Search

Lightning-AI is a machine learning system that addresses the critical gap in lightning safety by providing predictive warnings before the first strike occurs. The technology uses a combined CNN/LSTM architecture to identify atmospheric signals that lead to lightning initiation and converts them into short-term probabilistic forecasts. The model ingests four sequential WSR-88D radar scans spaced roughly 5 to 6 minutes apart, incorporating polarimetric variables including horizontal reflectivity, differential reflectivity, and correlation coefficient that reveal mixed-phase microphysics and graupel growth driving cloud electrification. The radar data is transformed into a uniform two-kilometer grid creating a consistent spatial framework. Before processing, data is normalized and filtered to remove non-meteorological clutter. The model uses lightning initiation points from the Geostationary Lightning Mapper as ground truth observations to learn physical signatures of developing storms that appear minutes before the first flash. The CNN identifies spatial electrification patterns while the LSTM interprets their temporal evolution across sequential scans. Together, they detect subtle microphysical cues of impending lightning initiation, even before precipitation reaches the surface. This capability transforms lightning safety from reactive to proactive, offering more accurate threat identification. Validation results demonstrate the system can forecast lightning 15 to 30 minutes in advance, achieving an 84% probability of detection with a 22% false alarm rate. The algorithm operates in near-real-time using existing radar infrastructure and can integrate into commercial weather applications, emergency management systems, and automated alert platforms. Currently at TRL 5, Lightning-AI is available for patent licensing.