Hurricane Harvey : a quantitative approach to assessing the accuracy of National Water Model forecasted inundation
The impact of Hurricane Harvey has stressed the need for accurate flood inundation forecasts to aid emergency response. The National Water Model uses predicted rainfall to produce forecasts of discharge and then generate flood inundation maps. To confidently employ flood forecast models for future events, we first need to assess the accuracy of these flood maps. The National Water Model combined maximum river discharge values for the affected Texas region and Height Above the Nearest Drainage (HAND) to calculate the extent of inundation from Hurricane Harvey. We compared this predicted inundation extent with a database of over 2000 high water marks gathered by the U.S. Geological Survey. At each high water mark location we calculated the difference of inundation depth between the National Water Model prediction and the measured high water mark. We find that the prediction has low bias with a mean difference of 26 cm, although there are far larger depth differences at individual locations. For approximately one-third of the comparisons made there is less than one meter of vertical difference between the National Water Model forecasted depths and the observed high water marks. We hypothesize that channel features such as slope, length, and stream level are factors which influence the accuracy of the forecasted inundation depths, and present relationships found between these features and the accuracy of their forecasts. From this analysis, we find that the presence of pluvial flooding and storm surge is likely to obscure significant trends between these channel factors and prediction accuracy as the National Water Model models only fluvial flooding. We find that in regions where the model over-predicts inundated depth, slope and stream level are strong predictors of accuracy.