VS30, the time-averaged 30-metre depth shear wave velocity (VS) for vertically propagating seismic shear waves, is among the most common parameters required for characterising sites in the context of geotechnical earthquake engineering. Direct measurement of VS30 requires invasive testing employing a geotechnical sounding (either drilled borehole or a pushed sCPT [seismic cone penetration test]). In recent years the use of indirect measurements (most commonly employing surface waves, generated and measured at the ground surface) has come into common practice for assessing VS30. Surface wave based methods require no soundings and are thus less expensive than direct measurement, but are also subject to caveats associated with the inherently inferential nature of surface wave data interpretation.

Regardless of the approach taken, obtaining high-quality VS30 measurements in the field entails deployment of specialised equipment and personnel. There are many situations where it is appealing to obtain VS30 estimates more cheaply by means of statistical or geophysical inference. These include preliminary “desktop studies” performed by engineers in the very early phases of project planning (when field budgets are low and precision is of relatively less importance than the overall scope of potential project challenges), zoning for residential building codes (when site-specific VS30 measurements are not economical), and ground motion simulation research (when a geographically continuous VS30 estimate is needed for applying simple empirical site amplification functions across large regions). In these and other situations, there is a need for VS30 estimates that are less expensive to obtain than field measurements, even at the expense of higher uncertainty. This thesis comprises two distinct approaches to addressing this need. The first entails using continuously-available spatial proxy variables (geology and terrain) to generate a statistically robust VS30 model in map form with values available across the whole of New Zealand. The second is a geophysical approach to estimating VS30 at strong motion stations using weak earthquake recordings.

In the first half of the thesis, a VS30 model is developed for New Zealand as a weighted combination of a geology-based and a terrain-based model. The one encoded with a estimate and the other with uncertainty quantified as lognormal standard deviation (σ). A Bayesian updating process allows local VS30 measurements to control model estimates where data exist, and uses model estimates developed for other parts of the world where local data are sparse or nonexistent. Geostatistical interpolation is performed on the geology- and terrain-based models using local VS30 measurements to constrain the model in the vicinity of data. Conventional regression kriging is compared with a flexible multivariate normal (MVN) approach that allows for arbitrary assumptions regarding measurement uncertainty at each data location. A modification to the covariance structure in the MVN application allows for more realistic estimates by reducing undesirable extrapolation across geologic boundaries. The results of kriging and MVN approaches are compared. The geology- and terrain-based MVN models are combined to produce a final model suitable for engineering applications. Validation is performed by comparing the model predictions with the worldwide slope-based VS30 model developed by Allen and Wald (2009).

In the second half of the thesis, VS30 is estimated at select New Zealand strong motion stations (SMS) using a novel geophysical method relying on many recordings of weak earthquakes. SMS with field VS30 measurements were selected so that validation could be
performed. The approach follows a number of re- cent studies that have used the continuum mechanical solution for the arrival of a P -wave at the surface of an elastic half-space to estimate shear wave velocity VS near the surface using small earthquake recordings, and thence (via empirical correlations) VS30. The approach requires as inputs the ratio of the radial and vertical components of ground velocity ( UR/UZ ) during the first P -wave arrival, and the ray parameter p. The latter is estimated using a regional-scale velocity model simplified to a two-layer, one-dimensional profile, and finding the location of refraction that minimizes travel time for the first arrival. In the present study the so-called “P -wave method” is applied to strong motion stations in New Zealand with VS30 measurements.

Since many stations have plentiful recordings of small earthquakes, a semiautomated approach to ground motion processing is desired. A subset of ground motions are selected for manual examination and used as a benchmark for validation of automated pulse-picking methods. Since the P-wave method uses the velocity (rather than acceleration) traces, baseline drift (nonzero local average velocity owing to cumulative integration of high- and/or low-frequency noise) is an issue. Automated methods for pulse-picking are effectively blind to baseline drift and may yield meaningless results. Some previous studies have been applied in relatively low-seismicity regions and hence the drift problem has been amenable to manual processing or culling. In this study, some effort is dedicated to assessing the degree to which an automated “local baseline correction” approach reduces the quality of VS30 estimates by comparison with manual processing and culling, and how this changes bias and variance.

Another avenue of investigation is the impact of one- versus two-dimensional representations of the regional velocity profiles used to generate VS30 estimates. The results indicate that if it is available, using a more detailed two-dimensional “slice” representation yields a significant benefit in terms of precision and accuracy of VS30 estimates. This is a unique contribution to the approach, although it can only be applied in regions where sufficiently detailed regional-scale P-wave velocity models are available.

Finally, some tentative investigations into the limitations of the P-wave method are presented. These focus on the degree to which the real world departs from the idealised representations assumed by the P-wave method, and how such departure can be quantified using information at hand, such as surface topography and the phase characteristics of the first radial and vertical wave arrivals.