Speaker: Lily Xu (IEOR)

Title: Quantifying the causal impact of ranger patrol deterrence on wildlife poaching using machine learning

Abstract: Rangers serve as a primary defense for wildlife conservation, patrolling 21.7 million km2 of protected areas globally to protect wildlife from poaching and other threats. The high cost of patrols is justified by the assumption that rangers deter poachers from returning in the future. While anecdotal knowledge and correlational studies support patrol-induced deterrence, there is currently no causal evidence of deterrence nor a quantification of the magnitude of its effects. Causal inference is especially difficult in this setting because poaching events are only partially observable, and the likelihood of observing poaching depends on patrol effort, thus entangling the treatment and outcome variables.
Using 7 years of ranger patrol data from Murchison Falls Conservation Area in Uganda, we present the first causal evidence of ranger-patrol deterrence. To enable this causal analysis, we leverage a past field evaluation of a machine learning tool for predicting poaching as a source of quasi-experimental variation, used machine learning to produce temporally and spatially granular estimates of site characteristics for matching, and designed a Bayesian inference model to impute true poaching outcomes, which were partially unobserved. We found that increasing patrol effort by 1 kilometer in a 1 km2 area reduced poaching probability in the next month by 45.0%. We further quantified that deterrence is most effective in areas that are more accessible to poachers and that optimizing patrol allocation can increase deterrence by up to 46.6%, creating possible cost savings of USD$351,494 annually in our study site alone.