Extended Evaluation of the National Water Model over Iowa

This study explores the performance of the analysis-and-assimilation configuration of the National Water Model over Iowa. This NWM model configuration runs hourly and produces a real-time snapshot of the most updated hydrologic states of the system. The resulting model states serve to initialize other configurations that produce short, medium and long-term range discharge forecasts. Consequently, the quality of these forecasts is subject to the accuracy of the analysis-and-assimilation configuration results, thus justifying the importance of studying and understanding its skill. The NWM model assimilates streamflow observations from the United States Geological Survey through a “nudging” procedure. Rigorous evaluation requires comparing model outputs with observations that are not assimilated. Quality-controlled streamflow observations from the Iowa Flood Center Bridge-Mounted River Stage Sensors (IFC BMRSS) provides the environment for these analyses. The authors compared NWM outputs between May 2016 and April 2017 with two datasets: 1) USGS streamflow and velocity observations, at sites where the model performs data assimilation; 2) streamflow data from the BMRSS locations. The spatial distribution of cross correlation (CC) and Nash Sutcliffe Efficiency (E) values provide quantification of model performance. The authors found that performance is related to the spatial scale of the basins. At USGS locations, small size basins have average CC=0.45 and average E=-0.25, while large basins have average CC=0.9 and average E=0.9. Evaluation at BMRSS shows average CC=0.2 for small basins, and average CC= 0.7 for larger basins. The assimilation procedure helps to improve forecast at downstream basins. Authors provide recommendations for selected locations where performance is low, therefore the model setup should be revised.