WRF-Hydro-RAPID Performance in Predicting Recent Texas Hill Country Flash Floods

Land surface models (LSMs) have been increasingly used in operational flood forecasting applications. Evaluating LSM performance under a variety of historical flood events is thus rather critical towards its operational goals. In this project, we evaluate the performance of the WRF-Hydro-RAPID model in simulating recent high-impact flash flood events in the Texas Hill Country and Balcones Escarpment areas, which are known as the “Flash Flood Alley”. Several flood events of different categories are simulated, including four rapid rate of rise events (10/30–31/2013, 5/23–24/2015, 5/25/2015, and 10/30/2015) and three moderate rate of rise events (10/13–14/2013, 10/24–25/2015, and 5/26–27/2016). The model simulation is configured in an offline mode driven by two types of rainfall data, namely the Next-Generation Radar (NEXRAD) StageIV quantitative precipitation estimates (QPE) and the Multi-Radar/Multi-Sensor (MRMS) QPE products, at 4-km/1-hourly resolution. The results are evaluated against 40 USGS gauge observations that are least affected by dam operations in our study area.

Preliminary results based on the 2015 Memorial Day flood (using StageIV data as input) suggest that 18, 7, and 15 out of the 40 gauges have good to excellent (0.4<NSE<0.82), acceptable (0<NSE<0.4), and poor (NSE<0) model performances, respectively. In order to truly inform model improvements, however, we will simulate the same events with the higher spatial/temporal resolution MRMS data, and systematically investigate multiple events of different magnitude. For gauge locations that continuously show excellent (poor) model performance in different simulations, the land surface physical properties including soil, vegetation, and geological conditions will be closely examined to understand LSM behaviors. Potential locations-of-interest will include Onion Creek, Bull Creek, and Blanco River, since previous studies have indicated possible land surface modeling challenges with surface heterogeneity for these locations. On the other hand, gauge locations with mixed model performance are likely subject to uncertainties in the rainfall forcing, suggesting the need for improved QPE and QPF at these locations. Findings from this project are expected to improve operational flash flood forecasting at WFO Austin/San Antonio (EWX). It may also help drive improvements to the National Water Model (NWM) that NWS RFCs and WFOs will be using operationally, which is founded on the WRF-Hydro modeling system.