Efficient agricultural water management is essential to address global water scarcity. Soil moisture serves as a key indicator for irrigation scheduling, crop yield forecasting, and hydrological modeling. However, traditional in-situ measurements are costly, labor-intensive, and spatially limited. This study develops a machine learning framework for predicting soil moisture at multiple depths (20, 40, and 80 cm) using readily available environmental variables. Türkiye was selected as the study area due to its pronounced climatic and pedological heterogeneity, providing an ideal testbed for training robust models. Seven machine learning algorithms were systematically evaluated using data from 201 meteorological stations spanning 2016–2024. Extreme Gradient Boosting (XGBoost) outperformed all alternatives following hyperparameter optimization, achieving strong accuracy across all depths (R² = 0.74, 0.69, and 0.66 at 20, 40, and 80 cm, respectively). Feature importance analysis indicated a depth-dependent shift in the controls on soil moisture. Near the surface (20 cm), short-term meteorological variability and seasonal dynamics were the primary drivers, whereas deeper layers were increasingly regulated by stable soil hydraulic properties, with clay dominating at intermediate depths, and large-scale spatial gradients represented by latitude and elevation. This pattern reflects a transition from weather-driven processes at shallow depths to long-term regulation by soil structure and regional climate patterns in the subsurface. The model maintained robust performance across diverse environmental conditions, with seasonal accuracy varying by nearly 10% between winter and autumn. Spatial analysis revealed regional variations in controlling factors: soil properties (particularly organic carbon) dominated in the northern regions, while their relative importance decreased southward where climatic and temporal variables contributed more substantially, reflecting heightened sensitivity of moisture dynamics to meteorological fluctuations. The framework provides a scalable, cost-effective solution for soil moisture monitoring in data-scarce regions, supporting irrigation optimization, drought early warning, and sustainable water governance.
Muhammed Sungur Demir (Tue,) studied this question.