Ensemble Subsetting of Convection-Allowing Ensemble Forecasts to Improve Thunderstorm Initiation Nowcasting

The Warn-on-Forecast project aims to extend severe thunderstorm and tornado warning lead times by incorporating high-resolution rapidly updating ensemble forecasts into the warning decision process. The volume of data from the ensemble forecasts likely requires that they be postprocessed to reduce the volume of data before being presented to the forecaster. The postprocessing typically involves creating products such as ensemble means and probabilistic guidance, so the forecaster does not need to individually examine all of the ensemble members. The objective of this work is to demonstrate and verify a system that generates guidance for warning decisions using a 36-member ensemble that is similar to the Warn-on-Forecast System. This postprocessing system uses ensemble sensitivity analysis (ESA) to determine where the forecast is most sensitive to perturbations in a state variable at an earlier time. The output is a subset of the original ensemble, containing members that best match the observations collected between the ensemble initialization time and the forecast time in areas with the strongest ensemble sensitivity. Our objective is to increase the skill of nowcasting thunderstorm initiation at lead times of 1-4 hours.

This system contains several innovations over prior ensemble subsetting systems. For example, forecast response regions are automatically defined in areas where the forecast is most uncertain using computer vision techniques, reducing the need for forecaster interaction in the ensemble postprocessing. Also, instead of only comparing ensemble members to a single state variable, several variables (temperature, moisture, sea level pressure, and the u and v components of the wind) are used to determine which ensemble members best match the observations. Preliminary results suggest that incorporating multiple variables when choosing the ensemble subset can result in forecast improvements over using a single variable. A description of the system and how it operates will be presented, along with verification using several severe thunderstorm events in the Southeast United States.