Integrating machine learning models with ground sensors to enhance soil moisture prediction in agroecosystems of Texas

This study aims to enhance soil moisture prediction by integrating in situ observations with machine and deep learning models. Soil moisture was monitored across crop and pasture agroecosystems of Prairie View A&M University’s research farm using TEROS soil moisture sensors. Bio-meteorological data, including evapotranspiration, soil temperature, rainfall, relative humidity, and soil heat flux, were obtained from Eddy Covariance Flux Towers at crop and pasture agroecosystems. These biometeorological variables were used as input features to predict soil moisture. The machine learning models included Random Forest, Support Vector Regression, Artificial Neural Networks, and Extreme Gradient Boosting, while the deep learning models comprised Deep Neural Networks and Long Short-Term Memory. Gini impurity and SHAP analyses identified air temperature, ecosystem respiration, and soil heat flux as key predictors of soil moisture in the machine and deep learning models. Hyperparameters for each model were optimized using the grid search method for each agroecosystem. Furthermore, the validation protocol employed 10-fold cross-validation (k = 10) with shuffled folds and a fixed random seed to ensure reproducibility. A bootstrapping approach was applied to quantify the uncertainty associated with each machine and deep learning model. The machine learning models demonstrated strong predictive capabilities, with performance assessed using Root Mean Square Error (RMSE), Mean Squared Error (MSE), and the coefficient of determination (R 2 ). Among machine learning models, Random Forest and Extreme Gradient Boosting exhibited superior performance, with R 2 values ≥ 0.90 and RMSE ≤ 0.01 m 3 m −3 . The Long Short-Term Memory has comparable soil moisture prediction skills (R 2 = 0.90 and RMSE = 0.021 m 3 m −3 ) to that of Random Forest and Extreme Gradient Boosting. The methodology and findings from this study can inform better irrigation practices, agricultural drought management, and agroecosystem management.