Content uploaded by Simone Barani
Author content
All content in this area was uploaded by Simone Barani on Jan 07, 2014
Content may be subject to copyright.
Disaggregation of Probabilistic Ground-Motion Hazard in Italy
by Simone Barani, Daniele Spallarossa, and Paolo Bazzurro
Abstract Probabilistic seismic hazard analysis is a process that integrates over
aleatory uncertainties (e.g., future earthquake locations and magnitudes) to calculate
the mean annual rate of exceedance (MRE) of given ground-motion parameter values
at a site. These rates reflect the contributions of all the sources whose seismic activity is
deemed to affect the hazard at that site. Seismic hazard disaggregation provides insights
into the earthquake scenarios driving the hazard at a given ground-motion level. This
work presents the disaggregation at each grid point of the Italian rock ground-motion
hazard maps developed by Gruppo di Lavoro MPS (2004),Meletti and Montaldo
(2007), and Montaldo and Meletti (2007). Disaggregation is used here to compute
the contributions to the MRE of peak ground horizontal acceleration (PGA) and
5%-damped 0.2, 1.0, and 2.0 sec spectral acceleration values corresponding to different
mean return periods (MRPs of 475 and 2475 yr) from different scenarios. These sce-
narios are characterized by bins of magnitude, M, source-to-site distance, R, and
number, ε, of standard deviations that the ground-motion parameter is away from its
median value for that MRpair as estimated by a prediction equation. Maps showing
the geographical distribution of the mean and modal values of M,R, and εare presented
for the first time for all of Italy. Complete joint M–R–εdistributions are also presented
for selected cities. Except for sites where the earthquake activity is characterized by
sporadic low-magnitude events, the hazard is generally dominated by local seismicity.
Moreover, as expected, the MRE of long-period spectral accelerations is generally con-
trolled by large magnitude earthquakes at long distances while smaller events at shorter
distances dominate the PGA and short-period spectral acceleration hazard. Finally, for a
given site, as the MRP increases the dominant earthquakes tend to become larger and to
occur closer to the site investigated.
Introduction
With his seminal work C. A. Cornell (1968) invented
probabilistic seismic hazard analysis (PSHA) by introducing
a method for evaluating the likelihood of exceedance (or
occurrence) of any level of earthquake ground shaking at a site
during a given period of time. This is achieved by combining
the effects of all the possible magnitudes and earthquake loca-
tions that could affect the hazard at the site investigated.
Although the original approach conceptually still holds, sub-
stantial progress has been made to refine the PSHA methodol-
ogy.For example, Power et al. (1981) proposed the use of logic
trees and, subsequently, Kulkarni et al. (1984) and Copper-
smith and Youngs (1986) suggested that logic trees provided
a convenient framework for the treatment of the uncertainty
both in the true parameter values in the mathematical models
adopted and also in the validity of the models themselves. This
uncertainty is often called epistemic (as opposed to aleatory)
because it is due to imperfect knowledge and, at least in prin-
ciple, could be reduced by further research. More recently,
many studies have emphasized the importance of acknowl-
edging and documenting uncertainties in PSHA (e.g., McGuire
and Shedlock, 1981;Rabinowitz and Steinberg, 1991;Cramer
et al., 1996; Senior Seismic Hazard Analysis Committee
[SSHAC], 1997;Grünthal and Wahlström, 2001;Cao et al.,
2005;Sabetta et al., 2005;Scherbaum et al., 2005;Barani
et al., 2007). Thus, sensitivity and uncertainty analyses have
become of primary importance in modern PSHA studies be-
cause they allow the identification of the models and parameter
values that have the highest influence on the hazard and drive
its uncertainty. The knowledge of those models and parameter
values is valuable in guiding focused research efforts to reduce
such an uncertainty and in facilitating a correct understanding
and use of PSHA results. In 1988 the National Research Coun-
cil (NRC,1988) pointed out the relevance of seismic hazard
disaggregation (or deaggregation) that becomes an insepa-
rable part of a PSHA. Important contributions to seismic hazard
disaggregation methods and applications were given by Bern-
reuter (1992),Chapman (1995),McGuire (1995),Cramer and
Petersen (1996),Bazzurro and Cornell (1999),Harmsen et al.
(1999),Sokolov (2000),Harmsen (2001),Harmsen and Fran-
kel (2001),Peláez Montilla et al. (2002),Halchuk and Adams
2638
Bulletin of the Seismological Society of America, Vol. 99, No. 5, pp. 2638–2661, October 2009, doi: 10.1785/0120080348
(2004),Hong and Goda (2006), and Cao (2007). The disag-
gregation process separates the contributions to the mean
annual rate of exceedance (MRE) of a specific ground-motion
value at a site due to scenarios of given magnitude M, distance
R, and, often, the ground-motion error term ε. This latter quan-
tity is defined as the number of standard deviations by which
the (logarithmic) ground motion generated by a given M–R
pair deviates from the median value estimated by a prediction
equation. Thus, this process reveals which earthquake scenar-
ios (defined by specific values of M,R, and ε) control the site
hazard. The knowledge of the M–R–εgroups dominating the
seismic hazard at a site is of paramount importance for many
engineering applications for which the hazard curves do not
suffice. For example, it is an important guide for simulating
or selecting appropriate ground-motion time histories for
nonlinear dynamic analyses of structures, liquefaction analy-
sis of soil, or slope stability studies. Appropriate here refers to
accelerograms that have the characteristics proper of those
scenario events that are more likely to cause the exceedance
of the target ground shaking level at the site. Considering ε
along with other disaggregation parameters (i.e., Mand R)
is also useful to better understand differences between prob-
abilistic ground-motion estimates and ground motions from
deterministic scenarios (Harmsen, 2001). The εpart of the dis-
aggregation results may also provide some guidance about the
minimum number of standard deviations, σ, at which the
ground-motion distribution for each scenario could be trun-
cated in the PSHA calculations. For example, severe trunca-
tions of the ground-motion distribution (e.g., at 1σ) are
never justified and are against empirical evidence of ground
motions 3σor more above the median in some historical
events. Such unwise truncations can cause severe underesti-
mation of the hazard (Abrahamson, 2006;Bommer and Abra-
hamson, 2006). Moreover, once the dominating magnitudes
are revealed, it might be desirable to examine in depth the in-
fluence of alternative magnitude distributions (e.g., exponen-
tial, characteristic) for the magnitude range relevant to those
dominant scenarios (McGuire, 2004). Thus, a comprehensive
PSHA should include sensitivity and disaggregation computa-
tions to explain what aspects of the PSHA drive the calculated
hazard (e.g., SSHAC,1997;Grünthal and Wahlström, 2001;
McGuire, 2004).
The work presented here deals with the disaggregation of
the Italian ground-motion hazard (Gruppo di Lavoro Mappa di
Pericolosità sismica [MPS], 2004;Meletti and Montaldo,
2007;Montaldo and Meletti, 2007). A new peak ground hor-
izontal acceleration (PGA) hazard map on rock for a mean re-
turn period (MRP) of 475 yr (the MRP is defined as the
reciprocal of the MRE1, thus an MRP of 475 yr is equivalent
to an MRE of 2×103) was produced for Italy by Gruppo
di Lavoro MPS in 2004. The term “rock”indicates sites with
average shear wave velocity, VS30, in the top 30 m of a soil
profile greater than 750 m=sec. Subsequently, many studies
were developed within the framework of the Dipartimento
di Protezione Civile–Istituto Nazionale di Geofisica e Vulca-
nologia (DPC–INGV) S1 project (see the Data and Resources
section) to supplement that map with a more comprehensive
documentation beyond the simple PGA hazard estimates.
Within the framework of that project, we performed the dis-
aggregation of the PGA hazard for nine MRPs (Spallarossa and
Barani, 2007). For information on the results, see the Data and
Resources section (Meletti et al., 2007). In this article, we
present maps showing the geographical distribution of both
mean (
M,
R, and
ε) and modal (M,R, and ε) values of
the 3D joint distributions of M,R, and εresulting from the dis-
aggregation of the values of PGA and 5%-damped spectral
acceleration, SaT, at periods, T, of 0.2, 1.0, and 2.0 sec that
are associated with 10% and 2% probability of exceedance in
50 yr (MRPs of 475 and 2475 yr, respectively). The maps were
created by considering a homogeneous grid of 0.05° spacing in
latitude and longitude. For a given MRP, disaggregating
the exceedance rates for multiple SaTallows us to identify,
for example, if common earthquake scenarios dominate
the hazard at all periods or, conversely, whether the hazard
at short- and long-period SaTis controlled by different
scenario earthquakes (e.g., McGuire, 1995;Barani et al.,
2008). Moreover, for 19 important cities in Italy, we tabulate
the values of both the triplets
M–
R–
εand M–R–εof the 3D
joint distributions of M,R, and εfor 0.2 and 1.0 sec spectral
acceleration values. For some of these cities we also present
and discuss example joint M–R–εdistributions. Note that
previous evaluations of the Italian seismic hazard based on
the application of the Cornell approach (e.g., Romeo and
Pugliese, 1998;Slejko et al., 1998;Albarello et al., 2000;
Romeo and Pugliese, 2000) did not provide any information
about the dominating earthquakes for any site. Finally,
although sensitivity analysis is not a central topic of this work,
the influence of different disaggregation procedures (i.e., dis-
aggregation by Mand Rrather than in terms of M,R, and ε) and
alternative ground-motion attenuation equations on disaggre-
gation results are also examined in some detail.
