Landslide Susceptibility Evaluation Based on Potential Disaster Identification and Ensemble Learning

Abstract

Catastrophic landslides have much more frequently occurred worldwide due to increasing extreme rainfall events and intensified human engineering activity. Landslide susceptibility evaluation (LSE) is a vital and effective technique for the prevention and control of disastrous landslides. Moreover, about 80% of disastrous landslides had not been discovered ahead and significantly impeded social and economic sustainability development. However, the present studies on LSE mainly focus on the known landslides, neglect the great threat posed by the potential landslides, and thus to some degree constrain the precision and rationality of LSE maps. Moreover, at present, potential landslides are generally identified by the characteristics of surface deformation, terrain, and/or geomorphology. The essential disaster-inducing mechanism is neglected, which has caused relatively low accuracies and relatively high false alarms. Therefore, this work suggests new synthetic criteria of potential landslide identification. The criteria involve surface deformation, disaster-controlling features, and disaster-triggering characteristics and improve the recognition accuracy and lower the false alarm. Furthermore, this work combines the known landslides and discovered potential landslides to improve the precision and rationality of LSE. This work selects Chaya County, a representative region significantly threatened by landslides, as the study area and employs multisource data (geological, topographical, geographical, hydrological, meteorological, seismic, and remote sensing data) to identify potential landslides and realize LSE based on the time-series InSAR technique and XGBoost algorithm. The LSE precision indices of AUC, Accuracy, TPR, F1-score, and Kappa coefficient reach 0.996, 97.98%, 98.77%, 0.98, and 0.96, respectively, and 16 potential landslides are newly discovered. Moreover, the development characteristics of potential landslides and the cause of high landslide susceptibility are illuminated. The proposed synthetic criteria of potential landslide identification and the LSE idea of combining known and potential landslides can be utilized to other disaster-serious regions in the world.

Landslide susceptibility evaluation based on active deformation and graph convolutional network algorithm

“…(6) Environmental criterion. Surface cover damage on a slope is sometimes associated with a landslide, e.g., vegetation damage and bare land expansion (Wang et al, 2022b). Moreover, a slope by a river more possibly evolves into a landslide under the cutting and erosion of river water on the slope foot (Richard et al, 1984).…”

“…The relation is measured according to the Pearson correlation coefficients passing a significance test with the significance value lower than 0.05 (Wang et al, 2022b). The relation to an earthquake is calculated by the Pearson correlation coefficient between deformation velocity and peak ground acceleration (PGA), the relation to precipitation is in terms of the Pearson correlation coefficient between cumulative displacement and cumulative rainfall, the relation to snowmelt is determined by the Pearson correlation coefficient between deformation velocity and temperature in the ice and snow melting season, and the relation to human engineering activity is according to the Pearson correlation coefficient between area proportion of active deformation and distance to road or building (Wang et al, 2022a;Wang et al, 2022b). In Wanzhou District, there is no obvious relation between landslides and the occurrence of snowmelt or earthquakes during the monitoring period; thus, only the criteria in precipitation and road construction are adopted in Wanzhou District.…”