Random Forest 알고리즘을 이용한 밭 토양 환경에서 토양 함수율 예측 = Estimation of soil Moisture in Upland Field Using Random Forest Algorithm

다국어 초록 (Multilingual Abstract)

Soil moisture is a key variable affecting crop growth in upland agricultural fields.
This study developed and analyzed a data-driven soil moisture prediction model using machine learning techniques. A Random Forest regression model was trained to predict soil moisture at four different depths (10 cm, 20 cm, 30 cm, and 40 cm) based on environmental variables including soil temperature, ambient temperature, humidity, rainfall, and solar radiation. To ensure data reliability, soil moisture measurements were verified and adjusted to realistic ranges observed in upland sandy loam fields (approximately 10–35%). The dataset was preprocessed through outlier removal and standardization, and model performance was evaluated using RMSE and R2. Strong predictive performance was observed across all depths, with R2 values exceeding 0.80. The model achieved the highest accuracy at 20 cm depth (R2 ≈ 0.87, RMSE ≈ 3.4%), while performance declined at 40 cm depth (R2 ≈ 0.81, RMSE ≈ 5.9%), reflecting the increasing influence of subsurface water dynamics. These findings demonstrate the robustness and applicability of the Random Forest-based soil moisture prediction model, providing a solid foundation for the development of smart and precise irrigation systems.