In order to divide runoff of urban river without upstream incoming water and further explore the impact of urban underlying surface on rainfall-runoff relationship， several automatic division methods of flow process line are introduced to divide the daily runoff of Zhongmu Hydrological Station from 2003 to 2016， then the urban river runoff is divided into annual base flow and annual direct runoff by using the selected optimal division method. The annual direct runoff coefficient is used to express the urban water catchment characteristics， and the five-phase land use data are combined to construct a multiple linear regression model of the annual direct runoff coefficient and various underlying surface areas， the annual direct runoff coefficients for 2017 and 2018 are predicted. The results show that the digital filtering method F1 has the best division effect on urban river runoff without upstream incoming water； the contribution rates of base flow and direct runoff to the total runoff growth are 83.96% and 16.04% respectively； the multiple linear regression model is significant at 0.01 level and has good performance， where NSE， R ²， and R_e are 0.891， 0.891 and 6.72% respectively. Moreover， the relative errors of annual direct runoff coefficient forecast for 2017 and 2018 are 4.80% and 9.75% respectively.