was calculated as individuals per 100 km². up–uninformed priors, pp–partially informed priors, cp–constrained priors. Error bars for density are 95%.