Scholarlink

Background <p>The prediction of landslide failure times is essential for effective disaster mitigation. However, current deterministic models frequently overlook the uncertainties associated with both aleatoric and cognitive factors.</p> Methods <p>This study introduces a novel Bootstrap-MLE framework that respectively integrates Residual, Wild, Pairs, and Moving Block Bootstrap methods with Maximum Likelihood Estimation (MLE) to construct prediction intervals. The framework was validated against 40 global landslide cases and assessed using several metrics.</p> Results <p>The findings indicate that the Bootstrap-MLE method significantly enhances the prediction interval coverage probability (PICP) for landslides (97.5%) and increases warning reliability, thereby extending the average warning time by over three days and improving efficiency by more than 130%. In scenarios where movement patterns are not differentiated, both the Wild Bootstrap-MLE (WB-MLE) and Residual Bootstrap-MLE (RB-MLE) demonstrate superior performance, achieving a warning probability of 97.5%, which surpasses that of conventional approaches. For varying movement patterns, WB-MLE remains particularly effective for fluctuating, step, and mutant landslides, attaining the highest PICP and symmetry among the evaluated methods, with RB-MLE performing comparably well. The Moving Block Bootstrap-MLE (MBB-MLE) method is particularly suitable for progressive landslides, demonstrating superior control over mean prediction interval center deviation. Conversely, the Pairs Bootstrap-MLE (PB-MLE) method underperformed across all movement patterns.</p> Conclusion <p>The Bootstrap-MLE framework enhances both the accuracy and reliability of landslide failure time prediction and provides a methodological basis for selecting appropriate approaches under different movement patterns.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A comparative analysis of Bootstrap-MLE methods for landslide failure time prediction intervals: evaluating differential applicability across movement patterns

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年9月30日

查看全文
  • Haojie Duan,
  • Zongxing Zou,
  • Shuwen Li,
  • Zhekai Jiang,
  • Yikai Niu,
  • Xinli Hu

摘要

Background

The prediction of landslide failure times is essential for effective disaster mitigation. However, current deterministic models frequently overlook the uncertainties associated with both aleatoric and cognitive factors.

Methods

This study introduces a novel Bootstrap-MLE framework that respectively integrates Residual, Wild, Pairs, and Moving Block Bootstrap methods with Maximum Likelihood Estimation (MLE) to construct prediction intervals. The framework was validated against 40 global landslide cases and assessed using several metrics.

Results

The findings indicate that the Bootstrap-MLE method significantly enhances the prediction interval coverage probability (PICP) for landslides (97.5%) and increases warning reliability, thereby extending the average warning time by over three days and improving efficiency by more than 130%. In scenarios where movement patterns are not differentiated, both the Wild Bootstrap-MLE (WB-MLE) and Residual Bootstrap-MLE (RB-MLE) demonstrate superior performance, achieving a warning probability of 97.5%, which surpasses that of conventional approaches. For varying movement patterns, WB-MLE remains particularly effective for fluctuating, step, and mutant landslides, attaining the highest PICP and symmetry among the evaluated methods, with RB-MLE performing comparably well. The Moving Block Bootstrap-MLE (MBB-MLE) method is particularly suitable for progressive landslides, demonstrating superior control over mean prediction interval center deviation. Conversely, the Pairs Bootstrap-MLE (PB-MLE) method underperformed across all movement patterns.

Conclusion

The Bootstrap-MLE framework enhances both the accuracy and reliability of landslide failure time prediction and provides a methodological basis for selecting appropriate approaches under different movement patterns.