A novel cluster-based framework for developing correlation model and its implementation for spectral acceleration

A novel framework is proposed in this paper for constructing a correlation model of spectral acceleration directly from the database (without using any GMPE). The proposed framework first numerically constructs the correlation structures contingent on the multidimensional clusters (bins). These numerically constructed correlation structures are next idealized accounting for any trends that may exist against the seismological parameters. This framework is implemented using the NGA-West2 database for five definitions of intensity measure (RotD100, RotD50, Geo-mean, GMRotD100, and GMRotD50). The correlation structure is observed to be independent of the definitions of the intensity measure. Systematic trends are also observed against the seismological parameters: a) reduction in correlation with increasing moment-magnitude; and b) reduction in correlation with increasing Joyner-Boore distance. These trends are expected due to the lowering of corner frequency in power spectra with increasing moment-magnitude; and amplitude decay of higher modes with increasing Joyner-Boore distance, respectively. An idealized correlation model using a few parameters is next proposed accounting for these observed trends. The existing correlation models (independent of seismological parameters) based on the NGA-West2 database (or its subsets) do not represent the correlation structure with sufficient accuracy. The influence of the proposed correlation model on the conditional mean spectrum is also discussed against the state-of-the-art correlation model (which is reported to be independent of the seismological parameters). Overall, the proposed model taking into account the possible influence of seismological parameters is expected to be a valuable input for various earthquake engineering applications.