Long series of natural runoff data under current underlying surface conditions are the basis of water resources assessment. However, changes of the underlying surface and human activities often affect the hydrological cycle process, bringing systematic deviation and great uncertainty to water resources assessment. Taking the Xishui River Basin as an example, a water resources assessment model is proposed combining the runoff component investigation method and the SWAT model. Different water balance equations are established for runoff reconstruction according to the relationship between the locations of the hydrological station and the reservoir, and the parameters of the model is calibrated and verified using the runoff data which is reconstructed by the years with similar underlying surface conditions. Then the model is applied to the simulation of a long-series natural runoff process. The results show that there are obvious deviations in the consistency of the rainfall-runoff series of the hydrological stations reconstructed only by the runoff component investigation method. However, by constructing the SWAT model based on the current underlying surface conditions, the correlation between rainfall and runoff of the simulated runoff values of each hydrological station is better than the reconstructed values, and the consistency of the rainfall-runoff relationship in different periods is significantly improved, which indicates that this approach can effectively correct the impact of underlying surface changes on runoff, and is helpful for the further study of water resources planning and risk assessment of flood and drought disasters.