Identifying bark beetle damage predictors in Austria using interpretable machine learning: Thresholds and management implications

The increasingly severe and frequent forest damage caused by the European spruce bark beetle ( Ips typographus ) has been systematically surveyed across Europe since the early 2000s, yet underlying drivers remain not fully understood. This study aims to identify key predictors and capture spatiotemporal dynamics of bark beetle damage by analysing annual damage time series from 70 Austrian forest districts (2005–2025) using extreme gradient boosting. We quantified the importance of 18 predictors reflecting preceding forest damage, climate, forest composition and topography. The model was evaluated at the national level and for eight regional clusters. Across Austria, it explained 62% of the variance (R 2 ) in cross-validation and identified 64% of true bark beetle outbreaks. Performance was highest in damage hotspots and lower in less affected regions. Preceding bark beetle damage, outbreaks in neighbouring districts and summer climatic water balance were among the strongest predictors. SHAP (SHapley Additive exPlanations) values revealed clear thresholds at which damage increased: negative summer climatic water balance (< 0 mm), warm spring temperatures (≥ 14.5 °C), and elevated preceding damage by abiotic events (≥ 2 m 3 ha −1 ) and I. typographus (≥ 1 m 3 ha −1 ). Exceeding these thresholds likely amplifies bark beetle damage in previously less affected regions. We conclude that machine learning-based modelling effectively captures the spatiotemporal dynamics and underlying drivers of bark beetle damage at forest district level. Our findings can advance threshold-based risk assessment systems from national to European levels, supporting bark beetle management and the preservation of ecosystem services under a changing climate.