Effects of different land-use types on soil detachment capacity in loess areas

•Hill curve is the optimal curve to describe root density and Dc.•Shrubland is the best land-use type in the study area.•The meta-model of Dc influencing factors was established by PLS-SEM.•A dimensionless comprehensive model for predicting Dc was established. Against the background of similar geomorphological characteristics and environmental temperatures, land-use types alter the soil detachment capacity (Dc) by changing soil properties and root traits. However, there is a lack of in-depth discussion regarding the response of Dc to land-use types in loess areas. The purpose of this study was to quantify the influences of different land-use types in loess areas on Dc, soil erodibility (kr), and critical shear stress (τc) and explore the key factors affecting Dc under runoff erosion based on the links between root traits and soil properties, and combined with PLS-SEM algorithm, a meta-model of Dc influence mechanism was established. Two hundred and seventy undisturbed soil samples were collected from six different land-use types and scoured under nine specific combinations of flow rate and slope (flow rate: 0.167–0.5 L s−1, slope: 10.5–20.8 %). The results showed that with the change in land-use type, Dc changed considerably. The average Dc of the shrubland was only 1/22 of that of the cropland. The change trend of kr was consistent with that of Dc, while the change trend of τc had no obvious regularity. The relationships between the root density parameters and Dc were simulated using the exponential function and Hill curve, respectively, and the predictive effect of the Hill curve was better than that of the exponential function. Compared with the other comprehensive hydraulic parameters, shear stress (τ) was Dc’s most effective predictive factor. Correlation and path analyses revealed that soil organic matter (SOM) and root length density (RLD) were the best indicators of Dc. Furthermore, a dimensionless Dc prediction model suitable for different land-use types was established based on τ, SOM, and RLD, and the model exhibited a high accuracy (R2 = 0.955, NSE = 0.911).