P4.14
Predictability of convection at 24-48h forecast range using a very-high resolution (6 km) NWP model
Michael A. Fowle, Univ. of Wisconsin, Milwaukee, WI; and P. J. Roebber
Warm season precipitation in the midwestern United States is dominated by convection. Historically, precipitation skill scores show dramatic seasonal fluctuations, with the lowest scores occurring in the warm season, as a result of the relatively low predictability of convection.
A forecast verification study was conducted to evaluate the occurrence of warm season convection in the Upper Midwest during the period 5 April 1999 through 20 September 1999. All days in the period were examined, using the high resolution WSR-88D precipitation dataset and the one and two day 6 km forecasts from the UWM quasi-operational version of the Pennsylvania State University – National Center for Atmospheric Research (PSU-NCAR) fifth-generation Mesoscale Model (MM5). Contingency measures showed skill in predicting convection at both forecast ranges [Kuipers skill score (KSS) of 0.837 and 0.754 respectively]. An examination of storm type (isolated, multi-cellular, and linear) was also conducted for the same dates. Contingency measures for storm type also showed high predictability for both periods [KSS of 0.914 and 0.858 respectively].
By grouping the convective events by storm type, a timing analysis was performed to chart storm evolution. Linear and multi-cellular type storms tended to form several hours late in the 00z –23z (day 1) MM5 model forecast as a result of the cold start initialization of the model.
A second component of this study is to examine the areal coverage of convective precipitation using a binary verification of precipitation. The spatial correlation for each day (00z – 23z) for which convection was both predicted and observed is computed using domain shifting to determine forecast position errors. The KSS is computed to determine errors in the forecast areal coverage given the correct position. The results from this analysis are then quantified by plotting each case as a single point on a KSS versus distance error scattergram. The implications of these results for future forecast strategies are highlighted.
Poster Session 4, Mesoscale Predictability and Ensembles—with Coffee Break
Wednesday, 1 August 2001, 2:30 PM-4:00 PM
Previous paper Next paper