M4: Next Generation AI-Based NRCS Water Supply Forecasting System
The US Department of Agriculture’s Natural Resources Conservation Service (NRCS) has monitored snowpack and issued seasonal water supply forecasts (WSFs) for close to 100 years. Today, NRCS operates the SNOTEL network and the largest stand-alone operational WSF system in the American West. Here, we introduce the next generation of this WSF system. The resulting prediction analytics engine, called M 4, is a careful blend of proven and fresh concepts, including multi-model ensembles, probabilistic prediction, and several directions in artificial intelligence. To our knowledge, it is the largest transition of machine learning into operational hydrologic forecasting so far, facilitated in part by a design ethos based on determining and meeting the specific and detailed needs of the operational WSF community. Hindcast testing, and live operational testing during the 2020 forecast season, demonstrate improved accuracy, objectivity, robustness, geophysical interpretability, and amenability to automation relative to existing methods.
Presenter Dr. Sean Fleming, Applied R&D Technical Lead, National Water and Climate Center, NRCS, USDA, Portland; College of Earth, Ocean, and Atmospheric Sciences, and Water Resources Graduate Program, Oregon State University, Corvallis; and Department of Earth, Ocean, and Atmospheric Sciences, University of British Columbia, Vancouver