Predicting thunderstorm risk probability at very short time range using deep learning

Abstract. Forecasting electrical activity within the atmosphere remains one of the most challenging predictions, especially due to the chaotic nature of thunderstorms. Lightning strikes are precisely located and occur very quickly, which makes this task particularly difficult. Additionally, these phenomena pose a significant risk to aviation, as they statistically strike each aircraft more than once per year. Over the years, several techniques have been employed for very short-term lightning forecasting (lower than 1 h and every 5 min), such as observation-based methods and, more recently, deep learning methods. Previous studies often face difficulties in accurately forecasting lightning probability, and even with AI-driven methods, it is still difficult to obtain calibrated outputs. To address this limitation, we propose a methodology that successfully predicts lightning risk using Convolutional Neural Networks (CNNs) with attention mechanisms. The network is fed with satellite observations and Numerical Weather Prediction (NWP) outputs formatted as a spatio-temporal sequence. Results show a F1 score of 0.65 for 5 min predictions and 0.5 for 30 min predictions with a very low Expected Calibration Error (ECE) of less than 10 %. Thanks to the well-calibrated outputs, risk probability maps can be plotted, showing areas with strong to low chances of having electrical activity.