Civil Engineering Reference
In-Depth Information
dynamically, a detailed model of the propagation medium is required in
order to defi ne the manner in which source effects are linked to site effects.
This link is achieved via Green's functions (Aki & Richards, 2002) and may
be theoretical Green's functions or empirical Green's functions.
Unfortunately, the computational effort that is required to generate sim-
ulated wave fi elds from seismological models at frequencies of relevance
for engineering applications are signifi cantly greater than that required for
either empirically based models or spectral models. More importantly, we
have stated that there is not a great deal of confi dence in the standard
deviations that are presented by developers of spectral-based models at the
present time. This statement also holds for the seismological models, but
the assertion is even stronger: these models require a larger number of
parameters and it is known that these parameters are strongly coupled with
each other. When one performs an inversion to recover, say, a model of the
slip distribution from a recorded event one is implicitly obtaining a set of
consistently coupled parameter values. However, from a forward modelling
perspective one must prescribe the covariance structure (or, more precisely,
the joint distribution of parameters)
a priori
. At this point, very little is
know about this joint distribution and given the nonlinear dynamic charac-
ter of the rupture process the implications of not adequately accounting for
a source effect, or for the appropriate relationships among variables can
have drastic consequences (Strogatz, 1994).
Therefore, while signifi cant advances have been made in fi elds related to
seismological simulation of high-frequency ground motions, we are not yet
at a point where these models satisfy the requirements for use within a
hazard or risk analysis framework. Given that empirically derived models
are by far the most commonly used within hazard and risk applications and
that the spectral-based models and seismologically based models have their
own (very different) issues related to the modelling of ground-motion
uncertainty, the rest of this chapter will focus solely upon empirically
derived ground-motion models.
2.3.2 Compilation of empirical strong-motion databases
One of the many decisions that a developer of empirical ground-motion
models must make is which records are to be included within a dataset. For
a long time this decision was not given a great deal of attention because
most modellers simply compiled as many records as they could. However,
now that databases of recordings are becoming larger, the decision related
to compilation of datasets becomes more important. The reason for this is
largely related to the regional composition of data as well as the heteroge-
neous quality of metadata. As will be discussed more formally in Section
2.4.2, the quality of the metadata directly impacts upon the model variance
Search WWH ::
Custom Search