Measuring landslide vulnerability status of Chukha, Bhutan using deep learning algorithms

Area of Science:

• Geosciences
• Natural Hazards
• Artificial Intelligence in Earth Science

Background:

• Landslides pose significant risks to human life, property, and the environment.
• Accurate landslide vulnerability mapping (LVM) is crucial for effective disaster risk reduction.

Purpose of the Study:

• To develop an improved framework for landslide vulnerability mapping in Chukha Dzongkhags, Bhutan.
• To integrate physical and social factors for more accurate LVM using advanced machine learning models.

Main Methods:

• Utilized 22 physical and 9 social conditioning factors for vulnerability modeling.
• Employed deep learning neural network (DLNN), artificial neural network (ANN), and convolution neural network (CNN) approaches.
• Validated models using receiver operating characteristics (ROC) and relative landslide density index (R-index).

Main Results:

• Generated nine landslide vulnerability maps, including landslide susceptibility, social vulnerability, and relative vulnerability.
• The Convolutional Neural Network (CNN) model integrating physical and social factors (CNN-RLV) achieved the highest accuracy (AUC = 0.921, 0.928).
• Deep learning neural network (DLNN) and artificial neural network (ANN) models also showed strong performance.

Conclusions:

• Combining physical and social factors provides a more accurate and appropriate landslide vulnerability map.
• The developed framework supports enhanced landslide prediction and management strategies.
• Deep learning approaches, especially CNN, are highly effective for complex landslide vulnerability assessment.