Comparison of soil erodibility factors in USLE, RUSLE2, EPIC and Dg models based on a Chinese soil erodibiity database

Soil erodibility (K-value) is a key parameter in erosion prediction and is important for conservation planning in the face of a rising need for protecting the limited land resources. This study investigated the predictive capability of the K-value estimated by Universal Soil Loss Equation (USLE), Revised Universal Soil Loss Equation (RUSLE), Erosion Productivity Impact Calculator (EPIC) and Dg models for different soil regions using a Chinese soil erodibility database covering 51 natural runoff plots. Model performance was evaluated using R2 (coefficient of determination), relative error (RE), Nash-Sutcliff efficiency (NSE) and P value (Mann Whitney U test) statistics. The results showed that the existing four models overestimated almost all the K-values for the Chinese erodibility database, with most observed values concentrated in the range of 0.015 0.035. Without calibration, only the USLE and Dg models could be reliable and directly applied for the black soil region and the loess soil region, respectively. The Dg OM model (R2 0.67, n 32) was established by the nonlinear best fitting techniques of multiple regression. In the Dg OM model, K-values accounted for the vibration in a combination of the Dg (geometric mean diameter) and OM (soil organic matter). NSE, R2 and the average RE was 0.94, 0.67 and 9.5% for the Dg-OM model’s calibration based on the Chinese erodibility database; similar results were found for the validation process, with NSE of 0.93, R2 of 0.66 and average RE of 6.5%. The model performances showed that the Dg OM model reached ‘good’ satisfactory level. Compared with the four existing erodibility models, the Dg-OM model permitted the best parameterization and accuracy, and was proved to be suitable for estimating soil erodibility values in China.