Systematic source, path and site effects on ground motion variability: the case study of Northern Italy

In this paper, we separate the residuals of ground motion prediction equations into different components, calculating the repeatable site-specific ( δ S 2 S s ), location-specific ( δ L 2 L r ) and path-specific ( δ P 2 P e r ) correction for the median model. In this way, we can reduce the total standard deviation (sigma) of the ground-motion model by Lanzano et al. (Bull Seismol Soc Am 106(1):73–92, 2016 ), removing these components and overcome the ergodic assumption. These repeatable terms are empirically estimated using a dataset composed by acceleration waveforms of Italian events mainly occurred in the Po plain and Eastern Alps (Northern Italy), in the time interval 1976–2015. The analysis is carried out on 2241 recordings, from 88 shallow earthquakes recorded by 168 sites. A local model, specifically tailored for Northern Italy, is used as the reference GMPE, which predicts the geometric mean of the horizontal peak ground accelerations and response spectral ordinates in the period range 0.04–4 s. Because most of the propagation paths sampled only once, the path-specific components and the resulting aleatory variability are also calculated from the spatial correlation of residuals, following the approach of Lin et al. (Bull Seismol Soc Am 101(5):2281–2295, 2011 ). We find a significant reduction of the overall ground motion variability to an extent comparable to what observed from previous studies on empirical and simulated datasets. The estimated median corrections and variance components can be used in probabilistic seismic hazard assessment for a single site and a single path in Northern Italy. As an example, we compute the hazard curves at some sites by means of three different approaches (ergodic, partially non-ergodic and full non-ergodic), in order to quantify the impact of relaxing the ergodic assumption.