Sie sind auf Seite 1von 13

1299

Seismic hazard estimates for sites in the stable


Canadian craton
Gail M. Atkinson and Scott N. Martens

Abstract: We address the challenging and economically important problem of seismic hazard estimation at low probabilities
for stable cratonic regions such as central Canada which are characterized by very low seismicity rates. At low probabilities,
large uncertainties in the seismicity rates result in calculated seismic hazards that are not negligible for practical purposes
and may have significant implications for design and retrofit of critical structures such as hydroelectric and tailings dams.
We examine the statistics of earthquake magnitude recurrence in such regions and their implications for hazard estimates
at low probabilities. Recommendations for low-probability ground motions are provided that are related to the design
life of the facilities.
Key words: seismic hazard, earthquake, uniform hazard spectrum, stable interior craton, central Canada, dam.

Rsum : Nous traitons ici de lestimation du risque de sisme de faibles probabilits pour les rgions cratoniques
stables, telles que le centre du Canada, qui sont caractrises par des taux de sismicit trs faibles, un problme
dimportance conomique et posant un grand dfi. de faibles probabilits, les grandes incertitudes dans les taux de
sismicit engendrent des risques de sismes non ngligeables toutes fins pratiques et qui peuvent avoir un impact important
sur la conception et la rhabilitation de structures importantes telles que les barrages hydrolectriques et les digues des
parcs rsidus. Nous examinons les statistiques de la rcurrence de la magnitude dans de telles rgions ainsi que leurs
implications dans lestimation des risques de faibles probabilits. Les recommandations pour les mouvements de sol
faible probabilit sont fournies; celles-ci sont connexes la dure de vie thorique des installations.

Mots-cls : risque sismique, sisme, gamme uniforme de risques, craton intrieur stable, Canada central, barrage, digue.
[Traduit par la Rdaction] Atkinson and Martens 1311

Introduction earthquakes in low-seismicity regions can act to produce


relatively large expected ground motions at very low proba-
Reliable estimation of seismic hazards is a challenging bilities; the reason for this is that uncertainty is an important
problem in regions of low seismicity such as the stable component of hazard estimates, and the low seismicity rates
Canadian craton (i.e., the Prairie provinces from western imply statistically large uncertainties.
Ontario to western Alberta). This issue is important because
critical structures such as hydroelectric and tailings dams
Motivation for this study
must be designed to withstand ground motions with proba-
The estimation of low-probability ground motions in stable
bilities as low as 1/10 000 per annum (pa). At low probabilities,
regions of Canada has been an issue of recent debate because
ground motions in stable craton regions are not insignificant
of the introduction of a stable craton model and associated
because moderate to large earthquakes have a probability
floor values of ground motion in the 2005 National Building
that is low but not negligible. There are significant economic
Code of Canada (NBCC) (NRCC 2005) seismic hazard
implications in designing for these motions, and thus their
maps, which are based on the Geological Survey of Canada
accurate estimation is of concern. Paradoxically, the rarity of
(GSC) fourth-generation seismic hazard model (Adams and
Halchuk 2003). For the first time, the national seismic hazard
Received 15 January 2007. Revision accepted 24 April 2007.
maps include an estimate of expected ground motions for
Published on the NRC Research Press Web site at cjce.nrc.ca regions in Canada where seismicity levels are too low to
on 30 October 2007. allow their reliable estimation based on regional historical
seismicity. This represents about half the Canadian landmass
G.M. Atkinson.1,2 Department of Earth Sciences, Biology and and includes several major cities, such as Winnipeg and
Geological Sciences Building, The University of Western Ontario, Edmonton. In previous versions of the maps, hazards to
London, ON N6A 5B7, Canada.
S.N. Martens. Klohn Crippen Berger Ltd., 114 6815 8th Street
these low-seismicity regions were assumed to come only
NE, Calgary, AB T2E 7H7, Canada. from distant external sources. International examples sug-
gest, however, that large (M6+) earthquakes can happen any-
Written discussion of this article is welcomed and will be received where, albeit with low probability (Johnston et al. 1994). To
by the Editor until 29 February 2008. address this problem, the latest national seismic hazard maps
1
Corresponding author (e-mail: gmatkinson@aol.com). (Adams and Halchuk 2003) include a minimum floor model
2
Present address: Department of Earth Sciences, Carleton for seismic hazard. The floor model characterizes the hazard
University, Ottawa, ON K1S 5B6, Canada. from a large background zone of seismicity, for which the

Can. J. Civ. Eng. 34: 12991311 (2007) doi:10.1139/L07-057 2007 NRC Canada
1300 Can. J. Civ. Eng. Vol. 34, 2007

