Spatial correlations of ground motion for non‐ergodic seismic hazard analysis

• Published in: Earthquake engineering & structural dynamics Vol. 49; no. 1; pp. 4 - 23
• Main Authors: Kuehn, Nicolas M., Abrahamson, Norman A.
• Format: Journal Article
• Language: English
• Published: Bognor Regis Wiley Subscription Services, Inc 01.01.2020
• Subjects: Correlation; Covariance; Data; Datasets; Distribution; Earthquakes; Ergodic processes; Gaussian process; Geological hazards; Ground motion; Hazard assessment; non‐ergodic PSHA; non‐stationary covariance function; Predictions; Seismic activity; Seismic analysis; Seismic hazard; spatial correlation; Spatial distribution; Variance analysis
• ISSN: 0098-8847; 1096-9845
• DOI: 10.1002/eqe.3221

Summary Traditional probabilistic seismic hazard analysis (PSHA) uses ground‐motion models that are based on the ergodic assumption, which means that the distribution of ground motions over time at a given site is the same as their spatial distribution over different sites. Evaluations of ground‐motion data sets with multiple measurements at a given site and multiple earthquakes in a given region have shown that the ergodic assumption is not appropriate as there are strong systematic region‐specific source terms and site‐specific path and site terms that are spatially correlated. We model these correlations using a spatial Gaussian process model. Different correlations functions are employed, both stationary and non‐stationary, and the results are compared in terms of their predictive power. Spatial correlations of residuals are investigated on a Taiwanese strong‐motion data set, and ground motions are collected at the ANZA, CA array. Source effects are spatially correlated, but provide a much stronger benefit in terms of prediction for the ANZA data set than for the Taiwanese data set. We find that systematic path effects are best modeled by a non‐stationary covariance function that is dependent on source‐to‐site distance and magnitude. The correlation structure estimated from Californian data can be transferred to Taiwan if one carefully accounts for differences in magnitudes. About 50% of aleatory variance can be explained by accounting for spatial correlation.