A Note on Models and Input Parameters
Before describing the models and input parameters, it is
important to provide an overview of the procedure adopted by
Gruppo di Lavoro MPS (2004) for the evaluation of the Italian
seismic hazard. According to the current practice in PSHA, the
hazard maps of Italy were developed using the logic tree
approach to account for the epistemic uncertainty in the mod-
els and in the parameter values of each model. Specifically,
1Incidentally, whenever possible in this article we chose to use annual
rates instead of annual probabilities because the definition of the latter
requires an assumption on the stochastic process of earthquake occurrence
(e.g., Poissonian), an assumption that is unnecessary for the topic discussed
here. However, because rates and probabilities when taking on small values
of engineering significance, such as those discussed here, are, for all prac-
tical purposes, numerically similar, these two terms are used somewhat in-
terchangeably. For example, the 475 yr map is also routinely called the 10%
in 50 yr exceedance probability map.
Disaggregation of Probabilistic Ground-Motion Hazard in Italy 2639
the logic tree adopted allows modeling the uncertainty in the
catalog completeness time for different magnitude ranges, in
the values of earthquake recurrence parameters and maximum
earthquake magnitude, and in alternative ground-motion
predictive equations. One earthquake catalog, called CPTI04
(Gruppo di Lavoro Catalogo Parametrico dei Terremoti
Italiani [CPTI], 2004; see the Data and Resources section),
and one seismogenetic zonation, named ZS9 (Gruppo di
Lavoro MPS,2004;Meletti et al., 2008; see the Data and
Resources section), were used both in the hazard evaluations
and in this study. Figure 1shows the geographical distribution
of epicenters based on the CPTI04 catalog and the ZS9 seis-
mogenetic zonation. Note that ZS9 assigns a prevalent
mechanism of faulting (interpreted as the mechanism with
the highest probability of generating future earthquakes) to
all its source zones for use in the ground-motion prediction
equations (Meletti et al., 2008). Fault sources were not con-
sidered in the hazard calculation because the distribution of
the epicenters appears to be inconsistent with the assumption
of line or planar sources, therefore preventing the attribution
of individual historical earthquakes to their causative fault.
Consequently, fault sources are not considered in our study
either. It is worth noting, however, that new and updated data
about known seismogenic faults in Italy have been recently
published (Basili et al., 2008). Should these new data after
careful evaluation be employed in future seismic hazard as-
sessments at a national scale, we will conduct a revision of
this disaggregation study to evaluate their effects.
Contrary to the recommendations of McGuire et al.
(2005) and Musson (2005) and to our opinion, but following
Abrahamson and Bommer (2005), the reference seismic haz-
ard maps of Italy were developed in terms of the median
hazard rather than in terms of the mean hazard. Thus, the
hazard values disaggregated here correspond to the fiftieth
percentile of the hazard distribution obtained by using a spe-
cific logic tree. Because the median hazard is not too sensi-
tive to all valid, properly weighted input assumptions
(particularly low-likelihood extreme assumptions that lead
to high hazard), the disaggregation is performed here using
the branches along the logic tree path that provided the
hazard values closest to the reference fiftieth percentile
hazard (path 921 of the logic tree adopted for the Italian
PSHA [V. Montaldo and C. Meletti, personal comm.,
2007]). In other words, the selected path corresponds to
the model and the parameter values that contribute the most
to the median hazard. This path considers the ZS9 seismo-
genetic zonation, a specific set of values of seismicity param-
eters (i.e., earthquake recurrence parameters and maximum
magnitude) for each source zone (see Table 1), and the rock
ground-motion attenuation equation by Ambraseys et al.
(1996). Note that this relationship was developed using
the larger of the two horizontal components of several
European ground-motion recordings, and it uses the surface
wave magnitude (MS) to quantify the size of earthquakes and
the Joyner and Boore distance (defined as the shortest dis-
tance to the surface projection of the fault rupture [Joyner
and Boore, 1981]) as a measure of the source-to-site dis-
tance. We have adjusted this equation to account for the style
of faulting by applying the average correction factors pro-
posed by Bommer et al. (2003). These multiplicative factors
were also used for the evaluation of the seismic hazard of
Italy. Finally, it is worth observing that we limited our cal-
culations to events with source-to-site distances up to 200 km
to avoid breaching the range of applicability of the ground-
motion prediction equation adopted.
Basics of 3D Seismic Hazard Disaggregation
In this section we give a brief description of the basic
concepts, definitions, and mathematics of the seismic hazard
disaggregation in terms M,R, and ε(3D disaggregation).
A detailed review of different disaggregation procedures
(e.g., 1D and 2D disaggregation, geographical disaggregation
in terms of latitude and longitude) can be found in Bazzurro
and Cornell (1999). However, it is worth specifying that the
disaggregation is performed here for the probability of ex-
ceeding the target ground-motion values rather than for
the probability of equaling them (McGuire, 1995).
Disaggregating seismic hazard means to evaluate the
contributions of different scenarios of given M,R, and ε
to the mean annual rate, λy, of exceeding a given
ground-motion parameter value yat a site. The contribution,
U,toλyof a particular (M,R, and ε) scenario, such as that
m1<M<m
2,r1<R<r
2, and ε1<ε<ε2, is given by
Um1<M<m
2;r
1<R<r
2;ε1<ε<ε2jY>y
PNS
i1υimminRm2
m1Rr2
r1Rε2
ε1fM;Rm; rfεεPY>y
jm; r; εdm dr dε
λy
;(1)
where
•NSindicates the number of potential earthquake sources;
•υimminis the annual rate of earthquake occurrence above
a minimum threshold magnitude mmin;
•fM;Rm; ris the joint probability density function (pdf) of
magnitude and distance;
2640 S. Barani, D. Spallarossa, and P. Bazzurro
•fεεis the pdf (standardized normal density function) for
the ground-motion error term, ε;
•PY>y
jm; r; εis the conditional probability of exceed-
ing a particular value, y, of a ground-motion parameter, Y,
for a given magnitude, m, distance, r, and εstandard de-
viations away from the predicted median ground motion.
Once all the contributions are evaluated, the disaggre-
gated hazard (i.e., the contribution to λyfrom different dis-
crete M–R–εbins) is represented by the joint probability
mass function (PMF)ofM–R–ε, which is usually summa-
rized by its mean (
M,
R, and
ε) and modal (M,R, and
ε) values. A closed-form solution for (M,R, and ε)is
not available but
M,
R, and
εcan be calculated as follows:
MPNS
i1υimminRRR mfM;Rm; rfεεPY>y
jm; r; εdm dr dε
λy
;(2)
RPNS
i1υimminRRR rfM;Rm; rfεεPY>y
jm; r; εdm dr dε
λy
;(3)
εPNS
i1υimminRRR εfM;Rm; rfεεPY>y
jm; r; εdm dr dε
λy
;(4)
where the triple integration covers all possible values of M,
R, and ε.
While the mode is a metric with the distinct advantage of
representing the event that most likely generates the exceed-
ance of the target ground-motion level at the site considered,
it has also the disadvantage of being sensitive to the bin size
adopted and to being dependent on the disaggregation
scheme (e.g., Bazzurro and Cornell, 1999;Abrahamson,
2006). The mean, on the other hand, is easier to understand
and to compute than the mode; it is independent of the bin-
ning size and of the disaggregation scheme but does not nec-
essarily represent the most likely M–R–εevent that contrib-
utes to the hazard. In some extreme cases the (
M,
R, and
ε)
bin may not even correspond to a feasible event. Because the
PMF representation is sensitive to the binning scheme, it
might be convenient to express the results in terms of pdf,
which is obtained by dividing the PMF contribution of each
bin by the bin’s size (Bazzurro and Cornell, 1999). The pdf
representation, indeed, is independent of the bin’s amplitude
but has the disadvantage that the ordinate of each cell does
not represent the real contribution to the hazard. In this study,
the results are represented in terms of PMF with distance and
magnitude bins of even size. A linear distance binning was
preferred here to other binning strategies, such as logarithmic
or pseudologarithmic, which tend to overemphasize the
contributions due to large Rvalues (Cramer and Petersen,
1996). The binning scheme applied is generic and appro-
priate for most cases. However, we acknowledge that
particular, nonstandard binning schemes may be prefera-
ble to those adopted here for some specific engineering
applications.
Disaggregation Results
The following sections present the results obtained from
the disaggregation of the Italian ground-motion hazard for
two MRPs of 475 and 2475 yr. The discussion of results
is organized as follows. First, for selected Italian cities, ex-
ample 3D joint PMFs of M–R–εare described and the mean
and modal values of M,R, and εtabulated and analyzed.
Subsequently, we discuss the geographic distribution (i.e.,
disaggregation maps) of
M,
R,
ε, and M,R,εas obtained
from a 3D disaggregation. Gridded values of Mand Rwill
be compared with those obtained by disaggregating the haz-
ard in 2D M–Rbins. This allows us to study the effect of
different disaggregation procedures on results. Finally, the
influence of alternative ground-motion predictive equations
on results is carefully examined.
Disaggregation of Seismic Hazard for Selected Cities
This section presents example joint M–R–εdistributions
and their central statistics resulting from the disaggregation
of the 5%-damped linear elastic SaTvalues at periods of
0.2 and 1.0 sec that correspond to MRPs of 475 and 2475 yr
for three Italian cities, L’Aquila, Rome, and Milan. These
three cities have been selected because they are representa-
tive of different hazard levels in Italy. Bins of width 0.5 in
magnitude, 10 km in distance, and 0.2 in εare used.
