Geoinformation-based landslide susceptibility mapping in subtropical area

Area of Science:

• Geosciences
• Environmental Science
• Disaster Management

Background:

• Landslide susceptibility mapping is crucial in rainfall-prone subtropical areas to mitigate societal damage.
• Existing methods require improvement for enhanced prediction accuracy.

Purpose of the Study:

• To propose an integrated hybrid methodology for improved landslide susceptibility mapping.
• To validate the methodology in Chongren, Jiangxi, a representative subtropical region.

Main Methods:

• Optimal discretization of geo-environmental factors using entropy.
• Application of Weight of Evidence (WoE) and machine learning (ML) models: Random Forest (RF), Support Vector Machine (SVM), and Logistic Regression (LR).

Main Results:

• The hybrid model achieved high prediction accuracy (82.35-91.02%) and AUC values (0.912-0.970).
• Random Forest (RF) demonstrated superior performance among the tested ML algorithms.
• WoE-based RF modeling is recommended for similar landslide risk predictions.

Conclusions:

• The proposed hybrid modeling approach is effective for landslide hazard mapping in subtropical zones.
• This research offers an operational reference for disaster reduction and prevention efforts by local governments.