Presented at the
Eighth Canadian Conference on Earthquake Engineering,
Vancouver, June 13-16 1999
Original paper published in Proceedings, Eighth Canadian Conference on Earthquake Engineering, Vancouver, 1999 p 83-88
Lowering the Probability Level -
Fourth Generation Seismic Hazard Results for Canada
at the 2% in 50 year probability level
Adams, John1, Weichert, Dieter2, and Halchuk, Stephen1
1 Earthquakes Canada,
Geological Survey of Canada,
7 Observatory Cres., Ottawa K1A 0Y3
2 Earthquakes Canada,
Geological Survey of Canada,
Sidney V8L 4B2
NOTE: some changes have been made to this web version of the text
since the original document was published.
In particular, the calculated spectral floor values have been recalculated in the Results section.
ABSTRACT
The Geological Survey of Canada's new hazard model for Canada, released for public
comment in 1996, is intended to form the basis for seismic design codes in the next edition of
the National Building Code of Canada. As such it will produce Canada's fourth generation of
official seismic hazard maps. The three probabilistic parts of the model use two complete
earthquake source models and a separate estimate for the stable part of Canada to represent
the uncertainty in where (and why) earthquakes will happen in the future. A deterministic
estimate is made for the Cascadia subduction earthquake. A "robust" method is used to
combine these probabilistic and deterministic estimates: the mapped value for spectral
parameters is the largest of the values determined from these four sources. The previous code
was based on median (50th percentile) ground motions for a 10% probability of exceedence
in 50 years. We present the 2% in 50 year (equivalent return period of approximately 2500
years) results for major cities, compare them to the 10%/50 year values, and demonstrate that
they provide a better basis for achieving a uniform level of building safety across Canada.
INTRODUCTION
Three generations of seismic hazard maps for Canada have been produced at roughly 15-year
intervals (1953, 1970, 1985), and a fourth generation is now justified because there is
sufficient new information available to improve the hazard estimates (Basham, 1995). As
described by Adams et al. (1995), the new hazard maps incorporate a significant increment of
earthquake data, recent research on source zones and earthquake occurrence, together with
recently-published research on strong ground motion relations. Spectral acceleration values
("PSA"; all 5% damped) are computed for the range of periods important for common
engineered structures, together with the peak ground velocity (PGV) and peak ground
acceleration (PGA) parameters of the current (1985) maps. Our previous publications on the
4th generation maps provided 10%/50 year values directly comparable to the 1985 maps.
However, we now present 2%/50 year (0.000404 per annum) values, and show why they
should form the basis of the revised building code (Heidebrecht, 1999).
METHOD
The present method builds upon the work of Basham et al. (1985) which established the third
generation of seismic hazard maps for Canada. We apply the same Cornell-McGuire
methodology using a customized version of the FRISK88 hazard code (FRISK88 is a
proprietary software product of Risk Engineering Inc.), which includes epistemic uncertainty
into the computation. Details of the seismicity model are contained in Adams et al. (1999).
The two probabilistic source zone models, intended to span the range of likely models, are
substantially unchanged from Adams et al. (1996), but we highlight the following two
changes.
Seismic Hazard for the "Stable" Part of Canada. About half of the Canadian landmass has
too few earthquakes to define reliable seismic source zones, and on prior maps the hazard
computed for these regions came only from distant external sources. However, international
examples suggest that large earthquakes might occur anywhere in Canada (albeit rarely). To
improve the reliability of the estimate of seismic hazard for the stable part of Canada we
combine the earthquake activity of those stable continental shields of the globe comparable to
the Canadian shield (Fenton and Adams, 1997) and then compute the hazard, using eastern
strong ground motion relations, at the centre of a large octagonal source zone with this per-area activity level. As our selection of comparable shield areas was conservative, these
values are expected to be the lowest likely for any part of Canada not included in a source
zone, and so form an appropriate "floor". This floor is also used for sites west of the
Rockies, where the activity rates are likely to be higher, but the attenuation is stronger.
