Merging Real-Time Channel Sensor Networks with Continental-Scale Hydrologic Models: A Data Assimilation Approach for Improving Accuracy in Flood Depth Predictions

This study proposes a framework that (i) uses data assimilation as a post processing technique to increase the accuracy of water depth prediction, (ii) updates streamflow generated by the National Water Model (NWM), and (iii) proposes a scope for updating the initial condition of continental-scale hydrologic models. Predicted flows by the NWM for each stream were converted to the water depth using the Height Above Nearest Drainage (HAND) method. The water level measurements from the Iowa Flood Inundation System (a test bed sensor network in this study) were converted to water depths and then assimilated into the HAND model using the ensemble Kalman filter (EnKF). The results showed that after assimilating the water depth using the EnKF, for a flood event during 2015, the normalized root mean square error was reduced by 0.50 m (51%) for training tributaries. Comparison of the updated modeled water stage values with observations at testing locations showed that the proposed methodology was also effective on the tributaries with no observations. The overall error reduced from 0.89 m to 0.44 m for testing tributaries. The updated depths were then converted to streamflow using rating curves generated by the HAND model. The error between updated flows and observations at United States Geological Survey (USGS) station at Squaw Creek decreased by 35%. For future work, updated streamflows could also be used to dynamically update initial conditions in the continental-scale National Water Model.