ANNOUNCEMENTS
Particularly in fire-prone areas like northern California, wildfires are becoming a greater hazard to human infrastructure, biodiversity, and ecosystems. Using multi-source geospatial data and machine learning, this work attempts to categorize fuel types, estimate aboveground biomass (AGB), and create a fire susceptibility model for the Yreka region. The method combines topographic, meteorological, and human-made variables with remote sensing data from MODIS, Landsat, Sentinel-2 and GEDI in the GEE platform. Using geographically stratified fire and non-fire samples labelled with FIRMS burned area data (2014–2023), a RF classifier was trained to create a 2025 fire susceptibility map. While AGB was calculated using a regression model trained on GEDI biomass data and Landsat-derived indices, the fuel type classification identified the primary vegetation categories as grass, shrub, and non-fuel. Five types of fire danger zones were identified in the resulting fire susceptibility map: very low, low, moderate, high, and very high. Strong predictive performance was demonstrated by the model validation, which produced an 88.5% fire classification accuracy and a Kappa of 0.77. The most important factors influencing fire risk, according to feature importance analysis, were population density, LST, distance to burned area, and AGB. Yreka is a perfect place for susceptibility modeling because of its diverse terrain, sensitivity to climate change, and wildfire history. The findings show how well machine learning and remote sensing work in dynamic fire risk assessment and provide a useful tool for early fire detection, land use planning, and wildfire prevention.
Keywords: Fire susceptibility, Machine learning, Remote sensing, Biomass estimation, Google Earth Engine.
