Landslide Susceptibility Mapping Using Machine Learning Algorithms and Remote Sensing Data in a Tropical Environment.

Area of Science:

• Geosciences
• Environmental Science
• Remote Sensing

Background:

• Landslides pose significant hazards globally, necessitating accurate predictive models.
• Effective landslide susceptibility mapping is crucial for land-use planning and disaster risk reduction.

Purpose of the Study:

• To spatially predict landslide susceptibility in Cameron Highlands, Malaysia.
• To evaluate the performance of AdaBoost (AB), Alternating Decision Tree (ADTree), and an ensemble AB-ADTree model.

Main Methods:

• Utilized a database of 152 landslides and 17 conditioning factors (e.g., slope, rainfall, land cover).
• Employed machine learning algorithms: AdaBoost (AB), Alternating Decision Tree (ADTree), and an ensemble AB-ADTree model.
• Validated models using Area Under the Receiver Operating Characteristic Curve (AUC) and various performance metrics.

Main Results:

• The AdaBoost (AB) model achieved the highest prediction accuracy with an AUC of 0.96.
• The ensemble AB-ADTree model (AUC = 0.94) and ADTree model (AUC = 0.59) showed lower performance.
• AB demonstrated superior capability in predicting landslide susceptibility compared to the other models.

Conclusions:

• AdaBoost is a highly effective tool for landslide susceptibility mapping.
• Findings contribute to developing accurate landslide prediction models for hazard mitigation and informed land-use management.