Two to Six Hour Severe Local Storm Probabilities: An Operational Forecasting System

Abstract

The National Weather Service has developed an operational objective system which produces 2?6 h probability forecasts of severe IOM storms. The probabilities Pertain to square areas of about 85 n mi (155 km) on a side which cover most of the United States east of the Rocky Mountains. The forecast probabilities are based on multiple linear regression equations. The independent variables (predictors) am derived from objectively analyzed surface atmospheric variables observed hourly, forecasts of atmospheric variables above the surface from a large-scale numerical prediction model, manually digitized radar data, and localized climatic frequencies of the predictand. Certain predictors, such as modified forms of standard stability measures, are derived from the latest observed temperature and moisture variables at the surface together with numerical forecasts of these same variables in the lower troposphere. Each predictor is positioned optimally (in a climatic sense) relative to the predictand areas and is derived to bear a linear relationship to the predictand relative frequency. The severe storm regression equations are found to contain many of the predictive variables used subjectively by operational severe storm forecasters. Predictors derived partly or totally from observed surface data are especially prominent in the equations. Separate equations for two geographical regions of the United States exhibit differences in the relative importance of certain predictors. The most notable single difference is that predictors derived partly or totally from observed surface data are more important in the Midwest and High Plains than they am near the Gulf Coast. The statistical predictor-predictand analysis also yielded evidence that the technique of deriving predictors from a combination of the latest observed data at the surface together with forecast data aloft produces more predictive information than when these data we used separately. Limited quantitative verification of the operational forecasts issued during the spring season of 1977 shows that the probabilities were superior to either climatology or persistence. However, comparison of the verification scores for this independent sample with those from the dependent sample indicates come instability in the system. This instability is evidently due to the small development and verification samples used and the low frequency of occurrence of the predictand. Subjective verification of the probabilities for six randomly selected severe storm days showed that the system's ability to forecast storm occurrences is generally respectable. However, the quality of individual forecasts is related to the organization and strength of the synoptic system that produced the storms.