Probabilistic seismic hazard analysis of civil infrastructure - Seismic Risk Analysis and Management of Civil Infrastructure Systems - page 13

interest. The procedure can be easily adjusted to include multiple source
zones. It is noteworthy that in repeating the subsequent simulation cycles,
different/alternative parameters/models can be considered as a part of epis-
temic uncertainty.
1.2.3 Current issues in modern PSHA
PSHA results are sensitive to model components/assumptions. Therefore,
any seismic hazard assessment should be scrutinised by testing various cases
regarding key model components. For many applications, the choice and
weighting of the GMPEs, including the assigned aleatory uncertainty, are
the most critical of the input parameters. This should not be surprising, as
the GMPEs control the ground motions at the site for every earthquake
considered; however, the GMPEs and their uncertainty often do not receive
suffi cient scrutiny in PSHA. Moreover, as discussed in Section 1.4, common
practices regarding the use of multiple GMPEs to represent epistemic
uncertainty, along with the use of regression statistics to defi ne aleatory
uncertainty, should be improved. Another signifi cant impact is the spatio-
temporal characterisation of seismic activities and seismotectonic features,
which controls the rates of seismicity in and around the site (Beauval et al. ,
2006; Atkinson and Goda, 2011). These characterisations may be particu-
larly uncertain in low-to-moderate seismicity regions. Other less critical
issues requiring careful consideration include ensuring internally-consistent
defi nitions for various variables in PSHA, such as magnitude, distance, and
orientation of ground motion parameters.
To illustrate some of the above-mentioned issues, GMPEs used for assess-
ing seismic hazard in western Canada are compared in Fig. 1.6 (see Atkin-
son and Goda, 2011, for more details). Figure 1.6 shows signifi cant differences
of the predicted SA at 0.2 s for the interface and inslab events; attenuation
characteristics over distance differ signifi cantly and variability is large at
short-distance range (where empirical data are scarce). These differences
are caused by several factors, such as different ground motion datasets,
adopted functional forms, and employed approaches (e.g. regression analy-
sis of empirical data versus simulation of ground motions). Uncertainties
are particularly large when ground motions are required for types of seismic
events that have not yet been recorded (e.g. Cascadia subduction events)
and when empirical data coverage is poor in the magnitude-distance ranges
of engineering interest (e.g. as is often the case in low-seismicity regions,
where instrumentation tends to provide only sparse coverage).
It should be emphasised that the controlling factors for PSHA results
depend signifi cantly on location, site condition, and probability level of
interest. Thus it is important to conduct a full range of sensitivity analyses
as a part of PSHA. A sensitivity analysis may be even more important than