[A-CG39] Biogeochemical cycles in Land Ecosystem

[ACG39-09] Automatic detection of subsidence landforms caused by thermokarst in Eastern Siberia using deep learning

*Kosuke Takaya¹, Noriko Kurata², Takeshi Ise¹, Yoshihiro Iijima³ (1.Kyoto Univ., 2.Yamaguchi Pref. Univ., 3.Mie Univ.)

Keywords:thermokarst, deep learning, satellite image, permafrost

The polar region is sensitive and vulnerable to climate change, and monitoring of Arctic regions is becoming important. In particular, the thawing of permafrost associated with rising temperatures accelerates the microbial decomposition of organic carbon in the soil, leading to greenhouse gas emissions and affecting climate change. Thermokarst is a landform formed by the thawing of ice-rich permafrost and subsidence of the ground surface. This landform is an indicator of permafrost degradation; thus, the distribution of thermokarst is essential for understanding carbon exchange and ecosystem impacts at regional and global scales. Although understanding subsidence landforms caused by thermokarst (polygons) distribution has been a labor-intensive task, automatic detection using deep learning and remote sensing techniques has been applied. However, the cost of creating training data for the specific area was challenging because polygon size and shape varied from each region. A recently developed method, the chopped picture method, is suitable for identifying ambiguous and amorphous objects such as the polygon landforms by thermokarst. This method is also appropriate for creating region-specific AI models because it efficiently produces the training images. The objective of this study was to evaluate the feasibility of thermokarst identification using this method. If this approach is valuable, region-specific AI models can be created easily and at a low cost. This study uses high-resolution panchromatic and pan-sharpened images (False color composite) in Eastern Siberia to reveal differences in the detection of thermokarst by satellite images. Results showed that our approach could clearly and automatically distinguish developed polygon from others, such as forests, lakes, and urban. Although the previous study showed higher accuracy for panchromatic images, we achieved highly accurate thermokarst detection for pan-sharpened images by improving the training data. The acquisition of appropriate training data as well as the type of satellite images is essential for thermokarst identification. This method will achieve low-cost automatic detection of polygon landforms by thermokarst through the use of satellite data and AI. With the increase of small satellites in the future, opportunities to use satellite images for observations in Arctic research will expand. Our approach will contribute to environmental monitoring in the Arctic by enabling the automatic mapping of subsidence landforms caused by thermokarst.
This study was supported by the Arctic Challenge for Sustainability Research Projects II (JPMXD1420318865) funded by the Ministry of Education, Culture, Sports, Science, and Technology of Japan.