11:00 AM - 1:00 PM
[HTT19-P03] Automatic detection of glacial lakes in mountainous regions of Asia using PlanetScope satellite data and deep learning
Keywords:Glacial lake outburst flood, Deep learning, Remote sensing
Recent global warming has caused the retreat of glaciers in the mountainous regions of Asia, and many glacial lakes filled with meltwater have formed in front of glaciers. These glacial lakes often collapse, and It has been reported that damage due to flooding caused by known as “Glacial Lake Outburst Floods : GLOF" .
In the Teskey Mountains (Kyrgyz Republic), located in the Tien Shan Mountains, there are many small glacial lakes, and 11 “GLOF" have occurred since 1998, resulting in about 150 casualties and the collapse of infrastructure facilities and houses (Narama et al., 2018). The glacial lakes in this region are not of the type that continuously expand as in the Eastern Himalayas, but of the type called short-lived glacial lakes that form and drain in only a few months ( Narama et al., 2010, Daiyrov and Narama 2021). In order to mitigate the damage caused by glacial lake outburst floods, high temporal resolution, high resolution, and wide area glacier lake monitoring is necessary to monitor the appearance of short-lived glacial lakes that expand in a short period of time. However, the detection of glacial lakes with wide area and high temporal resolution requires a lot of time and manpower. In addition, optical satellite images have a problem that the spectrum of the lake changes depending on the weather conditions and time when the image is taken, and the water area detection method using the NDWI normalized index is difficult to determine the threshold value when time series comparison is required, making automatic water area extraction extremely difficult.
In this study, we attempted to construct a method for automatic detection of small glacial lakes in the Tien Shan Mountains by using high temporal and high resolution optical satellite images of PlanetScope and deep learning. In addition, we evaluated the accuracy of the model in actual optical satellite images.
In the Teskey Mountains (Kyrgyz Republic), located in the Tien Shan Mountains, there are many small glacial lakes, and 11 “GLOF" have occurred since 1998, resulting in about 150 casualties and the collapse of infrastructure facilities and houses (Narama et al., 2018). The glacial lakes in this region are not of the type that continuously expand as in the Eastern Himalayas, but of the type called short-lived glacial lakes that form and drain in only a few months ( Narama et al., 2010, Daiyrov and Narama 2021). In order to mitigate the damage caused by glacial lake outburst floods, high temporal resolution, high resolution, and wide area glacier lake monitoring is necessary to monitor the appearance of short-lived glacial lakes that expand in a short period of time. However, the detection of glacial lakes with wide area and high temporal resolution requires a lot of time and manpower. In addition, optical satellite images have a problem that the spectrum of the lake changes depending on the weather conditions and time when the image is taken, and the water area detection method using the NDWI normalized index is difficult to determine the threshold value when time series comparison is required, making automatic water area extraction extremely difficult.
In this study, we attempted to construct a method for automatic detection of small glacial lakes in the Tien Shan Mountains by using high temporal and high resolution optical satellite images of PlanetScope and deep learning. In addition, we evaluated the accuracy of the model in actual optical satellite images.