Enhanced Spatiotemporal Landslide Displacement Prediction Using Dynamic Graph-Optimized GNSS Monitoring.

Area of Science:

• Geotechnical Engineering
• Geodesy
• Data Science

Background:

• Traditional landslide prediction methods struggle with complex, non-stationary spatiotemporal dynamics.
• Accurate monitoring of slope evolution is vital for effective disaster mitigation.

Purpose of the Study:

• To introduce an enhanced prediction framework integrating multi-scale signal processing and dynamic, geology-aware graph modeling.
• To improve the accuracy and robustness of landslide displacement prediction in real-world engineering environments.

Main Methods:

• Utilized Maximum Overlap Discrete Wavelet Transform (MODWT) for denoising Global Navigation Satellite System (GNSS) displacement data.
• Constructed a geology-aware graph using temporal displacement correlation to represent inter-node relationships.
• Implemented a dynamic graph optimization model with low-rank constraints to adaptively refine graph topology.

Main Results:

• The proposed DCRNN-based model achieved the highest accuracy among eight competing models.
• Recorded a Root Mean Square Error (RMSE) of 2.773 mm in the Vertical direction, a 39.1% reduction compared to the baseline.
• Demonstrated superior performance on a real-world dataset from an active open-pit mine.

Conclusions:

• The dynamic graph optimization approach provides a robust solution for landslide prediction.
• The framework effectively captures time-varying inter-sensor dependencies influenced by factors like mining activities.
• This study validates a significantly more accurate approach for landslide prediction in complex environments.