Presentation Information
[U06-P02]Enhance national water security and sustainability using SWOT wide-swath altimetry, multi-sensor satellite imaging systems, and GeoAI technology
*Hongxing Liu1, Jihee Seo1, Shujie Wang2, Lei Wang3, Dan Tian1, Haibin Su4, Naveenkumar Purushothaman1, Matthew LaFevor1 (1.University of Alabama, 2.The Pennsylvania State University, 3.Louisiana State University, 4.Texas A&M University-Kingsville)
Keywords:
water security and sustainability,SWOT,multi-sensor satellite imaging systems,GeoAI,machine learning
Reservoirs and lakes are foundational to national water-resource management, supporting municipal, agricultural, and industrial supply while buffering hydrologic extremes by regulating flows and enhancing resilience to droughts and floods. Their central role in water security and climate adaptation motivates the need for consistent, timely, and spatially comprehensive monitoring, especially for the many ungauged reservoirs across the contiguous United States. In this study, we present an integrated satellite remote sensing and GeoAI framework that delivers near–real-time, high-temporal-resolution estimates of reservoir water level, surface area, and storage change across the contiguous United States by combining SWOT wide-swath altimetry with multi-sensor SAR and optical imaging. We first develop a consistent, georeferenced national reservoir inventory derived from Sentinel-1 SAR imagery at 10 m resolution, enabling scalable, all-weather mapping of water bodies. Building on a deep learning foundation segmentation model (SAM2), we implement an automated software tool to extract reservoir water-surface extent from both SAR (Sentinel-1A/1C and NISAR) and optical (Landsat 8/9 and Sentinel-2A/B) observations. To translate frequent surface-area observations into hydrologically actionable variables, we exploit SWOT’s capability to observe inland-water elevations and construct reservoir-specific surface area–water level rating curves. These rating relationships enable conversion of high-frequency multi-sensor area time series into corresponding water-level and volume-change estimates, substantially increasing revisit frequency beyond SWOT alone while maintaining physical consistency. Finally, we fuse SWOT, SAR, and optical observations within a hybrid graph neural network (GNN) to hindcast and forecast reservoir level, surface area, and storage dynamics, leveraging spatial connectivity and shared hydroclimatic drivers across the reservoir network. We demonstrate that the proposed framework expands the spatial coverage and timeliness of reservoir monitoring, and enhances accuracy for key indicators of water availability and sustainability, supporting operational water management, ecological conservation, disaster response, and climate adaptation planning at national scales.