Ground Motion Models Using Machine Learning Techniques Based on the NGA-West2 Data
Description:
In this study, we derive ground motion models (GMMs) for the average horizontal component resulting from shallow crustal continental earthquakes in active tectonic regions by analyzing a subset of the NGA-West2 dataset. This subset encompasses 14,518 recordings from 285 earthquakes recorded at 2,347 different stations. Constructing these models involves employing four nonparametric supervised machine learning (ML) algorithms: the Artificial Neural Network, Kernel-Ridge Regressor, Random Forest Regressor, and Support Vector Regressor, each producing an individual model. Then, a weighted average ensemble approach is employed to merge these four models into a robust unified model for predicting various ground motion intensity measures, including peak ground displacement, peak ground velocity, peak ground acceleration, and 5%-damped pseudo-spectral acceleration. The model incorporates moment magnitude, rupture distance, VS30, and ZTOR as input parameters.
The ensemble modeling strategy seeks to reduce drawbacks or deficiencies inherent in different ML algorithms while leveraging their advantages, and thus, addressing epistemic uncertainty. Despite the absence of a predefined functional form, the model effectively captures prominent features observed in ground motions, such as saturation, geometrical spreading, anelastic attenuation, and nonlinear site amplification. The response spectra and scaling trends related to magnitude, distance, VS30, and ZTOR are consistent and comparable with the NGA-West2 GMMs, which include several additional input parameters. A mixed-effects regression analysis is applied to divide the total aleatory uncertainty into between-event, within-station, and event-site-corrected components. The model applies to earthquakes with magnitudes ranging from 3.0 to 8.0, rupture distances extending up to 300 km, and spectral periods spanning from 0 to 10 seconds.
Session: From Earthquake Recordings to Empirical Ground-Motion Modelling [Poster Session]
Type: Poster
Date: 5/2/2024
Presentation Time: 08:00 AM (local time)
Presenting Author: Farhad
Student Presenter: No
Invited Presentation:
Authors
Farhad Sedaghati Presenting Author Corresponding Author farhad.sedaghati@aon.com AON |
Shahram Pezeshk spezeshk@memphis.edu University of Memphis |
|
|
|
|
|
|
|
Ground Motion Models Using Machine Learning Techniques Based on the NGA-West2 Data
Category
From Earthquake Recordings to Empirical Ground-Motion Modelling