See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/242190028

Probabilistic Seismic Hazard Assessment

(Horizontal PGA) for Fennoscandia Using the
Logic Tree Approach for...

Article in Seismological Research Letters January 2001

DOI: 10.1785/gssrl.72.1.33

CITATIONS READS

15 71

2 authors, including:

Gottfried Grnthal
Helmholtz-Zentrum Potsdam - Deutsches GeoForschungsZentrum GFZ
113 PUBLICATIONS 2,616 CITATIONS

SEE PROFILE

All content following this page was uploaded by Gottfried Grnthal on 18 July 2014.

The user has requested enhancement of the downloaded file.

Originally published as:

Wahlstrm, R., Grnthal, G. (2001): Probabilistic seismic hazard assessment (horizontal

PGA) for Fennoscandia using the logic tree approach for regionalization and
nonregionalization models. - Seismological Research Letters, 72, 1, 33-45.
Probabilistic Seismic Hazard Assessment
(Horizontal PGA) for Fennoscandia Using the
Logic Tree Approach for Regionalization and
Nonregionalization Models
Rutger Wahlstrm and Gottfried Grnthal
GeoForschungsZentrum Potsdam

INTRODUCTION mark along with adjacent water areas. The grid has
nearly 600 points spaced with a density of 1 longi-
The most important practical aspect of regional seis- tude by 0.5 latitude, but the hazard is mapped only
micity is its potential for structural damage. Although at sites located on land.
Fennoscandia is characterized by a fairly low rate of Different global alternatives are used for the input
spatially scattered seismicity, the seismic hazard data containing (1) regionalized and (2) non-
must be considered for sensitive structures like dams, regionalized sources. In (1) the investigated area and
mines, nuclear power plants, underground deposito- its surroundings are divided into different source
ries for radioactive waste, oil platforms, etc. The regions, based on presumed seismic and/or geologi-
main objective of the present study is to provide a cal homogeneity. Three such regionalization models
detailed map of the median seismic hazard (horizon- are used: a revised version of the model by
tal PGA) for Fennoscandia for a mean return period NFR/NORSAR and NGI (1998) for GSHAP, and two
of 475 years using modern computational algorithms. models introduced by Wahlstrm and Grnthal
There are competing hypotheses as to the cause of (2000). In (2) the regionalization is replaced by event
earthquakes in Fennoscandia: release of stresses built size criteria.
up and propagated from the North Atlantic Ridge The computer program applied in the analysis,
versus stress adjustment connected to the postglacial FRISK88M (Risk Engineering, 1996), utilizes a logic
land uplift (e.g., see Wahlstrm, 1993). Occasional tree approach to account for uncertainties and weights
significant earthquakes have occurred, notably in of the various input parameters, i.e., the data set (magni-
1759 in Kattegat between Denmark and Sweden, tudes), regionalization model or completeness criteria,
MS = 5.6, and in 1819, 1866, and 1904 in Norway, attenuation function, maximum expected magnitude
MS = 5.8, 5.7, and 5.4, respectively (magnitudes from (Mmax), cumulative earthquake rate () and
NFR/NORSAR and NGI, 1998). Typical for an in- ln(frequency-magnitude slope) (-), and focal depth.
traplate region, fault mapping and understanding of The combination of magnitude sets with regionaliza-
the tectonics of seismicity patterns are incomplete. tion models or completeness criteria makes up the
Any regionalization of the investigated region, Fen- global alternatives. Two attenuation functions and a
noscandia, based on geological-structural elements set of discrete values for each seismicity parameter
and/or seismicity is therefore uncertain. In this study, and source region are used. The logic tree construc-
a combination of different source regionalization tion gives great flexibility in input parameter settings
models and region-independent models are used to and facilitates the merging of competing scientific
determine the regional seismic hazard. hypotheses in one hazard calculation. In our applica-
Wahlstrm and Grnthal (2000) used a technique tion, all regionalized and non-regionalized models
similar to that of the present paper to assign seismic are combined in one run.
hazard to Sweden, Finland, and Denmark. They also
reviewed previous studies of seismic hazard in Fen-
noscandia. Improvements and changes incorporated CATALOG DATA: Mw CONVERSION AND
in the present study include the addition of Norway, a SPATIOTEMPORAL COMPLETENESS
stronger influence of nonregionalization models, and
a statistically more prudent calculation of maximum The database used in the analysis, FENCAT, is a
expected magnitudes. continuously updated catalog of earthquakes in
Horizontal PGA values are calculated for a grid of Northern Europe compiled at the Institute of Seis-
points covering Norway, Sweden, Finland, and Den- mology, University of Helsinki (Ahjos and Uski,

Seismological Research Letters, Volume 72, No. 1, 2001 33

 Figure 1. Seismicity from FENCAT (Ahjos and Uski, 1992). Land and sea gross zones.

