Extensions to the Groningen ground-motion model for seismic risk calculations: component-to-component variability and spatial correlation

A bespoke ground-motion model has been developed for the prediction of response spectral accelerations, peak ground velocity and significant duration due to induced earthquakes in the Groningen gas field in the Netherlands. For applications to the calculation of risk to the exposed building stock, extensions to the model are required. The use of the geometric mean horizontal component in the ground-motion predictions and the arbitrary horizontal component for the building fragility functions requires the addition of component-to-component variability. A model for this variability has been developed that both reflects the strong horizontal polarisation of motions observed in many Groningen records obtained at short distances and the fact that the strong polarisation is unlikely to persist at larger magnitudes. The other extension of the model is the spatial correlation of ground motions for the calculation of aggregated risk, which can be approximated through simple rules for sampling the variance within site response zones. Making use of ground-motion recordings from several networks in the field and the results of finite difference waveform simulations, a Groningen-specific spatial correlation model has been developed. The new model also combines results from traditional variogram fitting approaches with a new method to infer spatial correlation lengths from observed variance reduction. The development of the new spatial correlation model relaxes the need to approximate spatial correlation through the sampling of site response, although the results obtained herein suggest that similar results could be obtained using either approach. The preliminary consideration of the numerical waveform modelling results in this study paves the way for significant extensions to be made for the modelling of spatial correlations and the decomposition of apparent spatial variability into systematic and random components within a fully non-ergodic framework .