To address this dilemma, NOAA Forecast Systems Laboratory has developed a convective probability forecast product based on the Rapid Update Cycle (RUC) model. Known as the RUC Convective Probability Forecast (RCPF), this product utilizes an ensemble approach to generate thunderstorm probability information, based on the hypothesis that at long lead times (> 2 h), a forecast of convection at a specific point from a single deterministic model run is statistically less likely to be correct than a well chosen ensemble of model predictors. Several methods for creating an ensemble exist, including the use of different model systems, perturbed initial and boundary condition, perturbed model physics, time-lagged forecast from a single model, and adjacent model gridpoints.
The initial RCPF product has utilized information from adjacent grid points to aggregate the model convective information to larger time and space scales. Given a distribution of model-predicted convective precipitation values, convective probabilities are determined at each model gridpoint by computing the fraction of model grid points within a specified distance that have convective precipitation exceeding a specified threshold. Using this technique, 3-, 5-, and 7-h RCPF products were created in a real time during the summer of 2003. Verification of RCPF-derived categorical forecasts against the National Convective Weather Diagnostic (NCWD) was performed within the Real-Time Verification System (RTVS). Analysis of the verification statistics for a two-month period (July and August 2003) indicates that the RCPF provides useful strategic thunderstorm likelihood information. Enhanced skill was noted for early morning forecasts of initial convective development. These early morning forecasts are extremely important to traffic flow managers, who must make decisions about strategic reroutes of transcontinental flights at this time.
Planned work for 2004 includes: 1) adding additional ensemble information including information from time-lagged ensembles, multiple-model ensembles and ensemble closure information from the cumulus parameterization; 2) utilizing additional model predictor fields, such as lifted index, vertical velocity, and terrain; and 3) including a potential convective storm top field (determined by the level of convective overshoot in the model sounding). At the conference, results from the summer 2004 real-time test will be presented. Long-term plans call for merging the RUC model-based convective probability forecast with shorter-range (0-2 h) radar-derived products to create a seamless 0-6 h convective probability product.