Land Subsidence Susceptibility Mapping in South Korea Using Machine Learning Algorithms.

Area of Science:

• Geosciences
• Environmental Science
• Machine Learning Applications

Background:

• Land subsidence poses significant geohazards globally.
• Accurate susceptibility mapping is crucial for risk mitigation.
• Machine learning offers advanced tools for geohazard assessment.

Purpose of the Study:

• To evaluate and compare four machine learning models for land subsidence susceptibility mapping.
• To identify key conditioning factors influencing land subsidence in the Jeong-am area, South Korea.
• To generate reliable land subsidence susceptibility maps (LSSM) for the study region.

Main Methods:

• Utilized Bayesian Logistic Regression (BLR), Support Vector Machine (SVM), Logistic Model Tree (LMT), and Alternate Decision Tree (ADTree).
• Incorporated eight conditioning factors: slope angle, distance to drift, drift density, geology, distance to lineament, lineament density, land use, and rock-mass rating (RMR).
• Trained and validated models using 24 historical land subsidence occurrences (70% training, 30% validation).

Main Results:

• All four models generated land subsidence susceptibility maps (LSSM).
• Bayesian Logistic Regression (BLR) exhibited the highest accuracy and reliability in susceptibility assessment.
• Statistical indices, AUROC, SR, and PR curves confirmed the performance of the models, with BLR outperforming others.

Conclusions:

• Bayesian Logistic Regression (BLR) is a highly effective model for land subsidence susceptibility mapping.
• The identified conditioning factors are critical for understanding subsidence mechanisms in the Jeong-am area.
• The generated LSSMs provide valuable data for land use planning and hazard management.