Disaggregation of Probabilistic Ground-Motion Hazard in Italy 2641
The 475 yr Sa0:2sechazard for L’Aquila (Fig. 2)is
controlled by nearby-distance earthquakes (from 0 to 10 km)
of low-to-moderate magnitude (from 4.5 to 6.5). For these
events, εis in the range of 0–2, meaning that the dominant
ground motions are within 2σof the median. Larger and
more distant events dominate the Sa1:0secvalue for
the same MRP. In particular, for Sa0:2sec,
Mis 5.8,
R
is 9.3 km, and
εis 1.13. The modal values of magnitude,
M4:75, and distance, R5km, are lower than the cor-
responding mean values, while ε, being equal to 1.5, is
greater (note that the modal values are the central values
of the M,R, and εbins used in the calculations; hence,
M4:75 implies that Mis comprised between 4.5 and
5.0). The modal value of Mof the 2D joint M–Rdistribution
is 0.5 units greater than that of the 3D PMF. For 1.0 sec re-
sponse,
M6:5,
R18:2km, and
ε0:91 while the modal
triplet is M6:75,R15 km, and ε0:5. For this peri-
od, the modal values of Mand Rof the 3D PMF coincide with
Figure 1. (a) ZS9 seismogenetic zonation and (b) distribution
of the national seismicity based on the CPTI04 catalog.
Table 1
Seismogenic Zone Parameters
Zone Ftype*mmin
†Mmax
†υmmin‡b§
901 undetermined 4.3 5.8 0.045 1.133
902 undetermined 4.3 6.1 0.103 0.935
903 undetermined 4.3 5.8 0.117 1.786
904 strike-slip 4.3 5.5 0.050 0.939
905 reverse 4.3 6.6 0.316 0.853
906 reverse 4.3 6.6 0.135 1.092
907 reverse 4.3 5.8 0.065 1.396
908 strike-slip 4.3 5.5 0.140 1.408
909 normal 4.3 5.5 0.055 0.972
910 reverse 4.3 6.4 0.085 0.788
911 strike-slip 4.3 5.5 0.050 1.242
912 reverse 4.3 6.1 0.091 1.004
913 undetermined 4.3 5.8 0.204 1.204
914 undetermined 4.3 5.8 0.183 1.093
915 normal 4.3 6.6 0.311 1.083
916 normal 4.3 5.5 0.089 1.503
917 reverse 4.3 6.1 0.121 0.794
918 undetermined 4.3 6.4 0.217 0.840
919 normal 4.3 6.4 0.242 0.875
920 normal 4.3 5.5 0.317 1.676
921 normal 4.3 5.8 0.298 1.409
922 normal 4.3 5.2 0.090 1.436
923 normal 4.3 7.3 0.645 0.802
924 strike-slip 4.3 7.0 0.192 0.945
925 strike-slip 4.3 7.0 0.071 0.508
926 strike-slip 4.3 5.8 0.061 1.017
927 normal 4.3 7.3 0.362 0.557
928 normal 4.3 5.8 0.054 1.056
929 normal 4.3 7.6 0.394 0.676
930 undetermined 4.3 6.6 0.146 0.715
931 strike-slip 4.3 7.0 0.045 0.490
932 strike-slip 4.3 6.1 0.118 0.847
933 reverse 4.3 6.1 0.172 1.160
934 reverse 4.3 6.1 0.043 0.778
935 strike-slip 4.3 7.6 0.090 0.609
936 undetermined 3.7 5.2 0.448 1.219
*Ftype is the prevalent mechanism of faulting.
†mmin and Mmax indicate the minimum and maximum
magnitudes, respectively. Magnitudes are given in terms of MS.
‡υmminis the annual rate of earthquake occurrence above a
minimum threshold magnitude mmin.
§bis the negative slope of the Gutenberg–Richter relation
(Gutenberg and Richter, 1944).
2642 S. Barani, D. Spallarossa, and P. Bazzurro
those of the 2D PMF. Regarding the disaggregation of the
2475 yr SaTat 0.2 and 1.0 sec values, results show that
earthquakes of larger size occurring at slightly shorter dis-
tances dominate. In other words, as the MRP increases, the
controlling earthquakes become larger in Mand occur closer
to the site investigated. In detail, events of magnitude from
5.0 to 7.0, again occurring at distances between 0 and 10 km,
control the 0.2 sec response while the larger contribution to
the 1.0 sec hazard comes from greater magnitude events from
6.0 to 7.5, at distances from 0 to 20 km. Mean values of εare
Figure 2. Joint M–R–εPMFs for L’Aquila obtained from the disaggregation of the Sa0:2secand Sa1:0secvalues corresponding to
MRPs of 475 (top row) and 2475 yr (bottom row).
Disaggregation of Probabilistic Ground-Motion Hazard in Italy 2643
greater than those obtained for an MRP of 475 yr while the
modal are either lower (for Sa0:2sec) or greater (for
Sa1:0sec). For both Sa0:2secand Sa1:0sec, the
modal event of the PMF of M–Rdiffers from that of the joint
distribution of M–R–ε.
For the same site, Figure 3shows the variation of mean
and modal values of M,R, and εwith spectral period (we
have taken into account additional periods for this exhibit).
Again, the acceleration values disaggregated corresponds to
MRPs of 475 (left-hand panels) and 2475 yr (right-hand pan-
els). As a comparison, modal values of both the 3D and 2D
PMFs are displayed. We denote the mode of a 2D distribution
as M
2D–R
2Dwhile that of a 3D one is simply indicated by the
triplet M–R–ε. For both MRPs considered, local earth-
quakes at distances up to 20 km dominate the hazard at both
lower and higher periods. Unlike
R,R, and R
2D, which vary
very slightly with period, the mean and modal values of M
increase more markedly as the spectral period increases. For
an MRP of 475 yr,
Mranges from 5.5 at 0.1 sec to 6.6 at
2.0 sec while Mand M
2Drange from 4.75 to 6.75. Note
that the values of
Mand M
2Dare larger for PGA than for
Sa0:1secand Sa0:15 sec.Mis lower than or equal
to M
2Dfor a period of up to 0.3 sec, becomes greater than
M
2Dfor a period up to 0.75 sec, and, for longer periods (1.0–
2.0 sec), coincides with M
2D. For an MRP of 2475 yr, both
the mean and modal magnitudes dominating the PGA hazard
are greater than those controlling the spectral accelerations
up to 0.15 sec. Then,
M,M, and M
2Dincrease progressively
with period. Mand M
2Dcoincide at 0.0, 0.4, 0.75, and
1.0 sec. Analyzing εbehavior with period does not reveal
any particular trend. That is, both
εand εneither increase
nor decrease regularly with increasing spectral period. As ex-
pected, both mean and modal values of εtend to increase as
the MRP increases. For an MRP of 475 yr, indeed, the domi-
nant ground motions are within 0:5σ–1:5σof the median
while; for an MRP of 2475 yr, they are within 1:1σ–1:9σ.
Figure 4shows the disaggregation plots for Rome. Here
the 0.2 sec spectral acceleration hazard is entirely dominated
by a single seismic zone (zone 922). Indeed, nearby-distance
events (up to 10 km) of low magnitude (from 4.0 to 5.0)
Figure 3. Variation of mean and modal values of M,R, and εwith spectral period for L’Aquila. Note that “STDV”in the label on the y
axis of the charts at the bottom of the figure stands for standard deviation.
2644 S. Barani, D. Spallarossa, and P. Bazzurro
contribute about 70% and 80% to the Sa0:2sechaz-
ard for MRPs of 475 and 2475 yr, respectively. Thus, mean
and modal values of M,R, and εare close to each other (only
the values of
εand εresulting from the disaggregation of the
Sa0:2secvalue with 10% probability of exceedance in
50 yr show significant differences). For both MRPs consid-
ered, ground motions with ε>1:0are necessary to exceed
the target acceleration values. Distributions for 1.0 sec
response are bimodal, clearly reflecting the contribution to
the hazard from both local (zone 922) and distant (zone
Figure 4. Joint M–R–εPMFs for Rome as obtained from the disaggregation of the Sa0:2secand Sa1:0secvalues corresponding to
MRPs of 475 (top row) and 2475 yr (bottom row).
Disaggregation of Probabilistic Ground-Motion Hazard in Italy 2645
923) sources. In such cases or, more generally, if distribu-
tions have more than one prominent peak (multimodal dis-
tributions), single summary statistics, like the mean and the
mode, are insufficient to fully describe the seismic threat at
the site. In these cases, it is desirable to document the dif-
ferent M–R–εgroups controlling the site hazard. For an
MRP of 475 yr, the local event contributing the most to the
1.0 sec spectral acceleration hazard corresponds to the modal
scenario M4:75,R5km, and ε1:5. The contribu-
tion from regional earthquakes is mainly associated with
magnitudes from 6.5 to 7.5 at distances comprised between
70 and 120 km. For both close and distant events the ex-
pected ground motions are within 2σof the median. Note,
finally, that the mean, because of its nature, corresponds
to a triplet having a very low contribution to the hazard
(<1:0%) and, therefore, represents an unlikely scenario.
Analogous observations can be made analyzing the 3D plot
obtained by disaggregating the Sa1:0secvalue for an
MRP of 2475 yr. It is worth noting, however, that, for
low-magnitude events at distances up to 10 km, εvaries from
1 to 3. Ground motions from larger magnitude earthquakes
occurring at longer distances (from 60 to 130 km), instead,
are generally within 2σof the median. Moreover, the modal
values of magnitude and distance of the 3D joint M–R–εdis-
tribution strongly differ from those of the 2D PMF. The for-
mer, indeed, corresponds to the pair M7:25,R95 km,
while the latter corresponds to M4:75,R5km.
Figure 5shows the variation of the mean and modal
values of M,R, and εwith spectral period for Rome. For
an MRP of 475 yr,
Mtends to increase progressively with
period, while Mand M
2Dassume the same constant value
of 4.75 (except for Sa0:1sec) up to 1.0 sec. Then, Min-
creases abruptly while M
2Ddoes not vary. An analogous be-
havior can be observed analyzing the variation of mean and
modal Rvalues. Thus, the large difference between mean and
modal values of both Mand Rfor periods greater than
0.5 sec indicates that more than one scenario contributes
significantly to the site hazard for those spectral periods.
