Machine learning integration of Sentinel-1 SAR and ground-based measurements for soil moisture estimation in temperate forested landscape of Somogy County, Hungary

Study region Somogy County is located in south-western Hungary (45.95°–46.50°N, 17.40°–18.00°E) and covers 6065 km². The region includes temperate forests, intensive agricultural land, and wetlands developed on diverse soils, including Eutric Luvisols, Cambisols, Albeluvisols, Acid Gleysols, and Aquic Gleysols. The climate ranges from subhumid to humid. Study focus This study integrated Sentinel-1 C-band SAR backscatter (VV and VH polarisations), soil hydraulic properties, SRTM-derived terrain variables, and meteorological data to predict soil water content using three machine-learning approaches: Random Forest (RF), Support Vector Machine (SVM), and K-Nearest Neighbours (KNN). In situ SWC observations were collected at 20 cm depth at three forest monitoring stations (Nagybajom, Barcs and Zselic) during the summers of 2024 and 2025 for model calibration and validation. Regional SWC maps were generated at 10 m resolution and further analysed using a Forest Vulnerability Index and a five-level management priority classification. New hydrological insight for the region Station-scale model performance was strong (R² = 0.866–0.913; RMSE = 0.018–0.022 m³ m⁻³; NSE = 0.838–0.891), while regional predictions showed moderate accuracy (R² = 0.532–0.648; RMSE = 0.033–0.041 m³ m⁻³), with KNN providing the best spatial transferability among the three models tested, though absolute spatial transferability remained limited under station-holdout cross-validation. Given the three-station, two-summer observation network, these county-scale maps serve as a pilot regional extrapolation rather than a fully validated product. VV backscatter, antecedent SWC, and saturated water content were the main predictors. Spatial patterns revealed higher drought sensitivity in sandy lowlands and greater moisture retention in silt-loam uplands, with a marked drying trend between 2024 and 2025, particularly in northern forest areas.