Collaborative Research: CAIG: AI-Enabled, Physics-Aware, Explainable Forecasting of Thermospheric Density with Uncertainty Quantification

Space weather can strongly disturb the upper atmosphere, but predicting space weather remains difficult because the Sun, solar wind, geomagnetic activity, and the Earth’s thermosphere interact in complex and uncertain ways. This project will develop new Artificial Intelligence (AI)-based methods to forecast thermospheric density from several hours to several days ahead and to explain when, why, and how much these forecasts can be trusted. The results will improve the understanding of upper-atmosphere variability and support national needs in space-weather preparedness and sustainable space operations. The project will also provide workforce training and strengthen the scientific workforce by training postdoctoral researchers, graduate students, and undergraduate students in AI, uncertainty quantification, solar image processing, and thermospheric science. Educational modules, open-source models, datasets, and analysis tools will be made publicly available. The project will also engage students and high-school teachers through hands-on research, seminars, tutorials, and outreach activities. This project develops an AI-enabled, physics-aware, uncertainty-driven framework for thermospheric density forecasting across multiple time scales. The first component advances short-term multi-hour probabilistic forecasting by coupling physics-aware density models with probabilistic predictions of solar and geomagnetic drivers. Driver uncertainty will be propagated through nonlinear forecasting mappings to produce calibrated density forecasts and uncertainty estimates. The second component enables long-term, multi-day ensemble forecasting by using solar imagery and heliospheric inputs, including solar wind and interplanetary magnetic field data, to generate probabilistic scenarios for geomagnetic drivers. These scenarios will drive ensembles of physics-aware density models when key drivers are difficult to predict deterministically. The third component applies explainable AI to quantify intrinsic predictability, identify dominant physical drivers, and attribute forecast uncertainty across different forecast horizons and space-weather conditions. The intellectual merit lies in establishing a unified, multi-scale framework for uncertainty-aware thermospheric density forecasting. Rather than treating short-term and long-term forecasts as separate problems, the project will examine how driver uncertainty, historical thermospheric states, empirical-model priors, solar imagery, heliospheric inputs, and spatiotemporal structure jointly affect forecast skill and uncertainty growth. The expected contributions include new probabilistic and ensemble forecasting methods, improved understanding of thermospheric predictability, and generalizable, eXplainable Artificial Intelligence (XAI) tools for uncertainty-aware geophysical forecasting beyond the thermosphere. 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.