4.1

**Forecasting lightning probability beyond nowcasting using NWP model output**

**William R. Burrows**, MSC, Edmonton, AB, Canada

Lightning is an episodic and highly variable phenomenon associated with convective clouds, and is not directly predicted in current operational weather prediction models. Due to the high required space and time resolution today's large scale operational numerical weather prediction models do not forecast convective clouds per se, but rather parameterize convection in terms of larger scale environment variables. For a now-casting time frame of a few hours reasonably good success can be achieved by extrapolating existing convective storms. However forecasting lightning occurrence beyond a few hours requires relating lightning occurrence to predictors derived from an NWP model output and sorting out the myriad decisions that arise in the determination of when and where lightning is likely to occur.

As an initial attempt to forecast lightning to 48 hours by statistical methods for Canada and adjacent portions of the United States, models valid May to September were developed by Burrows et al. (2006). These predict the probability of lightning in three-hour intervals using observations from the North American Lightning Detection Network and predictors derived from GEM model output at the Canadian Meteorological Center. Models were built with pooled 2000-2001 data using tree-structured regression. This is a modern statistical method which produces decision trees to explain predictand variance, is capable of handling many predictors, and can make probability predictions. GEM convective parameterization scheme variables were not used because they were not archived at the time, but it was still possible to build reasonably successful models using other predictors which are known to be related to convection and hence to lightning. Highest ranked predictors overall were the Showalter index, mean sea-level pressure, and troposphere precipitable water. Three-hour changes of 500 hPa geopotential height, (500-1000) hPa thickness, and MSL pressure, which are representative of frontal motion, were highly ranked in most areas. The three-hour average of most predictors was more important than the mean or maximum (minimum where appropriate). Several predictors outranked CAPE, indicating it must appear with other predictors for successful statistical lightning prediction models.

Predictions are available in real time at Canadian weather prediction centers. Verification of forecasts in 2003 and 2004 suggests a hybrid forecast method based on a mixture of maximum and mean forecast probabilities in a radius around a grid point and on monthly climatology will improve accuracy. The results showed tree structured regression to be a viable method for building statistical models to forecast lightning probability.

Forecaster feedback has been generally positive. More predictors are now available since the original study, in particular several that are output from the Kain-Fritsch convective scheme now used in GEM. Ongoing work at the time of this writing on new models that use these predictors shows prediction accuracy can likely be improved. A decision tree process will still be necessary because while the location of convection areas in GEM is generally good in a gross sense, the predicted areas of convection are often not coincident with observed lightning.

Recorded presentationSession 4, Assimilation of lightning data into forecast models

**Tuesday, 31 January 2006, 8:30 AM-9:45 AM**, A307** Next paper
**