1992). FENCAT contains a variety of magnitude formulae) are also given in FENCAT. A full homo-
types. The vast majority of the events have a local ML genization of magnitudes is not a trivial task and has
magnitude, mostly ML(UPP) (Wahlstrm and Ahjos, not been attempted in the present study.
1984) or some magnitude related to ML (UPP). Mag- NFR/NORSAR and NGI (1998) present hazard re-
nitudes compatible with ML derived from macroseis- sults for Norway based on a homogenized catalog
mic parameter data (Wahlstrm and Ahjos, 1984 (not FENCAT). This catalog covers only in part the

TABLE 1
Threshold Completeness Magnitudes, Maximum Magnitudes, Focal Depths and Values
Gross Year of Thresholda (Mw) Focal Depthsb
Mw Set Zone Completeness First Alt Mmax b (km)
(a) Land 1810 4.8 4.8 nonextended: 5.45, 5.95, 6.25, 6.55, 6.95 4, 8, 12, 17, 30 3.42
1880 2.8 3.3 extended: 4.95, 5.60, 6.15, 6.70, 7.50
1980 2.3
(a) Sea 1890 3.8 3.8 nonextended: 5.60, 6.00, 6.30, 6.55, 7.00 4, 10, 14, 21, 34 3.33
1955 3.3 3.3 extended: 5.45, 6.00, 6.40, 6.85, 7.60
1978 2.8
1986 2.3
(b) Land 1750 5.3 5.3 nonextended: 5.85, 6.10, 6.35, 6.60, 7.00 4, 8,12,17, 30 2.05
1880 2.8 3.3 extended: 5.90, 6.25, 6.60, 7.00, 7.70
1980 2.3
(b) Sea 1890 4.8 4.8 nonextended: 6.10, 6.25, 6.45, 6.65, 7.05 4,10,14, 21, 34 2.10
1955 4.3 4.3 extended: 6.15, 6.40, 6.70, 7.10, 7.75
1978 2.8 3.3
1986 2.3
a. The threshold values of the first column are used in both methods, while those of the second column are also used in the a lternative method.
b. For source regions covered by the small part of the sea gross zone, Mmax and focal depth sets from the land gross zone are used.

34 Seismological Research Letters, Volume 72, No. 1, 2001

 Figure 2A. Source regionalization models and source regions: NFR/NORSAR and NGI (1998) model revised.

area of interest for the present study. area treated by Kim et al. (1989), large historical
To comply with the applied attenuation functions earthquakes in FENCAT are often assigned MS mag-
(see below), the moment magnitude Mw, must be nitudes converted from macroseismic data. For some,
known. Two different sets of Mw magnitudes are used also generally large, earthquakes mb magnitudes
in this study. The first set (a) is based on Mw magni- have been calculated and are given in FENCAT. Av-
tudes converted from ML or from magnitudes com- erage global relations of seismic moment versus sur-
patible with ML. Kim et al. (1989) derived a first-order face-wave and body-wave magnitudes for stable con-
relationship between seismic moment and ML (UPP) for tinental region earthquakes presented by Johnston
earthquakes in the Fennoscandian shield in the magni- (1989) are combined with the relation by Hanks and
tude range 2-5.1. However, their data contain only a Kanamori (1979) to give
few events with magnitude above 4. For magnitudes
in the interval 4-5, a nonlinear behavior of the mo- Mw = 4.28 - 0.27MS + 0.09MS2 (2),
ment-magnitude relation has been observed in several
North American studies (e.g., Street et al., 1975; Ha- Mw = 4.85 - 0.85mb + 0.18mb2 (3).
segawa, 1983; Nuttli, 1986). A second-order regres-
sion based on the same data used by Kim et al. In the second input set (b) Equations (2) and (3) are
(1989) together with the Mw definition by Hanks and applied to events for which MS and mb are given in
Kanamori (1979) gives FENCAT The remaining events, the vast majority,
keep the same ML-based Mw values as in set (a). Mw
Mw = 1.2 + 0.28ML(UPP) + 0.06[ML(UPP)]2 (1). magnitudes calculated from Equations (2) and (3) are
generally larger than those calculated from Equation
Equation (1) is used for the generation of set (a). (1). Since mainly the higher-end magnitudes (large
Especially in and offshore Norway, outside the events) are calculated in the former way, set (b) gives

Seismological Research Letters, Volume 72, No. 1, 2001 35

 Figure 2B. Source regionalization models and source regions: Seismicity-based model.

