Civil Engineering Reference
In-Depth Information
risk analyses that have been conducted to date epistemic uncertainty in
ground-motion predictions has been accommodated by using multiple
ground-motion models and assuming that the differences among these
encapsulate the epistemic uncertainty. However, there are some obvious
issues that are associated with the development process that have been
discussed thus far that dictate that individual models also have their own
epistemic uncertainty. In the following sections the topic of individual
model uncertainty is discussed as well as how this can be dealt with in
hazard and risk analyses.
2.4.1 Model-specifi c epistemic uncertainty
Inspection of Fig. 2.1 revealed that empirical datasets used to develop
ground-motion models are not balanced with respect to magnitude-distance
scenarios. That is, some scenarios have far more empirical constraint pro-
vided by the data than others. This clearly has an implication for the uncer-
tainty that we should place upon the model predictions for different
scenarios. However, in current practice it is generally assumed that this
uncertainty will be captured via the different predictions that would be
obtained through the use of different ground-motion models. Unfortu-
nately, there is no guarantee that this would in fact be the case. For example,
it may be that for scenarios for which few recordings have been made, the
model developers constrain the scaling of their model using theoretical
considerations. If all developers are subscribing to the same underlying
theory then it could be the case that the scaling implied by the predictions
of a suite of models could be very similar.
In what follows, it is assumed that differences in predictions among dif-
ferent models may arise from the application of different hypotheses about
how ground-motions should scale with the primary predictor variables. It
is further assumed that these interpretations are all consistent with the
available data to at least some extent. In this case, the differences in model
predictions would be accounting for epistemic uncertainty. However, there
are other factors that infl uence the uncertainty in a model and these relate
to what dataset was used for the regression, what the quality of this dataset
is, how the regression was performed, what criteria was used to retain or
reject possible functional expressions, etc.
The issue of the quality of the dataset is discussed in Section 2.4.2 con-
cerning the infl uence of uncertainty of the independent variables. To some
extent this section also considers the issue of how the regression analysis is
conducted. The issue of what criteria are used to retain or reject functional
expressions is referred to in Section 2.4.3 on prediction intervals and this
section also deals with how a dataset is compiled in terms of the data cover-
age as well.
Search WWH ::
Custom Search