Automated Landslide-Risk Prediction Using Web GIS and Machine Learning Models.

Area of Science:

• Geographic Information Science
• Machine Learning
• Natural Hazard Prediction

Background:

• Spatial landslide prediction is crucial for inhabitant safety.
• Existing methods often lack dynamic risk assessment capabilities.
• Chiang Rai, Thailand, faces significant landslide risks.

Purpose of the Study:

• To develop an automated web Geographic Information System (GIS) for dynamic landslide risk prediction.
• To present a novel bidirectional long short-term memory (Bi-LSTM) algorithm for enhanced landslide forecasting.
• To integrate machine learning, web technologies, and APIs for real-time risk visualization.

Main Methods:

• Utilized Quantum GIS (QGIS) for geospatial landslide database construction.
• Employed machine learning models including Linear Regression (LR), Artificial Neural Network (ANN), LSTM, and Bi-LSTM.
• Developed a two-stage classification enhancement for LSTM and Bi-LSTM models.
• Integrated static factors (land cover, soil, elevation, slope) and dynamic factor (precipitation) into the prediction models.
• Trained and evaluated models using historical landslide data and real-time datasets.

Main Results:

• The bidirectional long short-term memory with Random Forest (Bi-LSTM-RF) model demonstrated superior landslide-risk prediction performance.
• Bi-LSTM-RF achieved significant improvements in Area Under the Curve (AUC) scores compared to LR, ANN, LSTM, and standalone Bi-LSTM.
• An automated web GIS was successfully developed, integrating trained models, APIs (rainfall, Google), and a geodatabase for interactive visualization.

Conclusions:

• The developed automated web GIS application effectively predicts landslide risk using the Bi-LSTM-RF model.
• The system provides a dynamic and visually intuitive platform for landslide risk assessment.
• This approach enhances the safety of inhabitants in landslide-prone areas through advanced technological integration.