Machine learning discerns large-scale atmospheric variability in rainfall δ¹⁸O over southwestern Europe

This repository contains the full computational workflow supporting the manuscript: “Machine learning discerns large-scale atmospheric variability in rainfall δ¹⁸O over southwestern Europe” The framework integrates ERA5 reanalysis data, GNIP precipitation isotope observations, IsoGSM isotope-enabled general circulation model simulations, large-scale atmospheric circulation indices (NAO and WeMO), and benchmark statistical approaches to quantify circulation–δ¹⁸O coupling across southwestern Europe. The repository includes ERA5 data download scripts (Copernicus Climate Data Store API), benchmark database construction routines, Random Forest model training and validation workflows, permutation-based feature-importance analyses, circulation–isotope diagnostics, figure-generation scripts reproducing all manuscript figures, and a fully specified Conda environment for reproducibility. All processed datasets, trained models, feature-importance tables, correlation diagnostics, and manuscript figures can be deterministically reproduced from the original data sources using the scripts provided. Due to redistribution policies and data volume constraints, ERA5 and IsoGSM datasets are not included in this repository and must be obtained independently from their original providers. The independent validation dataset compiled by Moreno et al. (2021) is publicly available at Zenodo (https://zenodo.org/records/4806413). Given identical input datasets and the provided environment file, all numerical results and figures are fully reproducible.