Creating Synergy between Land Surface and Rainfall-Runoff Models for Improved Flash Flood Predictions

In this contribution, we present efforts to develop synergy between operational numerical models of hydrologic processes, each with varying levels of physics representation complexity, resolution, and output generation cadence. Specifically, we aimed to enhance forecasts of surface flows generated by the hydrologic models within the Flooded Locations And Simulated Hydrographs (FLASH) system through the assimilation of state variables from the National Water Model (NWM). FLASH is the primary tool used by National Weather Service (NWS) forecasters for issuing flash flood warnings and watches across the United States. Conversely, the NWS has adopted the NWM as the primary model for coupling with storm surge/estuary models, providing inundation forecasts for land-falling hurricanes and urban flooding scenarios, and representing wildfire impacts on land surfaces and soils. The current version of the NWM is based on the WRF-Hydro modeling software, which utilizes the Noah-MP land surface model. This model has the advantage of incorporating physically based processes on the land surface, such as evapotranspiration, snow accumulation, ablation, and snowmelt. In contrast, FLASH handles these processes in a considerably simpler manner or not at all. Incorporating state variables and parameters from the NWM provides a direct way to represent computationally expensive physical processes without requiring additional resources and maintaining the same 1 km/10 min resolution forecasters enjoy with FLASH products. Presently, the NWM runs at the top of the hour with an additional 1 - 1.75 hrs needed for simulations to complete due to the added complexity of the model. On the other hand, FLASH runs every 10 min and incorporates the latest radar forcings from the Multi-Radar Multi-Sensor (MRMS) system to provide hydrologic forecasts of phenomena that respond very quickly to rainfall such as flash floods and debris flows emanating from rain on burn scars. We employed a data assimilation approach based on first-order sensitivity expressions of the numerical hydrologic model and compared it to simple nudging. This exercise was conducted for selected cases and regions where NWS forecasters identified areas with known deficiencies in FLASH, resulting in product values that were challenging for NWS forecasters.