This project focuses on the simulation of non-point source pollution (NPSP) of agro-forestry system in Lake Poyang Basin. Based on field survey and measured data, a forestry-growth model of changing density and mixed forestry have been built to improve the model of single plant-growth and biomass accumulation with averaged forestry density in SWAT model. This project also establishes necessary matching models such as the model of vegetation coverage fraction (VCF), the model of forest component abundance (FCA), the model of leaf area index (LAI) and extinction coefficient (EC) to get the relevant parameters of forest growth model by remote sensing techniques. Meanwhile, according to the Keating equation, a new model of radiant energy utilization of intercropping is added to original SWAT to modify its single daily-biomass accumulation model. The estimating methods of crop multi cropping index (MCI) and intercropping index (ICI), using remote sensing techniques, are all under investigation in this project. The temporal-spatial impacts of forest vegetation landscape on NPSP have been simulated, with these newly introduced variables (VCF, FCA and LAI of vegetation). The results show that the modified SWAT performs better in the aspect of flow simulation and nutrient loading, compared with the original SWAT. Based on the data from the field of Lake Poyang in the subtropical monsoon wetland and the red soil background, the effectiveness of the modified SWAT model was discussed. The results show that the modified SWAT performs better in the aspect of flow simulation and nutrient loading, compared with the original SWAT. The effectiveness of the simulations was improved by 7.8%, and the simulation of the peak flow rate was also improved to better reflect the actual situation of surface runoff. In simulating nutrient loading, the effectiveness was improved by 6.4%(total phosphorus)and 6.1%(total nitrogen).