Abstract:Taking the Lijiang River as the study area, a river pollutant flux estimation method based on "preprocessing - regression" procedure is proposed. The water quality data is preprocessed by box plot method, and the annual pollutant fluxes of NH3-N, CODMn, and TP at river sections of Guilin and Yangshuo hydrological stations are estimated based on the LOADEST model. The results show that: The concentration of NH3-N at Guilin river section is higher than that in Yangshuo river section, while the concentrations of both CODMn and TP in Yangshuo are higher.The correlation coefficient between pollutant flux and discharge is greater than those between pollutant flux and concentration, indicating that pollutant flux mainly depends on the variation of river discharge. After preprocessing the water quality data, the estimation values of the flux processes of NH3-N, CODMn, and TP are better coincided with the monitoring values, with the correlation coefficients of Guilin improve by 2.37%, 2.82%, and 0%, respectively; while increase by 2.89%, 1.72%, and 3.75%, respectively at Yangshuo river section. The annual average fluxes of NH3-N, CODMn, and TP at Guilin and Yangshuo river sections are 987.302 , 1254.781 t,6784.407 , 14964.549 t, 265.945 , 756.055 t, respectively. It is of great significance of this study to improve the estimation accuracy of pollutant flux and realize the accurate treatment of river pollutants.