This repository accompanies the paper Understanding the Role of Weather Data for Earth Surface Forecasting using a ConvLSTM-based Mode: link
code: it includes the code used to train the model (checkcode\readme.mdfor instructions on how to retrain a model)models: it includes the path to the trained model used to obtain most of the predictions reported in the paper and in this repository (checkmodels\readme.md)supl: supplementary materials which are not be included in the paper, checksupl\readme.mdfor a full descriptionsupl/predictions: RGB & NIR predictions for five random samples together with the corresponding ground truthsupl/simulations: RGB & NIR predictions for five random samples, shown as animations, obtained by individually perturbing each of the meteorological conditions. Additional to the results reported in the paper (which focused on rainfall only), we include here also those for sea level pressure and temperature.