**6. Conclusions**

In this work, we proposed a generated adversarial network (GAN) for mapping glacial lakes (GAN-GL) using Landsat-8 OLI imagery. This allowed for the extraction of glacial lake information quickly and effectively with less data dependency and postprocessing work. A complete glacial lake dataset was first created using random cropping, density cropping, and uniform cropping. We found that the density of glacial lakes in the training data was a factor that greatly impacted the final mapping accuracy. Then, we constructed a GAN-GL model for glacial lake mapping, which adaptively enhanced the potential lake information in a new water attention module. This module integrated the NDWI feature and spatial lake feature computed from two paralleled convolutional layers. The results of the ablation study show that our method, GAN-GL, could significantly improve the capacity to map glacial lakes, with an F1 score of 92.17% and an IoU of 86.34%. Moreover, by comparing our mapping results to those of classical global–local iterative segmentation algorithm and random forest for the entire Eastern Himalayas, the GAN-GL, with high evaluation scores, indicated that it could eliminate effects arising from mountain shadows, clouds, and melting glaciers, and automatically and precisely delineate glacial lakes. This delineation was eminently possible for many small glacial lakes under diverse environmental conditions. Our work provides a feasible way to systematically monitor and map glacial lakes over a large-scale area.

**Author Contributions:** Methodology, H.Z.; validation, H.Z. and M.Z.; formal analysis, M.Z.; writing, H.Z., M.Z. and F.C.; visualization, H.Z.; project administration, F.C.; funding acquisition, M.Z. and F.C. All authors have read and agreed to the published version of the manuscript.

**Funding:** This work was supported by the International Partnership Program of the Chinese Academy of Sciences (131551KYSB20160002/131211KYSB20170046) and the National Natural Science Foundation of China (41871345).

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**

