Forecasting Convective Gusts from WRF 15km Model Data Using the New Random Nearest Neighbor Data Mining

The recently developed Random Nearest Neighbor data mining technique is used to forecast convective gusts (CG) using WRF model output. CG are one of the most challenging weather hazards to forecast, as operational model horizontal resolution is typically insufficient to resolve thunderstorms. Furthermore, forecasting the likelihood of CG exceeding thresholds is beyond current operational model capability, therefore, combinations of model scale thunderstorm stability and dynamic parameters must be combined to infer the likelihood of severe criteria (25.72 m/s) for CG. The Random Nearest Neighbor methodology is seen to uncover non-linear interactions between model predictors, and is used to improve the operational convective gust forecast of the Air Force 557th Weather Wing.