Deep learning model weights for zonal wind statistical downscaling using different loss functions.

This repository contains trained deep learning model weights for the statistical downscaling of 10 m zonal wind speed. The models were developed as part of a benchmarking study on the impact of training loss functions in statistical downscaling. The weights correspond to different training using different loss functions, and are provided for the reproducibility of the experiments. The weights can be loaded directly into the provided PyTorch code. Please refer to the associated publication and Github repository for details on model architecture, training procedures, and evaluation. If you use these weights, please cite the related publication:El Montassir et al., "Training loss matters: exploring new fit-for-purpose loss combinations for statistical downscaling," submitted to JGR: Machine Learning and Computation (2025).