Artificial Neural Network-based Site Amplification Models for Central and Eastern North America

Published

June 19, 2025

Author(s)

Okan İLHAN, Youssef Hashash, Ellen Rathje, Jonathan Stewart, Aspasia Nikolaou, Kenneth Campbell

Abstract

This paper introduces deep learning/Artificial neural network (ANN)-based response spectrum (RS) and Fourier amplitude spectrum (FAS) site amplification models for Central and Eastern North America (CENA) using over 3.6 million one-dimensional (1D) site response simulations. ANNs are demonstrated to significantly reduce the bias in the estimations and to more accurately capture the features of site-specific amplification (e.g. the attributes of peak amplification) along with the site amplification at shallow sites as compared to their conventional counterparts using predetermined functional forms conditioned on same inputs with ANNs. The corresponding ANN models, each with varied inputs, are developed to encapsulate diverse simulated amplification datasets (e.g., linear versus nonlinear or RS versus FAS), suggesting that the reliance on distinct functional forms to estimate different types of amplification data might no longer be necessary. Regarding all these improvements by deep learning-based models, ANNs are proposed as an alternative models to the conventional functions to better represent the amplification data. A standard deviation is also suggested for ANN-based models.

Citation

Earthquake Spectra

Volume

41

Issue

4

Pub Type

Journals

Download Paper

https://doi.org/10.1177/87552930251343630

Local Download

Keywords

Site effects, site amplification, deep learning, artificial neural network

Citation

İLHAN, O. , Hashash, Y. , Rathje, E. , Stewart, J. , Nikolaou, A. and Campbell, K. (2025), Artificial Neural Network-based Site Amplification Models for Central and Eastern North America, Earthquake Spectra, [online], https://doi.org/10.1177/87552930251343630, https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=958013 (Accessed November 23, 2025)