Residual Matters: How Residual Learning Can Improve Climate Downscaling

Title: Residual Matters: How Residual Learning Can Improve Climate Downscaling Description:This repository contains the code and data used in the project "Residual Matters: How Residual Learning Can Improve Climate Downscaling." This project explores how residual networks, particularly models like EDSR and VDSR, can improve the downscaling of climate data. The goal is to enhance the spatial resolution of climate variables—such as ERA5 2m temperature data—using deep learning approaches. By leveraging residual learning techniques, this project achieves higher accuracy in climate downscaling, particularly in complex terrain regions, providing more reliable data for climate predictions and research. The repository includes: Python scripts implementing EDSR and VDSR architectures for temperature data downscaling. Requirements file (requirements.txt) listing all dependencies for reproducibility. Sample data preprocessing and model training scripts. Instructions for evaluating model performance. LICENSE file under the GNU General Public License (GPL) v3, permitting reuse and modification. License: GNU General Public License v3.0 Keywords: Climate Downscaling, Deep Learning, Super-Resolution, Residual Networks, EDSR, VDSR, ERA5 Usage Notes:To run the code, set up a Python environment using the dependencies listed in requirements.txt. The provided data files can be used as input examples, and the README offers detailed instructions on running the models. For further usage guidance, please refer to the README or the paper associated with this project. Acknowledgments: This work is conducted as part of research at IIT Mandi, focusing on improving climate downscaling techniques with deep learning.