rate of activity (magnituderecurrence relation), on a per particular, the assumption that the applied seismicity rate
area basis, is based on that observed globally for stable should be based on a global weighted average for all stable
craton regions such as central Canada (Fenton et al. 2006). cratons (including other regions of the world that are both
The advantage of this approach is that it allows substitution more and less active than central Canada) may not be the
of space for time to obtain more robust statistics on earth- best model to use for critical structures in central Canada.
quake occurrence rates in regions of very low seismicity. This study presents an analysis of uncertainty in seismicity
The disadvantage is that not all of the included regions may rates and their implications for seismic hazard in central
be truly comparable to the target region from a tectonic Canada.
point of view; our understanding of the geologic controls on The scope of the study is as follows. First, the region that
damaging earthquakes in stable cratons is poor, and so the can be characterized as the CC craton is defined as a subset
grouping together of various of these regions is speculative. of the larger NA craton. Then, the seismicity rates and their
Thus, the assessment of the hazard is uncertain and may be uncertainty within this region are characterized by a careful
inaccurate. examination of magnituderecurrence rates and their uncer-
The national seismic hazard maps provide ground motions tainty. Lastly, earthquake ground motions for annual proba-
for a probability of exceedance of 2% in 50 years, or bilities of 1/1000 to 1/10 000 for hard-rock site conditions
1/2500 pa. The documents supporting the maps state that are calculated. Parameter uncertainty is addressed in the analysis
the model is not intended for application to structures not using a logic-tree approach.
covered by the NBCC, and that more detailed investigations
are appropriate for specific projects or for lower probabilities.
Thus it is clear that the NBCC maps are not intended for Analysis methodology
application to critical structures, in the Prairie provinces or
elsewhere. However, because the GSC hazard maps are The probabilistic seismic hazard analysis performed for
publicly available and provide ground motions for all regions this study follows the CornellMcGuire method (McGuire
in Canada, they are often used as a starting point to estimate 1993). This well-known methodology is the same as that
earthquake ground motions for analysis of structures not used in the 2005 NBCC (NRCC 2005) national seismic
covered by the NBCC. Thus, they are often used in the analysis hazard maps. It is comprised of the following steps: (i) define
of dams in central Canada, if it is considered that site- the spatial distribution of earthquakes using source zones,
specific studies are not warranted. A probability of and in this application we use a single regional area source
1/10 000 pa is typically used to represent the maximum zone that is assumed to be spatially homogeneous in its
design earthquake for analysis of dams having a very high potential for future activity; (ii) describe the frequency of
consequence rating, and for some dams with a high conse- occurrence of earthquakes within each source zone using a
quence rating. As a first-level screening tool, it has been magnituderecurrence relationship; (iii) define the amplitude
common practice to extrapolate the GSC estimates to a prob- of ground motions as a function of magnitude and distance;
ability of 1/10 000 pa (note that this extrapolation is not rec- and (iv) calculate the frequency of exceeding a specified
ommended by the GSC or the authors of this study.) The level of ground motion at a site by integrating the hazard
ground motions implied by this extrapolation could be suffi- contributions from all source zones over all magnitudes and
cient to trigger liquefaction in loose sands and may lead to distances. Selection of the source zone and associated
low calculated factors of safety against slope instability magnituderecurrence relation is critical to the assessment
using pseudostatic analyses. A consequence is that detailed of seismic hazard in central Canada, where earthquakes from
seismic analyses may be required for many structures that which to compile statistics are sparse.
were previously below the screening level for such analy- Uncertainties in the input parameters, especially the
ses, resulting in an expense and level of effort that may not magnituderecurrence rates, are also critical to the evaluation.
be justified given the low historical seismicity levels. There are two types of uncertainty. Aleatory uncertainty, or
randomness in process, is due to the physical variability
inherent in future events, such as the scatter in ground motion
Purpose and scope of this study values about a median ground motion prediction equation.
In this study, we address the question of what ground Aleatory uncertainty is incorporated directly into the proba-
motions should be expected at low probabilities for sites in bilistic calculations by integrating over the statistical distri-
the central Canadian craton (CC craton), which we define as bution in the ground motion relations and by considering
the low-seismicity region within central Canada as shown in randomness in location. Epistemic uncertainty is a result of
Fig. 1. This area is more restrictive than the stable North incomplete knowledge about the physical processes that
American craton (NA craton) region defined by Fenton et control the random events. For example, we are uncertain as
al. (2006) (Fig. 1). We follow a methodology similar to that to whether the selected ground motion prediction equation is
used by Adams and Halchuk (2003) in the NBCC maps but a true representation of the median motions. Epistemic
look at the Canadian craton rates in more detail and extend uncertainty is incorporated in the analysis via the use of a
the calculations to lower probabilities. logic tree. In a logic tree, model uncertainties are treated by
A careful look at the constraints on seismicity rates in the considering a range of possible choices for input parameters
craton is warranted. The GSC stable craton model used a such as the ground motion prediction equations or the source-
number of arbitrary assumptions that, although reasonable zone models. The relative likelihood of each of the choices
for the NBCC application, should be revisited prior to appli- is weighted subjectively. Each calculated probability of exceed-
cation to low-probability estimates for critical structures. In ance of a specified ground motion is multiplied by the branch

2007 NRC Canada


Atkinson and Martens 1301

Fig. 1. Historical seismicity (to end of 2004), showing all Canadian events of M 2.5. Events in the USA are shown only where
included in the Geological Survey of Canada earthquake catalogue. Boundaries of the stable Canadian craton as defined by Fenton et
al. (2006) are shown, along with boundaries of the central Canadian (CC) craton as defined in this study.

probability. The sum of the weighted branch probabilities Magnituderecurrence statistics for cratonic
produces the mean-hazard probability curve, which is inter- regions
polated to provide the mean-hazard ground motion for each
probability level. The probabilities calculated from the branches Definition of source zones
are ordered to calculate the median-hazard ground motion Seismicity in Canada generally occurs on buried faults,
amplitude and various percentiles. Thus, for any probability whose locations and activity rates are not well known. Earth-
level we obtain a mean-hazard ground motion amplitude, quake activity is therefore grouped into seismic source zones,
along with the median-hazard value and other confidence which are regions that are assumed to have a spatially uniform
limits. This allows us to make statements such as I am 85% probability of future earthquakes. These source zones are
confident that the annual frequency of exceedence of 20%g defined based on historical seismicity patterns and regional
peak ground acceleration is less than 102 (McGuire 2004). geology and encompass a group of unknown subsurface
In this study, the key input parameters are included in the structures. Within the central region of Canada, seismicity
logic tree, and results are calculated at the mean and 16th, levels are very low, and thus the source zones must be
50th, and 84th percentile confidence levels. The mean-hazard defined in a broad way. We approached the problem by
ground motion values are generally preferred for site-specific defining a large cratonic region within Canada and the
studies (when not otherwise directed by regulatory guid- northern US and examining observed seismicity rates and
ance), as the mean is the expected value and takes account their variability within the region. The assessment incorpo-
of the influence of the lower probability branches of the rates the most recent seismicity information (to 2005) and
logic tree. The hazard calculations are made using the soft- thus updates previous work for the 2005 NBCC (which used
ware package EZ-FRISK (Risk Engineering Inc. 2005), seismicity to 1991).
which performs the integrations over source zones, magni- The definition of a stable craton region within North
tudes, and distances, to obtain the probability of exceeding America, which might be considered as a broad analogue for
any specified ground motion value for each combination of the seismicity levels that could be expected to occur in
input parameters. The EZ-FRISK code is run for each central Canada, was the starting point of our analysis. It
combination of input parameters to generate the multiple allowed us to substitute space for time in establishing better
branches of the logic tree; Visual Basic software was used to rate estimates in regions where the seismicity levels are too
automate processing of the EZ-FRISK output and to cal- low for reliable statistics to have been established in our
culate the mean and fractile probabilities from the multiple 100 year recording history. Unlike Adams and Halchuk (2003)
output results, following the logic tree weights. and Fenton et al. (2006), we did not consider other cratonic

