Abstract: Based on the experimental determination of nitrogen and phosphorus contents in 23 surface sediments of Xingyun Lake, combined with the nutrient data of the lake in different periods and the methods of machine learning K-Nearest Neighbor ( KNN ) and Inverse Distance Weight ( IDW ), Ordinary Kriging ( OK ) and Kernel Smoothing ( KS ) methods, the spatial distribution characteristics of nitrogen and phosphorus contents in surface sediments and the prediction accuracy of each interpolation model were analyzed. The influencing factors of the continuous increase of nitrogen and phosphorus concentrations in Xingyun Lake were studied, and the advantages of the machine learning KNN algorithm in the prediction of nitrogen and phosphorus contents in lake surface sediments were discussed. The results showed that the TN content in the surface sediments of Xingyun Lake ranged from 0.56% to 0.86%, with an average of 0.71%, and the TP content was between 0.57% and 0.91%, with an average of 0.78%.The spatial distribution of nitrogen and phosphorus predicted by the four algorithm models had a certain spatial similarity, but even under the conditions of different periods, the spatial interpolation prediction error of the KNN algorithm was the smallest, and the fitting accuracy was higher than that of the traditional interpolation model. It was found that the spatial prediction accuracy of the KNN model was higher when the concentration of nitrogen and phosphorus was lower.The results showed that the concentrations of nitrogen and phosphorus in the surface sediments of Xingyun Lake showed an overall upward trend, and the differences in different periods and spaces were mainly affected by land use types, agricultural non-point sources and natural factors of the lake.The research results will provide some reference for the spatial prediction of nutrients in surface sediments of low-latitude plateau lakes and the ecological protection of lakes.

Key words: 低纬高原湖泊, 表层沉积物, 营养元素, 机器学习, 空间插值