Deterministic Subduction Earthquake Ground Motions. Great earthquakes happen on the
Cascadia subduction zone on the average about every 500-600 years, so the median values
from our deterministic scenario have an annual probability about the same as for the 10%/50
year probabilistic values. However, those median values are not appropriate for the 2%/50
year hazard, since in circa 2500 years (i.e., approx 0.0004 p.a. or 2%/50 years) we can expect
to have 4-5 Cascadia subduction earthquakes with a suite of shaking levels. Hence, there is
an even chance one of the five will exceed the 75-80th percentile ground motions of the suite.
This percentile is very close to the 84th, so we have used the median plus one sigma ground
motions from our 10%/50 year calculations for the 2%/50 year deterministic hazard.
Strong Ground Motion Relations
For eastern Canada we continue to use the Atkinson and Boore (1995) relations, with the
same adjustment to take them from "hard rock" to "firm ground". While these relations are
fairly consistent with the majority consensus in the field, the excellent Saguenay data and
contrary opinions (e.g., Haddon, 1997) give us strong reservations about the shaking predicted
for the larger (magnitude about 6) earthquakes critical for hazard estimation. We hope that
this contentious issue will be resolved before the preparation of final maps for the National
Building Code. We would emphasize that no matter how good our source models, the
reliability of the final hazard values is highly dependent on the reliability of the extrapolations
within the attenuation relations used, as observational data from large eastern earthquakes is
sparse. For the western Canadian shallow source zones, including the subcrustal transition
zones west of Vancouver Island as well as the Queen Charlotte Fault, we continue to use our
adaptation of the ground motion relations of Boore et al. (e.g., 1997). For subcrustal source
zones deeper under Puget Sound and for the Cascadia subduction zone we have chosen to use
the Youngs et al. (1997) relations, adjusted to "firm soil", as we judge they are based on a
larger and better-selected data set than the Crouse relationship we previously used.
Ground Motion Parameters and Choice of Confidence Level
While the 1985 maps gave PGV and PGA values, we present spectral acceleration values for
0.1, 0.15, 0.2, 0.3, 0.4, 0.5, 1.0, and 2.0 second periods (denoted PSA0.1 etc) for both east
and west (note epistemic uncertainty is not available for PSA2 in the east). We also give
PGA values for both east and west, but PGV values for just the east (a PGV ground motion
relation is not available for the west). We provide values for two confidence levels, the 50th
percentile and the 84th percentile; the former is the median, and the latter includes a measure
of epistemic uncertainty arising from the incorporation of uncertainty in the model. Either
might be used for engineering design. The median is often chosen because it is a robust
parameter and can be expected to remain stable as the range of scientific opinion changes,
while the 84th percentile must be expected to fluctuate in future (hopefully decreasing over
the long term) as improved knowledge about epistemic uncertainty is incorporated into the
analysis.
Combining Hazard Estimates Using the "Robust" Approach
We combine the complete probabilistic hazard calculation from each of the two models,
together with the probabilistic "floor" level for the "stable" part of Canada and the
deterministic Cascadia model, in the fashion we term "robust" (Adams et al., 1995, 1999), i.e.
by choosing the highest value of the four sources for each grid point. The chief advantage of
the "robust" approach is that it preserves protection in areas of high seismicity but also
provides increased protection in low seismicity areas that are geologically likely to have
future large earthquakes.
RESULTS
Table 1 gives the 2% in 50 year probabilistic hazard values for selected Canadian cities,
itemizing separately the values for the two source zone models and their 50th and 84th
percentiles, together with the appropriate Cascadia values. Space precludes the presentation
of individual uniform hazard spectra, but these are given in Adams et al. (1999). The
"floor" hazard for the "stable" part of Canada, for firm-ground at the 2% in 50 year
probability level, is: PSA0.1=16%g; PSA0.2 =16%g; PSA0.3=12%g; PSA0.4=9.2%g;
PSA0.5=7.5%g; PSA1.0=2.9%g; PSA2%=1.0%g; PGA=11%g; PGV=0.045 m/s.