2007 NRC Canada


1302 Can. J. Civ. Eng. Vol. 34, 2007

regions around the globe in our assessment, as there may be craton, we defined an alternative CC craton, as shown in
a variety of tectonic factors, such as the most recent time of Fig. 1; note that the CC craton is a subset of the NA craton.
glaciation or other tectonic activity, which could differ from The CC craton is more representative of stable craton tectonics
the North American setting. Restricting our focus to North in central Canada than is the NA craton because it is more
America reduced the associated tectonic variability. Never- specifically restricted to cratonic regions that have not been
theless, any large zone that covers much of central North reactivated in about the last 500 Ma. The CC craton tectonic
America will include regions of varying levels of activity, environment is believed to be most representative of typical
some of which might be explainable by local geologic factors. seismicity levels for stable cratonic regions in central Canada.
However, following the logic of Johnston et al. (1994), we It has been postulated that there may be a linkage between
excluded regions with major geologic influences in the last the low rates of observed seismicity in the Canadian region
500 million years (Ma) or so, such as areas rifted during of the NA craton and the stresses imposed by glaciation in
early opening of the Iapetus Ocean. that region. However, Adams and Bell (1991) note that, al-
Our initial definition of the stable NA craton was adopted though deglaciation severely stressed the North American
directly from the NA craton model of Fenton et al. (2006); it plate about 10 000 years ago, the stresses associated with
includes the North American continent minus its actively deglaciation do not currently dominate the stress field and
deforming regions. Their intent was to remove (i) all regions have little residual effect on the stress field. It is therefore
of non-Precambrian crust; (ii) all passive margins, together not likely that the patterns of glacial cover over North America
with a zone of approximately 200 km inland from passive strongly affect the current observed patterns of seismicity.
margin coasts; and (iii) a 200 km zone around regions of
Phanerozoic (last 500 Ma) deformation. The idea was to Magnituderecurrence statistics
capture the range of stable craton behaviour observed in We considered the uncertainty in magnituderecurrence
Canada and the northern US. This region is not expected to parameters using these two alternative craton definitions (NA
be homogeneous in its seismicity levels, but merely repre- craton and CC craton). For general applications in central
sentative of the range of seismicity observed in typical NA Canada, the CC craton model is considered most appropriate.
craton environments. This was the broadest definition of po- The broader NA craton model might be more appropriate for
tential analogue craton that we considered. The Fenton et al. long-term facilities such as nuclear waste storage, as it takes
definition of the stable NA craton is shown in Fig. 2 along a longer geologic view of features that might be reactivated
with our discretization of the region as a grid of points (NA and provides more stable statistics for larger events.
craton). The earthquake catalogue used for this study, as plotted in
It has been pointed out that the Fenton et al. (2006) craton Fig. 1, was a compiled catalogue that included all events
(Fig. 2) actually includes parts of some Phanerozoic rifts from the GSC database for areas north of 40N (www.seismo.
(R.L. Wheeler, personal communication, 2006). These are nrcan.gc.ca) and data from US sources (Incorporated Research
the western part of the Oklahoma Aulocogen; parts of the Institution for Seismology (IRIS) and US Geological Survey
Rome Trough of central and eastern Kentucky, West Virginia, National Earthquake Information Center (NEIC); www.ncedc.
and southwestern Pennsylvania; and a few normal faults that org) for south of 40N, up to 2005. All event magnitudes
define the northwestern limit of the Iapetan passive margin were converted to moment magnitude (M) using the formu-
(Wheeler 1995). The main rifting in these areas preceded lae of Johnston (1996a, 1996b, 1996c) for events listed as
500 Ma, but it extended from Late Proterozoic to Early local magnitude (ML), body wave (mb), or surface wave
Cambrian, with later reactivation in some places (R.L. (MS). The empirical conversion of Atkinson (1993) was used
Wheeler, personal communication, 2006). For this reason, it for events with magnitude listed as Nuttli magnitude (MN).
could be argued that the Fenton et al. definition of the stable The catalogue is not complete in the regions outside of the
craton should be revised to exclude these more recently defined NA craton area of study.
active areas. For example, Johnston et al. (1994) chose to The earthquake activity of the craton can be characterized
distinguish Precambrian rifts, which are not more seis- using a magnituderecurrence relationship of the Gutenberg
mically active than the surrounding unrifted craton, from Richter form
Phanerozoic rifts, which are more active; they depict the
[1] log[N(M)] = a bM
Oklahoma Aulocogen and Rome Trough as Phanerozoic
structures. Similarly, the band of seismicity that is observed where N(M) is the cumulative annual rate of exceedance of
along the BoothiaUngava arch (northeast band within the magnitude M, a is the rate parameter, and b describes the
Fenton et al. craton, as seen in Fig. 1) could also be relative likelihood of large and small earthquakes. Typical b
excluded on tectonic grounds. The Boothia uplift was values range from 0.8 to 1.0 in most regions throughout the
geologically active from the Paleozoic to the Cretaceous world. Magnituderecurrence parameters a and b can be
(Okulitch et al. 1986) and was more recently reactivated in computed from the earthquake catalogue by compiling statistics
a CretaceousTertiary rifting episode (Kerr 1977, 1980). For on the rate of occurrence of each earthquake magnitude.
this reason, Adams and Basham (1989) treat the Boothia To calculate magnituderecurrence statistics within a
Ungava seismicity as being a tectonic zone that is distinct defined craton region, we first determined the completeness
from the craton seismicity environment. It is controlled by windows of the catalogue (in magnitude and time) across the
younger deformation features. study area. This need arises from the uneven instrumentation
To consider a more homogeneous tectonic environment, capabilities to detect earthquakes, in both space and time.
which excludes the previously noted younger tectonic features The completeness intervals vary across the large geographic
that lie within the boundaries of the Fenton et al. (2006) region considered, as the instrumental coverage is generally

2007 NRC Canada


Atkinson and Martens 1303

Fig. 2. Fenton et al. (2006) definition of the North American (NA) craton (lines), and grid of points used in this study to represent the
NA craton.

