Supplementing flash flood reports with impact classifications

In recent years, there has been an increase in flash flood impacts, even as our ability to forecast events and warn areas at risk increases. This increase results from a combination of extreme events and the exposure of vulnerable populations. The issues of exposure and vulnerability to flash floods are not trivial because environmental circumstances in such events are specific and complex enough to challenge the general understanding of natural risks. Therefore, it seems essential to consider physical processes of flash floods concurrently with the impacts they trigger. This paper takes a first step in addressing this need by creating and testing the coherence of an impact-focused database based on two pre-existing public and expert-based survey datasets: the Severe Hazards Analysis and Verification Experiment (SHAVE) and the US National Weather Service (NWS) Storm Data. The SHAVE initiative proposes a new method for collecting near-real-time high-resolution observations on both environmental circumstances and their disastrous consequences (material and human losses) to evaluate radar-based forecasting tools. Forecast verification tools and methods are needed to pursue improving the spatial and temporal accuracy of forecasts. Nevertheless by enhancing SHAVE and NWS datasets with socially and spatially relevant information, we aim at improving future forecast ability to predict the amount and types of impacts. This paper describes the procedures developed to classify and rank the impacts from the least to the most severe, then to verify the coherence and relevance of the impact-focused SHAVE dataset via cross-tabulation analysis of reported variables and GIS-sampled spatial characteristics. By crossing impact categories with socio-spatial characteristics, this analysis showed first benchmarks for the use of exposure layers in future flash flood impact forecasting models. The enhanced impact-focused datasets were used to test the capabilities of flash flood forecasting tools in predicting different categories of impacts for two extreme cases of flash flooding in Oklahoma, USA. Results showed a general tendency for the more severe impacts to be associated to higher mean exceedances over tool values. This means that, at least for these particular case studies, the tools were able to make a distinction between less severe and more severe impacts. Finally, a critical analysis of the NWS and SHAVE data collection methodologies was completed and challenges for future work were identified. (C) 2012 Elsevier B.V. All rights reserved.