Forecasting of Landslide Displacement Using a Probability-Scheme Combination Ensemble Prediction Technique

Area of Science:

• Geosciences
• Earthquake Engineering
• Data Science

Background:

• Data-driven models are widely used for landslide displacement prediction but often ignore predictive uncertainty.
• Deterministic models typically provide point estimates, failing to capture the full range of potential outcomes.
• Understanding and quantifying uncertainty is crucial for robust landslide risk assessment.

Purpose of the Study:

• To propose a probability-scheme combination ensemble prediction model for robust and accurate landslide displacement prediction.
• To develop a method for uncertainty quantification in landslide displacement forecasting.
• To evaluate the performance of the proposed model against traditional methods using a real-world case study.

Main Methods:

• Developed a Quantile Regression Neural Networks-Kernel Density Estimation (QRNNs-KDE) ensemble prediction model.
• Utilized QRNNs as base learning algorithms to generate multiple base learners.
• Integrated base learners through a probability combination scheme based on KDE for final predictions.

Main Results:

• The QRNNs-KDE approach demonstrated perfect performance in landslide displacement prediction.
• The proposed method significantly outperformed traditional models including BP, RBF, ELM, SVM, bootstrap-ELM-ANN, and Copula-KSVMQR.
• The model effectively quantified predictive uncertainty in landslide displacement.

Conclusions:

• The QRNNs-KDE ensemble model offers a robust and accurate solution for landslide displacement prediction and uncertainty quantification.
• The approach shows significant potential for medium-term to long-term horizon forecasting of landslide displacement.
• This method enhances landslide risk assessment by providing reliable predictions with quantified uncertainties.