a lower value than set (a) for several source re- tern Baltic Sea (Figure 1). Together, the two zones
gions. In all regionalized and non-regionalized mod- cover the area of all source regions. The threshold
els alike, set (a), with all magnitudes calculated in a values are based on the cumulative time distribu-
consistent way from Equation (1), is weighted at 0.75 tion of the number of earthquakes exceeding cer-
and set (b) at 0.25. The difference in maximum Mw tain magnitudes. Mw = 2.3 is the smallest accepted
magnitudes between the two sets for many source magnitude for the regionalized global alternatives,
regions is considered in the assignments of Mmax (see while 2.3 or 3.3 is the smallest for the non-
below). regionalized global alternatives.
Epicenters of earthquakes in FENCAT are plotted
in Figure 1. Two selection criteria are imposed on the
input data: ATTENUATION FUNCTIONS

1. Event independence. For earthquake sequences, There exist very few recorded ground acceleration
only one shock (main) is included in order to data from earthquakes in Fennoscandia. In the
maintain the assumption of a Poissonian distribu- present study, two relations valid for hard rock condi-
tion in time as required by the probabilistic ap- tions, i.e., suitable for the Fennoscandian shield cov-
proach. ering most of the investigated area, are selected:
2. Catalog completeness. Threshold completeness ln(A) = -2.143 + 0.751 Mw - (4),
magnitudes for the two Mw sets are specified for 1.04 x 10-3r - 0.815ln(r)
each of two gross zones, land and sea, and for
several time periods (Table 1). Offshore areas in log(A) = 1.79 + 0.298(Mw - 6) -
the vicinity of land belong to the land zone, and 5.36 x 10-2 (Mw - 6)2 + (5),
the sea zone has a small part also in the southeas- 1.35 x 10-3r - log(r)

36 Seismological Research Letters, Volume 72, No. 1, 2001

 Figure 2C. Source regionalization models and source regions: Geology-based + part of the NFR/NORSAR and NGI (1998) model.

where A is the horizontal PGA (m/s2) and r is hypo- seismically homogeneous. An adequate regionaliza-
central distance (km). Equation (4) is from Ambra- tion model can be based on geological units in each
seys et al. (1996) with MS converted to Mw using a of which the seismic potential is assumed to be con-
regression of seismic moment on MS for Central Eu- stant over the long term, even if the existing seismic
ropean data (GeoForschungsZentrum Potsdam, un- record may be insufficient to show this (cf. Adams et
published) and the Hanks and Kanamori (1979) Mw al., 1995a). Any regionalization not, or only poorly,
definition. Equation (5) is from Atkinson and Boore based on documented fault geography includes a
(1995) and was derived for conditions in eastern certain degree of uncertainty. In this study, the uncer-
North America, structurally similar to Fennoscandia. tainty is counteracted by using three different regio-
The standard deviation of the normal distribution of nalization models (Figure 2) from Wahlstrm and
ln(A) is 0.576 for Equation (4). This corresponds to a Grnthal (2000):
value of 0.25 for a log(A) relation, which is assigned
to Equation (5). In the present study, Equations (4) (A) A revised version of NFR/NORSAR and NGI
and (5) are equally weighted at 0.5. (1998) used for GSHAP, 31 source regions. The
revision mainly concerns the inclusion of addi-
tional regions to the southeast.
INPUT SETS
(B) A model with 21 source regions based on the
Regionalization Models seismicity distribution.
The applied computer program is based on a division
into source regions of the area of a potential threat to (C) A model with 14 source regions based mainly on
the investigated site or, as in the present case, a grid tectonic maps of Sweden, Finland, and Denmark
of sites. A source region should, in some respect, be in addition to 21 regions for Norway and its off-

Seismological Research Letters, Volume 72, No. 1, 2001 37

Figure 3. Example plot of a posterior likelihood function of maximum expected magnitude and the selection of five representative
values (at dashed lines) from a partition into five areas of equal size (separated by solid lines).

