Overview on current seismic hazard disaggregation studies

for Western Iberia and consequences for ground shaking

J.F.B.D. Fonseca & S.P. Vilanova

SUMMARY:

Probabilistic seismic hazard (PSHA) disaggregation is a tool that provides useful insights into the main sources
contributing to the hazard at a specific site. The importance of the technique in assisting the selection of ground
motion scenarios for seismic design and earthquake emergency planning has been established for more than a
decade. Recently, a new exercise on the PSHA disaggregation for Portugal was published. In contrast to previous
analysis, the study concludes that offshore seismic sources, located around 70km WSW of Cape St. Vincent,
dominate the contributions to seismic hazard for southern and central Portugal sites, including Lisbon, for both
10% and 5% exceedence probability in 50 years. We challenge those results by attributing the obtained hazard
pattern to biases in the input parameters used for the underlying PSHA assessment. In our opinion, PSHA
desegregation analysis is only useful if supported by a robust and reliable PSHA assessment, following the best
international practices regarding seismogenic sources, catalogue preparation, recurrence analysis, ground motion
prediction models, and epistemic uncertainty.

Keywords: seismic hazard, disaggregation, ground shaking, SW Iberia

1. PROBABILISTIC SEISMIC HAZARD ASSESSMENT AND ITS DISAGGREGATION

1.1. Probabilistic Seismic Hazard Assessment

Probabilistic seismic hazard assessment (PSHA) is a time-honoured technique whereby the probability
of exceedence of a specified level of ground motion over a certain period of exposure is estimated for
a site, from the available information on the regional seismic sources, the characteristics of the
attenuation of seismic waves, and local effects at the site (Cornell, 1968). Often, a probability of
exceedence deemed socially acceptable is specified by a building code – typically 10%, 5% or 2% in a
period of 50 years, depending on the type of structure – and for a particular site the PSHA must
estimate the level of shaking associated with that probability. While in more traditional studies the
ground motion was described via a macroseismic intensity scale, it became gradually more common to
adopt peak ground acceleration (PGA), and ultimately spectral acceleration (SA) at specified periods,
among other measures. The variable used is dictated by the ground motion prediction equation
(GMPE) adopted for the hazard calculations.

The choice of GMPE is of paramount importance for the final PSHA result. It should be derived from
instrumental strong motion data pertinent to the region under study, and used only within the range of
distances of that data. However, this rule cannot always be followed, particularly in continental areas
where strong motion data are very scarce. As the next best choice, limited data from the area of study
may be used to test the applicability of GMPE’s derived for regions of similar tectonic setting.
Alongside with the choice of GMPE, the characterization of seismic recurrence for the sources that are
relevant to the site is also of key importance. Frequently, the seismic activity rate – the a value of the
Gutenberg-Richter relation – can be well constrained from recent instrumental data, because it pertains
to the minimum magnitude still considered relevant for the hazard. For example, a may correspond to
the number of earthquakes in one year that have magnitude equal to or above M4, clearly dominated
by seismicity in a range of magnitudes for which the instrumental catalogues are easily complete in the
last decades. Moderate to high magnitude seismicity, on the other hand, is hard to estimate from
seismicity catalogues in intraplate areas, because the catalogue completeness periods tend to fail short
of the relevant return periods for PSHA. For a Poissonian model of seismic occurrence, the
probabilities given above as examples correspond to 475, 975 and 2475 years, respectively. The rate of
occurrence of moderate to large earthquakes is therefore inferred from low magnitude rates, through
the b value adopted for the Gutenberg-Richter relation. Such inference is extremely sensitive to errors
in b, given the logarithmic nature of the relation. If an error-free rate of occurrence of M4 earthquakes
is used to extrapolate for the M8 rate, an error of 0.25 in the b value leads to a relative error of 100%
in the M8 rate. Fortunately, it is widely confirmed by observation that b values tend to be in the
vicinity of 1, and therefore large errors are unlikely. Because large sources of error may be present in
any PSHA study, in particular in plate interiors, a great effort is put to quantify the uncertainties of the
final result, both aleatory and epistemic.

1.2. Disaggregation

The Cornell (1968) approach to PSHA is based on an area summation, covering all the relevant
sources. Restricting the summation to a given sub-area allows the estimation of the contribution of that
sub-area to the total hazard. This notion is the basis of the technique known as hazard disagreggation
(Bazurro and Cornell, 1999), which allows the identification of the seismic sources making dominant
contributions to the hazard at a particular site. This information can in turn be used to define the
ground shaking scenarios that are relevant for earthquake risk mitigation at the site. Gaining
information on the location of the relevant sources may also focus the research onto other
characteristics such as rupture mechanism and near-site effects (Bazurro and Cornell, 1999).

