In the 2024 I-GUIDE Spatial AI Challenge, our team built a lightweight U-Net model that maps wildfire threats along roads and railways — infrastructure often left out of wildfire risk modeling. We combined CAL FIRE threat zones, Sentinel-2 imagery, and transportation GIS data, to create a fast, spatially aware model that generates pixel-level wildfire risk maps, ready for real world use in areas like planning and disaster response.

This VCO spotlights how spatial AI can move wildfire risk modeling beyond forests and homes to focus on critical infrastructure. We’ll walk through our labeling workflow, model development, and GIS integration to show how this approach enables scalable, real-world decision support.