better in the south, and reported historical events begin completeness intervals of Table 1. The possible effect of
earlier in the east. Also, the completeness varies among the zero-rate grid points was accounted for by considering a
available catalogues. We evaluated completeness separately minimum seismicity observation of 0.5 events of M3 in the
for six areas based on the catalogue data. We judged this time window of record for each grid point (and correspond-
based on the year for which the number of events per decade ingly smaller values for higher magnitudes, assuming a
ceased to grow with time, and considering that the number regional b value of 1). This assigned minimum rate is the
of events should increase by about a factor of 10 as the mag- largest that would be consistent with the observed seismicity
nitude level decreases by 1 unit (based on the Gutenberg (i.e., if the true average rate in the time interval of observa-
Richter relation). The completeness intervals adopted for the tion is <0.5, then we would not expect to have detected any
recurrence statistics are shown in Table 1, where NY3 indi- such events, as 0.5 rounds to 0 whole earthquakes). It turned
cates, for example, the first year at which the reporting of out that this was not a significant factor, as the average rates
events of M 3 appears to become complete within that are very similar if no minimum counts are assigned. We also
region. The completeness intervals are a matter of judgment; examined the influence of the selected radius for the grid-
the procedure used does not guarantee complete reporting search algorithm by considering values from 50 to 250 km.
over all parts of each region over all time intervals used. It Note that these radii may overlap as we move from one grid
does, however, ensure a stable average rate of recurrence for cell to the next, depending on the selected radius. This was
each magnitude range, within each region, that does not also not a significant factor, as discussed later in the paper.
change with time over the period considered. Lastly, we also considered the effect of the selected grid
A moving-average grid algorithm was used to calculate geometry. In our calculations, we use a simple grid that is 1
seismicity rates and their uncertainty at various magnitude in latitude by 2 in longitude (about 100 km by 100 km);
levels throughout each defined craton region. The rates however, this means that the northern grid points are more
observed across the craton were used to define statistics at closely spaced than the southern grid points. We tested
each magnitude level, such as mean and standard deviation, whether this biases the results by using an area-weighted
on a per annum, per unit area basis. These statistics can be average rate over the grid points, instead of a simple average.
used to define uncertainty ranges for magnituderecurrence We confirmed that an area-weighted average would not
relations for central Canada. The NA craton was discretized significantly change the calculated rates. In summary, average
as a grid of points as shown in Fig. 2. The grid points were rates over the grid are robust, and not sensitive to the details
defined just inside the areal definition of the NA craton pro- of choices made in how the grid is defined, or to whether
posed by Fenton et al. (2006) (solid line) so as to avoid edge particular points are included.
effects. Levels of seismicity within this region are highly For each grid point, we counted the number of exceed-
variable, but generally low. We determined the occurrence ances of each magnitude level, normalized to a per century
rates for M3, M3.5, M4, M5, and M6 events within a given basis, using events within a specified radius that pass the
radius of each grid point, using the catalogue data within the completeness criteria. This provides an estimate of the rates

2007 NRC Canada


1304 Can. J. Civ. Eng. Vol. 34, 2007

Table 1. Earthquake catalogue, showing the magnitude completeness intervals.


Range of coordinates
Region NY3 NY3.5 NY4 NY5 NY6 Lat. (N) Long. (W)
Southwest 1990 1990 1900 1860 1800 3550 90105
Southeast 1990 1930 1870 1820 1800 3550 7590
Central west 1970 1960 1960 1900 1900 5060 90120
Central east 1980 1960 1940 1900 1900 5060 6090
Northwest 1980 1970 1960 1900 1900 6075 90125
Northeast 1970 1970 1960 1900 1900 6075 6590
Note: For example, NY3 indicates the first year at which the reporting of events of M 3 appears to
become complete within that region.

for that grid point, which we normalized to an area of Fig. 3. Cumulative recurrence statistics for the stable North
106 km2 (by dividing by the area enclosed by the given American (NA) craton (Fig. 2) and the central Canadian (CC)
radius, then multiplying by 106 km2). By repeating these craton (Fig. 1). Symbols show mean rates, and one-sided error
calculations for every grid point within the defined craton bars show standard deviation. Lines show least-squares relation
region, we obtained mean recurrence rates and their standard for NA craton (determined slope of 1.01 0.07; minimum rates
deviations (indicating variability of rates across the grid). assumed) and fitted relation for CC craton using assigned slope
The recurrence statistics for the entire NA craton are shown of 1.0. A radius R of 150 km about each grid point was used.
in Fig. 3 using a representative radius value of 150 km. The
plotted seismicity rates are the mean rates (N(M)) across
all grid points, per 100 years, per 106 km2. The line is a
least-squares fit to the data, which follow a well-behaved
GutenbergRichter magnituderecurrence relation:
[2] log N(M) = 5.35 1.0M

The calculated slope (= 1.01 0.07) is very close to the


typical b value of 1 that is commonly expected in low-
seismicity regions. Figure 3 also shows the rates that were
obtained by considering just the CC craton, as defined by
Fig. 1. These rates are lower than the average for the NA
craton by about a factor of three. Note that the M6 rate for
the CC craton is not reliable because of the limited statistics;
it is best estimated by considering the NA craton rate and
dividing by a factor of three. Thus the magnituderecurrence
relation for the CC craton (per 100 years, per 106 km2) is
[3] log N(M) = 4.88 1.0M

This is also a well-behaved GutenbergRichter relation-


ship; in this case, the slope has been fixed at the value of 1
based on the better constrained result of eq. [2]. The rate
parameter in the magnituderecurrence relations is a source
of considerable uncertainty in the hazard calculations. Based because they provide intermediate values for the seismicity
on the data plotted in Fig. 1, the rate parameter varies by rates. Neither the rates nor the slopes of the magnitude
about a factor of 3, with the greatest uncertainty being in the recurrence relation are sensitive to the choice; thus the com-
CC craton model, which is less constrained than the NA puted hazard would not differ significantly if a different
craton model. value of radius was adopted for the averaging. The small
The sensitivity of the results of Fig. 3 to details of the deviations in rates obtained when considering different aver-
grid-calculation algorithm was tested, considering sensitivity aging scales may reflect weak clustering properties of seis-
to the radius used for the grid recurrence calculations and micity within the NA craton.
sensitivity to the minimum rate assumptions. Figure 3 shows In computing the recurrence rates for each grid point, a
the results for a grid radius of 150 km (representing a mod- minimum rate of occurrence of events just below the thresh-
erate overlap of radii) and assuming minimum recurrence old for detection was assumed as an interpretation of the
rates for empty intervals. In Fig. 4, the results for grid radii empty intervals. Thus if the observed rate of events of M 3
of 50, 100, 150, 200, and 250 km are compared for the NA for a particular grid point in the period of observation was 0,
craton. It is seen that the radius selected is not a significant we assigned a fixed rate of 0.5 for M 3 (since an observa-
factor in the computed rates (especially in light of their tion of 0 would be expected to correspond to any rate <1).
uncertainties), and that the radius value of 150 km provides Based on the GutenbergRichter relation, the corresponding
results that are typical. We adopted the 150 km radius values assumed rates for M 4, M 5, and M 6 are 0.05, 0.005,

