An advective-statistical technique (ADSTAT) is currently being used within the National Weather Service (NWS) to operationally produce forecasts of rainfall amount and cloud-to-ground (CG) lightning in the 0-3 h timeframe. We are extending this approach to the prediction of severe local storm phenomena. The goal is to provide forecasters with automated monitoring of multiple remote-sensor platforms, surface observations, and numerical model output for the potential for hazardous convection events.

To develop the ADSTAT technique, forecasts of radar reflectivity and cloud-to-ground (CG) lightning flash rate were made for boxes on a 40-km map grid by advection of initial-time fields, for a large number of historical cases. The extrapolated reflectivity and flash rate values were then treated as candidate predictors of severe storm events in a statistical regression procedure. Additional candidate predictors from the NCEP Eta model and from surface observations were added to the remote sensor-based set. All predictors were then collated with the statistical predictands, which are severe storm events reported within the grid box during the valid period.

A forward-selection screening regression procedure was used to derive equations relating the predictors to the probability of severe weather. Predictors selected by the regression procedure included both radar-derived and storm environment information, such as the forecasted areal coverage by radar echoes greater than 55 dBZ, 500-mb wind speed, and the Total Totals index. The equations are applied to real-time extrapolation forecasts and Eta model output to produce probability forecasts, specifically the probability of any severe event, and the probabilities of large hail, damaging winds, or tornadoes as discrete events.

Input to the algorithm includes a 10-km national radar reflectivity mosaic, 15-minute lightning flash rates, and stability and humidity forecasts from the NCEP Eta Model. The 700-500 mb layer-mean wind field was used as the advection velocity field for radar and lightning features. Stability and humidity indices were obtained from the Eta model analysis or 6-h forecast valid closest in time to the latest radar image, and from the previous hour's surface observations.

Our aim is to produce probability forecasts conditional on the occurrence of convection within the grid box. Such forecasts are potentially more useful than absolute probabilities, since they provide information in situations where thunderstorms are unlikely, but any storms that do occur have a high potential for being severe. Therefore the sample of cases was limited to those map grid boxes in which CG lightning or severe phenomena were observed. An absolute severe storm probability forecast can be derived by multiplying the conditional severe storm probability by the 0-3 h lightning probability.

Initial evaluation indicates that the technique has useful skill; areas in which the absolute general severe storm probability exceeds 10% correspond well with watch boxes issued by the NWS Storm Prediction Center. Further statistical evaluation of the forecasts is ongoing. The algorithm is slated for operational implementation at a central site, with national dissemination of forecasts to NWS field offices and to external users.