Similar observations can be made for an MRP of 2475 yr.
Once again, examining the behavior of εindicates that both
Figure 5. Variation of mean and modal values of M,R, and εwith spectral period for Rome.
2646 S. Barani, D. Spallarossa, and P. Bazzurro
the mean and modal values increase as the MRP increases.
As an example, in the case of Sa0:2sec,
εincreases from
1.26 to 1.73 when the MRP changes from 475 to 2475 yr.
Similarly, for the same spectral period, εjumps from 0.7
to 1.5. Note that modal values of εare always lower than
the mean for both MRPs considered.
Figure 6shows the 3D PMF of M–R–εfor Milan. In the
first row, for response accelerations having 10% probability
Figure 6. Joint M–R–εPMFs for Milan as obtained from the disaggregation of the Sa0:2secand Sa1:0secvalues corresponding to
MRPs of 475 (top row) and 2475 yr (bottom row).
Disaggregation of Probabilistic Ground-Motion Hazard in Italy 2647
of exceedance in 50 yr, the left-hand figure (0.2 sec response)
shows that the dominating contributions come from low-
magnitude events (from 4.0 to 5.5) at distances ranging be-
tween 30 and 60 km. For these M–Rpairs, εis generally
lower than or equal to 2.0. The modal values of Mand R
of the 2D and 3D PMF coincide. The figure to the right
(Sa1:0sec) shows that the hazard is controlled by multiple
events that have similar contributions, generally lower than
4%. Again, the expected ground motions are approximately
within 2σof the median. In the second row, for an MRP of
2475 yr, the 0.2 sec response acceleration is again controlled
by low-magnitude events concentrated at about 30–60 km
from Milan while both local and distant earthquakes contrib-
ute to the 1.0 sec response. In this latter case, two peaks
can be identified. They corresponds to the cells M5:25,
R45 km (zone 907) and M6:25,R115 km (zone
906). The former corresponds to the modal M–Rpair of
the 2D PMF that, again, coincides with that of the 3D distri-
bution. For both the 0.2 and 1.0 sec response, εis in the range
of 1–3.
Figure 7shows the variation of the mean and modal values
of M,R, and εwith period for the city of Milan. For both the
MRPs considered,
Mand
Rincrease progressively with period
while the 2D and 3D modal Mand Rvalues show a stepwise
trend. As an example, for the 2475 yr case, Mand M
2Dare
equal to 4.75 up to 0.2 sec, then they increase to 5.25 at 0.3 sec,
remain constant up to 1.0 sec, and, finally, jump to 6.25 at
1.5 sec. For the same MRP,Rand R
2Dassume a nearly con-
stant value of 35–45 km up to 1.0 sec, then they increase sig-
nificantly and remain constant up to 2.0 sec. Once again, this
indicates that distant sources tend to control long-period spec-
tral accelerations while the local seismicity dominates at
shorter periods. As expected, for a given spectral period,
the mean and modal values of Rtend to decrease with increas-
ing MRP. Conversely,
εand εtend to increase.
As a summary of this disaggregation exercise, Tables 2
and 3list values of
M,
R,
εand M,R,εobtained from
the disaggregation of the Sa0:2secand Sa1:0secvalues
corresponding to MRPs of 475 (Table 2) and 2475 yr (Table 3)
for the main Italian cities (Fig. 8). The relative influence of
Figure 7. Variation of mean and modal values of M,R, and εwith spectral period for Milan.
2648 S. Barani, D. Spallarossa, and P. Bazzurro
Table 2
Mean and Modal Values of M,R, and εfor the Main Italian Cities for 0.2 and 1.0 sec Spectral Acceleration Values Corresponding to MRPs of 475 yr
City Sa0:2secSa1:0sec
MT0:2sec
RT0:2sec
εT0:2sec M
T0:2sec R
T0:2sec ε
T0:2sec
Zone
(T0:2sec)
MT1:0sec
RT1:0sec
εT1:0sec M
T1:0sec R
T1:0sec ε
T1:0sec
Zone
(T1:0sec)
Ancona 0.484 0.117 5.2 9.3 0.99 4.75 5 1.1 917 5.7 28.6 1.23 5.75 15 0.9 917
Aosta 0.248 0.051 4.9 16.4 1.01 4.75 5 0:1909 5.3 33.7 1.35 5.25 15 0.9 909
Bari 0.196 0.103 6.3 70.7 1.45 6.25 35 0.7 925 6.8 98.8 1.50 7.25 125 1.3 925
Bologna 0.416 0.102 5.0 9.6 1.11 4.75 5 0.5 913 5.4 18.6 1.40 4.75 5 1.3 913
Campo Basso 0.609 0.214 5.9 12.7 1.25 4.75 5 1.3 924 6.7 27.6 1.13 7.25 35 0.9 927
Catanzaro 0.637 0.231 6.0 10.4 0.98 5.25 5 0.9 929 6.7 22.8 0.83 6.25 15 1.3 929
Florence 0.346 0.091 5.1 14.5 1.40 4.75 5 0.5 916 5.7 34.1 1.54 6.25 25 0.7 915
Genoa 0.185 0.040 5.0 33.2 1.34 4.75 15 0.9 911 5.7 72.5 1.46 4.75 15 1.3 910
L’Aquila 0.648 0.212 5.8 9.3 1.13 4.75 5 1.5 923 6.5 18.2 0.91 6.75 15 0.5 923
Milan 0.139 0.026 5.1 65.0 1.88 4.75 45 1.9 907 5.5 109.0 1.58 5.75 125 1.7 907
Naples 0.412 0.141 5.2 14.1 1.25 4.75 5 1.1 928 6.6 61.6 1.51 7.25 65 1.1 927
Palermo 0.454 0.114 5.0 8.5 1.02 4.75 5 0.9 933 5.5 23.4 1.23 5.25 5 0.9 933
Perugia 0.471 0.125 5.2 10.3 1.40 4.75 5 1.1 919 6.0 33.9 1.47 6.25 15 0.7 919
Potenza 0.501 0.199 5.9 15.3 1.14 4.75 5 0.9 927 6.7 30.6 1.08 7.25 25 0.5 927
Rome 0.399 0.094 4.7 7.9 1.26 4.75 5 0.7 922 6.2 73.5 1.54 4.75 5 1.5 923
Turin 0.158 0.019 5.0 48.8 1.93 4.75 35 1.9 908 5.3 86.4 1.35 4.75 35 1.5 908
Trento 0.211 0.059 5.5 47.5 1.64 4.75 35 2.1 906 6.0 68.9 1.40 6.25 75 1.1 906
Trieste 0.293 0.084 5.1 23.2 1.51 4.75 15 1.5 904 5.9 53.9 1.65 5.25 15 1.5 905
Venice 0.193 0.070 5.7 63.1 1.70 5.75 45 1.3 905 6.1 73.2 1.50 6.25 45 0.7 905
Modal values are reported indicating the center of the M,R, and εbins used in the calculations (e.g., M4:75 indicates that Mis comprised between 4.5 and 5.0). The identification number of the source zone
contributing the most to the Sa0:2secand Sa1:0sechazard is also indicated for each city.
Disaggregation of Probabilistic Ground-Motion Hazard in Italy 2649
Table 3
Mean and Modal Values of M,R, and εfor the Main Italian Cities for 0.2 and 1.0 sec Spectral Acceleration Values Corresponding to MRPs of 2475 yr
City Sa0:2secSa1:0sec
MT0:2sec
RT0:2sec
εT0:2sec M
T0:2sec R
T0:2sec ε
T0:2sec
Zone
(T0:2sec)
MT1:0sec
RT1:0sec
εT1:0sec M
T1:0sec R
T1:0sec ε
T1:0sec
Zone
(T1:0sec)
Ancona 0.88 0.232 5.3 6.2 1.44 5.25 5 1.1 917 5.7 14.4 1.52 5.75 5 0.9 917
Aosta 0.445 0.103 4.9 7.5 1.25 4.75 5 0.9 909 5.3 19.4 1.69 5.25 5 0.9 909
Bari 0.339 0.200 6.5 50.7 1.73 6.75 35 1.1 925 6.9 76.8 1.98 6.75 35 1.3 925
Bologna 0.741 0.194 5.0 6.0 1.53 4.75 5 1.5 913 5.5 10.5 1.66 5.25 5 1.5 913
Campo Basso 1.066 0.396 6.1 9.1 1.59 5.25 5 1.7 924 6.8 20.8 1.47 6.75 15 1.3 927
Catanzato 1.274 0.533 6.4 6.7 1.28 6.25 5 0.9 929 7.0 14.0 1.11 7.25 15 0.9 929
Florence 0.584 0.170 5.1 9.2 1.72 4.75 5 1.3 916 6.0 24.2 1.79 6.25 25 1.5 915
Genoa 0.325 0.089 4.9 17.8 1.61 4.75 15 1.7 911 5.8 59.7 1.96 5.25 15 1.5 911
L’Aquila 1.218 0.456 6.2 6.7 1.45 5.75 5 1.3 923 6.7 12.6 1.24 6.75 15 1.5 923
Milan 0.218 0.056 5.2 51.8 2.19 4.75 35 2.3 907 5.9 103.0 2.02 5.25 45 2.1 906
Naples 0.739 0.257 5.1 6.6 1.68 4.75 5 1.7 928 6.8 53.5 1.97 7.25 55 1.7 927
Palermo 0.857 0.234 5.1 5.7 1.55 4.75 5 1.5 933 5.6 14.2 1.61 5.75 5 0.9 933
Perugia 0.813 0.235 5.3 7.1 1.81 4.75 5 1.7 919 6.1 25.0 1.79 6.25 15 1.5 919
Potenza 0.920 0.389 6.1 11.3 1.58 6.75 15 1.3 927 6.9 21.7 1.42 7.25 25 1.3 927
Rome 0.671 0.168 4.7 4.3 1.73 4.75 5 1.5 922 6.5 73.1 1.99 7.25 95 1.7 923
Turin 0.242 0.046 5.1 39.2 2.27 4.75 25 2.3 908 5.7 88.1 1.92 6.25 105 1.1 910
Trento 0.357 0.115 5.8 39.1 1.93 5.25 25 1.9 906 6.2 57.4 1.76 6.25 65 1.7 906
Trieste 0.503 0.160 5.1 14.7 1.90 5.25 5 1.3 904 6.0 46.8 2.10 6.25 45 1.7 905
Venice 0.336 0.141 6.1 55.3 2.06 6.25 45 1.5 905 6.3 62.9 2.00 6.25 45 1.5 905
Modal values are reported indicating the center of the M,R, and εbins used in the calculations (e.g., M4:75 indicates that Mis comprised between 4.5 and 5.0). The identification number of the source zone
contributing the most to the Sa0:2secand Sa1:0sechazard is also indicated for each city.
