NonErgodicGMM Project

Traditionally, due to the scarcity of regional data, ground-motion models (GMMs) were developed under the ergodic approach, which assumes the statistical properties of the ground-motion parameter do not change in space resulting in GMMs with stable median estimates but large aleatory variability. In recent years, as the size of available data has been increasing increased, the ergodic assumption can be relaxed. This led to the introduction of non-ergodic ground-motion models where repeatable effects related to the source, path, and site can be appropriately modeled, and separated from the aleatory variability. The reduced aleatory variability can impact the seismic hazard, especially at large return periods which is important for critical infrastructure. This reduction in aleatory variability is accompanied by epistemic uncertainty in regions with sparse recordings or a systematic shift in the median ground motion in regions with dense recordings.

The objective of this project is to introduce the Gaussian Process Regression (GPR) as a method for the development of non-ergodic ground motion models and to provide open‐source software tools and instructions for the derivation and application of NGMMs. The developed software packages were verified against synthetic data sets with known non-ergodic effects, and different implementations of the developed software were evaluated for scalability, universality, precision, and model complexity. The computer codes developed as part of this report and synthetic database used for the verification process are made available to the public through the UCLA NHR3 Git-hub repository