ProbSevere LightningCast is a part of the ProbSevere portfolio of machine-learning algorithms for convective hazards. The main LightningCast model predicts the probability of any lightning (as observed by GLM) in the next 60 minutes. It uses GOES-R ABI channels as predictors and utilizes a U-Net convolutional neural network architecture. The model was trained with TensorFlow and uses Python. It can be run in near-real time or on archived cases.
Training a model (opens a Google Colab Notebook)