Missing or inaccurate river bathymetry constrains modeling approaches and produces large errors in predicted water levels and inundation, with direct implications for flood forecasting and water management. The goal of this project is to develop a generative artificial intelligence (AI) framework with direct applications to river routing, flooding, inundation forecasting, and coastal coupling.
Why We Care
Most digital elevation models (DEMs) are hydro-flattened over rivers, and existing topobathy or survey datasets have extremely limited coverage, reducing channel representation and constraining forecasts. This gap constrains forecast accuracy and limits the development of a reliable hydrofabric for NOAA’s National Water model.
What We Are Doing
This project will deliver a generative AI framework to reconstruct missing river bathymetry in DEMs, directly addressing NOAA’s National Ocean Service and Office of Water Prediction priorities for improved channel representation in national hydrofabrics. The framework, trained on available topobathy and hydrographic survey data, will produce river-aware reconstructions and uncertainty-aware ensembles.
Benefits of Our Work
The anticipated outcomes of this project include improved representation of river channels in national DEMs, publicly available tools to produce bathymetry on demand, postdoctoral training, and advanced hydrologic prediction and hydrofabric development.
The project is led by Dr. Eric Anderson of the Colorado School of Mines and includes co-investigators Dr. Saeed Memari at the Colorado School of Mines and Dr. Vishnu Boddeti at Michigan State University. The project is part of the Cooperative Institute for Research to Operations in Hydrology (CIROH).