Data Repository for Next-Day Mean Wind Speed Forecasting in a High-Variability Coastal Microclimate: An Explainable Gradient-Boosting Benchmark and Predictability Analysis for Şarköy, Türkiye

This repository contains the derived model outputs, performance metrics, and feature importance analyses accompanying the manuscript "Next-Day Mean Wind Speed Forecasting in a High-Variability Coastal Microclimate: An Explainable Gradient-Boosting Benchmark and Predictability Analysis for Şarköy, Türkiye." Version 1.0.0 file descriptions: model_results_clean.csv: Contains the comprehensive performance metrics (RMSE, MAE, R², MaxAE) comparing the predictions of all tested models permutation_importance_rf.csv: Contains the permutation feature importances derived from the final Random Forest model on the training set. xgboost_feature_importance.csv: Contains the gain-based feature importances extracted from the optimized XGBoost model. rf_feature_importances_feature_selection_stage.csv: Contains the initial Random Forest feature importances calculated during the exploratory two-stage feature selection process. vif_results_feature_selection_stage.csv: Contains the Variance Inflation Factor (VIF) scores for the 23 features initially screened by the Random Forest threshold. It details the multicollinearity check where six features exceeded the VIF > 10 threshold, leaving the 17 features utilized in the reduced-set model testing. --- Version 2.0.0 adds the derived outputs of the updated/revised article: per-day forecasts of the 2024 hold-out (original_protocol_predictions.csv), of the nested rolling-origin evaluation over 2020–2024 (rolling_origin_predictions.csv) and of the predictor ablation (ablation_predictions.csv), the hyperparameters selected in each outer year (rolling_origin_params.json), every number quoted in the article (revision_results.json and results_NN.json), station metadata, and the tables written by the notebooks (tables.zip). The raw station records of the Turkish State Meteorological Service are not included.