Remote Sensing-Based Spatio-Temporal Fusion for Agricultural Drought Mapping in Central Anatolia

Agricultural drought is a major challenge in semi-arid Central Anatolia, Türkiye, where water scarcity, climate variability, and intensive farming affect crop productivity. Field-based monitoring is often spatially limited and cannot provide timely regional drought information. To address this limitation, this study developed a remote sensing-based spatio-temporal fusion framework for agricultural drought mapping in Ankara, Eskişehir, and Konya. Sentinel-2 surface reflectance data were used to derive monthly NDVI, NDWI, NDMI, and SAVI composites for the March–July growing season during 2020–2025, while cloud-contaminated pixels were removed using Sentinel-2 scene classification information. ERA5-Land precipitation, potential evapotranspiration, air temperature, and soil-moisture variables were integrated at their native grid scale, and Sentinel-2 indices were spatially summarized within ERA5-Land grid cells to ensure consistent spatial support. A standardized water-balance-based drought indicator was used to generate drought-stress labels. Random Forest, XGBoost, and LightGBM were evaluated using spatially blocked cross-validation and independent temporal validation on 2025 observations. LightGBM achieved the best independent 2025 performance, with a balanced accuracy of 0.655, macro-F1 of 0.658, ROC-AUC of 0.730, and PR-AUC of 0.653. Its macro-F1 exceeded that of XGBoost by 16.69 percentage points. The March 2025 drought-probability map represented spatial drought-risk patterns across the selected provinces, while permutation importance identified SAVI, NDMI, and month as the most influential predictors. Overall, the results show that integrating Sentinel-2 vegetation and moisture dynamics, ERA5-Land climate variables, and machine learning provides a scalable and interpretable approach for agricultural drought-risk mapping in semi-arid environments, supporting precision agriculture and climate-resilient land management.