2007 NRC Canada


Atkinson and Martens 1305

Fig. 4. Sensitivity of cumulative recurrence statistics for the Fig. 5. Comparison of North American (NA) craton and central
North American craton to the radius about each grid point used Canadian (CC) craton recurrence models of this study with those
in the calculations and to the assumption of minimum recurrence of the Geological Survey of Canada (GSC) and Fenton et al. (2006).
rates for empty calculation areas. Symbols show mean rates, and All relations converted to moment magnitude. F, floor model.
one-sided error bar shows standard deviation (R = 100 km case).
Line shows adopted regression result from Fig. 3 (R = 150 km
case; imposed minimum rates).

global values. Our NA craton model rates are very similar to


the corresponding NA craton rates given by Fenton et al.
and the GSC model. Our CC craton rates are considerably
higher than those of the central Canada model of the GSC,
and 0.0005, respectively. To test the effect of this assump- as the GSC central Canada model excluded all areas within
tion, in Fig. 4, the plotted points are based on an analysis their defined source zones, such as the area enclosing the
that does not include an assumed minimum rate (e.g., assigned activity near James Bay.
rate = 0 if no events observed within the grid radius), The US Geological Survey also uses a background seis-
whereas the plotted line does assume the minimum rates. It micity floor for American cratonic regions in its seismic
is evident that the computed rates in this case are very similar hazard computations, although it is implemented somewhat
to those found previously. We conclude that the results differently. Without going into the details, as their model is
presented in Fig. 3 are robust with respect to details of the complex and more applicable to the USA than to central
approach taken. Canada, it is interesting to note that the corresponding rates
(per 106 km2 per year) they apply are 0.15 for M4, 0.017 for
Floor models used for national hazard map calculations M5, and 0.0031 for M6 (with b value of 0.95) (C.S. Mueller,
The GSC fourth-generation seismic hazard model includes personal communication, 2006). These rates are very similar
the floor, or F model, for the central, stable region of to those for our NA craton model. However, they apply this
Canada where there have been too few earthquakes recorded model, with a 20% weight, only in areas where the historical
to define reliable source zones and activity rates. This model seismicity rate is lower than the background rate. Because of
is based on earthquake activity rates for three separate this implementation approach, and other differences in hazard
regions, namely central Canada, the portion of North America models, the US Geological Survey seismic hazard model, for
that is geologically similar to central Canada, and global quiet sites in northern Montana near the CanadaUSA border,
regions that are geologically similar to central Canada. The generally produces ground motion estimates that are slightly
activity rates from these regions were combined using lower than those reported in this study (C.S. Mueller,
weightings of 0.2, 0.4, and 0.4, respectively, in the deriva- personal communication, 2006).
tion of the GSC national seismic hazard maps. The magnitude
recurrence parameters for this study are compared with the Seismic hazard analysis parameters and
Fenton et al. (2006) data for various stable continental cratons weights
worldwide and the three components of the GSC fourth-
generation floor model in Fig. 5. All relations have been The input parameters typically required to calculate the
converted to moment magnitude or equivalent (such as MS) earthquake ground motions using the CornellMcGuire
for the comparison, using the conversions recommended by method include the source-zone geometry, the magnitude
the authors of each study. The b values determined in all recurrence parameters, focal depth, and ground motion rela-
cases are between about 0.8 and 1.0, which are typical tions giving amplitude as a function of magnitude and

2007 NRC Canada


1306 Can. J. Civ. Eng. Vol. 34, 2007

distance. This section describes the adopted input parame- Fig. 6. Logic tree for consideration of uncertainties in seismic
ters. Alternatives for each significant parameter were consid- hazard analysis. Weightings are given in parentheses. CC, central
ered, with the results being probabilistically combined using Canadian; NA, North American.
a logic-tree format. Figure 6 shows the input logic tree.

