Soil- and root properties in driving preferential flow of north subtropical forest stands: insights from machine learning

Abstract

As an important water infiltration property in the forest hydrological responses, the occurrence of preferential flow could be mainly explained by soil- and root properties. However, the mechanism of soil- and root properties influencing preferential flow at different forest stands remains unclear. In this study, dye-tracing experiments, machine learning algorithms, and shapley additive explanations approach were employed to analyze their relationships at the three forest stands (pine, bamboo, and oak). The results showed that preferential flow index (PFI) increased with increasing soil depth. PFI in the topsoils (0–20 cm) was in the following order: oak > pine > bamboo, while PFI in the subsoils (20–50 cm) was pine > oak > bamboo. The results also indicated that the Support Vector Machine (SVM) model demonstrated the excellent performance in explaining the complex relationships between soil properties, root properties, and preferential flow. The root volume showed the highest contribution to the predicted PFI at the pine forest stand, while soil capillary porosity and root biomass at the bamboo and oak forest stands, respectively. Soil bulk density with low values, also initial soil moisture, soil capillary porosity, soil non-capillary porosity, clay, and sand with high values could interact with root properties, thus influencing the changes in PFI jointly. The findings offer scientific basis for soil–water management strategies in forest ecosystems.