Weather overrides fuel memory: Seasonal dynamics of fire hazard and the limits of the fire return interval.

Abstract

Wildfire regimes in Mediterranean hotspots like Portugal are driven by anthropogenic factors and fuel build-up, causing severe socio-economic impacts. This study hypothesizes: 1) that seasonally-disaggregated models outperform an overall annual model by revealing shifts between weather- and fuel-driven regimes; and 2) that incorporating fuel memory, via the Fire Return Interval (FRI), enhances predictive accuracy across all weather conditions. Three fire seasons (High: Aug-Sep; Medium: Jul; Low: Oct-Jun) were statistically defined. Hazard maps were generated using a data-driven Spatial Multi-Criteria Evaluation (SMCE) and Simple Additive Weighting (SAW) approach. The FRI was calculated as the 10th percentile (P10) of re-burn intervals (1990-2024) per land cover class and geoclimatic zone; a temporal mask based on this threshold was applied to assign zero hazard to recently burned areas. All models were validated against independent 2019-2024 fire perimeters using the Area Under the Curve (AUC) method. The first hypothesis was partially validated: the seasonal High (AUC = 0.77) and Low (AUC = 0.81) models outperformed the Overall (AUC = 0.73). We found dominant hazard drivers shifted seasonally: the July model (AUC = 0.58) was linked to drought (low number of days per year with precipitation >10 mm, hereafter NDP10), whereas the High, Low, and Overall models were driven by topography, highly flammable land cover (shrublands) and high fuel load (proxied by high NDP10). However, the second hypothesis was refuted for the critical period, as the FRI mask decreased predictive performance in the High-season model. The findings indicate that while seasonality is essential for differentiating hazard drivers, this "fuel memory" is overridden by extreme weather conditions. Management strategies based solely on historical burn patterns are, therefore, insufficient to mitigate hazard in extreme events.