Simulations of strong ground motion in SW Iberia for the 1969 February 28 (M_s = 8.0) and the 1755 November 1 (M∼ 8.5) earthquakes - II. Strong ground motion simulations

Summary

This is the second paper of a series of two concerning strong ground motion in SW Iberia due to earthquakes originating from the adjacent Atlantic area. The aim of this paper is to use the velocity model that was proposed and validated in the companion paper for seismic intensity modelling of the 1969 (M_s = 8.0) and 1755 (M = 8.5–8.7) earthquakes.

First, we propose a regression to convert simulated values of Peak Ground Velocity (PGV) into Modified Mercalli Intensity (MMI) in SW Iberia, and using this regression, we build synthetic isoseismal maps for a large (M_s = 8.0) earthquake that occurred in 1969. Based on information on the seismic source provided by various authors, we show that the velocity model effectively reproduces macroseismic observations in the whole region. We also confirm that seismic intensity distribution is very sensitive to a small number of source parameters: rupture directivity, fault strike and fault dimensions. Then, we extrapolate the method to the case of the great (M = 8.5–8.7) 1755 earthquake, for a series of hypotheses recently proposed by three authors about the location of the epicentral region. The model involving a subduction-related rupture in the Gulf of Cádiz results in excessive ground motion in northern Morocco, suggesting that the source of the 1755 earthquake should be located further west. A rupture along the western coast of Portugal, compatible with an activation of the passive western Iberian margin, would imply a relatively low average slip, which, alone, would could not account for the large tsunami observed in the whole northern Atlantic ocean. A seismic source located below the Gorringe Bank seems the most likely since it is more efficient in reproducing the distribution of high intensities in SW Iberia due to the 1755 earthquake.

Iberian region, ocean—continent transition, strong ground motion, waveform modelling

1 Introduction

Most great tsunamigenic earthquakes are related to well-defined interplate convergence zones: the circum-Pacific seismic belt, the Sunda arc, the Hellenic arc and the Antilles arc. An exception is the massive earthquake that struck Iberia and Morocco on 1755 November 1, was felt in a large part of Europe, and produced a powerful tsunami that crossed the Atlantic Ocean. This earthquake occurred along a passive margin, in a region where plate boundaries are not unequivocally defined by bathymetry, and where the existence of an active subduction zone is not clearly supported by seismological evidence. Thus, despite the large size of the fault that is responsible for this extreme event (several hundred kilometres), the location of the epicentral region of the 1755 earthquake is poorly known; many hypotheses have been made by various authors for its source. Managing to discriminate between acceptable and unacceptable scenarios for this earthquake is a fundamental step in the improvement of seismic risk assessment.

In this paper, we concentrate on the comparison of macroseismic observations with synthetic seismic intensity for a set of possible source parameters. For this purpose, we have to take into account the major heterogeneities that affect, at various scales, the crustal structure in the region, and that largely condition the distribution of seismic intensities on land. A realistic velocity model, embedded into a wave propagation code, is an appropriate way to study source parameters of the 1755 earthquake, provided there is sufficient data. The finite difference method is well adapted to this strategy, since it is computationally efficient enough to enable a large number of tests to be run, and accurate in the low frequency range (f < 0.5 Hz), which is the most significant for the study of the destructive effects of large earthquakes at a regional scale. The velocity model used here was proposed and validated in the companion paper (Grandin et al. 2007). For this purpose, it is necessary to introduce a regression between simulated values of Peak Ground Motion Parameter (PGMP) and Modified Mercalli Intensity (MMI). The 1969 earthquake (M_s = 8.0) constitutes the most obvious test to calibrate this regression: source parameters are reasonably well constrained, and the distribution of macroseismic observations is uniform and of good quality.

After a short introduction to the seismo-tectonic setting in SW Iberia, we will establish this regression, and calibrate it using the 1969 earthquake. Finally, we will extrapolate to the case of the 1755 earthquake, and test three hypothetical models for the 1755 earthquake source. This will enable us to determine which source fits best the observed intensity distribution of the ‘Lisbon’ earthquake.