shore area from the NFR/NORSAR and NGI six global alternatives in the input. The regionaliza-
(1998) model. tion models (A) and (B) are given equal weight and
the geological models (C), established from the most
For each source region, FENCAT earthquakes are recent geological and seismicity data (references are
selected after the specified independence and com- given in Wahlstrm and Grnthal, 2000), is given the
pleteness criteria have been applied. Each region is added weight of (A) and (B).
assigned:
Maximum magnitude. Following the approach of
a gross zone, land or sea; Coppersmith (1994) and Cornell (1994) to obtain the
completeness tables, depending on the gross zone maximum expected magnitude, normal distribution
and the Mw set (Table 1); functions of maximum magnitudes in stable conti-
two {, } values for each Mw set, derived from nental regions globally, separated in two tectonic
data of the region itself and from data of all re- types, extended crust areas and non-extended crust
gions belonging to the same gross zone; areas, are multiplied by local likelihood functions
a focal depth distribution, calculated from data of based on the maximum observed Mw, on the value,
the appropriate gross zone (Table 1); and on the number of recorded large earthquakes. In
a tectonic type, extended or non-extended stable the present study, eight local likelihood functions are
continental crust, based on the distribution in Kant used by combining the two gross zones, the two tec-
(1994); and tonic types (extended and non-extended stable conti-
five Mmax values for each Mw set, depending on nental crust), and the two Mw sets. Each of these
the tectonic type and the gross zone (Table 1). functions is multiplied with the relevant global func-
tion (extended or non-extended crust) to result in
For some source regions, there are not sufficient data posterior distributions of the maximum expected
to calculate the regional recurrence parameters, and magnitude, Mmax. The small sea gross zone is consi-
data from one or more neighboring regions are in- dered to belong to the land gross zone in the Mmax
cluded in the derivation of . In cases where the bor- calculations. For each posterior distribution (see ex-
der between the land and sea gross zones or between ample in Figure 3), the plotted area is divided into
extended and non-extended crust cuts through a five subareas of similar size, the gravity points of
source region, the assignment for the region is based which specify five representative Mmax values (Table
on where most of the seismicity has occurred. The 1).
source regions are shown in Figure 2. Ascribing the gross zone and the tectonic type to
Each of the Mw sets, (a) and (b), is combined with each of the source regions, a set of Mmax values is
each of the three regionalization models to make up adopted for each Mw set to use as input in the seismic

38 Seismological Research Letters, Volume 72, No. 1, 2001

 Figure 4. Logic tree of input parameters with assigned weights. Equations and symbols are explained in the text.

hazard calculation. The five values of an Mw set are Focal depth. The largest known earthquakes in Fen-
there given a weight of 0.2 each. They are all larger noscandia occurred in pre-instrumental time and have
than the maximum observed Mw in the region. poorly estimated focal depth from macroseismic data.
The networks of seismograph stations in Fennoscan-
Minimum magnitude. The minimum magnitude dia have been sparse until recent years, and the focal
thought to cause structural damage must be specified depths of recorded events are often not well con-
with a fixed value. It is set at Mw = 4.0 in this study, strained. The depth distribution is therefore calcu-
the same for sets (a) and (b) and for every source lated only for the two gross zones and not for the
region. According to Equation (1), this corresponds individual source regions. For the land zone, the
to ML(UPP) = 4.9. The minimum magnitude is the depth distribution is calculated from events with
smallest magnitude contributing to the calculated Mw 2.8; for the sea zone, this limitation gives insuf-
seismic hazard, and the choice of this value has im- ficient data and events down to Mw = 2.3 are in-
pact on the hazard at small mean return periods, i.e., cluded.
the larger the minimum magnitude, the narrower the The computer algorithm requires fixed weights,
range of contributing magnitudes and thus the small- common for all regions, for each input seismicity
er the hazard (e.g., see Bender and Perkins, 1993; parameter. As for the Mmax, , and values, focal
Grnthal and Wahlstrm, 2001). depths must be assigned to predetermined weights
and not vice versa. Five representative depths, each
Seismicity rate and frequency-magnitude slope. Val- assigned a weight of 0.2, are selected from the distri-
ues of and , where is the annual rate of earth- bution in each gross zone (Table 1). In this case,
quakes with the selected minimum magnitude, there is no distinction between the two Mw sets. The
Mw = 4.0 or larger, are calculated for each source appropriate set of depths is then used for each source
region. If there are fewer than 40 events in a region, region. The source regions belonging to the small sea
data from one or more neighboring regions are in- gross zones are assigned the focal depth distribution
cluded in the calculation. values are also obtained of the land gross tone.
for the gross zones and kept fixed to calculate an
alternative set of values for the regions. A weight Nonregionalization Models
of 0.5 is assigned to each of the regional and gross For areas where the seismotectonics are not fully
zone values. understood, the specification of source regions is
uncertain. In this study, the introduction of more than

Seismological Research Letters, Volume 72, No. 1, 2001 39

Figure 5. Curves of annual seismic hazard. P: 16%, 37%, 50%, 63%, and 84% fractiles and the mean, for the grid points with the
highest median hazard in Norway, Sweden, Finland, and Denmark, respectively.

