Bayesian inference of empirical ground motion models to pseudo-spectral accelerations of south Iceland seismic zone earthquakes based on informative priors

The Icelandic strong-motion dataset is relatively small and the largest recorded earthquake magnitude is M(w)6.5, whereas events of around M(w)7.0 are known to have occurred in the South Iceland Seismic Zone (SISZ). As a result, the required features for use in probabilistic seismic hazard assessment (PSHA), such as deviation from self-similar magnitude scaling and magnitude-distance dependent saturation of ground motions at larger magnitudes, is challenging. Compounding the issue, GMMs from other seismic regions exhibit a strong bias to the available Icelandic strong-motions, underpredicting in the near-fault and overpredicting in the far-field regions. In this study, we approach this issue by considering several GMMs that have either been used or recommended for PSHA in Iceland or have functional forms that satisfy the minimum requirements of GMMs used in PSHA. We recalibrate these GMMs to fit the Icelandic data in the context of the Bayesian statistical framework and a Markov Chain Monte Carlo (MCMC) algorithm, where model inference is carried out using both non-informative and informative priors for selected model coefficients from the original GMMs. Moreover, we used a random effects model to partition the aleatory variability into inter-event and intra-event components. We show that the GMMs with informative priors for magnitude scaling and magnitude-distance scaling terms, not only capture the high near-fault amplitudes and rapid ground motion attenuation with distance, but also introduce a controlled saturation of large magnitude ground motions, which is consistent with observations in other interplate regions. In this study we also introduce a simple GMM, based on informative priors from a model for magnitude-dependent earthquake depth in the SISZ, that fully captures the salient characteristics of the recalibrated GMMs. The presented models thus form a suite of new, essentially hybrid, empirical GMMs that can be used with confidence in predicting PGA and PSA for Icelandic earthquakes, with particular implications for the reassessment of the seismic hazard of Iceland.