Weights for craton source models and recurrence


parameters
Two alternative source-zone definitions for the craton, and
their associated magnituderecurrence relations, were defined
in the previous section. These alternative models can be
weighted to represent epistemic uncertainty in future earth-
quake recurrence patterns in central Canada. The selected
weights may in part depend on the intended application. The
NA craton parameters would in general be conservative for
central Canada because there are regions in the NA craton,
such as the BoothiaUngava region, that are more active
than is typical for the craton due to more recent geological
deformation. It is unlikely that central Canada would experi-
ence such seismicity rates over the next 100 years or so
because the tectonic environment is different, being charac-
terized by less recent deformation. However, the NA craton
model might be appropriate for facilities with very long
Table 2. Suggested model weightings for varying design lives.
design lives, such as a nuclear waste repository. The CC
craton rate is likely appropriate for facilities in central CC craton NA craton
Canada with design lives of the order of 100 years or less, weighting (%) weighting (%) Suggested design life
as it considers the overall seismicity levels across a broad
90 10 Tens to hundreds of years
region that is tectonically similar and exhibits similar seis-
70 30 Hundreds to thousands of years
micity rates. The NA craton and CC craton models can be
50 50 Thousands to tens of thousands
combined in a logic tree to allow hazards to be estimated for
of years
a range of design lives.
In evaluating the weights that should be given to alterna- Note: CC, central Canadian; NA, North American.
tive source models, a relevant question concerns the stability
of current seismicity rates over time. In general, studies Within the CC craton model, uncertainty in the recurrence
undertaken in regions with very long historical seismicity rates is considered using a source-activity rate parameter
records, such as China, indicate that the best indication of that is a factor of 3 higher and lower than the best estimate,
activity levels in the next 50100 years is obtained by with a weighting of 0.1 for the high rate and 0.2 for the low
assuming the rates from the last 50100 years will be repeated rate (Fig. 6); a lower weight was accorded the higher esti-
(as opposed to using, say, an average rate for the last mate, as it is deemed unlikely that the CC craton rates would
1000 years) (McGuire and Barnhard 1981); the reason for be as high as the overall rates calculated for the NA craton
this is that, even in regions such as China that show time- (which coincide with the high CC craton estimate). The best
varying behaviour of seismicity rates, the rate of change in estimate rate was given a weighting of 0.7. A single rate
rates with time is slow. An analogy can be made with the parameter was used for the NA model, as this model is given
weather: in the absence of any other information, our best a lower or equal weighting in the logic tree, and the statistics
forecast for tomorrows temperature is that it will be the are better constrained for the NA craton model.
same as today.
A weight of 0.1 for the NA craton model and 0.9 for the Ground motion prediction equations and weights
CC craton model is suggested for facilities with a relatively Ground motion relations describe the amplitude of a ground
short design life in the range of 50100 years which pose no motion parameter, such as acceleration or velocity, as a func-
significant hazard following decommissioning. Over this time tion of earthquake magnitude, distance from the source, and
period, it is most likely that the future patterns of seismicity spectral period. The Atkinson and Boore (1995) ground
will be similar to the patterns of seismicity in the recent motion relations were used in this study for consistency with
past. For facilities with a longer design life, such as tailings the fourth-generation seismic hazard model of the GSC for
impoundments that are not operational but do present an eastern and central Canada. (Note that an updated version of
ongoing stability hazard and must be stable in perpetuity, these relations has been derived by Atkinson and Boore (2006)
more weighting should be given to the NA craton model. but was not yet available at the time this study was per-
This model may be a better estimate of the patterns of seis- formed.) The Atkinson and Boore (1995) relationship was
micity that may occur over long periods of time (thousands derived for hard-rock sites (National Earthquake Hazards
of years or more). Three different weighting schemes are Reduction Program (NEHRP) site class A). An important
suggested in Table 2 that correspond to a range of design aspect of the uncertainty in the hazard calculations is the
lives. Results for these three different weightings between statistical scatter about the median attenuation relationship,
the CC craton and NA craton are presented in the next section. which is characterized by the standard deviation of the natural

2007 NRC Canada


Atkinson and Martens 1307

Fig. 7. Uniform hazard spectrum for central Canada for a weighting of 90% central Canadian craton model and 10% North American
craton model (short design life): (a) mean, median, and 16th and 84th percentile values (horizontal component, hard rock) for 2% in
50 years (1/2500 pa); (b) corresponding results for 1/10 000 pa. PGA, peak ground acceleration; PGV, peak ground velocity.

logarithm of the residuals (slnk, where k is the ground motion Other input parameters and weights
parameter), often referred to as sigma. This aspect of uncer- The source models, recurrence parameters, and ground
tainty is incorporated directly in the hazard calculations by motion relations are the most important parameters for the
integrating over the variability about the median ground seismic hazard analysis for sites in central Canada. Other
motion relations. Larger sigma values lead to larger calcu- parameters have only a minor influence on the results. Focal
lated ground motions, particularly at low probability levels. depths of 5, 10, and 20 km were considered in the logic tree,
For this study, the sigma values recommended by Atkinson with weightings of 0.25, 0.50, and 0.25, respectively. These
and Boore were used. These period-dependent factors ranged depths and weightings are consistent with the GSC fourth-
from 0.55 to 0.62 natural logarithm units. generation model, and with the histograms of Wheeler and
Epistemic uncertainty in the Atkinson and Boore (1995) Johnston (1992), who compiled depths of moderate to large
attenuation relation was characterized by Atkinson (1995), earthquakes in eastern North America.
who provided recommended equation coefficients and weight- The minimum magnitude cutoff selected for inclusion in
ings for upper bound, median, and lower bound relations the hazard calculations was M4.75. Deaggregation analyses
based on Atkinson and Boore. The more accurate tabular of the hazard indicate that, numerically, some additional hazard
format of the Atkinson and Boore attenuation relations was does accrue from smaller magnitude events very close to the
used, rather than the quadratic approximation to these data site. It is empirically well accepted that below some threshold
used by Adams and Halchuk (2003) in the GSC hazard map magnitude, earthquakes do not generally cause damage to
calculations. To implement the upper and lower bound rela- engineered structures, even if they produce high acceleration
tions, the ground motion values were factored up or down amplitudes, as the ground motion will have a very short
based on the upper and lower equation coefficients listed in duration and little effect due to its lack of energy at interme-
Atkinson. This study used weights of 0.28 for the lower diate to long periods. The usual accepted solution is to accrue
bound, 0.44 for the median, and 0.28 for the upper bound hazard only from earthquakes having a magnitude above
relations, respectively, for all periods (simplified from Atkinson which some damage could potentially occur. This threshold
1995). magnitude is typically taken to be in the range of M4.5 to

2007 NRC Canada


1308 Can. J. Civ. Eng. Vol. 34, 2007

Fig. 8. Uniform hazard spectrum for central Canada for a weighting of 70% central Canadian craton model and 30% North American
craton model (long design life): (a) mean, median, and 16th and 84th percentile values (horizontal component, hard rock) for 2% in
50 years (1/2500 pa); (b) corresponding results for 1/10 000 pa. PGA, peak ground acceleration; PGV, peak ground velocity.