one regionalization model is intended to soften the to Frankels (1995) smeared out model (see above),
lack of knowledge. Techniques avoiding regionaliza- the sea gross zone is connected to extended crust and
tion have been suggested, e.g., Frankel (1995), the land gross zone to non-extended crust.
Rttener (1995), and Woo (1996). The former com- Irrespective of model, the weighting of the two
bines different event size classes and, with some ex- Mw sets is 0.75 for (a) and 0.25 for (b), as stated
ception, applies the same seismicity parameters but above. For each Mw set, the three non-regionalized
the seismicity rate to the whole investigated area models are assigned equal weight. The two attenua-
(central and eastern North America). Frankel (1995) tion functions and their weights, the sets of values
also has one model smearing out the seismicity over and weights of Mmax and the focal depth, and the min-
the whole investigated region. imum magnitude assumed to cause damage,
By introducing quasiregions of no seismotectonic re- Mw = 4.0, are the same as for the regionalized global
levance, the computer program can be applied also to alternatives.
global alternatives without seismic or geological regio-
nalization. In an approach similar to that of Frankel Regionalization and Nonregionalization Models Merged
(1995), small spatial windows of size 2 longitude by 1 In total there are twelve global alternatives, six with
latitude (quasiregions) are combined with the Mw sets source regions, four with quasiregions, and two with
(a) and (b) and also with subsets of (a) and (b), where only gross zones. The combination of all input para-
quasiregions without events of Mw 3.3, corresponding meters, with their weights, can be illustrated in a log-
to ML(UPP) = 4.0, are omitted. In these four additional ic tree (Figure 4). The total weight of the regionalized
global alternatives, only one seismicity rate, calculated global alternatives is the same (0.5) as the total
from the relevant gross zone value, is used for each weight of the non-regionalized. In practical applica-
window. tions, there must be a balance in the assignment and
In two further models, the two Mw sets have a weighting of parameters between the generous input
threshold of Mw = 3.3, and the seismicity is assumed options of the program algorithm and a caution to
to be equally distributed over each of the sea and land stick to simple and realistic values. The number of
gross zones. In these two models, which correspond values for any parameter must be the same for all

40 Seismological Research Letters, Volume 72, No. 1, 2001

Figure 6. Map of 90% probability of nonexceedence of horizontal PGA (m/s2) in 50 years, corresponding to a mean return period of 475
years. Median hazard values.

Seismological Research Letters, Volume 72, No. 1, 2001 41

Figure 7. Normalized contribution to the cumulative eartquake rate (seismic hazard), of earthquakes with different magnitudes and at
different distances, for a horizontal PGA of 1 m/s2, at the grid points with the highest hazard in each of the four countries. The lowest
magnitude class represents Mw 4.5 and the highest distance class represents r > 100 km. Mean hazard values.

regions within a global alternative, but the values As seen in Figure 4, there is a total of 900 branches
themselves may vary The number may also vary be- (2 Mw sets; 3 zonation models; 2 attenuation functions;
tween different global alternatives. A seismicity pa- 2 {, } sets; 5 Mmax values; and 5 focal depths for the
rameter can be combined with the same setting (de- regionalized models; and the same numbers but just
pendence) or every setting (independence) of this one {, } set for the non-regionalized models), giving
parameter for all source regions. 900 output hazard values which determine the distri-
The number of solutions, i.e., branches on the log- bution of the fractiles.
ic tree, for each global alternative is AMSBSHS,
where A is the number of attenuation functions, M
maximum expected magnitudes (or intensities), B {, RESULTS
} values, and H focal depth values. S is 1 if a para-
meter (M, B, H) is dependent and is equal to the Hazard values calculated for the selected grid of sites
number of source regions if it is independent. A is covering Norway, Sweden, Finland, and Denmark,
always dependent. If the number of source regions is with points spaced at 1 longitude by 0.5 latitude, or
large, there is evidently a drastic increase in the num- for selected sites of the grid, are presented in differ-
ber of branches (and computer time) when indepen- ent displays in Figures 5-7.
dent instead of dependent parameters are used. How- Figure 5 shows the annual probability of exceedance
ever, tests made an small samples show that the range of the horizontal PGA for the points of the selected grid
of output hazard values does not change significantly. which have the highest median hazard value in each of
In the present study, M, B, and H are all dependent. the four investigated countries. The 16%, 37%, 50%

