Landslide susceptibility assessment via the information value-coupled machine learning models

Area of Science:

• Geosciences
• Geological Hazard Assessment
• Machine Learning Applications in Earth Science

Background:

• Frequent landslides and collapses occur in the Tianshan Mountains, Xinjiang, posing significant risks.
• Accurate landslide susceptibility assessment is vital for geological hazard prevention and mitigation.

Purpose of the Study:

• To develop and evaluate models for precise landslide susceptibility assessment in the Tianshan northern slope economic belt.
• To identify the most reliable model for predicting landslide occurrences in the region.

Main Methods:

• Selected 10 landslide conditioning factors for multicollinearity analysis.
• Developed and compared Information Value-Maximum Entropy (I-MaxEnt) and Information Value-Logistic Regression (I-LR) coupled models.
• Utilized Receiver Operating Characteristic (ROC) curves and field validation for accuracy assessment.

Main Results:

• The Information Value-Logistic Regression (I-LR) coupled model achieved a superior Area Under the Curve (AUC) of 0.941.
• The I-LR model demonstrated higher accuracy and reliability compared to the I-MaxEnt model.
• Field validation confirmed the I-LR model's results align well with the actual landslide situation.

Conclusions:

• The I-LR coupled model is highly suitable for landslide susceptibility assessment in the Tianshan northern slope economic belt.
• This research provides a reliable basis for disaster prevention and mitigation strategies in the study area.