Enagnon97/TL-ORB: A climate similarity-based transfer learning framework using global Caravan dataset for enhancing streamflow prediction in Ouémé River Basin

Code, model configurations, and results accompanying the manuscript: A climate similarity-based transfer learning using the global Caravan dataset for enhancing streamflow prediction in the Ouémé River Basin (Benin, West Africa) This release archives the full workflow used in the study: the construction of a composite climate similarity index (CI), the selection and ranking of tropical donor catchments from the global Caravan dataset, the pre-training and fine-tuning of LSTM models on three cumulative donor subsets (50, 100, and 150 basins), and their evaluation against a local-only baseline at five stations of the Ouémé River Basin. Contents configs/ — neuralhydrology YAML configurations (pre-training, fine-tuning, and local baseline) notebooks/ — reproducible workflow (index construction, basin selection, training, evaluation, and figures) data/ — Ouémé catchment files and donor basin lists (external datasets referenced in data/README.md) results/ — performance metrics and manuscript figures Data sources Caravan global dataset (Kratzert et al., 2023): github.com ERA5-Land forcings via the Caravan framework LSTM implementation: neuralhydrology (github.com) Observed streamflow: Direction Générale de l'Eau (Benin) Reproducibility Trained models can be regenerated from the provided configurations and input data by running the notebooks in order. See README.md for the environment setup and execution steps. Citation If you use this repository, please cite both the manuscript and this archived release (DOI provided by Zenodo upon publication).