42 Seismological Research Letters, Volume 72, No. 1, 2001

(median), 63%, and 84% fractiles and the mean curve years, Figure 6, can be compared with those from
of each set of solutions are plotted, the fractiles corres- different models for Sweden, Finland, and Denmark
ponding to deviations from the median of -1, -0.5, by Wahlstrm and Grnthal (2000) and results for
zero, +0.5, and +1 respectively. The median can be Norway in NFR/NORSAR and NGI (1998). Taking
considered the best estimate from the given input para- into account that the latter study refers to the mean
meter values, and the 84% fractile is a conservative hazard, which usually is near the 64% fractile accord-
estimate consistent with general engineering practice ing to NFR/NORSAR and NGI (1998), whereas
(Naumoski and Heidebrecht, 1995). According to Wahlstrm and Grnthal (2000) and the present
Adams et al. (1999), the 84% fractile includes the con- study (Figure 6) refer to the median hazard, i.e., the
tribution of the epistemic uncertainty from all the input 50% fractile, the agreement among all obtained maps
parameters. For the median hazard, the 90% probability is generally good. The claim in NFR/NORSAR and
of non-exceedance in 50, 100, 500, and 1,000 years, NGI (1998) that the 64% curve is a fair approxima-
corresponding to mean return periods of 475, 950, tion of the mean hazard seems to be valid only for
4,745, and 9,490 years, respectively, occurs for peak relatively small mean return periods, judging from
ground accelerations of 0.7/0.3/0.2/0.2 (Norway/Swe- the results of the present study (Figure 5), where the
den/Finland/Denmark), 1.0/0.4/.3/0.3, 1.9/0.8/0.7/0.7, mean hazard curves approach the larger fractiles for
and 2.3/ 1.1/1.0/1.0 m/s2, respectively. long mean return periods. The discrepancies between
Figure 6 presents a map of the median hazard cal- the Norwegian results and those of the present study
culated for the whole area of investigation for a 90% are also larger for long mean return periods. The
probability of non-exceedance in 50 years. This cor- mean curves for the Norwegian and Danish points are
responds to a mean return period of 475 years. Two well above the 64% fractile curves even at small
coastal areas of Norway, around Bergen and in Nord- mean return periods.
land, display the highest hazard in Fennoscandia, up The obtained results give a good overall insight
to 0.7 m/s2 and 0.45 m/s2, respectively, followed by into the seismic hazard conditions of Fennoscandia,
the intermediate coastal area (0.35 m/s 2) and the but they should be regarded as rough regional esti-
Oslo-Vnern region in southeast Norway and adjoin- mates. For site-specific investigations of engineering
ing Sweden (0.3-0.35 m/s2). A second area of en- interest, local conditions, especially concerning the
hanced hazard in Sweden is along the Gulf of Both- regionalization and attenuation, need to be consi-
nia, with a maximum value in Vsterbotten. The dered.
maximum hazard levels in Denmark and Finland are There are two basic kinds of uncertainties in the
similar, about 0.25 m/s2. The highest hazard in Den- input parameters for seismic hazard calculations. The
mark, in northwest Jutland, can be ascribed to the aleatory uncertainty appears from the randomness of
offshore seismicity (Skagerrak and North Sea); cf. nature, accounted for as the standard deviation in the
the large contribution from large distances in Figure attenuation functions, and the epistemic uncertainty
7. No influence from the Tornquist zone is noticea- is due to the lack of full knowledge in parameter as-
ble. In Finland, the highest hazard is in the Kuusamo signment and regionalization, which is accounted for
district in the east-central part of the country, due to by the logic tree multiple-input option. The latter
seismicity both locally and on the Kola Peninsula. kind of uncertainty will decrease with improved
The smallest hazard values in the investigated area, knowledge of the tectonics, seismicity, and ground
below 0.1 m/s2 for a mean return period of 475 years, conditions. A general survey of the uncertainty of
are found in southern Finland, the Stockholm area, various input parameters and assumptions in seismic
and the islands of land, Gotland, and land in the hazard calculation is provided by Bender and Perkins
Baltic Sea. The spatial hazard distribution at longer (1993). Grnthal and Wahlstrm (2001) investigated
mean return periods shows, generally, approximately the sensitivity of individual input parameters.
corresponding results. The Mw sets are not optimum input data, lacking a
For the sites with the highest calculated hazard le- complete homogenization. However, all magnitudes
vels in each country, the contribution to the seismic in FENCAT, with the exception of a few entries
hazard of earthquakes with different magnitudes and based on Russian intensities, ultimately refer to the
at different distances is illustrated in Figure 7. This original Richter ML definition, although indirectly for
graph is related to a PGA level of 1 m/s2 and is based many types. There are obvious discrepancies between
on mean hazard values, the output from the computer different magnitude types in the catalog, but we may
program being so designed. not be too far off assuming magnitudes to be approx-
imately comparable. This motivates the higher
weight assigned to set (a). A similar assumption is
DISCUSSION made in hazard computations for western Canada,
where different magnitudes are mixed (Adams et al.,
The hazard map for a mean return period of 475 1999).

