Abstract:Dongjiang River Basin was selected as the study area, which is located in the south of Jiangxi Province, China. The SWAT model was used to evaluate the influence of different watershed delineation on runoff and sediment yield. Eleven scenarios including the sub-watershed thresholds were 25, 50, 100, 200, 400, 800, 1 100, 1 500, 2 000, 3 200, 4 000 hm2, were used to analyze the impact of soil spatial discretization on runoff and sediment yield. Terrain and land use data were input a single terrain and soil data to fix the effect of terrain and soil data spatial discretization. The results were as follow: (1) The discretization effects of soil parameter significantly increased with the increasing of sub-watershed, the largest area of latosol reduced significantly (P<0.05), the second largest area of osmotic paddy soil and the third largest area of yellow soils increased significantly (P<0.05). Lithosol soil, the smallest area soil type, was totally disappeared by soil discretization when the number of sub-watershed was under 267, while it increased when the number of sub-watershed was more than 524 (P<0.05). (2) The spatial discretization of soil parameters reduced the annual runoff (1.32%, P<0.05) and annual sediment yield (18.07%, P<0.05), and the sediment yield decreased greater than annual runoff. (3) The annual maximum one-day, annual maximum five-day, and annual maximum seven-day sediment yield decreased significantly with the increase of delineation degree (16.64%, 17.54%, 17.34%, P<0.01), while the sub-watershed delineation had less influence on the annual maximum one-day, annual maximum five-day and annual maximum seven-day runoff yield (0.59%, 0.89%, 0.83% P>0.05); (4) The peak sediment yield decreased significantly with the increase of delineation degree, while the effect for peak runoff was not significant. The soil parameters spatial discretization of watershed subdivision mainly affected the change of sediment yield through the change of K-factor. The research results will provide reference information for the selection of optimal computing units, improving the simulation accuracy and reducing the uncertainty of model simulation.