Identifying areas of high snaring risk in Kruger National Park: A novel citizen science approach for carnivore conservation
Description
We conducted an online search on the social media platforms Facebook and X, to identify posts documenting sightings of snare incidents involving the African large carnivore guild – lion, spotted hyaena (Crocuta crocuta), African wild dog (Lycaon pictus), cheetah (Acinonyx jubatus), and leopard (Panthera pardus) – in KNP from 2013 to 2023. Both platforms (Facebook and X) were found to be primarily used by local residents, tourists, and tourist guides for sharing detailed sightings of snared carnivores throughout the survey period. Occasionally, park officials would also use these social media platforms (e.g., ‘Latest Sightings’, a real-time sightings platform on Facebook and X) to share sightings and locations of snared carnivores that they required assistance in relocating for de-snaring operations. For data collection on X, we used combinations of keywords such as “Kruger”, “KNP”, “wire-snare”, “snare”, “snares”, “snared” and “snaring” alongside the names of the respective species and the study period to standardise our search. For data collection on Facebook, we followed the methodology described by Chowdhury et al., (2024) by first identifying relevant Facebook groups used for sharing KNP wildlife sightings. Subsequently, we extracted data using the keywords “wire-snare”, “snare”, “snares”, “snared” and “snaring”, including species names and the study period, using the Facebook search function within each group. Only posts containing exact locations (coordinates or detailed description e.g., at a specific road junction) were included in further analyses. Repeated sightings of the same individuals (identified using pelage patterns, distinctive individual characteristics and/or the location of the snare wound) and sightings of healed/treated injuries were excluded to the best of our ability. Each snaring event was assigned to the first recorded sighting in the dataset. If the same individual was observed at different locations, the snaring event was attributed to the location where it was first documented. For each recorded incident, we noted the date of the sighting, species, location, snare placement on the animal’s body, source of the sighting and whether the post included photos of the snare or snare injury. In adherence to privacy and ethical considerations, no personal information was recorded and we have removed hyperlinks to the sightings from the shared dataset to protect user privacy. Subsequently, each recorded snare incident was georeferenced in QGIS software (QGIS Development Team, 2023) using the location description from the social media post to obtain coordinates for running the subsequent occupancy model.
Files
Steps to reproduce
We implemented our Bayesian single-species occupancy model using Pólya-gamma data augmentation for computational efficiency using the package ‘spOccupancy’ (Doser et al., 2022). We used the default vague prior values on occurrence and detection logit-scale parameters with a mean of zero and variance of 2.72 (Doser et al., 2022). To compute posterior distributions of the parameters, we ran three Markov chain Monte Carlo (MCMC) chains, each with 50,000 iterations, a burn-in of 5,000 samples, and a thinning rate of 20. We confirmed chain convergence with the Gelman-Rubin statistic (R-hat ≤ 1.1; Gelman and Rubin, 1992) and visual inspection of trace-plots. We ensured that the minimum effective sample sizes for all parameters were > 500. To check that our model fit the data adequately, we conducted posterior predictive checks and estimated Bayesian p-values for both sites and detection histories using the Freeman-Tukey statistic, with Bayesian p-values close to 0.5 indicating adequate model fit and values <0.1 or >0.9 suggesting poor fit (Hobbs and Hooten, 2015). We calculated the posterior mean, standard deviation, and 95% Bayesian credible intervals (BCI) for all coefficients and considered parameters to have strong support when 95% BCIs did not overlap with zero. To provide evidence for the direction of a given parameter estimate, we calculated the proportion of the posterior distribution that was greater than or less than zero, given the direction of the effect.
Institutions
- Nelson Mandela University