Evaluation of the Inter-frequency Correlation of New Zealand CyberShake Crustal Earthquake Simulations

The inter-frequency correlation of ground-motion residuals is related to the width of peaks and troughs in the ground-motion spectra (either response spectra or Fourier amplitude spectra; FAS) and is therefore an essential component of ground-motion simulations for representing the variability of structural response. As such, this component of the simulations requires evaluation and validation when the intended application is seismic fragility and seismic risk. This article evaluates the CyberShake NZ [1] crustal earthquake ground-motion simulations for their inter-frequency correlation, including comparisons with an empirical model developed from a global catalogue of shallow crustal earthquakes in active tectonic regions, and with results from similar simulations (SCEC CyberShake; [2]). Compared with the empirical model, the CyberShake NZ simulations have a satisfactory level of total inter-frequency correlation between the frequencies 0.1 – 0.25 Hz. At frequencies above 0.25 Hz, the simulations have lower (statistically significant at 95% confidence level) total inter-frequency correlation than the empirical model and therefore require calibration. To calibrate the total correlation, it is useful to focus on the correlation of the residual components. The between-event residual correlations, physically related to source effects (e.g., stress drop) which drive ground motions over a broad frequency range, are low at frequencies greater than about 0.25 Hz. Modifications to the cross-correlation between source parameters in the kinematic rupture generator can improve the inter-frequency correlations in this range [3]. The between-site residual correlations, which represents the correlation between frequencies of the systematic site amplification deviations, are larger (statistically significant at 95% confidence level) than the empirical model for frequencies less than about 0.5 Hz. We postulate that this relates to the relative simplicity of site amplification methods in the simulations, which feature less variability than the amplification observed in the data. Additional insight would be gained from future evaluations accounting for repeatable path and basin effects, using simulations with refined or alternative seismic velocity models, and using simulations with a higher crossover frequency to deterministic methods (e.g., 1 Hz or higher).