Abstract: Nonlinear rainfall-runoff processes limit the accuracy of flood forecasting in semi-arid and semi-humid regions,which can not meet the requirement of flood regulation and management.Taking the Liulin experimental watershed in Xingtai City,Hebei Province as the study area,analysis of the observed “rainfall-discharge-soil moisture profile” data indicates that saturation-excess runoff and infiltration-excess runoff rarely occur at large scales during most flood events,while interflow serves as the dominant runoff component.Runoff generation is influenced by soil moisture content and exhibits pronounced spatiotemporal variability.Considering these characteristics,a mixed rainfall-runoff model considering spatiotemporal variability of interflow is developed based on the Xin’anjiang model.In the model,a virtual free water reservoir is introduced to calculate infiltration-excess runoff,free water content is employed to derive outflow coefficients for water sources dividing,and a P-order parabolic function is used to describe the heterogeneity of underlying surface conditions.Simulation results of 24 typical flood events demonstrate that the proposed model outperforms the Xin’anjiang model,with the average Nash-Sutcliffe efficiency (NSE) improved by 0.05,the average correlation coefficient (R) improved by 0.02,the average relative flood peak error reduced by 2%,and the average relative runoff volume error decreased by 3%.The improvements are more significant for small to medium floods,and the model also better reproduces soil moisture dynamics.The findings of the study could provide theoretical significance and practical value for flood forecasting and regulation in semi-arid and semi-humid regions.