Abstract: Soil and water loss has been one of the most serious hazards to the basin ecology in recent years. An accurate sediment transport model can greatly contribute to the control of soil and water loss in the basin. This study aims to explore the variation trend of sediment transport by Mann-Kendall. The runoff and sediment transport data was also collected from the 19 hydrological stations in the Loess Plateau. Annual sediment transport model was then established to take the runoff erosion power as the core factor in Pearson correlation analysis. Random forest variables were selected to evaluate the influence of factors (such as annual runoff erosion power, check-dam index, check-dam relative index, normalized difference vegetation index (NDVI) and impervious land area) on the annual sediment transport of the basin. The parameters were also determined in the annual sediment transport model by nonlinear least square method. The accuracy of the model was compared with different factor combinations. The annual sediment transport model was proposed with the strong applicability for the Loess Plateau. The attribution analysis was finally carried out on annual sediment transport. The results showed as follows: 1) It was reasonable to choose the runoff erosion power as the core factor in the annual sediment transport model. But there was a significant negative correlation between the area of watershed and the accuracy of the single factor annual sediment transport model with only annual runoff erosion power that built by the power function. The correlation coefficient was -0.505 (P<0.05). The accuracy of the single factor model decreased with the increase of the watershed area. As such, the applicability of the model was poor in watersheds larger than 7 000 km². 2) The best performance was achieved in the multi-factor annual sediment transport model with the combination of annual runoff erosion power, check-dam index and impervious area factors. The average NSE and RMSE values of the model were 0.84 and 0.21 10⁸t at the calibration rate, 0.79 and 0.27 10⁸t at the verification period. Compared with the single-factor annual sediment transport model that only considering the annual runoff erosion power, the mean NSE value increased by 339.39%, whereas, the mean RMSE value decreased by 81.88% during the verification period. The mean NSE value increased by 7.69%, while the mean RMSE value decreased by 38.64%, compared with the annual sediment transport model considering erosion power, the relative index of check-dam and NDVI combination. 3) The influencing factors on the annual sediment transport were as follows: annual runoff erosion power, impervious area and check-dam index. Their average contribution rates to the annual sediment transport were 79.52%, -38.89%, and 15.97%, respectively. The runoff of most watersheds decreased, where the inter-annual distribution of runoff tended to be uniform, which was greatly reduced the probability of large flow concentrated sediment transport. The simulation and prediction of sediment transport can provide the theoretical support for the prevention and control of soil and water loss in the ecological management of the Loess Plateau.