2 Seismo-Tectonic Setting

2.1 Tectonic context

The current tectonic regime at the boundary of the African and Eurasian plates varies with longitude (Fig. 1), as a result of the rotation of Africa, with respect to Eurasia, around an Euler pole located offshore Morocco, close to (20°N; 20°W) (Argus et al. 1989; DeMets et al. 1994). This boundary is commonly divided into three sections (Buforn et al. 1988). To the west, between the Africa—Eurasia—North America triple junction (35°W) to the eastern end of the Terceira Ridge (24°N), the regime is transtensional with an extension rate of 4.4 mm yr⁻¹ (Borges et al. 2007), and is responsible for the active volcanism found in the Azores archipelago. In the central section, the relative motion of the two plates seems to be accommodated by a single right-lateral fault, the Gloria fault, although significant seismic activity is observed in a broad region off the fault (Lynnes & Ruff 1985). East of 16°W, the bathymetric continuity of the Gloria fault cannot be followed, and the definition of the boundary is less clear. A transpressive tectonic regime dominates, with a very low convergence rate of 4 mm yr⁻¹ (Argus et al. 1989; McClusky et al. 2003) trending NW—NNW, consistent with the observed maximum horizontal stress direction (Ribeiro et al. 1996; Borges et al. 2001; Stich et al. 2003; Carrilho et al. 2004). Deformation is distributed over an increasingly large area that can reach a N—S width of 300 km near the continental margin of Iberia (Chen & Grimison 1989). In this section, seismicity is scattered, most events concentrating along a 100-km wide band that trends ESE—WNW from 16°W to 9°W (Fig. 1). A progressive shift of focal mechanisms, from strike-slip mechanisms in the west, to predominant reverse faulting in the east, has also been reported by Buforn et al. (1988), and can be interpreted as an increasing plunge of the minimum compression axis. In the Gulf of Cádiz, seismicity is denser to the north, around the Guadalquivir Bank. East of the Strait of Gibraltar, in the Alborán domain and in Betic-Rif Arc, seismicity (Calvert et al. 2000) and GPS data (Fadil et al. 2006) both suggest that active deformation is spread over several hundreds of kilometres. In this study, we focus on the central section, which is believed to be the host of the great 1755 earthquake.

Figure 1.

Bathymetric map of the region of study. Inset shows the location of the region of study, and a sketch of the kinematic regime at plate boundaries. Isobaths are traced with a 1000 m interval below −1000 m, with a 200 m interval above (data provided by the British Oceanic Data Centre 2003). Dots show the location of earthquakes epicentres recorded from 1995 January to 2005 December (data provided by the Instituto de Meteorologia, Lisbon). Small light grey dots are earthquakes with 3 < M_w < 4, medium dark grey dots are earthquakes with 4 < M_w < 5, and large black dots are earthquakes with M_w > 5. White triangles indicate the location of large historical earthquakes (the possible location of the source of the 1755 ‘Lisbon’ earthquake is discussed in the text). Major offshore active fault are also shown (from Hayward et al. 1999; Gràcia et al. 2003a). Submarine canyons are identified by dashed lines.

In the area, a series of topographic structures trending WSW—ENE occur (Fig. 1). The Horseshoe scarp and the Marquês de Pombal scarp, parallel to the São Vicente canyon, have experienced deformation since, at least, the Miocene (Zitellini et al. 2001; Gràcia et al. 2003b; Terrinha et al. 2003). The most prominent seamount is Gorringe Bank, an uplifted thrust block of crustal and upper-mantle rocks that reaches 25 m below sea level and was emplaced during the Miocene (12 Ma) by plate breakage, differential loading and NW—SE shortening of the Jurassic (150 Ma) oceanic lithosphere (Hayward et al. 1999). North of the Coral Patch Seamount and Coral Patch Ridge, scarps affect the seafloor, with the same trend, and large amplitude folds involve the sedimentary cover down to the basement (Hayward et al. 1999). The transition between the localized and diffuse plate boundary may be due to a change in the nature (and age) of the lithosphere (Jiménez-Munt et al. 2001). On the other hand, by analogy with the Macquarie Ridge Complex at the Australia—Pacific Plate boundary, Massel et al. (2000) suggest that the proximity of the Europe—Africa pole of rotation may induce significant changes in boundary conditions over small timescales (a few million years) thus resulting in a transitional state of deformation. The latter scenario is supported by the occurrence of unusually large oceanic earthquakes inside the area of scattered seismicity in 1969 (M_s = 8.0; Fukao 1973) and 1975 (M_s = 7.9; Lynnes & Ruff 1985). Moreover, focal mechanisms of these two events shared no obvious resemblance with bathymetric structures in the epicentral area.

In the Alboran region, seismic activity is scattered, and strike-slip mechanisms are consistent with a NW—SE compression, and horizontal extension in the NE—SW direction (Bezzeghoud & Buforn 1999). East of 1°E, as the least compressive axis shifts to a vertical direction, seismicity becomes more localized along the Algerian coast, and the compressive stress regime is expressed by somewhat simpler thrust mechanisms (Bezzeghoud & Buforn 1999).

