Scholarlink

<p>Ground Motion Prediction Equations (GMPEs) are crucial in seismic hazard analysis. However, most existing GMPEs for the Himalayan region are based on limited or simulated datasets, affecting their reliability. This study develops new, region-specific GMPEs using 371 strong-motion records from 56 real earthquakes (<i>M</i><sub><i>wg</i></sub>: 3.5–7.7) with hypocentral distances up to 1600&#xa0;km. Key parameters include peak ground acceleration (PGA), magnitude, average shear wave velocity to 30&#xa0;m depth (<i>V</i><sub><i>s30</i></sub>), and distance (<i>X</i>). A non-linear multiple regression approach is employed to derive the “DC24 physical model,” which demonstrates superior predictive performance compared to existing models. The model achieves a higher coefficient of determination (R<sup>2</sup>), and lower Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE), capturing regional attenuation patterns more accurately. In addition, a hybrid GMPE is developed by integrating the DC24 physical model with a correction term derived from the Random Forest machine learning algorithm. This “DC24 hybrid model” yields further improvements in predictive accuracy, especially for lower-to-intermediate magnitude events. Notably, the study introduces the world’s first GMPE based on the generalized moment magnitude scale (<i>M</i><sub><i>wg</i></sub>), which outperforms the traditional <i>M</i><sub><i>w</i></sub>-based model for smaller earthquakes in the Himalayas. These advancements significantly enhance the reliability of seismic hazard assessments in the region and provide a foundation for future models covering a broader range of spectral periods and incorporating larger datasets.</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

Ground motion prediction equations for the Himalayan region based on earthquake data using conventional and random forest approaches

  • 期刊论文

  • 发表日期：

    2025年9月20日

查看全文
  • Arindam Das,
  • Ranjit Das,
  • Deepankar Choudhury

摘要

Ground Motion Prediction Equations (GMPEs) are crucial in seismic hazard analysis. However, most existing GMPEs for the Himalayan region are based on limited or simulated datasets, affecting their reliability. This study develops new, region-specific GMPEs using 371 strong-motion records from 56 real earthquakes (Mwg: 3.5–7.7) with hypocentral distances up to 1600 km. Key parameters include peak ground acceleration (PGA), magnitude, average shear wave velocity to 30 m depth (Vs30), and distance (X). A non-linear multiple regression approach is employed to derive the “DC24 physical model,” which demonstrates superior predictive performance compared to existing models. The model achieves a higher coefficient of determination (R2), and lower Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE), capturing regional attenuation patterns more accurately. In addition, a hybrid GMPE is developed by integrating the DC24 physical model with a correction term derived from the Random Forest machine learning algorithm. This “DC24 hybrid model” yields further improvements in predictive accuracy, especially for lower-to-intermediate magnitude events. Notably, the study introduces the world’s first GMPE based on the generalized moment magnitude scale (Mwg), which outperforms the traditional Mw-based model for smaller earthquakes in the Himalayas. These advancements significantly enhance the reliability of seismic hazard assessments in the region and provide a foundation for future models covering a broader range of spectral periods and incorporating larger datasets.