Abstract:Based on the measured flood data from 2016 to 2018 in the Wufeng River Basin in Pingxiang City, Jiangxi Province and the baseflow rate obtained by the baseflow straight bevel cutting method, the applicability and accuracy of baseflow separation using test methods of cumulative anomaly, Pettitt, sliding T and Mann-Kendall were compared. Optimization of nonparametric tests by set pair analysis was selected. The results showed that, compared with the pre-construction of sponge city(2016), the baseflow of the standard recession curve increased by 65. 2%, the underground runoff increased by 47. 3%, the proportion of underground runoff to the total runoff increased by 53% after the construction of sponge city(2018). The relative error of baseflow obtained by the Pettitt method and straight bevel cutting method was the smallest, with an average annual relative error of 1. 7%; the relative error of variation coefficient is the smallest, which is 5%; and the connection degree obtained by set pair analysis is the highest, which is 0. 81. The Pettitt method is the most suitable non-parametric test method for testing the conversion of direct surface and underground binary runoff components in sponge city.