Comparison of 36-60 Hour Precipitation Forecasts from Convection-Allowing and Convection-Parameterizing Ensembles

NOAA/Hazardous Weather Testbed Spring Forecasting Experiments (SFEs) have typically emphasized “day 1” forecast guidance, which corresponds to the 12-36 hour forecast period of a 0000 UTC initialized forecast. However, during the 2014 SFE, an attempt was made to generate “day 2” forecast products using 36-60 hour guidance from a convection-allowing ensemble provided by the Center for Analysis and Prediction of Storms at the University of Oklahoma. Because numerous previous studies have shown that convection-allowing guidance significantly improves upon coarser convection-parameterizing guidance in the day 1 forecast period, the goal of this study is to objectively evaluate whether the advantage of convection-allowing models translates into the day 2 forecast period. For this evaluation, the ensemble run by CAPS, known as the Storm Scale Ensemble Forecast (SSEF) system, is compared to NCEP's operational Short Range Ensemble Forecast (SREF) system. Perturbations extracted from the SREF system members are used for initial condition perturbations in the SSEF system, and the SREF forecasts are used as lateral boundary conditions in the SSEF. Thus, the comparisons are somewhat “clean”, but of course many other aspects of the ensemble configurations are different. Precipitation forecasts are examined and a variety of deterministic and probabilistic metrics are used for the evaluations. The first metric that was used to evaluate the forecast was the Equitable Threat Score (ETS). The ETS was computed for five different precipitation thresholds ranging from 0.10" to 1.00" for each individual SSEF and SREF member for all 60 forecast hours in 3 hour intervals. It was found that the SSEF forecasts were clearly superior to the SREF forecasts, especially at the thresholds of 0.50", 0.75", and 1.00". However, not much forecast skill (ETS<0.15) was observed at the higher thresholds. Next, a qualitative analysis of Hovmoller diagrams of diurnally averaged rainfall was completed for each ensemble member of the SSEF and SREF. In addition, spatial correlation coefficients (Clark et al. 2007) were computed for each ensemble member in time-longitude space. It was found that there was a three hour phase lag between most of the SREF member forecasts and observed precipitation, while the SSEF member forecasts depicted the diurnal cycle very well, even in the 36-60 hour time frame (with the exception of some members slightly overestimating precipitation). Examination of other metrics is currently underway.