Seismological Research Letters, Volume 72, No. 1, 2001 43

 Some subjectivity is inevitable in selecting seis- Besides the studies by NFR/NORSAR and NGI
mological, geological, and other criteria for the re- (1998) and Wahlstrm and Grnthal (2000) men-
gionalization. Borders between source regions are tioned above, only a few applications of probabilistic
usually not sharp with respect to seismic activity. seismic hazard calculation using the logic tree tech-
Furthermore, for any area where active faulting is not nique have been published for Europe so far, e.g., by
fully manifested, the lack of a complete understand- Labk et al. (1998) for a site in Slovakia and by Mus-
ing of the long-term tectonic processes implies an son and Winter (1997) for the UK. The latter study
uncertainty in the specification of regional borders. concludes that the generous input options require
This general border problem is discussed by Bender great consideration in the assignments of the parame-
(1986). In the Canadian approach (quasiprobabilistic ter values. A great effort in the selection of input data
robust model; Adams et al., 1995a), the hazard is has been made in the present study, and we are con-
calculated separately for smaller regions based on fident that the obtained hazard values are of the right
seismic homogeneity and larger regions based on order. Sensible alternative input sets are fully possi-
geological homogeneity, and the highest of the ob- ble and defensible, but neither these nor the applica-
tained values at each site is selected. The option of tion of other methods will likely give very different
the logic tree approach to use several regionalization results.
models makes the border effect less prominent. The
present study keeps a fully probabilistic approach by
combining all applied models, regionalized and non- ACKNOWLEDGMENTS
regionalized, in one large computation.
Models developed to avoid regionalization, such Ch. Bosse provided valuable assistance in the calcu-
as those of Frankel (1995) and Rttener (1995), have lations. G. Michel helped in interpreting the new tec-
their main drawback in the deliberate omission of tonic maps for the regionalization model (C). The
distinct geological and seismological knowledge. It is extension to include Norway was proposed by col-
doubtful if the objectivity of such techniques can leagues at NORSAR. The critical review by John E.
compensate for this. For a critical discussion of Fran- Ebel was helpful in improving the paper.
kels (1995) method, which currently is a standard
procedure in USGS hazard calculations, see Adams
et al. (1995b). REFERENCES
Assignments of the maximum expected magnitude
Mmax. are uncertain in regions where the dimensions Adams, J., P. W. Basham, and S. Halchuk (1995a). Northeas-
of active faults are poorly known, as is generally the tern North American earthquake potential: New challenges
case in intraplate environments. On the other hand for seismic hazard mapping, in Current Research 1995-D,
Geol. Survey of Canada, 91-99.
the influence of Mmax is minor if the mean return pe- Adams, J., S. Halchuk, and P. W. Basham (1995b). Prelimi-
riod is not too long (e.g., see Grnthal and nary Results from an Application of the Frankel Method of
Wahlstrm, 2001). To combine global (from tectoni- Seismic Hazard Estimation for Canada, 1995 April 12
cally similar regions) and regional information on Geophys. Div., Geol. Survey of Canada Internal Report, 19
maximum observed magnitudes appears to be a sens- pp.
ible way to generate the Mmax sets. Adams, J., D. H. Weichert, and S. Halchuk (1999). Trial
Seismic Hazard Maps of Canada 1999: 2%/50 Year Values
Paleoseismic evidence extends observational re-
for Selected Canadian Cities, Geol. Survey of Canada
currence periods for large earthquakes and has im- Open-File Rep. 3724, 114 pp.
pact on the Mmax setting. In northern Fennoscandia, Ahjos, T. and M. Uski (1992). Earthquakes in northern Eu-
several earthquakes estimated at magnitudes up to rope in 1375-1989, Tectonophysics 207, 1-23.
and above MS = 8 occurred at the latest phase of deg- Ambraseys, N. N., K. A. Simpson, and J. J. Bommer (1996).
laciation some 9,000 ya (Lagerbck, 1990). Whatever Prediction of horizontal response spectra in Europe, Earthq.
the underlying cause of these events (see Wahlstrm, Eng. Struct. Dyn. 2, 371-400.
Atkinson, G. M. and D. M. Boore (1995). Ground-motion
1993), the dose spatial and temporal connections to relations for eastern North America, Bull. Seism. Soc. Am.
an occasional geological process-in the time frame 85, 17-30.
concerned-make their influence on the Mmax setting Bender, B. K. (1986). Modelling source zone boundary un-
irrelevant. certainty in seismic hazard analysis, Bull. Seism. Soc. Am.
To be based on a more reliable data set, values 76, 329-341.
are often kept constant over large areas in hazard Bender, B. K. and D. M. Perkins (1993). Treatment of para-
calculations (e.g., Musson and Winter, 1997; meter uncertainty and variability for a single seismic hazard
map, Earthq. Spectra 9, 165-195.
NFR/NORSAR and NGI, 1998). In the present study, Coppersmith, K. J. (1994). Conclusions regarding maximum
the gross zone values are given equal weight to the earthquake assessment, in The Earthquakes of Stable Con-
regional values for the six global alternatives contain- tinental Regions, Vol. 1: Assessment of Large Earthquake
ing source regions. Potential, Electric Power Research Institute (EPRI) TR-

