[Influencing factors of non-point source pollution of watershed based on boosted regression tree algorithm.].

Non-point source (NPS) pollution has become a key water pollution problem under the condition of point source pollution was controlled. The complexity and uncertainty research of NPS pollution influential factors has always been important and difficult. This paper simulated NPS pollution of the Fanhe River watershed in 2003-2012 by the soil and water assessment tool (SWAT) and analyzed its spatial distribution. Meanwhile, the boosted regression tree (BRT) method was proposed to quantitatively analyze the corresponding influential factors including land use, soil, elevation and slope. The results showed that NPS pollution in the Fanhe River watershed had high spatial heterogeneity. The spatial distribution of total nitrogen (TN) had greater difference than that of total phosphorus (TP). The three pollutants, TN, TP and sediment, were all positively related to slope gradients (P<0.01). The slope gradients played the strongest role in determining the sediment and TP output with the contribution rate of 46.5% and 38.2%, respectively. Land use had important influence on sediment and TP loads, with the contribution rate of 27.2% and 35.3%, respectively. TN was produced abundantly in low-elevation and steep-slope locations and with cultivated land use. Cinnamon soil was most vulnerable to the TN load while meadow soil took the second place in terms of soil erosion and TP load. The paper overcame the complexity of influential factors for NPS pollution by BRT, and deepened the understanding of NPS pollution mechanism.