Data Assimilation Improves the Performance of the Iowa Flood Center Real-Time Streamflow Predictions.

The Iowa Flood Center (IFC) operates a real-time streamflow forecasting system that provides forecasts for about 1,000 locations in the state of Iowa, including small communities and stream gages. The core element of the system, the Hillslope-Link Model (HLM) is forced with radar rainfall inputs in open loop mode, and simulates the most up to date values of hydrologic state variables, including soil moisture, groundwater storage, and water storage in the channels. Observed streamflow at USGS gages have been used in the past to assess the performance of the open-loop HLM, and establish a benchmark reference for future improvements of the model. Recently, the IFC started applying direct data assimilation (DA) of streamflow data into the operational model. The assimilation is done by forcing observed streamflow data as the initial conditions of the ordinary differential equations describing the flux of water in the channels. The effect of DA propagates downstream of the assimilated gages in the stream network. In this study, we analyze the impact of such DA on the performance of the model. For that purpose, we compare open loop simulations of HLM with and without DA, to observed streamflow. We assess the performance using different metrics that are relevant to flood simulations: Kling Gupta Efficiency, Volume Bias, and Peak Difference in magnitude and time. We analyze how performance changes with the increase of distance between the assimilation and downstream evaluation points. Our results indicate significant improvements of all statistical measures of performance. For example, when assimilating the information of upstream USGS gages, the percentage of evaluation points with KGE >0.5 increased from 24% to 70%; for KGE>0.75 increased from 1% to 34%. Including DA from IFC Bridge Sensors installed at small streams, increases even more the performance.