Scholarlink

<p>Landslide time-of-failure prediction is crucial for the implementation of early warning strategies. Velocity is recognized as a reliable predictor of landslide time-of-failure, and several velocity-based models have been widely examined in post-hoc analyses. However, substantial challenges faced by these methods in forward prediction have rarely received attention, particularly in defining the onset of acceleration (OOA) and selecting appropriate prediction models. Addressing these critical issues is essential to support real-time early warning in practice. Therefore, this study proposes an acceleration stage detection method and a dynamic model selection method for real-time landslide time-of-failure predictions. The method includes a generalized OOA point identification procedure, which determines the OOA point in real-time by analyzing the statistical characteristics of all available velocities. Using Bayesian theory, the most appropriate model is selected from candidate models by evaluating its ability to model acceleration observations and its complexity, while quantifying its prediction uncertainty, thereby enabling probabilistic time-of-failure predictions. For practical early warning scenarios, sequential Bayesian updating and parallel computing are further incorporated into the above method, facilitating efficient and dynamic updates of time-of-failure prediction as new monitoring data become available. Ultimately, the proposed method is validated through a synthetic data case and 15 historical landslide cases. The results demonstrate that the approach accurately identifies OOA points in real-time and produces reliable, dynamic predictions of landslide time-of-failure. Compared to single-model prediction methods, this approach achieves a reasonable balance between prediction accuracy and uncertainty, with an average absolute prediction error of only 0.58 days for predicting the most likely time-of-failure. Overall, this study presents a practical and probabilistic framework for real-time landslide time-of-failure predictions, aimed at supporting proactive landslide risk management.</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

Acceleration stage detection and dynamic model selection for real-time landslide time-of-failure predictions

  • 期刊论文

  • 发表日期：

    2025年12月16日

查看全文
  • Bing Feng,
  • Peng Zeng,
  • Tianbin Li,
  • Xiaoping Sun,
  • Xing Zhu,
  • Xuanmei Fan,
  • Qiang Xu

摘要

Landslide time-of-failure prediction is crucial for the implementation of early warning strategies. Velocity is recognized as a reliable predictor of landslide time-of-failure, and several velocity-based models have been widely examined in post-hoc analyses. However, substantial challenges faced by these methods in forward prediction have rarely received attention, particularly in defining the onset of acceleration (OOA) and selecting appropriate prediction models. Addressing these critical issues is essential to support real-time early warning in practice. Therefore, this study proposes an acceleration stage detection method and a dynamic model selection method for real-time landslide time-of-failure predictions. The method includes a generalized OOA point identification procedure, which determines the OOA point in real-time by analyzing the statistical characteristics of all available velocities. Using Bayesian theory, the most appropriate model is selected from candidate models by evaluating its ability to model acceleration observations and its complexity, while quantifying its prediction uncertainty, thereby enabling probabilistic time-of-failure predictions. For practical early warning scenarios, sequential Bayesian updating and parallel computing are further incorporated into the above method, facilitating efficient and dynamic updates of time-of-failure prediction as new monitoring data become available. Ultimately, the proposed method is validated through a synthetic data case and 15 historical landslide cases. The results demonstrate that the approach accurately identifies OOA points in real-time and produces reliable, dynamic predictions of landslide time-of-failure. Compared to single-model prediction methods, this approach achieves a reasonable balance between prediction accuracy and uncertainty, with an average absolute prediction error of only 0.58 days for predicting the most likely time-of-failure. Overall, this study presents a practical and probabilistic framework for real-time landslide time-of-failure predictions, aimed at supporting proactive landslide risk management.