Safeguarding biodiversity through predictive modelling of illegal wildlife poisoning hotspots

Illegal wildlife poisoning is a high-impact but cryptic wildlife crime, requiring proactive, evidence-based targeting of limited enforcement resources. We developed a patrol-oriented spatiotemporal risk framework for illegal poisoning across Greece using poisoning incidents recorded during 2014-2024. Incidents were aggregated to a 5-km grid and modelled as monthly counts using negative binomial models with covariates representing accessibility, conflict and motivation, governance context, and environmental conditions. To enhance operational relevance, we further modelled incident severity (affected animals per incident) and baiting complexity (distributed baiting) and produced severity-weighted expected impact surfaces. We analysed 915 incidents (1620 affected animals; 2295 poison baits) and found that predicted risk was spatially concentrated, with markedly higher baseline risk in Crete-Dodecanese region and additional high-priority areas across the mainland. Monthly risk increased with road density and livestock depredation losses, and decreased with Natura 2000 presence, while being higher closer to Natura boundaries. Risk exhibited clear seasonality with peaks in early spring and autumn. Models based on incidents investigated by specially trained dogs revealed substantially higher bait recovery and distributed baiting than reports not investigated by dog units. Severity and distributed-baiting models further distinguished frequency-based priority areas from impact-based priorities, and severity-weighted maps re-ranked some areas relative to incident probability alone. Our framework provides evidence-based layers for seasonal patrol planning, uncertainty-aware prioritization, and impact-oriented prevention.