Abstract: Antecedent moisture condition and rainfall intensity are key factors that affect the runoff prediction accuracy of the Soil Conservation Service Curve Number (SCS-CN) model. Analyzing how the model behaviors depend on the two factors for different basins plays a crucial role in model improvement. This study uses partial correlation analysis and K-means cluster analysis to improve the SCS-CN model, based on the data of rainfall and runoff from five basins in semi-arid, semi-humid and humid regions. The results show that regrouping the antecedent moisture condition intervals for these basins improves its prediction ability greatly, and reduces its average deviation significantly, increasing the Nash-Sutcliffe efficiency by 42.8% on average. This model can also be improved by introducing the maximum 10 min rainfall intensity, which increases the efficiency to a certain extent and shows an improvement for the semi-arid and semi-humid basin better than that for the humid basin. The revised models have achieved good results for the study basin with an average deviation of less than 7 mm. Except for the Chengcun basin in the non-flood period, the Nash-Sutcliffe efficiency is over 0.93 or an average increase by 89%, and the Root Mean Square error is reduced by 29.2 mm.

Key words: runoff simulation, Soil Conservation Service Curve Number model, rainfall intensity, antecedent moisture conditions, curve number