M5.0. For this study, M4.75 was selected for consistency of the source is not affected by the location of the source
with the fourth-generation GSC model. boundaries.
The best estimate of the maximum magnitude is based on Calculated ground motions are presented for three combi-
Johnston et al. (1994) and Fenton et al. (2006); these studies nations of weighting between the CC craton model and the
concluded that M7.0 is the approximate maximum magnitude NA craton model, as listed in Table 2. Ground motions are
that occurs in unrifted stable continental cratons throughout presented for the random horizontal component, for site class
the world. This maximum magnitude is consistent with the A (hard rock) conditions. The uniform hazard spectra for the
GSC floor model. Upper and lower bounds of M7.2 and mean, median, 16th percentile, and 84th percentile results
M6.8, respectively, were included in the logic tree, following are presented in Fig. 7 for the 90% CC craton and 10% NA
Fenton et al. The best estimate, lower bound, and upper craton weighting. Corresponding spectra for the 70% CC
bound were given weightings of 0.6, 0.3, and 0.1, respec- craton and 30% NA craton weighting and the 50% CC craton
tively, in the logic tree. The lower estimate receives a higher and 50% NA craton weighting are shown in Figs. 8 and 9,
weighting than the upper, based on the observation that the respectively. Mean and median uniform hazard spectra, peak
largest stable craton event worldwide during the period of ground velocity (PGV), and peak ground acceleration (PGA)
historic record had M6.8 (Johnston et al. 1994). are presented in Tables 35 for each of the three weighting
combinations of the CC craton and NA craton models. It
Seismic hazard results should be noted that the case giving 90% weight to the CC
craton model produces mean ground motion predictions that
The seismic hazard calculations were performed at the are not much higher than would be obtained by giving 100%
centre of an arbitrary octagonal source with an area of 1 weight to the CC model. Specifically, the 10% weight
106 km2 (approximate radius of 570 km). The earthquake accorded to the more active NA craton model acts to increase
activity rates within the area source are assumed to be uni- predicted mean ground motions by about 10% (relative to
form, as defined by the magnituderecurrence relations. This predictions made giving no weight to the NA craton model)
area source is sufficiently large that the hazard at the centre while having little effect on median motions.

2007 NRC Canada


Atkinson and Martens 1309

Fig. 9. Uniform hazard spectrum for central Canada for a weighting of 50% central Canadian craton model and 50% North American
craton model (very long design life): (a) mean, median, and 16th and 84th percentile values (horizontal component, hard rock) for 2%
in 50 years (1/2500 pa); (b) corresponding results for 1/10 000 pa. PGA, peak ground acceleration; PGV, peak ground velocity.

Comparison with Geological Survey of Canada fourth- higher at short periods, although they are similar at longer
generation model periods. In our view, the GSC model is conservative for
Table 6 compares the mean results of this study with the application to regions of central Canada that do not include
median ground motions for the floor model in the GSC any of the geologically younger features included in the
fourth-generation model, at a probability of 2% in 50 years. Fenton et al. (2006) definition of NA craton. This is a signif-
We use our mean values for the comparison, as these are the icant reason why we might expect the results for the GSC
recommended values; we compare these with the GSC median floor model to be higher than ours. Another important rea-
because that is what is adopted in the national seismic son for the difference is that the GSC model used a simple
hazard maps. To make the comparisons, the GSC values, quadratic approximation of the Atkinson and Boore (1995)
which are given for site class C (firm ground) by Adams and ground motion prediction equations which is known to over-
Halchuk (2003), are converted to the site class A (hard rock) estimate expected ground motions for low-seismicity sites.
equivalent used in this study. The conversion is made by For such sites, the more accurate tabular form is recom-
dividing the GSC firm-ground values by the reference ground mended by Atkinson and Boore and was used in this study.
condition factors that they used to convert their original cal- A sensitivity calculation reveals that use of the quadratic
culations for site class A to the C values that they provide in approximation would increase the computed values that we
the national hazard maps (see Adams and Halchuk 2003 for obtained by about 50% at short periods while having little
details). The most directly relevant comparison is to the case effect on long-period values. We therefore conclude that
that gives 50% weight to the NA craton model (with 50% to much of the difference in results between this study and the
the CC craton model). Note that the GSC model accords GSC values can be attributed to the adoption of the less
40% weight to the NA craton model, 20% weight to a CC accurate quadratic approximation of the AB95 relations in
craton model (similar but not identical to our CC model), the GSC calculations. It is worth noting that our results,
and 40% to a global model. although lower than the corresponding GSC results, are con-
Compared with our results (for 50% CC craton weight), servative relative to the US Geological Survey results for
the GSC 2% in 50 year ground motion values are about 50% quiet cratonic sites near the CanadaUSA border. This pro-

2007 NRC Canada


1310 Can. J. Civ. Eng. Vol. 34, 2007

Table 3. Peak ground velocity (PGV), peak ground acceleration (PGA), and uniform hazard spectra for 90% central Canadian craton
model and 10% North American craton model.
Return Confidence PGV PGA Sa(0.05) Sa(0.1) Sa(0.2) Sa(0. 5) Sa(1) Sa(2)
period level (m/s) (% g) (% g) (% g) (% g) (% g) (% g) (% g)
1/1 000 Mean 0.004 0.9 2.1 2.2 2.0 1.1 0.4 0.2
1/1 000 Median 0.002 0.8 1.9 1.9 1.6 0.6 0.2 <0.1
1/2 500 Mean 0.009 2.0 4.6 4.3 3.6 1.9 0.8 0.3
1/2 500 Median 0.005 1.7 4.0 3.7 2.8 1.2 0.4 0.1
1/10 000 Mean 0.020 5.1 12.4 10.3 8.1 4.2 1.9 0.7
1/10 000 Median 0.012 4.5 11.0 8.8 6.1 2.4 0.9 0.3
Note: SA, spectral acceleration.

Table 4. Peak ground velocity (PGV), peak ground acceleration (PGA), and uniform hazard spectra for 70% central Canadian craton
model and 30% North American craton model.
Return Confidence PGV PGA Sa(0.05) Sa(0.1) Sa(0.2) Sa(0.5) Sa(1) Sa(2)
period level (m/s) (% g) (% g) (% g) (% g) (% g) (% g) (% g)
1/1 000 Mean 0.005 1.2 2.6 2.7 2.4 1.3 0.5 0.2
1/1 000 Median 0.003 0.8 1.9 2.0 1.6 0.6 0.3 0.1
1/2 500 Mean 0.010 2.3 5.6 5.2 4.3 2.2 1.0 0.3
1/2 500 Median 0.005 1.7 4.1 3.8 2.8 1.2 0.4 0.2
1/10 000 Mean 0.022 6.0 14.8 12.0 9.3 4.8 2.1 0.8
1/10 000 Median 0.012 4.6 11.2 9.2 6.3 2.5 0.9 0.3

