Learn More

USRA partnered with Meta to support the U.S. Geological Survey (USGS) water observation system, which is used for flood and drought forecasting, reservoir management, and water management decisions. In 2019, the U.S. Congress funded USGS to launch the Next Generation Water Observing System (NGWOS) pilot, deploying new monitoring stations and sensors in several medium-sized watersheds. USGS and USRA's RIACS collaborated to explore how AI can complement traditional water analysis.

USRA utilized Meta's Segment Anything Model (SAM), a foundation model for image segmentation that can segment objects in images and videos with minimal training. USRA research scientists developed a fine-tuned version of SAM 2, enhancing its ability to identify water bodies through targeted adaptation. The fine-tuned version was trained using USGS and other datasets, combining drone and satellite remote sensing data (including high-resolution PlanetScope imagery from NASA's Commercial Satellite Data Acquisition Program and USGS 3D Hydrography Program data) to identify streams and detect water extents. The method showed promising results in case studies and demonstrated potential for national scalability.

Dr. David Bell, Director of USRA, noted that SAM is used to automate a critical part of the workflow, which USGS colleagues described as a major bottleneck in real-time image analysis—previously relying on manual digitization of water boundaries, which is crucial for rapid response and national scalability. The automated method can detect changes in water conditions to indicate floods and offers significant cost savings compared to traditional monitoring methods.

Data from NOAA's National Centers for Environmental Information shows that from 1980 to 2024, over 400 weather disasters in the U.S. caused losses exceeding $1 billion each (CPI-adjusted to 2024), with total losses nearing $3 trillion. Water-related disasters (floods, cyclones, droughts) accounted for over $2 trillion in losses, with severe floods contributing $200 billion. The five-year average loss cost has more than tripled over the past 20 years. FEMA's National Risk Index also indicates that riverine flood risk is a national concern.

With the release of SAM 3, USRA and USGS will be able to achieve unified detection, segmentation, and tracking of objects in images and videos using text, example, and visual prompts, thereby advancing research. The collaboration demonstrates how public-private partnerships (combining open-source AI innovation, research expertise, and federal agency knowledge) can drive progress in disaster preparedness and resilience.