← Back to Sessions

Development of a Machine Learning-based Ground Motion Model for Induced Earthquakes in the Central and Eastern United States

Description: 

In this study, a Ground Motion Model (GMM) is developed using machine learning regression techniques to predict Peak Ground Acceleration (PGA) and Pseudo-Spectral Acceleration (PSA) values across 17 periods up to 3 seconds. The model utilizes data from 31,000 induced earthquakes in the Central and Eastern United States (CEUS), focusing on small-to-moderate magnitude events with moment magnitudes (M_W) ranging from 3.0 to 5.8, hypocentral distances (R_hypo) up to 200 km. Input parameters for the proposed GMM include M_W, R_hypo, and the time-averaged shear-wave velocity of the upper 30 m of soil (V_S30), while the output variables are PGA and PSA across different periods. Machine learning methods, including Artificial Neural Networks (ANN), Kernel Ridge Regression (KRR), Random Forest Regression (RFR), and Gradient Boosting Regression (GBR) are employed. GBR demonstrates the best performance among the individual models.

An ensemble technique is applied to combine the outputs of individual models, resulting in a robust and smoothed final prediction. The use of machine learning addresses the limitations of traditional GMMs, such as the need for predefined functional forms, and the ensemble approach further enhances model robustness. Comparisons with conventional GMMs show that the proposed method achieves higher accuracy, particularly when sufficient training data are available. The results highlight the potential of machine learning-based GMMs for reliable seismic hazard assessments, especially for small-to-moderate earthquakes in the CEUS region.

Session: From Physics to Forecasts: Advancements and Future Directions of Induced Seismicity Research [Poster]

Type: Poster

Date: 4/15/2025

Presentation Time: 08:00 AM (local time)

Presenting Author: Najme

Student Presenter: Yes

Invited Presentation: 

Poster Number: 105

Authors