Uncertainties in ground motion prediction in probabilistic seismic hazard analysis (PSHA) of civil infrastructure - Seismic Risk Analysis and Management of Civil Infrastructure Systems - page 34

events and short source-to-site distances is not consistent with what we
would expect. As a consequence, various modifi cations are required in
order to obtain sensible predictions from the near-fi eld scenarios (e.g., Toro,
2002; Boore, 2009).
Given the subject of the present chapter, an important feature of stochas-
tic-based ground-motion models is that the standard deviations that are
provided for these models are not directly constrained by empirical data.
Instead, the standard deviation is computed by attempting to attribute
uncertainties to the various parameters of the underlying spectral model
and to then propagate these through the simulation process. While signifi -
cant efforts have been made in order to make this approach transparent
(Toro et al ., 1997), the standard deviations that are currently reported
for these stochastic models are not regarded as being particularly robust.
This is a very important point within the context of hazard and risk
assessment.

Models for subduction zones

Subduction zones are responsible for the largest earthquakes that we
observe, yet despite this far less attention has been devoted to deriving
empirical ground-motion models for these types of events. One reason for
this is that data from these events is not as readily accessible as it is for
shallow crustal events. That is not to say that there are not abundant sources
of data regarding subduction zone ground motions, but rather that fewer
efforts have been made to compile this data in some routine way and to
make it publically available. One possible reason why this is the case is that
there is arguably more reason to suspect that there could be regional dif-
ferences in ground motions coming from different subduction zones. This
results from the different geometries that are encountered in the various
subduction zones around the world.
When efforts are made to compile datasets for subduction zone events
the resulting databases tend to be rather rich in recordings from large mag-
nitude events, particularly in comparison with the datasets compiled for
stable continental regions and shallow crustal regions. Figure 2.3 shows the
magnitude-distance distribution of the data used for the development of
the subduction zone prediction model of Atkinson & Boore (2003). The
same axis ranges are used in this fi gure as in Fig. 2.2 and it is clear that that
datasets almost complement each other.
For the subduction zone models that do exist there are three key features
that distinguish them from their stable continental and shallow crustal
counterparts:

1. Subduction zone models typically include functional terms that account

for the depth of the earthquake event.