Mechanical and Civil Engineering Seminar

Mechanical and Civil Engineering Seminar Series

Title: A non-ergodic ground motion model for the Groningen, Netherlands: Merging physics-based and recorded data

Abstract: We present a new non-ergodic ground motion model (GMM) for the Groningen natural gas field in the Netherlands, aimed at improving the region's seismic hazard characterization. Non-ergodic GMMs offer the ability to reduce aleatory variability, which often controls the seismic hazard, especially at longer return periods. This reduction, however, is accompanied by epistemic uncertainty in areas with sparse observations, while it shifts the median ground motion in regions with available recordings. To further constrain epistemic uncertainties, we conduct a series of 3D wave propagation simulations using a detailed 3D velocity model for Groningen. These simulations employ point sources to illuminate seismic wave paths absent in the empirical datasets.

We propagate ground motions up to a maximum usable frequency of 5 Hz by using a dense spectral element mesh and a modified Burne source time function to mitigate spatial aliasing. The non-ergodic GMM is based on a Gaussian Process Regression framework. We represent systematic source and site effects using isotropic kernels, while a new vector-integral kernel function is proposed to capture the systematic path effects. This kernel accounts for correlations along entire seismic rays, incorporating source and site locations, azimuthal dependence, while honoring reciprocity.

To handle large empirical and simulated datasets efficiently, we implement a sparse approximation of the precision matrix through a Kullback-Leibler (KL) divergence minimization. We estimate hyperparameters through a cross-correlation Kernel Flow scheme, which enhances model predictability for previously unobserved scenarios. Our formulation achieves a 40% reduction in aleatory variability, significantly improving the accuracy of seismic hazard assessments for Groningen.

Bio: Grigorios (Greg) Lavrentiadis is a postdoctoral research associate at the Mechanical and Civil Engineering Department at the California Institute of Technology. His current research focuses on developing machine-learning methodologies for ground motion synthesis, post-disaster reconnaissance, and the development of physics-informed non-ergodic ground motion models (NGMMs). He received his Ph.D. in Civil and Environmental Engineering from the University of California, Berkeley, in 2021. Before his PhD, he worked for two years at Fugro USA as part of the Global Services - Earthquake Engineering group. He earned his bachelor's degree from Aristotle University of Thessaloniki, Greece in 2014 and his masters of science degree from the University of California, Berkeley, in 2015.

NOTE: At this time, in-person Mechanical and Civil Engineering Lectures are open to all Caltech students/staff/faculty/visitors.