Calibration of Wellington 3D ground shaking model

Abstract

This thesis calibrates the Hutt Valley section of the previously developed Wellington 3D ground shaking model, using weak motion records from seven earthquakes recorded during the Lower Hutt deployment of seismometers. The Hutt Valley section is approximately 1/3 of the whole 3D model, but exhibits the highest shaking for a presumed Wellington Fault earthquake. We use 24 weak motion sites in Lower Hutt. The sites sample the full range of soil types and depths in the region, from bedrock to thick soft sediments. In this research, the focal mechanisms of the 15 events recorded by the four portable deployments are solved by the combined amplitude ratio and first motion method, using all the available data from New Zealand Standard Network (NZSN), SNZO, and the portable deployment. A new method named the 1D+3D hybrid modeling technique was developed to simulate the ground motion in Hutt Valley to compare with the recorded ground shaking data from the Lower Hutt portable deployment. The method combines 1D modeling between the source and the bottom of the valley with 3D modeling from the valley bottom to the surface. The discrete wavenumber (DWN) method and general reflection and transmission coefficient matrices are used with a 1D velocity model to calculate the stress and velocity wavefield at the bottom of the Hutt Valley sediments. A double couple, point source model with a modulated ramp time function is used as the earthquake focus in the 1D modeling. The finite difference (FD) scheme is then used with the 3D Hutt Valley shaking hazard model to calculate the velocity wavefield at the free surface of the Hutt area, using the time domain stress and velocity wavefield at the bottom of Hutt Valley sediments as input. Stress and velocity synthetics are determined at each 40 m of the grid. To compare with the observed seismogram data in the Lower Hutt deployment, the points within the model corresponding to the recording sites were selected and the velocity time series for those sites were calculated. Through comparing the synthetic and observed ground shaking in both time and frequency domains, we conclude the following: 1. The newly developed (1D+3D) forward modeling technique works. This is verified by the 1D/3D and 1D/(1D+3D) tests. However, the 40 m grid and the low surface velocities used may be too coarse for the technique to be useful beyond 2.5 Hz. 2. Using the 1D forward modeling technique alone, but with a local 1D model for each station can reproduce some of the characteristics of the ground motion observed, e.g., some of the increase of the peak ground velocity (PGV) and some of the increase duration time of basin motions relative to rock motions. 3. The (1D+3D) forward modeling technique can match more features of the ground motion observed than the 1D modeling technique. The synthetics from the (1D+3D) modeling match the recorded data better in waveform, valley resonance frequency and site response. But the (1D+3D) forward modeling technique consumes much more computational time, which may take over 100 times of the time needed by 1D modeling, and needs a more powerful computer. 4. Two ratios are obtained for calibration of the Hutt 3D shaking hazard model. One is cpvr, i.e., the ratio of the measured to synthetic peak velocity; another is cpsr, i.e., the ratio of measured to synthetic peak spectral ratios. cpvr and cpsr are obtained for different shaking hazard zones classified by Dellow et al. (1992) based on source types (dip slip and strike slip earthquakes). The ratio of the measured to synthetic peak velocity in zone 5 and basin edge averages 0.6±0.3 for dip slip and 0.7±0.3 for strike slip earthquakes. The ratio of the measured to synthetic peak spectral ratios in zone 5 averages 2.0±0.8 for dip slip and 1.5±0.9 for strike slip earthquakes. The results for the other zones are similar. Thus, the peak velocity ratios are overestimated with the Hutt 3D shaking hazard model, while the peak spectral ratios are underestimated.