AI-Based Poaching Risk Prediction

Abstract

Wildlife poaching causes an estimated 7,000-28,000 elephant deaths, 1,000-3,000 rhinoceros deaths, and 100 million shark kills annually, driving multiple species toward extinction, yet anti-poaching patrol resources are severely limited relative to the vast protected areas they must monitor. This study developed, trained, and validated an AI-based poaching risk prediction system (PAWS-AI v2.0) integrating random forest, gradient boosting, and spatiotemporal neural network models across 42 protected areas in Sub-Saharan Africa and Southeast Asia (2018-2024), using 2,840 confirmed poaching incident records, 28 environmental and socioeconomic predictor variables (terrain, vegetation, distance to roads/settlements, rainfall, moon phase, patrol history, local poverty index), and real-time satellite imagery (Sentinel-2; Planet Labs). The ensemble model achieved AUC = 0.884 +/- 0.024 for predicting poaching incident location (grid cell level; 1 km2 resolution) and AUC = 0.864 +/- 0.028 for predicting incident timing (daily, 30-day horizon) in 10-fold cross-validation. Prospective validation -- comparing PAWS-AI patrol recommendations against ranger judgement over 12 months at 18 study sites -- showed 2.84-fold higher poaching interception rate for AI-directed patrols (42.4 +/- 8.4% interception) vs. ranger-directed patrols (14.8 +/- 4.8% interception; p < 0.001). Feature importance analysis identified distance to nearest settlement (22.4%), previous patrol gap duration (18.4%), terrain ruggedness (14.4%), and moon illumination (12.4%) as the four strongest predictors. PAWS-AI predictions were communicable to field rangers through a simple smartphone interface providing daily 'heat maps' of predicted risk zones. These results demonstrate that AI-based patrol optimisation can substantially improve anti-poaching effectiveness without increasing ranger deployment, providing a scalable technology solution for wildlife law enforcement under the International Consortium on Combating Wildlife Crime (ICCWC) framework.