Comparative Evaluation of Machine Learning Algorithms for Predicting Soil Wetting Front Dynamics Under Drip Irrigation System

Accurate prediction of wetted width and wetted depth is essential for optimizing water use efficiency in drip irrigation systems. Existing empirical models are often restricted to specific soil textures and cannot adequately capture the complex nonlinear interactions among soil hydro-physical and chemical properties, irrigation variables, and different soil textures. This study evaluated four machine learning algorithms—Linear Support Vector Machine (Linear SVM), Medium Gaussian Support Vector Machine (Medium Gaussian SVM), Matern 5/2 Gaussian Process Regression (GPR), and Boosted Tree Regression—for predicting wetted width and wetted depth in sand and sandy loam soils. Model inputs included emitter discharge, irrigation duration, and selected soil hydro-physical and chemical properties. Models were developed using a 70% training dataset and validated with the remaining 30%. The Matern 5/2 GPR achieved the highest training accuracy for wetted width (R2 = 0.99; RMSE = 0.74) and wetted depth (R2 = 0.98; RMSE = 0.90), but validation errors increased to RMSE values of 2.27 and 3.84, respectively. Medium Gaussian SVM yielded the lowest validation RMSE (2.11) for wetted width, whereas Boosted Tree Regression achieved the best wetted depth prediction (RMSE = 2.11; MAE = 1.69). These findings demonstrate the importance of model-specific selection for reliable irrigation management.