Kizilirmak Delta Wetland Forage-Yield Dataset and Reproducible Analysis Code

This repository contains the field observations, satellite-derived predictor variables, reproducible analysis code and model outputs associated with the study “Improving Forage-Yield Estimation for Grazing Management in Wetland Rangelands Using Multi-Sensor Satellite Data Fusion and Machine Learning.” The dataset was collected in the Kızılırmak Delta, northern Türkiye, during the 2022 and 2023 grazing seasons. The original dataset comprised 468 field observations. Following satellite matching and complete-case filtering, 331 observations were retained for modelling, including 251 training observations and 80 independent test observations. The repository includes: the complete Google Earth Engine export; the model-ready dataset; corrected Google Earth Engine preprocessing and export code; reproducible Python scripts for data validation, Random Forest, Gradient Boosting Regressor and Extreme Gradient Boosting models; cross-validation and independent test-set results; correlation and permutation feature-importance analyses; observed-versus-predicted and residual diagnostic figures; data dictionary, methodological documentation and file checksums. The analysis compares Sentinel-2, Landsat-8/9 and multi-sensor fusion scenarios for estimating forage yield in heterogeneous wetland rangelands. The complete workflow can be reproduced by following the instructions provided in the README and QUICKSTART files.