Landslide susceptibility modelling using GIS-based machine learning techniques for Chongren County, Jiangxi Province,...

Area of Science:

• Geosciences
• Environmental Science
• Computational Geography

Background:

• Landslide susceptibility maps are crucial for hazard mitigation and land-use planning.
• Advanced machine learning techniques offer potential for improving landslide susceptibility modelling.

Purpose of the Study:

• To assess and compare four machine learning models: Bayes' net (BN), radical basis function (RBF), logistic model tree (LMT), and Random Forest (RF).
• To identify the most effective model for landslide susceptibility modelling in Chongren County, China.

Main Methods:

• Utilized 222 historical landslide locations and 15 conditioning factors.
• Employed the information gain (IG) method for factor selection.
• Trained and validated BN, RBF, LMT, and RF models using a 70/30 data split.
• Evaluated models using receiver operating characteristic (ROC) curves and statistical measures (sensitivity, specificity, accuracy).

Main Results:

• The Random Forest (RF) model achieved the highest sensitivity (0.787), specificity (0.716), and accuracy (0.752) on the training dataset.
• RF demonstrated an optimized balance between training and validation performance based on AUC values and statistical metrics.
• The study confirmed the benefit of selecting optimal machine learning techniques and conditioning factors.

Conclusions:

• The Random Forest model is highly effective for landslide susceptibility modelling.
• Optimal selection of machine learning algorithms and conditioning factors significantly enhances landslide hazard assessment accuracy.