Deep Learning for Near-Real-Time Wildfire Monitoring

About the Project

Geostationary satellites such as GOES observe the continual U.S. every few minutes but at kilometer-scale pixels, while polar-orbiting sensors such as VIIRS resolve fires at 375 m yet revisit only a few times per day. Neither alone supports timely, detailed wildfire monitoring.

This project develops deep learning models trained on spatiotemporally paired GOES and VIIRS observations. The models jointly super-resolve GOES imagery and detect active fire pixels, with preprocessing steps that reduce noise, correct artifacts, and improve alignment between the two satellite sources. The result is a pipeline that estimates fire location and brightness temperature at VIIRS-like resolution from each GOES acquisition, supporting more accurate and timely wildfire detection, monitoring, and response.

Flare Tracker

An interactive map viewer showing super-resolved satellite imagery and predicted fire masks with time-series playback for recent fire events.

Live Fire Viewer

Publications