44 Seismological Research Letters, Volume 72, No. 1, 2001

 102261-V1, 6-1--6-24. Naumoski, N. and A. C. Heidebrecht (1995). Implications of
Cornell, C. A. (1994). Statistical analysis of maximum mag- preliminary seismic hazard spectra ordinates for design val-
nitude, in The Earthquakes of Stable Continental Regions, ues in the National Building Code of Canada, in Proceed-
Vol. 1: Assessment of Large Earthquake Potential, Electric ings, 7th Canadian Conference an Earthquake Engineer-
Power Research Institute (EPRI) TR-102261-V1, 5-1--5-27. ing, Montreal, June 1995, 1,017-1,024.
Frankel, A. (1995). Mapping seismic hazard in the central NFR/NORSAR and NGI (1998). Seismic Zonation for Nor-
and eastern United States, Seism. Res. Lett. 66(4), 8-21. way, Report for the Norwegian Council for Building Stan-
Grnthal, G. and R. Wahlstrm (2001). Sensitivity of para- dardization, 162 pp. + appendix.
meters for probabilistic seismic hazard analysis using a log- Nuttli, O. W. (1986). Average seismic source parameter rela-
ic tree approach, J. Earthq. Eng. (in press). tions for mid-plate earthquakes, Bull. Seism. Soc. Am. 73,
Hanks, T. C. and H. Kanamori (1979). A moment magnitude 519-535.
scale, J. Geophys. Res. 84, 2,348-2,350. Risk Engineering, Inc. (1996). FRISK88M Users Manual,
Hasegawa, H. S. (1983). Lg spectra of local earthquakes Version 1.79, 4155 Darley Ave., Suite A, Boulder, Colora-
recorded by the Eastern Canada Telemetered Network and do, USA.
spectral scaling, Bull. Seism. Soc. Am. 73, 1,041-1,061. Rttener, E. (1995). Earthquake Hazard Evaluation for Swit-
Johnston, A. C. (1989). Moment magnitude estimation for zerland, Contributions to Geology of Switzerland, Geo-
stable continental earthquakes, Seism. Res. Lett. 60, 13. physics 29, 106 pp. + appendices.
Kant, L. R. (1994). Tectonic interpretation of stable continen- Street, R. L., R. B. Herrmann, and O. W. Nuttli (1975). Spec-
tal crust, in The Earthquakes of Stable Continental Region, tral characteristics of the Lg wave generated by central
Vol. 1: Assessment of Large Earthquake Potential, Electric United States earthquakes, Geophys. J. R. Astr. Soc. 41, 51-
Power Research Institute (EPRI) TR-102261-V1, 2-1--2-98. 63.
Kim, W.-Y., R. Wahlstrm, and M. Uski (1989). Regional Wahlstrm, R. (1993). Fennoscandian seismicity and its
spectral scaling relations of source parameters for earth- relation to the isostatic rebound, Glob. Planet. Change 8,
quakes in the Baltic Shield, Tectonophysics 166, 151-161. 107-112.
Labk, P., A. Bystrick, P. Moczo, W. Campbell, and L. Wahlstrm, R. and T. Ahjos (1984). Magnitude determina-
Rosenberg (1998). Preliminary probabilistic seismic hazard tion of earthquakes in the Baltic Shield, Ann. Geophys. 2,
assessment for the nuclear power plant Buhonice (Slovakia) 553-558.
site, 11 th European Conf. Earth. Eng., Balkema, Rotter- Wahlstrm, R. and G. Grnthal (2000). Probabilistic seismic
dam, 12 pp. hazard assessment (horizontal PGA) for Sweden, Finland
Lagerbck, R. (1990). Late Quaternary faulting and paleo- and Denmark using different logic tree approaches, Soil
seismicity in northern Fennoscandia, with particular refer- Dyn. Earthq. Eng. (in press).
ence to the Lansjrv area, northern Sweden, Geol. Fren. Woo, G. (1996). Kernel estimation methods for seismic ha-
Stockh. Frb. 112, 333-354. zard area source modelling, Bull. Seism. Soc. Am. 86, 353-
McGuire, R. K. (1995). Probabilistic seismic hazard analysis 362.
and design earthquakes: Closing the loop, Bull. Seism. Soc.
Am. 85, 1,275-1,284. GeoForschungsZentrum Potsdam
Musson, R. M. W. and P. W. Winter (1997). Seismic hazard
Telegrafenberg
maps for the UK, Natural Hazards 14, 141-154.
D-14473 Potsdam
Germany

Seismological Research Letters, Volume 72, No. 1, 2001 45

View publication stats