Due to the vast expanse of Alaska covered with forests, fires in the sparsely populated state have the potential to reach tremendous sizes, capable of devastating vast areas of wilderness. Since lightning is a major cause of the fire starts, it is important to have accurate predictions as to where and when thunderstorms capable of sparking fires will form. With accurate predictions, fire prevention and fighting agencies will be able to respond more quickly and efficiently to potential fire threats.

To make these forecasts, the perfect prognosis (PP) technique was employed to produce gridded probabilistic forecasts for cloud-to-ground (CG) lightning. PP has the advantages of being model independent and is a computationally efficient method to quickly produce forecasts. In the PP approach, several hundred candidate predictors are reduced to a dozen primary factors using principal component analysis. Logistic regression is then used to produce the forecasts for each grid point.

Eight years of training data were used to develop the forecast equations. These came from North American Regional Reanalysis data archived at NCDC from 2000 to 2007, and a lightning climatology developed from the Bureau of Land Management lightning detection network data covering the same time period. For this work, real-time prediction data from the Global Forecast System (GFS) model from June to August 2008 is used as input into the forecast equations. Forecasts are made for each model cycle every 3 hours out to 180 hours for ≥1, ≥3, and ≥10 CG flashes on a 45 x 45 kilometer grid. Equations for ≥30 and ≥100 flashes were also developed, but the forecast probabilities for these relatively rare events were judged to have little discrimination between events and non-events. Results and verification for the 2008 Alaska lightning season will be presented at the conference.