2650 S. Barani, D. Spallarossa, and P. Bazzurro
local zones and of regional sources farther from the site might
be deduced from the data of Tables 2and 3. For example, at
cities such as Ancona, Bologna, Catanzaro, Palermo, Perugia,
and Rome, the similarity of
Mand M, and that between
R
and R(Table 2), indicates that a single source zone surround-
ing these cities dominates the Sa0:2sechazard at the 475 yr
MRP level. The events controlling the Sa0:2sechazard
at these locations are generally characterized by low-to-
moderate magnitudes (from 4.5 to 6.0) at distances up to about
10 km. For these cities, the dominant ground motions are gen-
erally within 1:4σof the median. On the other hand, for the
remaining sites, the differences between the mean and modal
values of Mand Rindicate that more than one event contrib-
utes significantly to the hazard. In particular, as the discrep-
ancy between
Rand Rincreases, the hazard tends to be
controlled by both the local and regional seismicity, suggest-
ing a bimodal distribution. Thus, the noticeable difference be-
tween
Rand Rat Bari, Genoa, Milan, Turin, Trento, and
Venice indicates that both close and distant earthquakes affect
the Sa0:2sechazard and should be taken into account to
properly derive time histories for engineering analysis and de-
sign at a given MRP. Furthermore, for these cities, the dominant
ground motions are generally within 2σof the median. For the
same MRP of 475 yr, variations with spectral period of both the
mean and modal values of M,R, and εreveal that different
scenario events control the hazard at short (T0:2sec) and
moderate (T1:0sec) periods. The difference is evident, for
example, at Campobasso, Florence, L’Aquila, Milan, Naples,
Figure 8. Map of Italy with the location of the cities considered in Tables 2and 3.
Disaggregation of Probabilistic Ground-Motion Hazard in Italy 2651
Perugia, Potenza, and Trento where closer, low-to-moderate
magnitudes dominate the 0.2 sec hazard but larger magnitude
events from distant sources control the Sa1:0secvalues.
Similar observations can be made from Table 3, which shows
the mean and modal values of M,R, and εcontrolling the
SaThazard at 0.2 and 1.0 sec for the 2475 yr MRP.As
observed previously, the hazard at long MRPs tends to be
controlled by closer and larger earthquakes.
Disaggregation Maps
Figures 9,10, and 11 present maps showing the geo-
graphic variation of both the mean and modal values of
M,R, and εextracted from the 3D joint PMFs conditioned
on exceeding the 475 yr values of PGA and the 5%-damped
Sa0:2sec,Sa1:0sec, and Sa2:0sec. The maps are
contours based on values computed for each node of a grid
with spacing of 0.05° in latitude and longitude. Specifically,
the left- and right-hand columns of these three figures display
the geographic distribution of the mean and modal values of
M,R, and ε, respectively, resulting from the disaggregation
of the four ground-motion parameters mentioned previously.
These maps should be inspected in conjunction with prob-
abilistic seismic hazard maps (for more information, see
the Data and Resources section) and with the seismicity dis-
tribution depicted in Figure 1b.
Maps of
Mand M(Fig. 9) confirm the results presented
in the previous section, which show that the values of both
parameters tend to increase as the spectral period increases.
Given the similar behavior for Sa0:2secand PGA (i.e.,
PGA behaves as a high-frequency ground motion), as well
as for the 1.0 and 2.0 sec response, in the following discussion
we will focus on maps for Sa0:2secand Sa1:0seconly.
Except for two areas in the eastern sector in Northern Italy,
mapped values of
Mfor the 0.2 sec response vary between
4.5 and 5.5, while Massumes a nearly constant value of
4.5–5.0. The largest values of both these Mstatistics are con-
centrated in two low-seismicity areas located in the northeast
where the hazard is controlled by source zones 905 and 906
(see Fig. 1), two tectonic provinces where strong earthquakes
with magnitudes larger than 6.0 occurred in the past (the
largest event is the well-known Friuli earthquake that occurred
on 6 May 1976 with magnitude MS6:4). Note that the largest
differences in the distributions of the mean and modal values
are generally concentrated in areas that in the past were af-
fected only by sporadic low-magnitude earthquakes (e.g., east
of zone 908) and within source zones 905 and 906 where the
hazard is dominated by local seismicity. For the same spectral
period, an analogous spatial distribution can be observed in
central Italy where, again,
Mand Mvary from 4.5 to 5.5 and
from 4.5 to 5.0, respectively, almost everywhere. In southern
Italy, the geographic distribution of
Mresults are substantially
smoother than that of Mbecause averaging takes into account
the contribution of both the local and regional seismicity.
Generally,
Mranges from 5.5 to 7.0 while Mis strongly local,
assuming values from 5.0 to 6.5 in areas (e.g., zones 927 and
929) that in the past were struck by strong earthquakes (the
largest one is the 1908 Messina earthquake of magnitude
MS7:2) and values from 6.5 to 7.5 in some onshore and off-
shore areas characterized by low-seismic activity. For a 1.0 sec
response,
Mvaries from 4.5 to 6.0 in northwestern Italy
where, except for the Ligurian Sea and adjacent areas, it
can be up to 0.5 units greater than M. The distribution of
the mean and modal values, instead, is similar in the northeast
where the dominant magnitudes are up to 6.5.
Mand Mas-
sume larger values in central Italy where they are up to 7.0 and
7.5, respectively.
Mis between 6.5 and 7.0 almost everywhere
in southern Italy and reaches its maximum (7:0<
M<7:5)in
the southern Tyrrhenian Sea where the hazard is controlled by
distant sources (e.g., zone 929). Modal values are slightly
greater with their maxima in areas (e.g., near Bari, south of
Potenza, central Sicily southeast of Palermo, Tyrrhenian
Sea, and Ionian Sea) that historically have had only small
earthquakes. Both
Mand Mare lowest in northern Sicily near
Palermo where both local sources (zone 933) and events at dis-
tances up to 60 km contribute significantly to the Sa1:0sec
hazard.
Again, maps of the mean and modal distances (Fig. 10)
show that
Rand Rincrease with increasing spectral period
(this behavior is more evident for
R). Analogous to what is
observed in Figure 9, maps for PGA and a 0.2 sec response
and for Sa1:0secand Sa2:0secare similar to each other.
Except for very mildly seismic areas, the Sa0:2sechazard is
generally controlled by local seismogenetic zones. Close-
distance sources (e.g., R≤20 km) dominate the hazard both
in highly seismic areas (e.g., zones 905, 923, 927, and 929) and
at sites where seismic activity is prevalently characterized by
small-to-moderate but quite frequent events (e.g., zones 912,
917, 918, and 921). Sources at larger distances dominate the
Sa0:2sechazard in the northwest (east of Turin), northeast
(north of Trento), near Bari in southern Italy, in Sicily south-
east of Palermo, and offshore. For a 1.0 sec response, the geo-
graphic distributions of
Rdiffer significantly from that of R
because the mean, contrary to the mode, is strictly affected by
contributions from the regional seismicity.
Ris generally be-
tween 10 and 60 km and is larger, as much as 200 km, in areas
where earthquakes are rare (e.g., near Bari). On the other hand,
except for low-seismicity areas, Ris smaller than or equal to
10 km almost everywhere in the north and exhibits slightly
greater values in central and southern Italy.
Figure 11 shows the geographical variation of the mean
and modal values of ε. Contrary to the maps of the mean and
modal Mand R, which show very similar distributions for
PGA and Sa0:2sec, those of
εand εpresent some differ-
ences and, therefore, are both discussed. For PGA,
εis be-
tween 0.5 and 1.0 both at sites with higher PGA hazard
(e.g., zones 905, 923, 927, 929, and 935 where the PGA
values for an MRP of 475 yr are greater than 0.225g) and
in areas affected by quite frequent, small magnitude events
(e.g., zones 912, 917, 921, and 933 where the PGA values
corresponding to an MRP of 475 yr ranges from 0.125 to
0.20g). Mean values of εrise up to 2.0 (and even to 2.5
2652 S. Barani, D. Spallarossa, and P. Bazzurro
Figure 9. Maps of mean (left-hand column) and modal (right-hand column) magnitude, M, values obtained from the 3D disaggregation
of the 475 yr values of PGA,Sa0:2sec,Sa1:0sec, and Sa2:0sec.