Table 5. Peak ground velocity (PGV), peak ground acceleration (PGA), and uniform hazard spectra for 50% central Canadian craton
model and 50% North American craton model.
Return Confidence PGV PGA Sa(0.05) Sa(0.1) Sa(0.2) Sa(0.5) Sa(1) Sa(2)
period level (m/s) (% g) (% g) (% g) (% g) (% g) (% g) (% g)
1/1 000 Mean 0.006 1.4 3.1 3.1 2.7 1.4 0.6 0.2
1/1 000 Median 0.004 0.8 2.0 2.0 1.6 0.8 0.3 0.1
1/2 500 Mean 0.012 2.7 6.5 5.9 4.8 2.5 1.1 0.4
1/2 500 Median 0.007 1.8 4.2 3.8 2.9 1.5 0.6 0.2
1/10 000 Mean 0.025 6.9 16.9 13.5 10.4 5.3 2.3 0.9
1/10 000 Median 0.016 4.7 11.5 9.4 6.5 3.4 1.3 0.3

Table 6. Comparison of mean results from this study with those from the Geological Survey
of Canada (GSC) floor model (median) for a 2% in 50 year probability for hard rock (site
class A).
PGA Sa(0.1) Sa(0.2) Sa(0.5) Sa(1) Sa(2)
Case (% g) (% g) (% g) (% g) (% g) (% g)
GSC median 4.2 6.3 6.2 2.4 0.9 0.2
Mean this study (50% CC, 50% NA) 2.7 5.9 4.8 2.5 1.1 0.4
70% CC, 30% NA 2.3 5.2 4.3 2.2 1.0 0.3
90% CC, 10% NA 2.0 4.3 3.6 1.9 0.8 0.3
Note: CC, central Canadian craton; NA, North American craton.

2007 NRC Canada


Atkinson and Martens 1311

vides an additional check that our results are in reasonable Atkinson, G., and Boore, D. 2006. Ground motion prediction equa-
agreement with other international standards. tions for earthquakes in eastern North America. Bulletin of the
Seismological Society of America, 96: 21812205.
Acknowledgements Fenton, C., Adams, J., and Halchuk, S. 2006. Seismic hazards
assessment for radioactive disposal sites in regions of low seis-
We acknowledge the helpful and constructive comments micity. Geotechnical and Geological Engineering, 24: 579592.
from the external reviewers on the project, Dr. Russell L. Johnston, A.C. 1996a. Seismic moment assessment of earthquakes
Wheeler and Dr. Charles S. Mueller, both of the US Geolog- in stable continental regions I. Instrumental seismicity. Geo-
ical Survey. The comments received from members of the physical Journal International, 124(2): 381414.
Fort McMurray Seismic Hazard Working Group and from Johnston, A.C. 1996b. Seismic moment assessment of earthquakes
Dr. John Adams of the Geological Survey of Canada were in stable continental regions II. Historical seismicity. Geo-
also appreciated. The review comments from two anony- physical Journal International, 125(3): 639678.
mous referees resulted in significant improvements in the Johnston, A.C. 1996c. Seismic moment assessment of earthquakes
manuscript. This study was funded by Albian Sands Energy, in stable continental regions III. New Madrid 18111812,
Canadian Natural Resources, Suncor Energy, and Syncrude Charleston 1886 and Lisbon 1755. Geophysical Journal Interna-
tional, 126(2): 314344.
Canada, who made up the industry representatives of the
Johnston, A., Coppersmith, K., Kanter, L., and Cornell, C.A. 1994.
Fort McMurray Seismic Hazard Working Group. The interest
The earthquakes of stable continental regions. Vol. 1. Report
and support of all members of the working group were
TR-102261-V1, Electric Power Research Institute, Palo Alto,
appreciated. Calif.
Kerr, J.W. 1977. Cornwallis Fold Belt and the mechanism of base-
References ment uplift. Canadian Journal of Earth Sciences, 14: 13741401.
Adams, J., and Basham, P. 1989. Seismicity and seismotectonics of Kerr, J. 1980. Structural framework of Lancaster Aulocogen, Arctic
eastern Canada. Geoscience Canada, 16: 316. Canada. Geological Survey of Canada, Bulletin 319.
Adams, J., and Bell, J.S. 1991. Crustal stresses in Canada. In McGuire, R.K. 1993. Computations of seismic hazard. Annali di
Neotectonics of North America. The geology of North America. Geofisica, 36(34): 181200.
Edited by D.B. Slemmons, E.R. Engdahl, M.D. Zoback, and McGuire, R. 2004. Seismic hazard and risk analysis. EERI Mono-
D.D. Blackwell. Geological Society of America, Boulder, Colo. graph MNO-10, Earthquake Engineering Research Institute (EERI),
Decade Map Vol. 1, pp. 367386. Oakland, Calif.
Adams, J., and Halchuk, S. 2003. Fourth generation seismic hazard McGuire, R., and Barnhard, T. 1981. Effects of temporal variations
maps of Canada: values for over 650 Canadian localities in- in seismicity on seismic hazard. Bulletin of the Seismological
tended for the 2005 National Building Code of Canada. Geolog- Society of America, 71: 321334.
ical Survey of Canada, Open File 4459. NRCC. 2005. National building code of Canada. National Research
Atkinson, G. 1993. Source spectra for earthquakes in eastern North Council of Canada (NRCC), Ottawa, Ont.
America. Bulletin of the Seismological Society of America, 83: Okulitch, A.V., Packard, J.J., and Zolani, A.I. 1986. Evolution of
17781798. the Boothia Uplift, Arctic Canada. Canadian Journal of Earth
Atkinson, G. 1995. Ground motion relations for use in eastern Sciences, 23: 350358.
hazard analyses. In Proceedings of the 7th Canadian Conference Risk Engineering, Inc. 2005. EZ-FRISK software, version 7.0.
on Earthquake Engineering, Montral, Que., 57 June 1995. Risk Engineering, Inc., Boulder, Colo.
Canadian Association for Earthquake Engineering, Montral, Que. Wheeler, R. 1995. Earthquakes and the cratonward limit of Iapetan
pp. 10011008. faulting in eastern North America. Geology, 23: 105108.
Atkinson, G., and Boore, D. 1995. New ground motion relations Wheeler, R., and Johnston, A. 1992. Geologic implications of earth-
for eastern North America. Bulletin of the Seismological Society quake source parameters in central and eastern North America.
of America, 85: 1730. Seismological Research Letters, 63: 491514.

2007 NRC Canada

Das könnte Ihnen auch gefallen