USC researchers are coupling VIIRS and GOES data with physical simulations to move from reactive observation to fire front prediction.
The CD3 group led by Professor Assad Oberai at the USC Viterbi School of Engineering, together with PhD student Bryan Shaddy and collaborators from several universities, has developed a tool that reconstructs and anticipates the progression of a forest fire. The approach combines generative AI, satellite imagery and simulations based on fire physics, trained on real fires.
The advanced version of the model, described in an article published in April 2026 in the journal Remote Sensing, aggregates several complementary sources: the polar-orbiting VIIRS satellite for its fine spatial resolution and its two daily passes, the geostationary satellite GOES, whose images every five minutes make it possible to date the start of the fire, as well as the slope and altitude data that govern the speed of propagation.
The resulting reconstructions are closely aligned with the infrared perimeters recorded by aircraft, a field reference in firefighting. The operational interest lies in this changeover: instead of noting where the fire has gone, the teams have a real-time prediction of where it will go — the data that conditions evacuation decisions and the positioning of resources.
Read the full article on the original source
Go to the source →