Suspended sediment is a key driving factor in river evolution and ecosystem stability. However, there is a significant change in the annual suspended sediment loads in many basins under a current rapidly changing environment. Therefore, it is highly demanding to clarify the influencing factors and their attributions for the suspended sediment, thereby improving control measures of soil and water loss in basins. Taking the Ganjiang River in the Poyang Lake Basin of China as an example, the forest coverage rate increased by nearly 40% in the last 30 years, where there were 139 large and medium-sized reservoirs in 2014. In this study, the annual suspended sediment of the Ganjiang River Basin was firstly determined during the period of 1958-2014. Then, time-varying moment models were constructed for different types of probability distributions, where the factors of rainfall erosivity, forest coverage, and reservoir regulation were taken as explanatory variables. Finally, the Bayesian Model Averaging (BMA) was selected to combine four time-varying moment models for the simulation of annual suspended sediment load, and thereby to separate the attributions of different factors. The results were as follows: 1) Rainfall erosivity, forest coverage, and reservoir regulation were all key factors affecting the suspended sediment load. The time-varying moment model was constructed using four probability distributions (Gamma, Pearson-Ⅲ, Weibull, Lognormal) considering the time-varying of location and scale parameters, where the rainfall erosivity (Re), forest coverage rate (Fc), and reservoir regulation coefficient (Rc) were utilized as physical covariables. The distribution characteristics and process changes of annual sediment load were reasonably described in the study areas, with a Nash-Sutcliffe Efficiency coefficient (NSE) of 0.93; 2) BMA improved the simulation effect, while reduced model uncertainty, compared with the single time-varying moment model. The Root Mean Square Error (RMSE) and Relative Error (RE) of simulated value were 125.96×104 t/a, and 11.41%, respectively. The Containing Ratio (CR) and average band-width (B) of uncertainty interval were 94.47%, and 481.38×104 t/a, respectively. A fixing-changing method was selected to perform the attribution analysis of suspended sediment load using BMA under various scenario combinations. 3) The annual suspended sediment load showed a local trend in the study area from 1958 to 2014, with the change point year of 1989. There was a significant decreasing trend of annual suspended sediment load at the 5% significance level during the period from 1990 to 2014, due possibly to the increase of forest coverage during the regulation of large and medium-sized reservoirs. Specifically, the decreased sediment loads were 414.87×104 and 261.45×104 t/a, for the increase of forest coverage and reservoirs in 1958-1989, indicating a great contribution of 59.72% and 37.63%, respectively. The factor of rainfall erosivity increased the suspended sediment load by about 42.99×104 t/a, whereas, the forest coverage and reservoir regulation caused the decrease of 61.40×104t/a. Furthermore, the decrease of suspended sediment load was actually caused by human activities, indicating better agreement with the conclusions obtained by the double cumulative curve and sediment yield function. The dynamic characteristics of the first and second moment were well elucidated in the observed samples using the BMA based on time-varying moment models, with an exact description for the mean value and variance. Therefore, the attribution analysis on the suspended sediment load was more objective and reliable in practice. Some recommendations were made here to strengthen the protection of forest resources, and the scientific regulation of large and medium-sized reservoirs, further to improve the control ability of water and soil loss in the Ganjiang River Basin.