Uncertainties in ground motion prediction in probabilistic seismic hazard analysis (PSHA) of civil infrastructure - Seismic Risk Analysis and Management of Civil Infrastructure Systems - page 49

this is larger depends upon how well constrained the regression coeffi cients
are as well as how strongly the functional for depends upon these coeffi -
cients. What is ordinarily found when implementing Equation (2.16) is that
the infl uence of the uncertainty in the regression coeffi cients becomes
larger as one considered scenarios that are further away from the centroid
of the data used to develop the model. In effect, the formulation of Equa-
tion (2.16) knows that the model is being pushed toward its limits of appli-
cability and increases its model-specifi c uncertainty accordingly.

2.5 Future trends

The approaches used to develop ground-motion models have been rela-
tively stable for many years now. However, there have been an increasing
number of contributions in the literature that demonstrate issues that are
associated with these existing approaches and it is likely that ground-motion
model development will take more of these issues into account in the
coming years. In the present section I outline what I anticipate to be the
major changes that are either already being considered or that are likely to
attract further attention in the near future.

2.5.1 Removal of the ergodic assumption

In order to constrain empirical models for the wide range of magnitude and
distance scenarios that are considered within probabilistic hazard and risk
analyses it is generally necessary to compile datasets of ground-motions
from different regions (see Fig. 2.1). However, as mentioned earlier, what
we ideally desire is a prediction of the distribution of motions that should
be expected when a particular event occurs in a particular location. Any
differences that exist between the propagating media, the source excitations
or the site responses in the regions from where the data has been obtained
will manifest as increased variance in the ground-motion model. One of the
key challenges that remains to be properly solved is how to remove the
ergodic assumption that is currently made. Studies have already been con-
ducted on this issue and have usually been undertaken in data-rich regions.
However, the vast majority of these studies have fundamental issues that
leave the solution of this problem unresolved. Unfortunately, it is not pos-
sible to enter into a full discussion of these issues here. However, from the
work that has been completed to date it is clear that a signifi cant reduction
in the site-specifi c standard deviation can be obtained once the ergodic
assumption is removed. That said, the removal of this assumption requires
a certain level of understanding about the specifi c characteristics of the
target site and region that is often not currently available. The net result is
that while the aleatory variability can be reduced quite signifi cantly, this