[H-TT17] Environmental Remote Sensing

[HTT17-P02] Detection Characteristics of Glacial Lakes in Mountainous Asia Using PlanetScope Satellite Data and Deep Learning

*Naho Yamada¹, Chiyuki Narama², Yusuke Iida³ (1.Graduate school of science and technology ,Niigata university, 2.Faculty of Science,Niigata university , 3.Faculty of Engineering,Niigata university)

Keywords:Glacial lake outburst flood, Deep learning, Remote sensing

Currently glacial lakes identify using remote sensing methods widely over the world. Developed some methods for automatic mapping of glacial lakes. However, it is difficult to automatically detect water bodies/glacial lakes using NDWI and other methods because of their small size in the study region. It wasn’t developed preferable methods yet for accurately identify lakes in the Tien Shan Mountains of Kyrgyzstan. Therefore, it is important to map glacial lakes using automatic accurate methods.
In this study, we developed a method to automatically detect small glacial lakes in the Tian Shan Mountains using PlanetScope optical satellite images and deep learning. The applied method clarified the minimal size of glacial lakes and differences in their surrounding environments by comparing with the ground-truth data.
Our results show that the accuracy was about 80% which is the high adequately to extract small glacial lakes where lake area less than 0.01 km2 in the study sites.
Also present method can be suitability for analysis either the lake exist or not on the glacier moraines. It can allow us to analyses the easonal changes of glacial lakes and lead us to understand the character of glacial lakes.
This method has shown that it is suitable for accurate identification of glacial lakes, especially in small glacial lakes and non-glacial lakes using deep learning. Therefore, it can be used to understand the seasonal changes of glacial lakes and to quickly detect the locations where glacial lakes have emerged and disappeared.