Mesoscale lightning threat guidance for operational use at NWS offices

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Wednesday, 26 January 2011
Mesoscale lightning threat guidance for operational use at NWS offices
Washington State Convention Center
Pete Saunders, Florida State University, Tallahassee, FL; and H. E. Fuelberg, S. J. Hodanish, J. Mittelstadt, A. I. Watson, and S. Zubrick

We have used the perfect prognosis (PP) scheme to develop lightning forecast guidance products for three diverse regions of the country (Colorado, Washington, D.C., and Oregon) during the warm season (May-September). The guidance products were developed on a 10 10 km grid at 3-hourly intervals. Four seasons of RUC20 analyses and NLDN data were used to develop binary logistic regression (BLR) equations for predicting one or more flashes (PROB ≥ 1). And, negative binomial regression (NB) was used to develop equations to predict the amount of lightning (PROB ≥ T) conditional on one or more flashes occurring during each 3-h period. Our procedure is based on Shafer and Fuelberg (2009) who considered Florida where the sea breeze is dominant forcing mechanism for warm season thunderstorms. This proposed paper develops and evaluates the procedure when applied to areas with less well defined forcing.

RUC analyses of geopotential heights were used with a map typing procedure to develop candidate predictors of lightning frequency for five dominant flow regimes. The goal was to capture small-scale lightning features due to local forcing that are not well resolved by NWP models. The map-type frequencies were used as candidate predictors for both the BLR and NB models. Numerous RUC-analyzed parameters describing moisture, temperature, wind, and stability also were considered as candidate predictors in the PP equations. A principle component analysis was employed as an objective method to select a subset of the most physically relevant predictors that were not mutually correlated. Finally, a combination of forward stepwise regression and cross-validation was used to select the best set of predictors that would generalize to the dependent data set.

We evaluate the results for each region using independent data during the 2009 warm season. Output from the National Centers for Environmental Prediction (NCEP) 13-km RUC (RUC13) and the NCEP 12-km North American Mesoscale Model are employed. Evaluation parameters include Brier Scores, Brier Skill Scores, and reliability diagrams. The objective is to determine whether forecasts from our procedure are superior to those from a combination of climatology and persistence alone, and thus exhibit skill. Results show that there is good agreement between the model forecasts and observed lightning verification for most 3 h forecast periods in all three regions. Brier score analysis indicates that our models beat climatology in forecasting one or more flashes, as well as the amount of lightning. However, the reliability diagrams indicate that while our scheme performs well for forecasting one or more flashes, the amount of lightning often is over- or under-forecast. Thus, results for the three regions are not as good as those from Florida; however, this was expected since Florida has a much better defined forcing mechanism, the sea breeze. We currently are working to make the technique a fully operational guidance product that can be input to the NWS's Graphical Forecast Editor (GFE) in the three study regions. Our proposed paper will give a complete description of our procedures and findings.

Shafer, P. E., and H. E. Fuelberg, 2008: A perfect prognosis scheme for forecasting warm-season lightning over Florida. Mon. Wea. Rev., 136, 1817-1846.