On the same token, it can be concluded that inadequate disaggregation of seismic hazard may be a
distracting factor, allocating the research effort to sources that are not dominant and underestimating
others that deserve attention. We contend that seismic hazard disaggregation proposed recently for SW
Iberia is an example of this type of shortcoming, and discuss the subjective factors that may be at the
origin of such bias. Finally, we discuss the implications for the hazard zonation adopted in the EC8
National Annex for Portugal.

2. ATTEMPTS OF DISAGGREGATION OF SEISMIC HAZARD IN SW IBERIA

To our knowledge, two attempts were conducted so far to disaggregate the seismic hazard in SW
Iberia. The earlier study (Montilla et al., 2002) adopted macroseismic intensity as the measure of
ground motion, and fitted empirical relations to the available intensity data to derive the GMPE’s.
More recently, Sousa and Costa (2009) use the same ground motion variable, and adopt the same
procedure to derive the GMPE’s. Surprisingly, the outcomes of the disaggregations are radically
different. We will take the all-important result for Lisbon to illustrate the differences.

Montilla et al. (2002) conclude that, at a return period of 475 years, 52% of the hazard in Lisbon is due
to sources located to the north of the city, within distances of 50 km. Sources within 60 km to the
south of Lisbon account for another 35%, according to the same authors. Only the remaining 13% of
the hazard are attributed to sources beyond 60 km.

Sousa and Costa (2009), on the other hand, conclude that the dominant rupture scenario for Lisbon at
the same return period is located offshore, 70 km WSW of Cape St. Vincent. In the proposed
disaggregation, the modal 10 km by 10 km bin contributes with 52.9o/oo of the hazard in Lisbon,
whereas the onshore bins that comes closer contributes with 14.3o/oo. This can be regarded as a mirror
image to the conclusions of Montilla et al. (2002). The authors conclude that the modal scenario for
475 (and 975) years of return period is “dominated by the effect of the large magnitude Lisbon 1755
earthquake”.

In view of the disparate results of the two analyses, we are inclined to conclude that the available
information does not warrant the disaggregation of the hazard in a robust and reliable way (Fonseca
and Vilanova, 2011). Already for that reason, our own results for PGA-based PSHA (Vilanova and
Fonseca, 2007), shown in Fig. 1, were not subject to disaggregation. That said, the pattern shown
around Lisbon for a return period of 475 years – with high hazard “wavelengths” of the order of 50km
to 100 km in the Lisbon region, depending on orientation – points to a dominant contribution of local
sources, as opposed to dominant distant offshore sources that would result in a smoother pattern. The
results of Vilanova and Fonseca (2007) are therefore more in line with the conclusions of Montilla et
al. (2002). A similar conclusion may be derived from the analysis of other hazard maps such as the
ESC-SESAME Euro-Mediterranean hazard map (e.g., Solomos et al, 2006, their Figures 6.2 and 6.3).

Blank line 10 pt

Figure 1. Mean peak ground acceleration in rock with a probability of 10% of being exceeded in 50 years, in
units of g. Also shown are the 15th and the 85th percentiles. Increment between different shades of gray is 0.01g.
After Vilanova and Fonseca (2007).
3. DISCUSSION

The conflicting results presented above highlight the difficulty inherent to the disaggregation of
seismic hazard when the starting point is an analysis already affected by large uncertainties. Fonseca
and Vilanova (2011) identified and discussed in detail several causes of bias in Sousa and Costa
(2009). In particular, we showed that the proposed source 70 Km WSW of Cape St. Vincent is an
artefact due to strong contrast in GMPEs and b values across the converging borders of three source
zones. We refer the reader to Fonseca and Vilanova (2011) for a detailed discussion of the inadequacy
of GMPE's (intensity differences of up to 2.5 for the same distance and magnitude, depending on
source zone) and b values (ranging from 5.9 to 1.22, lowest value associated with the proposed source)
used in Sousa and Costa (2009). Here we will highlight two aspects: 1) the danger that a large
historical disaster such as the Lisbon earthquake may subjectively “force its way” into hazard
assessment; 2) the effect of this “traditional”view on the zonation associated with the EC8 National
Annex.

3.1. The “invisible” role of the 1755 earthquake.

Whereas PSHA is desired objective, despite more than 250 years elapsed since its occurrence the 1755
earthquake still carries a strong weight in Portuguese cultural references. Somehow this may have
influenced also the seismic hazard assessment. As an example of this, we examine two aspects of the
input data used in Sousa and Costa (2009). Arguing in favour of the ad hoc empirical GMPE's that
would be adopted in the paper, Sousa (2006) states that no relation published in the literature could
predict the high intensities of the 1755 earthquake, while the adopted relations had the advantage of
being derived from (mainly) those high intensities. It is therefore not surprising that the resulting
analysis should point to a remarkable contribution from very distant (~200 km) sources, since the
GMPE's were designed so that they would reproduce that unusual effect. However, alternative
explanations have been put forward for the extreme intensities reported for Lisbon in 1755, including a
secondary rupture in the vicinity of the city (Vilanova et al., 2003).

