Real-time AI-Assisted Forecasting of Reservoir Capacity Across the United States

Moiyyad Sufi , Alejandra Botero-Acosta , Carlos Erazo, Amanda L Cox, Paul Boyd, Marian Muste, Ibrahim Demir

Reservoir sedimentation progressively reduces storage capacity, threatening water supply, flood control, hydropower, and ecosystem services across the United States. However, systematic sedimentation monitoring remains sparse, limiting the feasibility of national scale assessment using traditional methods. This study develops a web-based end to end data driven framework for evaluating reservoir capacity loss across the contiguous United States by integrating historical sedimentation surveys, geospatial feature extraction, and interval based machine learning. Reservoir sedimentation is formulated as capacity change over time, enabling scalable prediction and cumulative reconstruction of storage trajectories. Results demonstrate reliable performance for medium and large federally managed reservoirs, with predictable limitations outside the training domain. The framework is operationalized through a web based decision support application that provides transparent access to model inputs and outputs. This study demonstrates how scalable, data driven systems can support national prioritization of sedimentation risk and guide strategic reservoir management under limited monitoring conditions.