Improving Predictions of Convective Storm Wind Gusts through Statistical Post-Processing of Neural Weather Models

This dataset contains relevant data used for post-processing neural weather model as part of above named article. Included are:

• for CNN, ANN, and ViT, the five set of parameters fitted using a k-fold cross validation scheme on years 2016 to 2020. Each pickle file contains a list of 5 dictionaries of paramaters, where each dictionary corresponds to one fold (first item corresponds to model trained on year 2017 to 2020 and validated on year 2016, etc...)
• for VGAM, a compressed folder with RData files whose names are "{fold}{leadtime}{cluster}.rds", which each contains the VGAM parameters for the corresponding cluster and lead time, with again a k-fold cross validation scheme.
• for all models, the files "{model}_crps.pkl" contain numpy.ndarray of shape (folds, clusters, lead time, date) with the computed crps for each case.