Disaggregation of Probabilistic Ground-Motion Hazard in Italy 2653
Figure 10. Maps of mean (left-hand column) and modal (right-hand column) distance, R, values obtained from the 3D disaggregation of
the 475 yr values of PGA,Sa0:2sec,Sa1:0sec, and Sa2:0sec.
2654 S. Barani, D. Spallarossa, and P. Bazzurro
Figure 11. Maps of
ε(left-hand column) and ε(right-hand column) obtained from the 3D disaggregation of the 475 yr values of PGA,
Sa0:2sec,Sa1:0sec, and Sa2:0sec. Note that “STDV”in the legend stands for standard deviation.
Disaggregation of Probabilistic Ground-Motion Hazard in Italy 2655
at some sites in the northwest) in areas that are almost aseis-
mic (e.g., northwestern Italy near Turin, northeastern Italy
near Trento, near Bari, Tyrrhenian, and Ionian Sea) where
the PGA is lower than 0.05g. The distribution of ε, although
it appears more grained, does not differ substantially from
that of
ε. The main differences concentrate in the southern
Tyrrhenian Sea and the Ionian Sea where εis about 0.5 units
lower than
ε. For a 0.2 sec response, the mean and modal
values of εare slightly greater (no more than 0.5 units) than
for PGA. Again, the largest values (from 1.5 to 2.5) concen-
trate in areas characterized by very low-seismic activity
while in higher seismic hazard regions,
εand εrange be-
tween 0.5 and 1.5. The mean and modal values of εdecrease
for Sa1:0secand increase again for Sa2:0sec.2For
these spectral periods, maps of the mean and modal εdiffer
significantly from each other. Except for regions character-
ized by a high-seismic hazard level (e.g., Calabrian arc and
southern Sicily near the Etna volcano), for a 1.0 sec response,
εis generally between 1.0 and 1.5 almost everywhere in
Italy. The geographical distribution of ε, instead, varies
widely from site to site. As was the case for
ε,εis also
minimum in high-seismic hazard areas where it is generally
lower than 1.0. For a 2.0 sec response, the geographical dis-
tribution of εvalues is very similar to that for Sa1:0sec.
The map of
ε, instead, presents some differences mostly in
northern and central Italy where mean values of εare slightly
greater than those for a 1.0 sec response.
Influence of 2D and 3D Disaggregation Scheme
on Results
As mentioned previously, for some sites the bivariate
(M
2D,R
2D) values do not coincide with the Mand Rmodal
values extracted from the 3D joint PMF. To quantify the effect
of the 2D and 3D disaggregation procedures at a national
scale, we have calculated and mapped the ratio of the 2D to
the 3D modal values of Mand R(Fig. 12). The left- and
right-hand columns of the figure correspond to maps result-
ing from the disaggregation of the Sa0:2secand
Sa1:0secvalues for an MRP of 475 yr while rows show
the geographical distribution of M
2D=Mand R
2D=R. The
maps are very grained, indicating that both magnitude and
distance ratios vary significantly from one site to another.
Therefore, it is difficult to relate the geographical variation
of M
2D=Mand R
2D=Rto the seismicity distribution
(Fig. 1b) and the hazard maps for each spectral period. In
general, the modal Mand Rvalues of the 2D PMF coincide
with those of the corresponding 3D distribution at most of
the gridded points. Specifically, for a 0.2 sec response, the
bivariate (M
2D,R
2D) differs from the Mand Rmodal values
of the 3D PMF at approximately 25% of the points. For this
period, the largest differences between Mand M
2Dconcen-
trate in southern Italy near l’Aquila, where M
2Dis up to 1
unit greater than M, in the southern Tyrrhenian Sea, where
M
2Dis about 0.5–1 units lower than M, along the Calabrian
arc (i.e., zones 927 and 929), where M
2Dcan be either lower
(zone 927) or higher (zone 929) than M, and near the Etna
volcano (zone 935) in Sicily. Differences between Rand
R
2Ddo not concentrate in specific areas. Note that about
80% of sites where R
2D=Rdiffers from unity show values
of Rgreater than R
2D. For a 1.0 sec response, modal M–R
pairs of the 2D and 3D PMFs differ in about 32% of the cases.
Again, the main differences in modal magnitudes are concen-
trated in central and southern Italy where M
2Dis generally
(0.5–1.0 magnitude units) lower than or equal to M.Ris
greater than R
2Dat 86% of the sites, most of which are
located in areas characterized by a medium-to-high seismic
hazard level (0:125g<S
a1:0sec<0:25g). At most of
these points, R
2Dis about half the value of R.
Effect of Alternative Attenuation Relationships
As specified in the section titled A Note on Models and
Input Parameters, the ground-motion predictive equation of
Ambraseys et al. (1996), which contributes the most to the
reference median hazard, was employed to derive reference
disaggregation maps. However, other attenuation relation-
ships based on national (Sabetta and Pugliese, 1996) and re-
gional (De Natale et al., 1988;Patanè et al., 1994,1997;
Malagnini et al., 2000,2002;Morasca et al., 2006) ground-
motion data were used to assess the Italian seismic hazard. In
this study, we evaluate the influence of an alternative attenu-
ation equation on disaggregation results. Specifically, we
compare the mean and modal values of M,R, and εcalcu-
lated by applying the Sabetta and Pugliese (1996) equation
with the reference values mapped in Figures 9,10, and 11 for
Sa0:2secand Sa1:0sec. More specifically, the compar-
ison is based on the ratio of the mean and modal values of M,
R, and εobtained from Sabetta and Pugliese (1996) (denoted
as SP96) to those from Ambraseys et al. (1996) (see Fig. 13).
The mean values of Mand Robtained by using the two pre-
dictive equations are similar at most of the gridded points
while the values of
εSP96 and
εdiffer substantially in some
regions. In the Sa1:0seccase, for example,
MSP96 and
M
assume nearly the same value almost everywhere in central
and southern Italy and never differ by more than 0.4 magni-
tude units. The largest differences are concentrated offshore
and in low-seismicity areas in northern Italy. At these sites,
using the equation by Sabetta and Pugliese (1996) provides
magnitude values that are slightly greater than the reference
ones. For the same spectral period,
RSP96 and
Rdiffer by
more than 10 km in only 1.2% of cases. Except for very
few sites south of Palermo and off the Sicilian coast, the
Sabetta and Pugliese (1996) equation provides mean distance
values that are lower than those calculated with the Ambra-
seys et al. (1996) relationship. As mentioned previously,
unlike the mean values of Mand R,
εSP96 and
εdiffer from
2Note that, for these two spectral periods, the maps show values of
εand
εlower than 2:5. These values, which should be disregarded, correspond
to sites where the spectral acceleration values provided by the seismic hazard
maps were erroneously set to zero.
2656 S. Barani, D. Spallarossa, and P. Bazzurro
one another at most of the gridded points. The
εSP96=
εratio is
always lower than unity and reaches the lowest values in
areas characterized by high-seismic hazard level, such as
the Calabrian arc (source zones 927 and 929) and the area
of the Etna volcano in southern Sicily (zone 935), where dif-
ferences between
εSP96 and
εare up to about 0.4.
Maps for modal values are not presented because they are
highly grained and, therefore, do not allow any meaningful
generalizations about any possible relation with the seismicity
distribution. It is worth noting, however, that also modal
values of εstrongly depend on the attenuation equation used
while those of Mand Rexhibit only minor variations that
are generally smaller than the bin amplitude considered in
the calculations.
Summary and Conclusions
The article has presented the results of the disaggrega-
tion of the Italian ground-motion hazard maps for rock con-
ditions developed by Gruppo di Lavoro MPS (2004),Meletti
and Montaldo (2007), and Montaldo and Meletti (2007).
These maps have been officially adopted as reference maps
by the Italian government (Ministero delle Infrastrutture e dei
Trasporti, 2008). We identified the contributions of earth-
quake scenarios to the mean rates of exceeding PGA and 0.2,
1.0, and 2.0 sec spectral acceleration (SaT) values asso-
ciated with MRPs of 475 and 2475 yr at a grid of locations
covering the entire Italian territory. The maps showing
the geographical distribution of the mean (
M,
R, and
ε) and
modal (M,R, and ε) values of the 3D joint PMFs of
M–R–εhave been presented and compared with those
obtained from a 2D disaggregation. It is worth noting, how-
ever, that these maps show the mean and modal values of
M–R–εonly. The complete joint M–R–εdistributions in
some cases can have multiple peaks, some of which contrib-
ute about the same amount as the mode to the hazard. There-
fore, it is suggested that joint 3D PMFs be evaluated for
site-specific analyses because they are more informative than
their central statistics.
As observed by other researchers (e.g., Harmsen et al.,
1999;Peláez Montilla et al., 2002), disaggregation results
depend strictly on the seismicity distribution—that is, on the
proximity to sources where the seismicity is mainly concen-
trated. Distant, larger magnitude events dominate the hazard
in areas characterized by a low-seismic activity. Conversely,
nearby seismicity is often the major contributor to the hazard
at sites located in high-seismicity regions or in zones where
the seismic activity is characterized by weak-to-moderate but
quite frequent events. At these sites, joint M–R–εdistribu-
tions tend to be unimodal and values of
εand εare generally
smaller than those observed in low-seismicity areas. More-
over, the larger the MRP (or, conversely, the smaller the
Figure 12. Maps showing the ratio of the 2D to 3D modal values of magnitude (top panels) and distance (bottom panels).
Disaggregation of Probabilistic Ground-Motion Hazard in Italy 2657
MRE), the greater the contribution from closer, higher mag-
nitude events. It was also observed that as the MRP increases,
both the mean and modal values of εtend to slightly increase.
