For anyone who has been forced to evacuate during a wildfire, the experience of not knowing how fast the fire is moving or which direction it will take is terrifying. That unpredictability is exactly what a team of researchers at USC Viterbi's Department of Aerospace and Mechanical Engineering is working to eliminate.
The new model, published in the journal Remote Sensing, is an advancement of earlier work from the same team. The original system relied solely on VIIRS, a polar-orbiting satellite that detects heat signatures with high spatial resolution but only passes over the same location twice a day. That creates precise but intermittent snapshots — gaps of many hours in which a fire can change dramatically and unpredictably.
The updated model closes that gap by adding data from GOES, a geostationary satellite that monitors the same region continuously and refreshes every five minutes. GOES doesn't match VIIRS for spatial detail, but it provides something the previous model lacked: a reliable estimate of exactly when a fire ignited. That single piece of information turns out to be critical. A fire's current footprint could reflect two hours or ten hours of growth, and knowing which tells responders how rapidly the fire is spreading — information that directly shapes the decisions they make.
The model also factors in the physical landscape. It accounts for how slope and elevation affect fire behavior — fires are well known to move faster uphill — and it is trained not on generic scenarios but on simulations of actual historical wildfires, capturing the real variability of weather, vegetation, and topography that governs how fires behave in practice.
The result is a tool that can reconstruct a fire's progression with high accuracy. Reconstructions closely match the high-resolution infrared perimeters that aircraft have historically been used to measure — a meaningful benchmark for how useful the model would be in an operational setting.
Access the complete research paper published in Remote Sensing here.
