Sensitivity Analysis of Rainfall and Streamflow Thresholds for Forecasting Flash Floods

Flash flood impacts are characterized by their high specificity in space and time. A common approach to flash food forecasting is in the use of proxies based on threshold analysis of rainfall and streamflow to define the location and time frame of the threat. Verification of these forecasts is challenging because of the limited amount of direct hydrometeorological observations and the displacement between the forecast proxy and the actual flash flood impact. Flash Flood Local Storm Reports (FFLSRs) collected by the National Weather Service (NWS) represent a useful asset for forecast verification because they include information about the location and time of impacts. However, there are difficulties in using FFLSRs given the uncertainties introduced by their inherent human input in terms of the time and location of the events. Another source of uncertainty comes from the mismatch between the scale of the observations and the scale that the forecast product can resolve. Therefore, methodologies employed to verify these forecasts need to be able to account for and mitigate the effects of these uncertainties. A common strategy to address these effects is to perform space and time aggregations of both the forecast and observation data and generate simpler objects that share the same representative scale of information. Nevertheless, the spatiotemporal parameters of these aggregations can have an influence on the computed forecast skill, ultimately misguiding the assessment of the utility of flash flood forecast products. In this contribution, we present a study of the use of an object-oriented method to verify dichotomous (yes/no) spatial forecasts of flash floods based on proxies derived from quantitative precipitation estimates (QPE) and simulated streamflow products of the Multi-Radar Multi-Sensor (MRMS) Flooded Locations And Simulated Hydrographs (FLASH) suite. The products include gridded QPE-to-FFG (Flash Flood Guidance) ratio and Average Recurrence Interval (ARI) from various rainfall accumulations (1-hour, 6-hour, etc.). Central to the study, a sensitivity analysis is performed to evaluate the impact of aggregation parameters on the quantification of the forecast skill. The results from this study will provide guidance on the use of these products for flash flood forecasting in NWS operations.