Non-point source pollution is defined to be the runoff transport of pollutants from the land to streams because of rainfall and thaw and ground water. With the effective controlling of point source pollution, non-point source pollution had becomed to the main pollution in many area.The non-point source pollution decreased the quality of water environment and threaten health of drinking water.

Tranditonal calculation of non-point source pollution mainly choosed coefficient method to statistic different category of non-point soure pollution,which was inaccurate and could not get the spatial distribution of non-point source pollution and could not actualize concrete method to control non-point source.

A distributed hydrological model SWAT was developed to study the non-point source pollution in the Muxihe watershed.Through establishing attribute databases,the non-point source pollutions were simulated. Management scenarios were builded up according to the BMPs (Bset Management Practices) in order to find the best management practices.The main conclusions were as follows:

(1) The soil physics characteristics were estimated by the SPAW program and MATLAB program to build up soil attribute database.A method to estimate solar radiation was proposed on the basis of relationship between sunshine duration and solar radiation.It is proved that the SWAT model can be used in the Muxihe watershed.

(2) The sensitive parameters were identified through sensitive analysis and the calibration and validation for streamflow and nutrients were performed. The spatio-temporal distribution features of NPS (non-point source) pollution in the Muxihe watershed was revealed.NPS pollution mainly took place in flood season and was heavy in plain and koppie.

(3) The critical areas of nutrient loss were identified. The concentration of nitrate was calculated at the outlets of each subbasin.The result showed that the water quality of Muxihe watershed was safe according to the quality of water environmental criteria.

(4) The remote sensing image of Landsat-5 TM in 2000 and 2004 were used as information source, and the method of supervised classification combined with the expert knowledge were used for the classification.By calculating the transform matrix of land use,the result showed that large numbers of forestland were planted from 2000 to 2004.Beside, through actualizing the policy of “farm land conversion to forestry land”,the NPS pollution in 2004 was decendent comparing with the NPS pollution in 2000.

(5)There management scenarios were build up according to the BMPs in order to find the best management practices, one or different combination of these scenarios may be selected to prevent NPS pollution in the future.