The GeoDS lab received a new NSF Grant on Collaborations in Artificial Intelligence and Geosciences (CAIG)!
Investigators: Qunying Huang (Principal Investigator), Song Gao (Co-PI), Robert B Pierce (Co-PI)
Abstract: Wildfires are increasing in frequency, intensity, and societal impact as warming temperatures, increasing aridity, and more extreme weather alter fire-prone environments across the United States and globally. These changes threaten communities, ecosystems, air quality, and critical infrastructure. Despite advances in wildfire observation and modeling, major scientific questions remain regarding how weather, vegetation, topography, and human activities interact to influence wildfire ignition, spread, and suppression. This project will develop an integrated geospatial artificial intelligence framework to improve wildfire monitoring, forecasting, and suppression planning while advancing understanding of wildfire–environment–human interactions. The project will produce open datasets, models, and computational tools to support future wildfire research, contribute to training students in geospatial data science and Earth system science, and accelerate translation of research advances into operational wildfire management.
This project will develop an integrated geospatial artificial intelligence framework for wildfire detection, spread prediction, and cooperative wildfire control through three research objectives. First, it will fuse high-temporal-resolution Geostationary Operational Environmental Satellite (GOES) observations with high-spatial-resolution Visible Infrared Imaging Radiometer Suite (VIIRS) imagery using a diffusion-based super-resolution model to generate enhanced active fire observations and enable rapid fire detection from time-series imagery. Second, it will develop a physics-guided spatial–temporal graph convolutional network that combines physical constraints with data-driven learning to improve wildfire spread prediction. Third, it will develop knowledge-guided deep reinforcement learning models that integrate fire management knowledge with weather, fuels, terrain, transportation networks, and population distributions to support adaptive suppression strategies and resource allocation. Using observations and case studies from more than 100 large wildfire events across the United States, the project will generate new insights into wildfire dynamics and improve understanding of the climatic, environmental, geographic, and human factors shaping wildfire behavior.