A new computer model produces a dynamic wildfire risk map,
starting with the state of California
Washington, D.C., December 9, 2025 – Wildfires pose a significant threat across the southwestern United States, due to the region's unique topography and weather conditions. Accurately identifying locations at the highest risk of a severe wildfire is critical for implementing preventive measures.
With this goal in mind, scientists from the University at Buffalo have developed GeoFlame VISION, a proposed computer model that uses AI and satellite imagery to produce a dynamic wildfire risk map at a granular spatiotemporal scale for the entire United States. The authors will present a case study of California using their model on Dec. 9 at the Society for Risk Analysis Annual Meeting in Washington, D.C.
"This novel approach of integrating remote sensing data with machine learning and AI will not only help with efficient wildfire mitigation, but also aid in decision-making related to land management and the controlled expansion of Wildland-Urban-Interface (WUI) regions – which in turn can lower the risk of wildfire-induced damages to the critical infrastructures and WUI communities in the future", says Sayanti Mukherjee, assistant professor of industrial and systems engineering at the University at Buffalo and corresponding author of the study.
Preliminary findings from the dynamic wildfire risk map of California:
The eastern, southwestern, and northwestern parts of California are significantly more vulnerable to wildfires than other regions. This is mainly because the impact of the dry, warm Santa Ana winds is particularly pronounced in California's southwestern and northern regions, increasing the likelihood of wildfires there.
According to the map, the California counties of Mono, Inyo, Mariposa, Ventura, and Tulare are at the highest risk of wildfire. The 2022 Airport Fire in Inyo and the 2017 Thomas Fire in Ventura reinforce the credibility and accuracy of this finding.
The model shows that Los Angeles and Mono counties are in wildfire hotspots due to their dry climates, unique vegetation, and prevailing wind directions.
The key predictors of wildfire risk are climate, topographical, and vegetation factors.
Wind speed plays a significant role in the occurrence of a severe wildfire, followed by the normalized difference vegetation index (NDVI) and precipitation.
"The wildfire risk in a region not only depends on the topographical, landcover, and weather-related variables, but also on the built environment, such as the buildings and power grid infrastructure, which is often overlooked in the traditional physics-based wildfire spread models," says Poulomee Roy, lead author of the study and a doctoral candidate at the University at Buffalo. "Thus, the interactions among all these factors -- which we include in our study -- are instrumental in modeling the dynamic wildfire risk."
To create the map, the researchers used satellite imagery data from NASA's Moderate Resolution Imaging Spectroradiometer (MODIS) to extract information on historical burned areas from 2015 to 2022 at a weekly time scale (with small areas marked burned or unburned). This dataset was integrated with information on variables such as topography, elevation, climate, vegetation, windspeed, and the locations of critical infrastructure, including residential buildings and power stations. Using advanced vision-based AI and other technologies, a pixel-based predictive analysis of the wildfire-burnt areas was then performed. The map is about 92% accurate at predicting dynamic wildfire risk at a granular spatiotemporal scale (based on real data from past wildfires).
