Flash Flood Verification Problems: Pondering Precipitation Proxies

Greg Herman

Colorado State University

Tuesday, Jul 17, 2018, 10:00 am
DSRC Room 1D403

Abstract

The prediction of flash floods is a notoriously challenging forecast problem, requiring not only accurate prediction of heavy rainfall magnitudes, but also on the spatiotemporal distribution of that rainfall; the hydrologic interactions between precipitation, terrain, and land surface; and also on antecedent precipitation and its effects on soil conditions. Further exacerbating the flash flood forecast problem is the considerable difficulty in verifying flash flood events, an essential component to forecasting any phenomenon. Flash flood reports and warnings are subject to population bias, with report databases often missing transient floods in very rural areas, and also to varying reporting and report encoding practices in different regions of the United States. It is often attractive to consider flash flood potential in a simplified framework considering only the quantitative precipitation estimates (QPEs) or forecasts (QPFs), depending on the context. There are numerous ways to estimate risk of flash flooding directly from QPEs, such as exceedances of some precipitation threshold within a prescribed time interval (e.g. 3”/3 hr.); by exceedances of average recurrence intervals (ARIs), which determine amounts required for a fixed frequency of exceedance given the local precipitation climatology; or by products such as Flash Flood Guidance (FFG), featuring dynamically varying thresholds that account for the antecedent conditions and hydrologic characteristics at the location. Through these come many different ways to use QPE to assess whether a flash flood is likely to occur (in the forecasting context) or has occurred (in the analysis context). Factors to consider include the precipitation accumulation interval length, which can vary from subhourly to over an entire day; threshold source and magnitude; and QPE source. Three popular QPE sources—the Multi-Radar/Multi-Sensor System (MRMS), NCEP Stage IV Precipitation Analysis (ST4), and the Climatology Calibrated Precipitation Analysis (CCPA)—are compared over a 2.5 year record from January 2015 through June 2017 for exceedances of different thresholds—including fixed amounts, ARIs, and FFG—which are used as candidate proxies for flash flooding. These threshold exceedances are then evaluated against flash flood warnings and reports to assess degree of correspondence. A synthesis of recurring systematic issues with the different QPE sources is presented, as well as which exceedances provide the best and worst correspondence with flash flood observations. Some results reconfirm conventional wisdom in flash flood analysis, while others prove contrary to what one might expect.