Conference Agenda

Wildfires in central Chile have intensified in frequency and impact, yet operational models often remain limited to correlational predictors, hindering robust interpretation and decision support. This study aims to discover and validate the causal structure underlying wildfire occurrence in the Maule Region, Chile, identifying the most plausible drivers and their directed dependencies. We compile a multi-source dataset integrating wildfire records with meteorological, vegetation, topographic, and anthropogenic variables, harmonized to a common spatiotemporal grid. Causal discovery is performed using constraint-based and score-based structure learning under a linear non-Gaussian assumption, complemented with stability selection and sensitivity analyses to assess robustness against sampling variability and confounding. The inferred graphs consistently recover a sparse, stable causal backbone in which short-term meteorological conditions and fuel-related proxies act as primary upstream drivers, while human-accessibility variables exhibit mediated effects through ignition likelihood. Across multiple resampling runs, key directed edges remain stable, and the resulting causal model improves out-of-sample interpretability and risk attribution relative to purely associative baselines. This causal framework provides actionable insight into drivers likely governing wildfire dynamics in Maule, enabling more defensible early-warning indicators, targeted prevention policies, and a principled basis for scenario analysis through interventions on controllable factors.