Advancing environmental resilience to wildfires: A high-performance stochastic simulation approach for real-time risk...

Effective wildfire management requires rapid and accurate spread predictions to support real-time decision-making. This study presents a high-performance stochastic simulation framework based on Cellular Automata (CA), specifically optimized for "urgent computing" through Python's Numba Just-In-Time (JIT) compilation and thread-level parallelism. The system integrates multivariate data fusion, harmonizing multi-resolution geospatial datasets (30-100 m)-including Digital Elevation Models (DEM), Corine Land Cover (CLC) fuel maps, and multi-temporal vegetation indices-with a dedicated firebrand spotting module. Computational benchmarking on Intel Xeon Gold 6526Y architecture demonstrates that the framework executes an ensemble of 1000 stochastic iterations in approximately 22 s for medium-scale events (<5000 ha), while scaling to 27 min for high-complexity, multi-day scenarios. This performance range ensures that statistically robust burn probability maps are delivered within critical operational windows relative to the fire's rate of spread. Following initial calibration on the Spetses (1990) event, the framework was independently validated against four high-impact Mediterranean wildfires (Parnitha, 2007; Rhodes, 2008; Tornos, 2022; and Córdoba, 2024), exhibiting high spatial concordance (Sørensen-Dice index up to 81.98%) with area deviations ranging from <1% to -9.35%. Statistical convergence analysis confirmed that the 1000-iteration standard stabilizes the mean burned area with a Relative Standard Error (RSE) below 1.5%. By balancing predictive fidelity with computational agility, this CPU-based framework offers a cost-effective solution for integration into Decision Support Systems (DSS). The tool empowers environmental managers to shift the paradigm from crisis-driven response to pre-emptive tactical planning, reinforcing landscape-scale resilience in alignment with the goals of international frameworks such as FIREPOCTEP+.