Abstract: Supraglacial lake is an important part of glacier and an indicator of glacier ablation, which not only responds rapidly to global climate change, but also has important significance for the understanding and mastering of regional water resources information. In this paper, the Sentinel-2 images and random forest algorithm are employed to identify and extract the supraglacial lakes of Baltoro Glacier. Furthermore, the spatial distribution characteristics of supraglacial lakes and the relationship between the area/number of supraglacial lakes and glacier elevation are analyzed. The accuracy, completeness and error rates of supraglacial lakes extraction in this paper are 96.07%, 92.18% and 11.59%, respectively. A total of 567 supraglacial lakes are identified in Baltoro Glacier, with an area ranging from 249.46 to 37134m², and most of them are distributed between 3 and 26km from the glacier terminus. Moreover, the number of supraglacial lakes is the largest between 3800 and 4300m with generally large area, and the average area is 1922m². With the increase of elevation, the number and area of supraglacial lakes gradually decrease. In particular, there are only 15 supraglacial lakes with an average area of 356m²in the elevation of above 5300m. The decrease in glacier surface temperature due to the increase in elevation is the main reason for the sharp decline in the number and area of supraglacial lakes.

Key words: Baltoro Glacier, supraglacial lakes, Sentinel-2 images, random forest algorithm