A Record-Fusion Ground-Motion Model for Spectral Acceleration Prediction from 0.01 to 16 s

Introduction

Most existing ground-motion models provide spectral acceleration predictions only up to approximately 10 s, which limits their applicability to seismic response analysis and hazard assessment of long-period structures such as super high-rise buildings, long-span bridges, and large liquid storage systems. This study develops a record-fusion ground-motion modeling framework for 5%-damped spectral acceleration over an extended period range of 0.01–16 s, with particular attention to ultra-long-period ground motions beyond 10 s.

Methods

A database of 7804 collocated acceleration–velocity record pairs from 126 earthquakes in Southern California was compiled. Spectral accelerations derived from acceleration and velocity records were fused using period-dependent weights optimized through Bayesian optimization. The fused observations were then used to develop a ground-motion model through staged term search and mixed-effects regression. Source, path, site, rupture-geometry, hanging-wall, anelastic-attenuation, and basin-depth effects were incorporated into the candidate model structure.

Results

The resulting model reproduces physically reasonable magnitude scaling, distance attenuation, and site-condition effects over the full period range. Residual analyses show that between-event, site-to-site, and remaining within-event residuals are generally centered around zero, with no evident systematic trends against major predictor variables. Compared with acceleration-only and previous fusion-weighting schemes, the proposed framework reduces residual variability over short-to-long period ranges. Comparisons with NGA-West2 models further show broadly consistent predictions within the common 0.01–10 s range, while stable prediction behavior is maintained from 10 to 16 s.

Conclusions

The proposed record-fusion framework provides a practical approach for extending empirical ground-motion modeling into the ultra-long-period range and supports improved seismic input characterization for long-period engineering structures.