Abstract: The non-stationary nature of watersheds due to changing environments has invalidated the assumption that hydrological model parameters reflecting the physical characteristics of watersheds were time-invariant. To improve the performance of a hydrological model, it is thus necessary to consider the time variation of the model parameters. In this paper, the linear and nonlinear functions of five environment-related indicators (Normalized Difference Vegetation Index, population, gross domestic production, grain production, and irrigated area) were employed to describe the temporal variation of the parameters of the two-parameter monthly water balance model to reflect the impacts of vegetation change and human activities on hydrological processes. Impacts of four different scenarios of the temporal variation of two parameters were investigated for the upstream catchments of six hydrological stations, i. e., Zhimenda, Zhangjiashan, Xianyang, Hengjiang, Danjiangkou, and Xiajiang, by comparing the model performances under different scenarios. Results showed that the hydro-meteorological characteristics of the study catchments was non-stationary during the period 1982-2006. Compared with the baseline scenario with all time-invariant parameters, the model with time-varying parameters could improve the monthly runoff simulations and provide more skillful predictions. Specifically, the Nash-Sutcliffe efficiency coefficient (N_S) of the model with time-varying parameters improved by 10.3% and 8.8% in the calibration and validation periods, respectively, at the Zhimenda hydrological station. This research can provide the theoretical basis and technical support for reservoir operation and water resources planning and management.