Landslide Susceptibility Assessment Using Integrated Deep Learning Algorithm along the China-Nepal Highway.

Area of Science:

• Geosciences
• Remote Sensing
• Machine Learning

Background:

• Mountain hazards, particularly landslides, pose significant risks to the vital China-Nepal Highway in the Kush-Himalayan region.
• Accurate and real-time hazard assessments are essential for infrastructure safety and disaster prevention in this geologically active area.

Purpose of the Study:

• To develop and compare data-driven algorithms for predicting landslide susceptibility along the China-Nepal Highway.
• To evaluate the effectiveness of machine learning models in utilizing temporal and spatial sensor data for hazard assessment.

Main Methods:

• Ten landslide instability factors were analyzed: elevation, slope angle, aspect, plan curvature, vegetation index, built-up index, stream power, lithology, precipitation intensity, and cumulative precipitation index.
• Four machine learning algorithms were implemented and compared: Decision Tree (DT), Support Vector Machines (SVM), Back Propagation Neural Network (BPNN), and Long Short Term Memory (LSTM).

Main Results:

• The Long Short Term Memory (LSTM) model achieved the highest prediction accuracy at 81.2%.
• Support Vector Machines (SVM) followed with 72.9% accuracy, while Back Propagation Neural Network (BPNN) and Decision Tree (DT) showed lower accuracies of 62.0% and 60.4%, respectively.
• LSTM's superior performance is attributed to its ability to capture long temporal dependencies in time-series data.

Conclusions:

• Machine learning, particularly LSTM, shows significant promise for accurate landslide susceptibility prediction using dynamic geological and geographical parameters.
• The study highlights the importance of considering the temporal evolution of factors for effective landslide risk management along critical transportation routes.