A Multi-Species Occupancy Modelling Approach to Access the Impacts of Ecological Covariates on Terrestrial Vertebrates in a Tropical Hotspot in Central, Cameroon

Abstract

Mammalian communities living in tropical forests, particularly those in ecological transition zones, are under constant threat from human activities. In many regions, baseline data on mammal richness, occupancy, detection probability and the environmental factors that influence these metrics are lacking. As a key metric for guiding conservation decisions, species richness can be underestimated due to varying detection probabilities, leading to species being overlooked. Advances in technology and methodology have revolutionised wildlife monitoring, fostering the increase of multi-species occupancy models (MSOMs) for efficient studies of community, shifting focus from single species to entire communities. MSOMs, hierarchical models that share information across species via random effects, address imperfect detection to provide accurate and unbiased species richness estimates. To fill this information gap, we used cameras trap data from Mpem and Djim National Park, Central Cameroun. We used generalised linear models and a model selection approach to evaluate factors affecting species detection events. Similarly, we used MSOMs within a Bayesian hierarchical framework to evaluate our initial species richness estimate at each camera trap location and to understand the influence of environmental covariates on the occupancy and detection probability of 19 vertebrates recorded in the area to inform management decisions for these species. From 915 independent photographic events obtained over 1700 days of capture, the study results highlight the importance of habitat, distance to river, normalised difference vegetation index and topographic position index (TPI) in explaining patterns of detection events. We found that forest (GLM: B 0.31, IRR 1.36, CI 0.14–0.48, p < 0.001), NDVI (GLM: B 0.31, IRR 1.36, CI 0.21–0.41, p < 0.001) and TPI (GLM: B 0.17, IRR 1.19, CI 0.08–0.26, p < 0.001) were positively associated with species detection events, whereas distance to river (GLM: B −0,19, IRR 0.83, CI 0.0.27 to (−0.11), p < 0.001) was negatively associated with species detection events. However, the mean probability of community occupancy was 0.33 ± 0.10 [2.5%–97.5% CI: 0.17, 0.54], while the mean probability of community detection was 0.07 ± 0.02 [2.5%–95% CI: 0.04, 0.12], indicating that, on average, approximately 33% of the sites are likely to be occupied by the community of interest, with a 7% probability of detection at occupied sites. After accounting for imperfect detection, the maximum occupancy and detection probability estimated from the MSOMs were 0.88 ± 0.07 (2.5%–97.5% CI: 0.71–0.98) and 0.22 ± 0.2 (2.5%–97.5% CI: 0.18–0.27) for Philantomba monticola, respectively. Globally, the community responses were close to zero and relatively weak, probably due to mixed responses at the species level. Despite their weak effect, distance to road (β: −1.53 ± 1.97 [2.5%–97.5% CI: −1.72–5.62]) and NDVI (β: −0.09 ± 0.22 [2.5%–97.5% CI: −0.50–0.38]) had a negative significant effect on occupancy. However, there were significant responses at the species level with Cephalophus nigrifrons, for example, exhibiting a strong response to NDVI. This study contributes to baseline information on the ecology of mammal communities in Central Cameroon and supports the need for future multi-season surveys to understand the influence of temporal factors on community occupancy and richness in the area.