An analysis of the trend of the disaggregation results with
spectral period (T) shows that the mean rate of exceeding
long-period spectral accelerations is generally controlled
by earthquakes that are larger in size and farther from the
site than those dominating the PGA and short-period spectral
acceleration hazard.
As discussed in many articles (e.g., Bommer et al.,
2005;Sabetta et al., 2005;Scherbaum et al., 2005), ground-
motion predictive equations have a large impact on final
hazard estimates. Hence, in this study we have also analyzed
the sensitivity of disaggregation results to the choice of the
attenuation relationship. Although limited in scope, our re-
sults revealed that applying different attenuation relation-
ships induces changes in the disaggregation results that, at
least in this case, are not very significant (perhaps with
the exception of the values of ε). We found that differences
in the mean and modal values of Mand Rare generally smal-
ler than the bin amplitude considered in the calculations
(Δm0:5,Δr10:0km). Larger differences, often twice
as large as the bin amplitude (Δε 0:2), were observed in
the mean and modal values of εthat, therefore, depend
strongly on the attenuation equation used.
To summarize, the M–R–εdisaggregation maps pre-
sented here can be helpful to both private and public stake-
holders for many applications. For example, they can be used
Figure 13. Sensitivity maps showing the ratio of mean M,R, and εvalues resulting from the disaggregation of the 475 yr Sa0:2sec
and Sa1:0secvalues obtained using the Sabetta and Pugliese (1996) (indicated by SP96) and the Ambraseys et al. (1996) ground-motion
predictive equations.
2658 S. Barani, D. Spallarossa, and P. Bazzurro
by structural engineers to select appropriate ground-motion
records for testing the adequacy of the design of new struc-
tures or the response of existing ones. Geotechnical engi-
neers can also select ground-motion records consistent with
the Mand Rvalues shown here for site amplification, lique-
faction, and slope stability studies. Local governments can
use such maps, for instance, to select scenarios for assessing
their preparedness to conduct postevent emergency re-
sponses (e.g., the Shake Out exercise conducted in Southern
California on November 2008, see the Data and Resources
section for more information).
To conclude, it is worth mentioning again that the seis-
mic hazard maps that we used for our disaggregation exer-
cise do not account for the contribution of either faults or
subduction zones. The recently released version of the Data-
base of Individual Seismogenic Sources (DISS)(Basili et al.,
2008), which collects data (e.g., fault length, fault width,
slip rate, and maximum magnitude) about known active
faults in Italy, was not adopted in the development of the
hazard maps considered here. This is because of the large
uncertainty affecting such data (e.g., for most faults in DISS,
the slip rates vary from 0.1 to 1:0mm=yr) and the missing
information about which historical events were caused by
which fault (DISS indicates only the largest historical event
for each fault). Studying the effects of faults on the dominant
earthquake scenarios is beyond the scope of this article. In-
terested readers are referred to the work of Barani and Spal-
larossa (2008) that incorporates faults and area sources into
an updated seismic source model for PSHA in Italy. That
preliminary study revealed that at sites near seismogenic
faults (up to a distance of about 30 km) the contribution
from area sources is low and the hazard is almost exclu-
sively controlled by fault sources. Thus, at these sites, incor-
porating faults into the earthquake source model provides
(mean and modal) values of Mand Rthat are, respectively,
higher and lower than those presented here. As the source-
to-site distance increases, the fault contributions tend to de-
crease rapidly and the area sources dominate the hazard.
These are only preliminary considerations and further re-
search is needed before fault data can be confidently used
in future PSHA studies for Italy. Research efforts should
focus on analyzing the sensitivity of the hazard to different
sets of fault characteristic parameter values (e.g., fault
length, fault width, slip rate, and maximum magnitude) and
alternative magnitude distributions (e.g., exponential and
characteristic).
Data and Resources
The CPTI04 catalog can be downloaded at the following
Web address: http://emidius.mi.ingv.it/CPTI04/ (last ac-
cessed June 2009). A file providing the geographical coor-
dinates of the ZS9 seismogenetic zonation can be
downloaded at the following Web address: http://
zonesismiche.mi.ingv.it/elaborazioni/ (last accessed June
2009). For information on the DPC–INGV S1 project, see
the Web site http://esse1.mi.ingv.it (last accessed June
2009). Results of the disaggregation of the PGA harzard
for nine MRPs can be consulted on the interactive Web site
http://esse1-gis.mi.ingv.it/ (last accessed June 2009). Infor-
mation on the Shake Out exercise is available at http://
www.shakeout.org (last accessed June 2009).
Acknowledgments
We thank an anonymous reviewer and the associate editor, M. C.
Chapman, for their thorough review and helpful comments. We wish to ac-
knowledge the use of the Generic Mapping Tools software package by Wes-
sel and Smith (1991) to produce most of the figures in this article. This study
was supported and funded by the DPC-INGV S1 Project (http://esse1.mi
.ingv.it).
References
Abrahamson, N. A. (2006). Seismic hazard assessment: Problem with
current practice and future development, in Proc. of the 1st European
Conf. on Earthquake Engineering and Seismology (ECEES),
Geneva, Switzerland, 3–8 September 2006, Paper Number: Keynote
Address K2.
Abrahamson, N. A., and J. J. Bommer (2005). Probability and uncertainty in
seismic hazard analysis, Earthq. Spectra 21, 603–607.
Albarello, D., V. Bosi, F. Bramerini, A. Lucantoni, G. Naso, L. Peruzza,
A. Rebez, F. Sabetta, and D. Slejko (2000). Pericolosità sismica del
territorio nazionale: Carte di consenso GNDT e SSN, in Le Ricerche
del GNDT nel Campo della Pericolosità Sismica (1996–1999),
F. Galadini, C. Meletti, and A. Rebez (Editors), http://gndt.ingv.it/
Pubblicazioni/Meletti/3_02_Albarello.pdf (last accessed June 2009).
Ambraseys, N. N., K. A. Simpson, and J. J. Bommer (1996). Prediction of
horizontal response spectra in Europe, Earthq. Eng. Struct. Dyn. 25,
371–400.
Barani, S., and D. Spallarossa (2008). On the use of fault sources for prob-
abilistic seismic hazard analysis in Italy, Presented at the 31st General
Assembly of the European Seismological Commission (ESC08),
Hersonissos, Greece, 7–12 September 2008, http://www.dipteris
.unige.it/persone/BaraniS/Barani&Spallarossa_ESC08.pdf (last ac-
cessed June 2009).
Barani, S., R. De Ferrari, G. Ferretti, and C. Eva (2008). Assessing the
effectiveness of soil parameters for ground response characterization
and soil classification, Earthq. Spectra 24, 565–597.
Barani, S., D. Spallarossa, P. Bazzurro, and C. Eva (2007). Sensitivity
analysis of seismic hazard for western Liguria (northwestern Italy):
A first attempt towards the understanding and quantification of hazard
uncertainty, Tectonophysics 435, 13–35.
Basili, R., G. Valensise, P. Vannoli, P. Burrato, U. Fracassi, S. Mariano,
M. M. Tiberti, and E. Boschi (2008). The Database of Individual Seis-
mogenic Sources (DISS), version 3: Summarizing 20 years of research
on Italy’s earthquake geology, Tectonophysics 453, 20–43.
Bazzurro, P., and C. A. Cornell (1999). Disaggregation of seismic hazard,
Bull. Seismol. Soc. Am. 89, 501–520.
Bernreuter, D. L. (1992). Determining the controlling earthquake from prob-
abilistic hazards for the proposed Appendix B, Lawrence Livermore
National Laboratory Report UCRL-JC-111964, Livermore, California.
Bommer, J. J., and N. A. Abrahamson (2006). Why do modern probabilistic
seismic-hazard analyses often lead to increased hazard estimates?,
Bull. Seismol. Soc. Am. 96, 1967–1977.
Bommer, J. J., J. Douglas, and F. O. Strasser (2003). Style-of-faulting in
ground-motion prediction equations, Bull. Earthq. Eng. 1, 171–203.
Disaggregation of Probabilistic Ground-Motion Hazard in Italy 2659
Bommer, J. J., F. Scherbaum, H. Bungum, F. Cotton, F. Sabetta, and
N. A. Abrahamson (2005). On the use of logic trees for ground-motion
prediction equations in seismic hazard analysis, Bull. Seismol. Soc.
Am. 95, 377–389.
Cao, T. (2007). Deaggregation of seismic hazard extended to disaggregation
of annualized loss ratio, Bull. Seismol. Soc. Am. 97, 305–317.
Cao, T., M. D. Petersen, and A. D. Frankel (2005). Model uncertainties of
the 2002 update of California seismic hazard maps, Bull. Seismol. Soc.
Am. 6, 2040–2057.
Chapman, M. C. (1995). A probabilistic approach for ground-motion
selection for engineering design, Bull. Seismol. Soc. Am. 85,
937–942.
Coppersmith, K. J., and R. R. Youngs (1986). Capturing uncertainty in prob-
abilistic seismic hazard assessment with intraplate tectonic environ-
ments, in Proc. of the 3rd U.S. National Conf. on Earthquake
Engineering, Charleston, South Carolina, 1,301–312.
Cornell, C. A. (1968). Engineering seismic risk analysis, Bull. Seismol. Soc.
Am. 58, 1583–1606.
Cramer, C. H., and M. D. Petersen (1996). Predominant seismic source dis-
tance and magnitude maps for Los Angeles, Orange, and Ventura
counties, California, Bull. Seismol. Soc. Am. 86, 1645–1649.
Cramer, C. H., M. D. Petersen, and M. S. Reichle (1996). A Monte Carlo
approach in estimating uncertainty for a seismic hazard assessment of
Los Angeles, Ventura, and Orange counties, California, Bull. Seismol.