(A.1) (A.2) (B)

Figure 2. Seismic zonation for a distant scenario (A1) and for a nearby scenario (A2), in the EC8 National
Annex. Colours correspond to different PGA values, and at each site the highest of the two values is adopted. For
Lisbon, PGA is 1.5 ms-2 in the distant scenario and 1.7 ms-2 in the nearby scenario. However, in combination with
the different shapes of elastic response spectrum to be used with each scenario, for all sites to the south of, and
including, Lisbon the distant scenario (B, thick red line) dominates with respect to the nearby scenario (B, thin
red line) except at very low periods of marginal engineering interest. This effect is stronger than with LNEC
(1983), the building code previously in force (B, blue lines, same convention). After Carvalho (2011).
Also “boosting” the effect of the 1755 earthquake scenario in the hazard assessment of Sousa and
Costa (2009) is the low b value of 0.59 adopted for the area encompassing its offshore epicentral
region (Vilanova and Fonseca (2007) found 0.94 and 1.09, with a more robust analysis).

3.2. The dual zonation of the Eurocode 8 National Annex

Contrary to general practice, the Portuguese National Annex of the Eurocode 8 adopts two zonations,
one for the near-source scenario and the other for a distant-source scenario. This is in line with the
approach taken in the previous building code (LNEC, 1983). It can be seen in Fig. 2A.1 that the distant
scenario zonation is strongly dominated by the 1755-type earthquake scenario, as should be expected.
At each site, the higher value given by the two zonations is adopted. However, taken in conjunction
with the regulatory elastic response spectra, the nearby scenario for Lisbon and all sites to its south
exceeds the distant scenario only at periods below 0.3 s, of marginal engineering relevance (Fig. 2B).

4. CONCLUSION

The identification of rupture scenarios through hazard disaggregation as proposed by Sousa and
Campos (2009) for SW Iberia should only be performed starting with robust hazard results. Otherwise,
it may lead to biased ranking of seismic sources, as we contend was the case in the cited study.
This bias may have contaminated the seismic zonation adopted in the National Annex of
Eurocode 8, and may distract research from relevant seismogenic sources at short distances. The EC8
zonations should therefore be revised taking into account the best practices recommended at
international level. In particular, the recommendations of FP7 project SHARE (Seismic Hazard
Harmonization for Europe) should be taken on board for the revision of the National Annex.

REFERENCES

Bazzurro P. and Cornell CA (1999). Disaggregation of seismic hazard, Bull. Seismol. Soc. Am., 89, 501–520

Carvalho, E. C. (2011). Aplicação do EUROCÓDIGO 8 à concepção e projecto de edifícios – Aspectos gerais e

Cornell C.A. (1968). Engineering Seismic Risk Analysis, Bull. Seismol. Soc. Am., 58, 1583-1606.

Fonseca, J.F.B.D. and Vilanova, S.P. (2011). Comment on Sousa, M. L. and Costa, A. C., “Ground motion
scenarios consistent with probabilistic seismic hazard disaggregation analysis. Application to Mainland
Portugal”, Bull. Earthquake Eng., 9, 1289-1295

LNEC (1983). Regulamento de Segurança e Acções para Estruturas de Edifícios e Pontes, vol. Decreto-Lei N.º
235/83, Imprensa Nacional Casa da Moeda, Lisbon, 114 pp. (in Portuguese)

Montilla, J.A., Casado, C,L and Romero, J.H. (2002). Deaggregation in magnitude, distance, and azimuth in the
South and West of the Iberian Peninsula. Bull. Seismol. Soc. Am., 6, 2177–2185

Solomos, G., Pinto, A. and Dimova, S. (2008). A review of the seismic hazard zonations in national building
codes in the context of Eurocode 8, JRC Technical and Scientific Report EUR23563EN, European Commission.

Sousa, M. L. (2006). Risco Sismico em Portugal, Ph.D. Thesis, Instituto Superior Tecnico, Lisbon.

Sousa, M.L. and Costa, A.C. (2009). Ground motion scenarios consistent with probabilistic seismic hazard
disaggregation analysis. Application to Mainland Portugal, Bull. Earthquake Eng., 7, 127-147.

Vilanova, S.P., Nunes, A.C. and Fonseca, J.F.B.D. (2003). Lisbon 1755: a case of triggered onshore rupture?,
Bull. Seismol. Soc. Am., 93, 2056–2068.

Vilanova, S.P. and Fonseca, J.F.B.D. (2007). Probabilistic seismic hazard assessment for Portugal, Bull. Seismol.
Soc. Am., 97, 1702-1717