Sensitivity of ELMFIRE to real-world input datasets for WUI fire modeling

Wildland–urban interface (WUI) fires pose significant threats to communities, and computational models are critical for their mitigation. These models depend strongly on input data, yet the impact of real-world input variability on operational models remains unquantified. To address this gap, we integrated an urban fire spread model into ELMFIRE and simulated three California WUI fires (Tubbs, Thomas, and Camp) to assess sensitivity to commonly used inputs, including wind, fuel moisture content (FMC), structures, and roadways. Real-world wind data varied by up to 40%, resulted in differences of over 50% in simulation accuracy, while downscaling had minor effects ( < 20%). FMC was equally influential, with multi-stage processing increasing uncertainty and reducing the accuracy of burned area (40%) and structure damage (70%). Simplified structure representations minimally affected burned area ( < 15%) but reduced structure damage accuracy ( > 40%), while misrepresented road firebreaks could significantly reduce accuracy (60%). Overall, wind and FMC effects are dominant; wind direction and speed control directionality and extent, while FMC governs vegetation ignitability. Some inputs can offset inaccuracies through compensation effects, while others (e.g., wind direction) have unique, non-compensable effects. Despite limitations such as variability in structure properties, this study provides practical guidance for selecting input data toward standardized operational WUI fire modeling. • Real-world wind data differences can reach up to 40% and cause simulation discrepancies over 50%. • Downscaling wind data had minor effects, altering results by less than 20%. • Multi-stage FMC processing increased uncertainty, reducing accuracy by up to 70%. • Misrepresenting types of road considered as firebreak could lower accuracy by up to 60%. • Some variables like wind direction capture key fire behavior; others can have compensation effects.