The 21 June 1997 Flood: Storm-Scale Simulations and Implications for Operational Forecasting

Abstract

On 20?21 June 1997, a convective outbreak in Nebraska, Iowa, Illinois, and Wisconsin resulted in two fatalities, eight injuries, and approximately $104 million in damage. The majority of the damage ($92 million) was the result of flooding in southeastern Wisconsin owing to nearly 250 mm of rain produced by training convection and to a lesser extent the passage of a persistent, elongated convective system. The flood event is analyzed and storm-scale (5- and 1.67-km grid spacing) resolution model simulations at 0?24-, 12?36-, and 24?48-h ranges are produced to study the evolution and predictability of the rainfall. Synoptic conditions corresponded closely to the mesohigh pattern frequently associated with heavy rainfall events. Despite the recognition by National Weather Service forecasters of the potential for heavy rainfall, uncertainty concerning event magnitude and affected areas, exacerbated by poor operational model guidance, resulted in a failure to issue flash flood watches prior to the onset of flooding. Simulations of the event using 5-km grid spacing show that short-range forecasts (0?24 h) are able to suggest focused precipitation in southeastern Wisconsin. Initiation of the convection is tied to lift at the leading edge of a low-level jet (LLJ), with a focus along a weak frontogenetic boundary. A rapid loss of predictability at the 12?36- and 24?48-h ranges results in part from poorer representations of the LLJ. Increased resolution (1.67-km grid spacing) is necessary to capture the magnitude of the event. The increased resolution led to little change in available moisture or lift, but substantial increases in precipitation efficiency, arising from the capability to resolve the details of convective organization in an aggregate, time-averaged sense. With 1.67-km grid spacing, the probability of detection (false alarm ratio) for critical precipitation thresholds of 25, 50, 75, and 125 mm in the heavy rainfall region are 0.79 (0.01), 0.60 (0.21), 0.51 (0.23), and 0.36 (0.32), respectively. Precipitation forecast skill as measured by the Kuiper skill score for these same thresholds is 0.30, 0.36, 0.39, and 0.33. However, on a county warning area basis for flash flooding, these same forecasts yield probability of detection (false alarm ratio) of 0.71 (0.55). The results of this study suggest that the consistent use of storm-scale model output in conjunction with observations in such events will likely lead to improved forecasts and greater forecast value, although significant constraints related to real-world application of such models and procedures would still remain. In order to firmly establish these points, it is necessary to analyze a number of such cases within an operational context.