Multi‐Annual Inventorying of Retrogressive Thaw Slumps Using Domain Adaptation

• Published in: Journal of geophysical research. Machine learning and computation Vol. 2; no. 1
• Main Authors: Lin, Yiling, Hu, Xie, Lu, Haoyu, Niu, Fujun, Liu, Gengnian, Huang, Lingcao, Zhang, Shanghang, Liu, Jifu, Liu, Yunhuai
• Format: Journal Article
• Language: English
• Published: Wiley 01.03.2025
• Subjects: domain adaptation; Qinghai‐Tibet plateau; retrogressive thaw slumps (RTSs); semantic segmentation; temporal generalization
• ISSN: 2993-5210; 2993-5210
• DOI: 10.1029/2024JH000370

Retrogressive Thaw Slumps (RTSs), a form of thermokarst hazards, pose risks to hydrological and ecological environments and the safety of the Qinghai‐Tibet Engineering Corridor. We still lack the knowledge about the geographic locations of RTSs and their dynamically changing spatial margins. However, visual interpretation is labor‐intensive while the present‐day deep learning methods become ineffective when the model trained in one year is directly transferred to another. To enhance the model's generalization ability, here we implemented and compared three domain adaptation methods, that is, the classic supervised fine‐tuning method and two proposed unsupervised methods: Image StyleTransfer Domain Adaptation (ISTDA) and the Tversky Adversarial Domain Adaptation (TADA) network. In our proposed ISTDA, we uniformed the contextual information of multi‐temporal images by Cycle Generative Adversarial Network (CycleGAN). We introduced the Tversky loss and the automatic adjustment of weights for multiple loss functions to suppress false positives and to improve the generalization of TADA. We tested three methods' performance in Beiluhe region over the Qinghai‐Tibet Plateau using PlanetScope optical images during 2019–2022. The three domain adaptation methods are successful in generating regional, multi‐annual RTS inventories. Remarkably, TADA sustains good performance in complex transfer scenarios without additional label cost, achieving an F1 increase of 14.32%–24.17% compared to classic methods. Our work is the first to apply an unsupervised domain adaptation to automatically map the RTSs on a multi‐annual timescale, demonstrating a strong potential of its applicability for monitoring large‐scale, multi‐temporal evolution of geomorphological features. Plain Language Summary The Retrogressive Thaw Slumps (RTSs) are one of the thermokarst hazards that keeps expanding during warming climate. Knowledge about the geographic locations and the dynamic changes in the exterior boundaries of RTSs is crucial for assessing their impact on the natural environment and for protecting transportation corridors nearby. However, when it comes to the multi‐annual inventorying, the present‐day RTS mapping algorithms might be ineffective. Our proposed methods, the Image Style Transfer Domain Adaptation and the Tversky Adversarial Domain Adaptation, significantly improve our ability to recover the evolution of RTSs accurately. In this study, the RTS boundaries are delineated on an annual basis from 2019 to 2022 in Beiluhe region in the Qinghai‐Tibet Plateau. RTSs mostly occurred on gentle slopes facing to the north and loam soil with moderate to high moisture. Key Points Semantic segmentation of Retrogressive Thaw Slumps (RTSs) is limited by the generalization ability Our proposed Unsupervised Domain Adaptation method achieved 14.32%–24.17% increase of F1 in RTS mapping compared with classic methods The RTSs in Beiluhe are rapidly expanding and their distribution correlates with the topographic and geological environments