Physics-informed optimization for a data-driven approach in landslide susceptibility evaluation
Citation
Liu, S., Wang, L., Zhang, W., Sun, W., Wang, Y., Liu, J. (2024). Physics-informed optimization for a data-driven approach in landslide susceptibility evaluation. Journal of Rock Mechanics and Geotechnical Engineering, 16(8): 3192-3205. Link to paper
Abstract
Landslide susceptibility mapping is an integral part of geological hazard analysis, with recent emphasis on data-driven models, notably those derived from machine learning, owing to their aptitude for tackling complex non-linear problems. However, the prevailing models often disregard qualitative research, leading to limited interpretability and mistakes in extracting negative samples, i.e. inaccurate non-landslide samples. To address these shortcomings, this paper introduces Scoops 3D, a three-dimensional slope stability analysis tool, for the qualitative assessment of slope stability in the Yunyang section of the Three Gorges Reservoir area. By exploring different threshold values of the factor of safety (FoS), the relationship between the number of negative samples and the predictive performance of the random forest model is analyzed. The obtained FoS of 1.5 can be used as a recommendation for extracting high-quality negative samples in the study area. Furthermore, the feasibility of using the relative FoS ranking as a weight is explored to improve the performance of the random forest model for landslide susceptibility mapping. The results demonstrate that negative samples from areas with higher FoS rankings yield better prediction accuracy. The proposed methodology demonstrated a notable increase of 29.25% in the evaluation metric, the area under the receiver operating characteristic curve (ROC-AUC), outperforming the prevailing benchmark model. The physics-informed landslide susceptibility map shows better agreement with the field investigation of the potential landslide areas, which can better serve for landslide risk reduction.