Soil Water Dynamic Modeling Using the Physical and Support Vector Machine Methods

Abstract

Comparison of long‐term daily prediction of soil water content at different depths of soil coming from two models, one physically based (HYDRUS‐1D) and the other being soft‐computing technique (Support Vector Machine), has been done. The models used provide quite accurate prediction for soil moisture content, especially in the most important layers for crop growth—those close to the surface. The RMSE of estimations of the volumetric soil water content in topsoil made by the Support Vector Machine (ν‐SVM) based model was 0.035, 0.030, and 0.021 cm3 cm−3 for depths of 5, 10, and 25 cm, respectively, while the physically based model had RMSE values of 0.042, 0.049, and 0.045 cm3 cm−3 for the same depths. Our results showed that the ν‐SVM modeling approach can be used for estimations of soil water content changes in time with the accuracy comparable of to the one of the physically based models.