Development and Demonstration of Ensemble Hydrologic Data Assimilation Strategies for a Real-Time Distributed Regional Hydrologic Forecast System

The operational streamflow forecasting enterprise has been working to develop ensemble predictions that quantify forecast uncertainty to support risk-based decisionmaking in a variety of contexts (from flood mitigation to water resources management). Ensemble techniques used in hydrologic forecasting include not only the downscaling and use of meteorological ensemble forecasts, but also ensemble-based hydrologic data assimilation methods to improve model states for forecast initialization. The ensemble methods used in operational hydrology have mostly been developed in the context of applications at the watershed scale, and often with lumped conceptual models. Recent forecast model development effort, however, has focused on high-spatial resolution distributed modeling, which complicates the use of some ensemble techniques. The updating of distributed model states during data assimilation (DA), for instance, is problematic when the assimilated observations are point measurements (such as streamflow). To address this forecast science challenge, we evaluate a DA strategy that uses an ensemble particle filter to assimilate streamflow into an intermediate-scale HUC-based configuration of a SUMMA hydrologic model for the Pacific Northwest region of the US. The model and DA technique run within the NCAR System for Hydromet Analysis Research and Prediction Applications (SHARP), which was designed to demonstrate real-time over-the-loop ensemble flow forecasting techniques. The DA approach relies on a 100-member ensemble of meteorological model forcings to initialize an ensemble of watershed moisture states, from which a subset are used to initialize 7-day lead flood predictions driven by ensemble weather forecasts from the NCEP GEFS. This presentation describes the SUMMA forecasting implementation within SHARP and presents preliminary results and findings from the data assimilation component. The project is a collaboration between NCAR and the University of Washington and is sponsored by the Bureau of Reclamation and the US Army Corps of Engineers.