Enhancing Agricultural Sustainability using AI-Driven Soil Moisture Modeling: A Soil-Type and Depth Approach with SHAP Interpretability

The accurate prediction of soil moisture content (SMC) is important for optimizing irrigation, reducing water wastages and enhancing sustainability in agriculture. This study developed a Random Forest Regression model for soil-depth-specific prediction of SMC during two vegetation seasons. The model was applied to two soil types (loam and silt loam) at five depths with two different scenarios based on the used inputs: the first used only vegetation indices and the second integrated meteorological data with the vegetation indices. The results showed a significant rise in model’s accuracy in the second scenario in both soil types at all depths, highlighting the importance of integrating meteorological features. In loam soil, R2 increased from 0.65, 0.61, and 0.82, in the first scenario, to 0.94, 0.83 and 0.87, in second scenario, at 5, 20 and 40 cm depth, respectively. Similarly in silt loam, at 5, 20 and 40 cm depths the R2 in the second scenario improved to get an R2 of 0.97, 0.96 and 0.94, respectively, compared with an R2 of 0.88, 0.94 and 0.82 in the first scenario at same depths, respectively. SHAP (SHapley Additive ExPlanations) research revealed that the most influential features on SMC prediction were precipitation, humidity, Normalized Difference Vegetation Index (NDVI) in loam soil, and solar radiation and NDVI in silt loam. These results emphasized that integrated meteorological data increases the model’s performance in SMC prediction, and the importance of SHAP explainability for enhancing model interpretability and support real-time irrigation decision making. This research allows for better water resource management and enhances sustainability.