Data and code for: Daily Streamflow Forecasting through Multi-Source Data Fusion and Ensemble Learning: The Role of Gauged-Record Completeness in Multi-Station Modelling

Processed data and code of a study on one-day-ahead daily streamflow forecasting in the Marmara, Susurluk, and Meriç-Ergene basins (northwestern Türkiye), in which the full analysis was repeated for four minimum levels of record completeness (revised manuscript ERC-106225, Environmental Research Communications). The archive contains three zip files. data.zip holds the four modelling datasets (60%, 70%, 80% and 90% minimum record completeness; 71, 66, 59 and 45 gauges; 1 October 2004 to 30 September 2021; Parquet format); the digitized DSİ daily discharge of all 156 candidate gauges, with dry days and days without measurement coded separately; gauge metadata, drainage-area checks and catchment polygons; the Landsat-7/Sentinel-2 harmonization coefficients; and a data dictionary. code.zip holds the complete workflow: digitization of the DSİ Streamflow Observation Yearbooks, catchment delineation, retrieval of the NASA POWER, Landsat-7, Sentinel-2 and ERA5-Land predictors, construction of the four datasets, model training, SHAP-based predictor reduction, source ablation, and the statistical analyses, tables and figures of the article. outputs.zip holds the daily forecasts of persistence, climatology, XGBoost, LSTM, Transformer, stacking, and blending for eight predictor configurations, together with the model summaries, SHAP results, and selected hyperparameters read by the analysis scripts. A README and SHA-256 checksums are included. The data are licensed under CC BY 4.0, and the code under the MIT License.