Comparative study of thaw settlement susceptibility mapping for the Qinghai-Tibet Plateau based on index and machine learning models

Abstract

Climate warming has caused frequent thaw settlement in the permafrost region of the Qinghai-Tibet Plateau (QTP), significantly threatening the ecological environment and infrastructure. This study assesses thaw settlement susceptibility using index and machine learning (ML) models and compares their accuracies. The settlement index (Is), risk zonation index (Ir), and geohazard index (Ia) models were selected to map thaw settlement susceptibility, and their results were combined to construct a comprehensive index (Ic) model using a majority vote criterion. Based on 12 conditioning factors related to topography, soil, vegetation, and climate, susceptibility studies using artificial neural network (ANN), K-nearest neighbor (KNN), support vector machine (SVM), and random forest (RF) models were conducted. The results indicate that although the Ic model improves the accuracies of the Is, Ir and Ia models, it remains limited, with 75.06% of thaw settlements occurring in low and moderate susceptibility areas. Conversely, the ML models demonstrated superior accuracy, with the RF model performing the best, which remained only 13.87% of thaw settlements in low to moderate susceptibility regions, effectively pinpointing the Qiangtang Plateau (QP) and Three Rivers Source (TRS) region as high susceptibility areas. Notably, the Budongquan-Beiluhe sections of the Qinghai-Tibet Highway (QTH) and Qinghai-Tibet Railway (QTR) were identified as potential high-risk regions for thaw settlement. These findings offer valuable insights for thaw settlement susceptibility evaluation and disaster risk management in the QTP.