Constructing a Bayesian Spatial Presence-Absence Model for Animals in the Serengeti National Park

Abstract

Ecologists have been spearheading the Snapshot Serengeti project in Tanzania's Serengeti National Park for the past few years, a large camera trap project with the intent of discovering more about how animals interact with their environment. We fit a hierarchical logistic model to predict the presence or absence of a particular species in a specific area using environmental covariates. Additionally, we utilized centered spatially dependent terms; a term accounting for species dependence from neighboring sites, and a term accounting for cross-species dependence from neighboring sites. Our model also includes a latent variable for detectability; the model accounts for the fact that we are working with imperfectly observed data by including a latent variable for detectability. We set out to investigate a few questions. First, what is the relationship between body size and attraction to high NDVI levels? Second, are herbivores more likely to avoid areas where they have recently seen lions or where lions have historically visited? Third, what are the different cross-species dependence levels? Our model has currently only been applied to pairs of species simultaneously, but something that makes it different is the fact that, aside from computational difficulty, nothing prevents it from running as many species as we want at once and seeing how they interact with each other.