Characterization and prediction of soil hydraulic and solute transport parameters using a random forest model

To effectively control nonpoint source pollution and predict its transport and load, understanding water and solute transport processes, patterns and mechanisms is essential. However, the measurement of water movement and solute transport parameters is usually a laborious and time-consuming task. It is important to predict water movement and solute transport parameters from more readily available soil physical and chemical properties. In this study, a database of soil hydraulic and solute transport parameters containing information retrieved from 83 published studies on soil properties, land use, management measures, etc. was established to characterize and simulate soil water movement and solute transport processes. Our results showed that the soil particle composition was closely related to all soil hydraulic and solute transport parameters. As the soil texture changed from sand to clay, the soil residual water content (theta r) obviously increased. The soil porosity was significantly positively correlated with theta r, saturated water content (theta s), van Genuchten parameter alpha, dispersity (lambda) and retardation factor (R) and negatively correlated with van Genuchten parameter n (p < 0.01). The use of random forest models allowed the prediction of soil hydraulic and solute transport parameters by inputting common or even incomplete soil property parameters. The bulk density and particle composition jointly contributed 66% to the prediction of the soil hydraulic parameters and 44% to the prediction of the solute transport parameters. The pH exerted a notable impact on the solute transport parameters, especially dispersion coefficient (D), lambda and R. The results may be useful in providing data support and facilitating the development of watershed nonpoint source pollution models.