Cloud-to-Ground Lightning Nowcasting from Remote Sensing and Numerical Weather Prediction Modeling Data

In this paper a multidisciplinary approach to develop a regression algorithm to nowcast cloud-to-ground lightning activity one hour in advance is presented. The algorithm is developed using atmospheric variables retrieved from remote sensing, numerical weather prediction model outcomes and lightning location systems’ data. Gaussian Process Regression is used to estimate the possible dependence among the available meteorological variables and the lightning activity. Separate models are created for positive and negative strokes and for strokes over land or sea areas. The performed tests suggest that the proposed model quite accurately estimates low numbers of strokes whereas larger numbers of strokes are estimated with errors, mostly of underestimation. Nonetheless, since the presence of strokes is correctly nowcasted, the detection of their presence may be considered a useful information for activities related to the weather hazard management.