Search

The densely populated Po Plain, a very deep sedimentary basin in northern Italy, is prone to heavy shaking during earthquakes. Seismic hazard assessment must account for local variation in wave amplification. Standard ground motion prediction equations may fail to picture the complexity of strong lateral gradients in seismic response, due to sharp structural heterogeneity. For this reason, there is an increasing demand for full waveform predictions for engineering applications. Here, we present an implementation of a hybrid broadband simulation based on the method of Mai et al. (Bull Seismol Soc Am 100(6):3338–3339, 2010), to obtain complete broadband seismograms of 0.1–10 Hz. With this method, low frequency (< 1 Hz) and high frequency (1–10 Hz) seismograms are simulated separately using a deterministic and a stochastic method, respectively. We apply the method to four events recorded within the Po basin, with magnitude ranging from Mw = 4.4 to Mw = 5.6. The low frequency (LF) simulation is performed using SPECFEM3D on a few test subsurface velocity models. The three-dimensional velocity model MAMBo (Molinari et al. in Bull Seismol Soc Am 105(2A):753–764, 2015)—consisting of a detailed structural description of the basin, based on extensive active-source data, embedded within a regional 3D crustal model—provided the best results for broadband simulations that most closely corresponded with the observations. It performed better than an ambient noise tomography model with more accurate S-wave velocities but less well defined layer topographies, emphasizing the importance of first order velocity discontinuities. The high frequency (HF) seismograms are simulated using the multiple scattering approach of Zeng et al. (J Geophys Res Solid Earth 96(B1):607–619, 1991). The scattering coefficients are obtained by performing a non linear inversion for each station to find best fitting synthetic envelopes. HF energy is then combined at ∼ 1 Hz to match the amplitude and phase spectra of the LF signal. We are able to simulate full waveforms throughout the Po Plain, of which shaking duration matches observed data for stations located in the basin. Shaking amplitudes are generally overestimated in the low frequency simulation by the MAMBo velocity model. Updating the MAMBo velocity model with more accurate S-wave velocity information of the ambient noise tomography model should improve the fit in future simulations. [ABSTRACT FROM AUTHOR]