Soc. Am. 86, 1681–1691.
De Natale, G., E. Faccioli, and A. Zollo (1988). Scaling of peak round
motion from digital recordings of small earthquakes at Campi Flegrei,
southern Italy, Pure Appl. Geophys. 128, 37–53.
Grünthal, G., and R. Wahlström (2001). Sensitivity of parameters for prob-
abilistic seismic hazard analysis using a logic tree approach, J. Earthq.
Eng. 5, 309–328.
Gruppo di Lavoro CPTI (2004). Catalogo Parametrico dei Terremoti Italiani,
versione 2004 (CPTI04), INGV, Bologna, http://emidius.mi.ingv.it/
CPTI/ (last accessed June 2009).
Gruppo di Lavoro MPS (2004). Redazione della mappa di pericolosità sis-
mica prevista dall’Ordinanza PCM 3274 del 20 marzo 2003, Rapporto
conclusivo per il dipartimento di Protezione Civile, INGV, Milano–
Roma, Aprile 2004, 65 pp plus 5 appendici, http://zonesismiche.mi
.ingv.it/documenti/rapporto_conclusivo.pdf (last accessed June 2009).
Gutenberg, B., and C. R. Richter (1944). Frequency of earthquakes in
California, Bull. Seismol. Soc. Am. 34, 185–188.
Halchuk, S., and J. Adams (2004). Deaggregation of seismic hazard for
selected Canadian cities, in Proc. of the 13th World Conf. on Earth-
quake Engineering, Vancouver, Canada, 1–6 August, Paper Num-
ber 2470.
Harmsen, S. (2001). Mean and modal εin the deaggregation of probabi-
listic ground motion, Bull. Seismol. Soc. Am. 91, 1537–1552.
Harmsen, S., and A. Frankel (2001). Geographic deaggregation of seismic
hazard in the United States, Bull. Seismol. Soc. Am. 91, 13–26.
Harmsen, S., D. Perkins, and A. Frankel (1999). Deaggregation of probabi-
listic ground motions in the central and eastern United States, Bull.
Seismol. Soc. Am. 89, 1–13.
Hong, H. P., and K. Goda (2006). A comparison of seismic-hazard and risk
deaggregation, Bull. Seismol. Soc. Am. 96, 2021–2039.
Joyner, W. B., and D. M. Boore (1981). Peak horizontal acceleration and
velocity from strong-motion records including records from the
1979 Imperial Valley, California, earthquake, Bull. Seismol. Soc.
Am. 71, 2011–2038.
Kulkarni, R. B., R. R. Youngs, and K. J. Coppersmith (1984). Assessment of
confidence intervals for results of seismic hazard analysis, in Proc. of
the 8th World Conf. on Earthquake Engineering, San Francisco 1,
263–270.
Malagnini, L., A. Akinci, R. B. Herrmann, N. A. Pino, and L. Scognamiglio
(2002). Characteristics of the ground motion in northeastern Italy, Bull.
Seismol. Soc. Am. 92, 2186–2204.
Malagnini, L., R. B. Herrmann, and M. Di Bona (2000). Ground motion
scaling in the Apennines, Bull. Seismol. Soc. Am. 90, 1062–1081.
McGuire, R. K. (1995). Probabilistic seismic hazard analysis and design
earthquakes: Closing the loop, Bull. Seismol. Soc. Am. 85, 1275–1284.
McGuire, R. K. (2004). Seismic Hazard and Risk Analysis, EERI Mono-
graph MNO-10, Earthquake Engineering Research Institute, Oakland,
California.
McGuire, R. K., and K. M. Shedlock (1981). Statistical uncertainties in seis-
mic hazard evaluations in the United States, Bull. Seism. Soc. Am. 71,
1287–1308.
McGuire, R. K., C. A. Cornell, and G. R. Toro (2005). The case of using
mean seismic hazard, Earthq. Spectra 21, 879–886.
Meletti, C., and V. Montaldo (2007). Stime di pericolosità sismica per di-
verse probabilità di superamento in 50 anni: valori di ag, Progetto
DPC-INGV S1, http://esse1.mi.ingv.it/d2.html/ (last accessed June
2009).
Meletti, C., F. Galadini, G. Valensise, M. Stucchi, R. Basili, S. Barba,
G. Vannucci, and E. Boschi (2008). A seismic source zone model
for the seismic hazard assessment of the Italian territory, Tectonophy-
sics 450, 85–108.
Meletti, C., F. Meroni, F. Martinelli, M. Locati, A. Cassera, and M. Stucchi
(2007). Ampliamento del sito web per la disseminazione dei dati del
progetto S1, Progetto DPC-INGV S1, http://esse1.mi.ingv.it/d8.html/
(last accessed June 2009).
Ministero delle Infrastrutture e dei Trasporti (2008). Norme tecniche per
le costruzioni, D.M. 14 Gennaio 2008, Supplemento ordinario
alla Gazzetta Ufficiale No. 29, 4 Febbraio 2008, http://www
.infrastrutture.gov.it/consuplp/ (last accessed June 2009).
Montaldo, V., and C. Meletti (2007). Valutazione di ordinata spettrale a 1 sec
e ad altri periodi d’interesse ingegneristico, Progetto DPC-INGV S1,
http://esse1.mi.ingv.it/d3.html/ (last accessed June 2009).
Morasca, P., L. Malagnini, A. Akinci, D. Spallarossa, and R. B. Herrmann
(2006). Ground-motion scaling in the western Alps, J. Seism. 10,
315–333.
Musson, R. M. W. (2005). Against fractiles, Earthq. Spectra 21, 887–891.
National Research Council (NRC) (1988). Probabilistic seismic hazard
analysis, in Report on the Panel on Seismic Hazard Analysis, National
Academy Press, Washington D.C.
Patenè, D., F. Ferrucci, and S. Gresta (1994). Spectral features of microearth-
quakes in volcanic areas: Attenuation in the crust and amplitude
response of the site at Mt. Etna (Italy), Bull. Seismol. Soc. Am. 84,
1842–1860.
Patanè, D., F. Ferrucci, E. Giampiccolo, and L. Scaramuzzino (1997).
Source scaling of microearthquakes at Mt. Etna volcano and in the
Calabrian arc (southern Italy), Geophys. Res. Lett. 24, 1879–1882.
Peláez Montilla, J. A., C. L. Casado, and J. H. Romero (2002). Deaggrega-
tion in magnitude, distance, and azimuth in the south and west of the
Iberian peninsula, Bull. Seismol. Soc. Am. 92, 2177–2185.
Power, M. S., K. J. Coppersmith, R. R. Youngs, D. P. Schartz, and
R. H. Swan (1981). Seismic exposure analysis for the WNP-2
and WNP-1/4 site, Appendix 2.5K to Amendment No. 18 Final
Safety Analysis Report for WNP-2, Woodward-Clyde Consultants,
San Francisco, 63 pp.
Rabinowitz, N., and D. M. Steinberg (1991). Seismic hazard sensitivity
analysis: A multi-parameter approach, Bull. Seismol. Soc. Am. 81,
796–817.
Romeo, R., and A. Pugliese (1998). A global earthquake hazard assessment
of Italy, in Proc. of the 11th Conf. on Earthquake Engineering,6–11
September 1998, CNIT, Paris la Défense, France.
Romeo, R., and A. Pugliese (2000). Seismicity, seismotectonics, and seismic
hazard of Italy, Eng. Geol. 55, 241–266.
Sabetta, F., and A. Pugliese (1996). Estimation of response spectra and si-
mulation of nonstationary earthquake ground motions, Bull. Seismol.
Soc. Am. 2, 337–352.
Sabetta, F., A. Lucantoni, H. Bungum, and J. J. Bommer (2005). Sensitivity
of PSHA results to ground-motion prediction relations and logic-tree
weights, Soil Dyn. Earthq. Eng. 24, 317–329.
Scherbaum, F., J. J. Bommer, H. Bungum, F. Cotton, and N. A. Abrahamson
(2005). Composite ground-motion models and logic trees:
2660 S. Barani, D. Spallarossa, and P. Bazzurro
Methodology, sensitivities and uncertainties, Bull. Seismol. Soc. Am.
95, 1575–1593.
Senior Seismic Hazards Analysis Committee (SSHAC) (1997). Recommen-
dations for probabilistic seismic hazard analysis: Guidance on uncer-
tainty and the use of experts, NUREG/CR6372, Nuclear Regulatory
Commission, Washington, D.C., 256 pp.
Slejko, D., L. Peruzza, and A. Rebez (1998). Seismic hazard maps of Italy,
Ann. Geofis. 41, 183–214.
Sokolov, V. Y. (2000). Hazard-consistent ground motions: Generation on the
basis of uniform hazard Fourier spectra, Bull. Seismol. Soc. Am. 90,
1010–1027.
Spallarossa, D., and S. Barani (2007). Disaggregazione della pericolosità
sismica in termini di M–R–ε, Progetto DPC-INGV S1, http://esse1
.mi.ingv.it/d14.html/ (last accessed June 2009).
Wessel, P., and W. H. F. Smith (1991). Free software helps map and display
data, EOS Trans. AGU 72, 441.
Dipartimento per lo Studio del Territorio e delle sue Risorse
University of Genoa
Viale Benedetto XV
5-16132 Genova, Italy
barani@dipteris.unige.it
daniele@dipteris.unige.it
(S.B., D.S.)
AIR Worldwide Company
San Francisco, California
pbazzurro@air‑worldwide.com
(P.B.)
Manuscript received 1 December 2008
Disaggregation of Probabilistic Ground-Motion Hazard in Italy 2661
