<![CDATA[Soil Water Holding Capacity (WHC) is a crucial hydrological parameter for tea plant productivity in hilly terrains. Conventional WHC mapping on a large scale is generally constrained by high operational costs and lengthy analysis time. This study aims to evaluate the performance of Random Forest Regression (RFR) and Multiple Linear Regression (MLR) algorithms in predicting the spatial distribution of WHC at the Wonosari Tea Plantation, Malang. Soil sampling was conducted at 16 observation points using a stratified purposive sampling method based on Land Map Units (LMU). To represent water retention capacity in the effective root zone, undisturbed soil samples were collected vertically at depths of 0–20 cm, 20–40 cm, and 40–60 cm at each point, analyzed using a pressure plate apparatus, and integrated into a single profile average value. Six spectral indices were extracted from Sentinel-2A imagery (NDVI, NDSI, NDWI, LSWI, MSI, NMDI) based on their sensitivity to surface moisture and canopy density, then combined with slope data (DEMNAS) as predictor variables. Given the limited sample size, the RFR model validation was performed using the Leave-One-Out Cross-Validation (LOOCV) method to ensure predictive stability. Results showed that the RFR model with a combination of three key variables (NDSI, MSI, and slope) achieved higher accuracy (R²cv = 0.423; RMSEcv = 1.47%) compared to the MLR model (R² = 0.371; RMSE = 1.59%). Feature importance analysis revealed that slope was the most dominant controlling factor (68.5%). This evaluation concludes that the RFR algorithm is more reliable than MLR for modeling the spatial complexity of WHC in hilly areas. The resulting prediction map effectively divides the plantation into three management zones (High, Medium, Low) to support precision irrigation strategies and soil conservation, potentially increasing operational cost efficiency by 30–40%.]]>
Copyrights © 2026
