Landslide susceptibility assessment based on frequency ratio and semi-supervised heterogeneous ensemble learning...

Area of Science:

• Geosciences
• Environmental Science
• Data Science

Background:

• Data-driven landslide susceptibility assessment faces epistemic uncertainty from individual model inaccuracies and non-landslide sample selection.
• Effective landslide prediction is crucial for disaster management and urban sustainability.

Purpose of the Study:

• To propose a heterogeneous ensemble learning method incorporating frequency ratio (FR) and semi-supervised sample expansion to address uncertainties in landslide susceptibility assessment.
• To develop a more accurate landslide susceptibility map for Ji'an City, China, and evaluate its economic benefits.

Main Methods:

• A heterogeneous ensemble learning approach was developed, combining frequency ratio (FR) with semi-supervised sample expansion.
• Twelve environmental factors were used to calculate frequency ratios (FFR).
• LightGBM, Random Forest (RF), and Convolutional Neural Network (CNN) models were employed for prediction, with ensemble strategies integrating results to expand samples.

Main Results:

• The negative sample based on FFR sampling proved more accurate than random sampling.
• The proposed FR-SSELR model achieved high performance, with an Area Under the Curve (AUC) of 0.971 and Accuracy (ACC) of 0.941.
• The model generated a more reliable landslide susceptibility map, concentrating landslides in high-risk zones and improving economic benefits by 3.82-14.2%.

Conclusions:

• The FR-SSELR model effectively reduces uncertainty in landslide susceptibility assessment.
• The study provides valuable guidance for landslide management and promotes the sustainable development of Ji'an City.
• Semi-supervised ensemble strategies and FFR-based sampling significantly enhance landslide prediction accuracy and reliability.