Insights into Hydrometeorological Factors Constraining Flood Prediction Skill during the May and October 2015 Texas Hill Country Flood Events

Improving forecasts and impact-based decision support for flooding is critical because it is the deadliest severe weather hazard in the United States, causing over 90 fatalities a year over the past decade. The Texas Hill Country and the Austin-San Antonio corridor is one of the deadliest flash flood-prone regions of the contiguous United States due to relatively frequent episodes of heavy rainfall with rapid river rises due to the region’s complex terrain, soil characteristics, and rapid population growth. Although existing operational hydrological models generally struggle with accurately simulating the Hill Country’s rapid response to heavy rainfall, the United States National Water Model (NWM) and similar model configurations may improve regional flood prediction for extreme rain events. This mature study evaluates the 23-24 May and 30 October 2015 flood prediction skill of a physically-based model resembling the NWM (i.e., WRF-Hydro-RAPID) over the Texas Hill Country for a pair of extreme rainfall events that both had annual exceedance probabilities below 0.2%. In addition, we investigate several hydrometeorological factors that contributed to a record flood along the Blanco River at Wimberley (WMBT2) in the 23-24 May 2015 event that resulted in 13 fatalities and significant damage.

Using two radar-based quantitative precipitation estimation (QPE) products – Stage IV and Multi-Sensor Multi-Radar (MRMS), it is shown that the event precipitation accuracy dominates the flood prediction skill. The finer-resolution MRMS QPE outperforms ST4 across all basins during the October event and for basins with small drainage areas during the May event despite underestimating QPE due to algorithmic dependencies on the Maximum Estimated Size of Hail (MESH). Overall, the model exhibits good performance at gauges with fast flood response from causative rainfall and gauges that are not forecast points in the National Weather Service’s Advanced Hydrometeorological Prediction System, showing great promise for forecasts, warnings, and emergency response. However, the model suffers from poor prediction skill over regions without rapid flood response and regions with human-altered flows, suggesting the need to revisit the channel routing algorithm and incorporate modules to represent human alterations. Examining both flood events at WMBT2 with similar meteorological characteristics in greater detail reveals that the location of intense rainfall combined with land physiographic features are key to the flood response differences. Model sensitivity tests further show the record flood peak on 23-24 May 2015 could be better obtained by tuning the deep-layer soil wetness and the flow velocity field in the river network, which offers hydrometeorological insights into the causes and the complex nature of such a flood and why the model struggles to predict the record flood peak.