Collaborative Research: CAIG: Leveraging Artificial Intelligence for understanding and forecasting flash drought development and propagation

Flash droughts develop quickly, often in a few weeks to months, leaving farmers, water managers, ecosystems, and communities with little time to prepare. These rapidly intensifying droughts can reduce crop yields, stress forests, increase wildfire risk, and strain water supplies. As flash droughts become more common across the United States and globally, society needs better tools to anticipate when and where they will occur and to understand why they develop so quickly. This project will support this need by using artificial intelligence (AI) to improve flash drought prediction while also making the predictions transparent and scientifically meaningful. The project will identify the environmental signals that help predict drought onset, including how vegetation, soil moisture, heat, and atmospheric conditions interact across space and time. The project will improve scientific capacity for drought early warning, information that can support agricultural and water-resource decision-making, and open tools and knowledge that can benefit researchers, agencies, and communities working to reduce risks from hydroclimate extremes. The project will develop a robust, interpretable, multi-modal, multi-task (RI-M³T) deep learning framework to improve flash drought prediction and advance understanding of vegetation–atmosphere interactions. The model will integrate multiple sources of information, including vegetation conditions, atmospheric drivers, soil moisture, and other environmental variables, and will be designed to capture spatial and temporal interactions within and across these variables. The project has three main goals: to build an AI model that can provide more reliable flash drought forecasts under changing climate and ecological conditions; to use the model’s interpretability to quantify how vegetation and atmospheric processes, memory effects, and spatial connectivity influence flash drought development and spread; and to compare AI-derived relationships with those represented in Earth system models. These comparisons will help identify major sources of model error and guide improvements in the simulation and prediction of drought extremes. Collectively, the research will provide new scientific benchmarks, improve understanding of flash drought mechanisms, and lay the foundation for future AI-enabled drought forecasting and risk-management systems. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.