Project and Position Details: AI models are increasingly used to map floods, land cover, and urban change from satellite imagery and other geospatial data. But these models can be confidently wrong, and their errors are often concentrated in particular places, such as rural or data-scarce regions, where decisions may matter most. This project develops methods to measure, explain, and correct where and why geospatial AI fails, so that its outputs can be used responsibly.

The undergraduate researcher will help (1) prepare and explore public geospatial datasets (e.g., satellite-based flood and land-cover maps), (2) run and evaluate existing AI models, (3) analyze how model errors and uncertainty vary across space, and (4) create maps and visualizations that communicate when a model can and cannot be trusted. The student will also read and discuss short research papers with the PI. Training in Python, geospatial data analysis, and machine learning will be provided. Students who make substantial contributions will have opportunities to present at the UI Undergraduate Research Festival and contribute to publications, and the PI will support motivated students in applying for an OUR ICRU Fellowship.