Civil Engineering Reference
In-Depth Information
by inspecting the fi tness of the POT data to the GP model. Another reason
for decreasing right tail may be attributed to the limitation of the adopted
earthquake-engineering-based model, which does not incorporate, for
example, the demand surge effects on repair costs and the indirect ripple
effects across local and regional economy. Thus the total economic seismic
loss is likely to be underestimated, when very severe seismic damage occurs
to a large number of buildings and infrastructure in a region. Based on
preliminary investigations, a threshold level corresponding to 0.995 annual
probability of non-exceedance (i.e. return period of 200 years) is chosen for
the seismic loss data for Group 2 (i.e. 6.579 million CAD).
To show the marginal distribution fi tting of the GP model to the POT
data for Group 2, a Q-Q plot is presented in Fig. 28.5, and the estimated
parameter values are indicated in the fi gure. The Q-Q plot of the POT data
shows that the marginal distribution fi tting based on the GP model for
Group 2 is adequate. This is an improvement, as good fi tting is achieved
using a simpler model with statistical foundation rooted in extreme value
theory, in comparison with the previous investigation by Goda and Ren
(2010), where diffi culty was encountered in fi nding a suitable parametric
distribution function for an entire data range. (Note that it was found during
preliminary investigations that the gamma distribution fi ts the POT data
equally well.) The Q-Q plot shows that the chosen threshold value, related
to Fig. 28.4, was indeed adequate.
10
9
Group 2:
Generalised Pareto distribution
x
= 0.130 and
b
= 12072
Threshold (
m
): 6.579
10
6
(CAD)
×
10
8
10
7
10
6
10
5
10
5
10
7
Quantile based on original data
10
6
10
8
10
9
28.5
Quantile-quantile plot of seismic loss data for Group 2.
Search WWH ::
Custom Search