Sampling extreme wildfire events from LPJmL-SPITFIRE large ensemble simulations

Extreme fire weather conditions are becoming increasingly unprecedented worldwide, yet the full range of potentially high-impact extreme wildfires remains difficult to assess. Here we generate a large ensemble of wildfire simulations by forcing the process-based model LPJmL-SPITFIRE with a 40-member bias-adjusted and statistically downscaled climate model (ACCESS-ESM1-5). This enables robust sampling of extreme wildfire events and allows comparison against single realizations (using forcing from climate reanalysis GSWP3-W5E and from an individual climate model ensemble member, r1i1p1f1). We show that wildfire ensemble maxima typically exceed single realizations maxima, suggesting that using a single climate forcing misses a substantial portion of the plausible extreme wildfire events due to internal climate variability. Extreme fire impacts (carbon emissions and burned area) respond more strongly to internal climate variability than fire weather conditions, suggesting a strong vegetation-fire feedback sensitivity to the climate forcing. Additionally, the large ensemble simulations capture climate driver-fire relationships not captured by single realizations, where maximum impacts occur without maximum fire danger, and vice-versa, highlighting the critical role of other factors beyond weather conditions that contribute to whether fires become extreme. These findings demonstrate that modelling a large range of possible wildfire events using the full distribution of climate realizations can help identify the mechanisms leading to the most extreme events.