Loading...

Preparing Landslide Susceptibility Map of Retrogressive Thaw Slumps Using Machine Learning Algorithms (Case Study: Qinghai-Tibet Plateau)

Bandamiri, Sepehr | 2025

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 58539 (09)
  4. University: Sharif University of Technology
  5. Department: Civil Engineering
  6. Advisor(s): Moghim, Sanaz
  7. Abstract:
  8. Retrogressive thaw slumps (RTS) are among the most significant forms of surface instability in permafrost regions, whose intensity has increased due to climate change and rising temperatures. This phenomenon not only poses a threat to the stability of local ecosystems but also has direct impacts on hydrological, geomorphological, and environmental cycles. The present study aimed to map the susceptibility of RTS in the central part of the Qinghai–Tibet Plateau (QTP) using machine learning algorithms. For this purpose, a set of multi-source data was collected, including topographic indices (elevation and slope), climatic indices (thawing degree days, thawing degree nights, and diurnal temperature range), permafrost-related variables (active layer thickness and water content), and vegetation indices (absolutely local changes in vegetation cover—ALICE-NDVI—). These data were preprocessed, standardized, and introduced into the models. Three algorithms—Random Forest (RF), Support Vector Machine (SVM), and XGBoost—were selected for modeling and optimized through grid search and five-fold cross-validation. The performance of the models was evaluated using metrics such as accuracy and the area under the ROC curve (AUC). Results indicated that the SVM model achieved the best performance with an accuracy of 0.945 and an AUC of 0.987, outperforming RF and XGBoost. Feature importance analysis further revealed that water content, the ALICE-NDVI index, and thermal indices contributed the most to RTS prediction. The final susceptibility map showed that lowland areas with gentle slopes, high water content, and moderate active layer thickness had the highest probability of RTS occurrence. Conversely, higher and drier regions were mostly classified into low susceptibility zones. These results highlight the complex interplay of climatic, topographic, and vegetation factors in RTS formation. Beyond providing a scientific framework for RTS hazard assessment in the QTP, this study offers a practical tool for risk management, environmental planning, and forecasting future changes under different climate change scenarios
  9. Keywords:
  10. Machine Learning ; Remote Sensing ; Climate Change ; Permafrost Regions ; Retrogressive Thaw Slump ; Landslide Susceptibility Mapping

 Digital Object List

 Bookmark

No TOC