ANNOUNCEMENTS
Wildfires continue to pose a significant threat to infrastructure, biodiversity, and ecosystems across fire-prone landscapes such as Santa Barbara and Ventura Counties in Southern California. This study integrates multi-source geospatial data and machine learning to classify fuel types, estimate aboveground biomass (AGB), and develop a fire susceptibility model tailored to this region. Leveraging the capabilities of Google Earth Engine (GEE), the approach synthesizes topographic, climatic, and anthropogenic variables with remote sensing inputs from MODIS, Landsat, Sentinel-2, and GEDI. A Random Forest classifier was trained using spatially stratified fire and non-fire samples, labelled with FIRMS burned area data from 2014 to 2023, to generate a fire susceptibility map for the year 2025. Fuel classification delineated vegetation into grass, shrub, and non-fuel types, while AGB was estimated through a regression model combining GEDI biomass observations and Landsat-derived vegetation indices. The final fire susceptibility output categorizes the landscape into five risk zones: very low, low, moderate, high, and very high. The model achieved an overall accuracy of 93%, a macro-average F1 score of 0.85, and a ROC AUC of 0.956, indicating robust predictive performance. Feature importance analysis identified mean temperature, proximity to built-up areas, and terrain complexity as key drivers of fire risk.
To support decision-making and communication, an interactive dashboard was developed. This dashboard includes a story map section and a three-panel interface: the first panel allows for location selection, the second displays fuel classification and forest biomass layers, and the third features supporting environmental and anthropogenic layers. For each layer, raster statistics are dynamically displayed, providing users with on-the-fly spatial summaries. An integrated AI chatbot offers real-time descriptions of selected layers, including legends, color palettes, and contextual specifications. This platform enhances accessibility, transparency, and utility of the wildfire susceptibility model, making it a valuable tool for proactive fire management, early-warning systems, and landscape-level planning.
