Enhanced detection of freeze‒thaw induced landslides in Zhidoi county (Tibetan Plateau, China) with Google Earth Engine and image fusion

• Published in: Advances in Climate Change Research
• Main Authors: Jia-Hui Yang, Yan-Chen Gao, Lang Jia, Wen-Juan Wang, Qing-Bai Wu, Francis Zvomuya, Miles Dyck, Hai-Long He
• Format: Article in Journal/Newspaper
• Language: English
• Published: KeAi Communications Co., Ltd. 2024
• Subjects: Permafrost degradation; Random forest; Thaw slump; Spatial distribution; Tibetan Plateau; Meteorology. Climatology; QC851-999; Social sciences (General); H1-99; permafrost
• Online Access: https://doi.org/10.1016/j.accre.2024.03.002; https://doaj.org/article/e592124de58a47abb9e5074a6a2836a6

Freeze‒thaw induced landslides (FTILs) in grasslands on the Tibetan Plateau are a geological disaster leading to soil erosion. These landslides reduce biodiversity and intensify landscape fragmentation, which in turn are strengthen by the persistent climate change and increased anthropogenic activities. However, conventional techniques for mapping FTILs on a regional scale are impractical due to their labor-intensive, costly, and time-consuming nature. This study focuses on improving FTILs detection by implementing image fusion-based Google Earth Engine (GEE) and a random forest algorithm. Integration of multiple data sources, including texture features, index features, spectral features, slope, and vertical‒vertical polarization data, allow automatic detection of the spatial distribution characteristics of FTILs in Zhidoi county, which is located within the Qinghai‒Tibet Engineering Corridor (QTEC). We employed statistical techniques to elucidate the mechanisms influencing FTILs occurrence. The enhanced method identifies two schemes that achieve high accuracy using a smaller training sample (scheme A: 94.1%; scheme D: 94.5%) compared to other methods (scheme B: 50.0%; scheme C: 95.8%). This methodology is effective in generating accurate results using only ∼10% of the training sample size necessitated by other methods. The spatial distribution patterns of FTILs generated for 2021 are similar to those obtained using various other training sample sources, with a primary concentration observed along the central region traversed by the QTEC. The results highlight the slope as the most crucial feature in the fusion images, accounting for 93% of FTILs occurring on gentle slopes ranging from 0° to 14°. This study provides a theoretical framework and technological reference for the identification, monitoring, prevention and control of FTILs in grasslands. Such developments hold the potential to benefit the management of grassland ecosystem, reduce economic losses, and promote grassland sustainability.