← Back to Sessions

NGA-Subduction Region-Specific Ground Motion Models Using Machine Learning Algorithms

Description: 

In this study, we derive data-driven ground motion models (GMMs) on a global scale and region-specific basis for subduction earthquakes. We employ a weighted average ensemble model that combines four distinct nonparametric supervised machine learning algorithms: Artificial Neural Network, Kernel-Ridge Regressor, Random Forest Regressor, and Support Vector Regressor. To accomplish this, individual models are trained using a subset of the NGA-Sub dataset, comprising 9,559 recordings from 153 interface and intraslab earthquakes recorded at 3,202 different stations. Hyperparameter optimization is carried out through a grid search for each model. Subsequently, the four models are combined using an equal-weighted average ensemble approach, leveraging ensemble modeling to capitalize on the strengths of diverse machine learning algorithms and mitigate their respective weaknesses. The ensemble model incorporates moment magnitude, rupture distance, V_S30, Z_TOR, tectonic, and regional parameters as input variables. GMMs predict horizontal component ground motion intensity measures, including PGD, PGV, PGA, and 5%-damped PSA values at spectral periods ranging from 0.01 to 10 seconds in a logarithmic scale. While no specific functional form is defined, the response spectra, as well as the scaling trends for distance and magnitude in the weighted average ensemble model, are consistent and comparable to the other NGA-Sub GMMs, but with slightly lower standard deviations.

To analyze variability, a mixed effects regression is employed to partition total aleatory variability into between-event, between-station, and event-site-corrected components. The resulting global GMMs are applicable to interface earthquakes with magnitudes ranging from M4.9 to M9.12, rupture distances from 14 to 1,000 km, and Z_TOR values up to 47 km for sites with V_S30 values between 95 and 2,230 m/sec. For intraslab events, the global GMMs are applicable to magnitudes ranging from M4.0 to M8.0, rupture distances from 28 to 1,000 km, and Z_TOR values from 30 to 200 km for sites with V_S30 values between 95 and 2,100 m/sec.

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