2.2 Instrumental seismicity

On land, seismicity is scattered, with a maximum focal depth of 15–20 km (Fig. 1), and a large variety of focal mechanisms (Borges et al. 2001). In the Lusitanian Basin, microseismicity reveals a complex pattern of focal mechanisms, and the depth of earthquakes, between 14 and 23 km, suggests that deformation is controlled by tectonic activity deep in the basement (Fonseca & Long 1991). Seismic activity in the region of Évora is associated with unmapped structures (Borges et al. 2001). In the Monchique range, the steady microtremor activity probably involves pre-existing tectonic features located at depth (Dias 2001). Further south, the activity of the Portimão Fault can be inferred from the N to S trending active zone (Carrilho et al. 2004), consistent with average focal mechanisms showing N—S strike-slip fault-planes (Bezzeghoud & Borges 2003), as suggested by Terrinha (1997). In the Guadalquivir foreland basin, seismicity is low, and seems to be limited to the blind Subbetics frontal thrust.

Offshore, the poor azimuthal coverage and the lack of stations at short distances from epicentres do not allow the depth distribution of earthquakes, even close to the coast, to be constrained. However, most events occurring between 14°W and 7°W are located at depths between 5 and 25 km, and can reach 90 km (Buforn et al. 1988) in the oceanic domain that lies west of 10°W. The widening of the diffuse plate boundary goes together with an increase of the maximum depth of earthquakes: from 35 km at the eastern termination of the Gloria fault (Grimison & Chen 1986), to 60–90 km in the Gulf of Cádiz (Buforn et al. 2004), and 150 km east of the Strait of Gibraltar, with occasional deep earthquakes (650 km) below the Alborán Sea (Calvert et al. 2000).

The largest instrumental earthquake, M_s = 8.0 (USGS), occurred on 1969 February 28 in the eastern Horseshoe Abyssal Plain. From the study of depth phases (pP, sP and pwP), Grimison & Chen (1986) determined a focal depth of 50 km for a strong aftershock (m_b = 5.5) that occurred on 1969 May 5. They also deduced that the old Jurassic oceanic lithosphere (155 Ma) has a brittle thickness of approximately 40–70 km, limited by the 600 °C isotherm, thus justifying the intermediate depth seismicity observed in the area. In the Gulf of Cádiz, seismicity concentrated around the Guadalquivir Bank, as well as the occurrence of the 1964 earthquake (Udías & López-Arroyo 1970), with magnitude M_w = 6.2 (Buforn et al. 1988), suggest that this structure, that is clearly identified on seismic profiles and free-air gravity anomaly maps (Gràcia et al. 2003a), is active.

In the eastern Azores-Gibraltar Fault Zone, the largest events have thrust-type focal mechanisms, with a small strike-slip component. However, since the 1969 sequence, only one earthquake has reached a magnitude of 5² (2003 July 29, M_w = 5.3, discussed in detail in the companion paper); hence, most events are too small to be of regional significance.

2.3 Historical seismicity

The Lower Tagus Valley was struck by several important historical earthquakes: the 1531 January 26 earthquake (Justo & Salwa 1998), with magnitude M∼ 6.5–7.0 (Justo & Salwa 1998; Vilanova 2003) and the 1909 April 23 earthquake (Moreira 1985; Teves-Costa et al. 1999), with magnitude M_w = 6.0 (Teves-Costa et al. 1999), are the best known cases. The source of the 1858 November 11 Setúbal earthquake, with magnitude M∼ 7.1 (Vilanova 2003), might be a blind thrust fault, trending NNE—SSW, located below to the Setúbal canyon, in the Sado Valley (Ribeiro 2002), possibly connecting with an E—W fault on land (Moreira 1985). In the Algarve, the Loulé Fault, which accommodated a significant part of the shortening across the Algarve basin, due to the presence of halokinetic structures in its footwall, could be responsible for the 1856 December 1 earthquake (Moreira 1985; Terrinha 1997), with magnitude M∼ 5.5 (Vilanova 2003). The source of the tsunamigenic 1722 December 27 Tavira earthquake, with magnitude M∼ 7.0 (Vilanova 2003), might be located offshore the southern Portuguese coast, possibly near the Guadalquivir bank area (Baptista et al. 2000).

In short, seismic activity is low on the continent, and thrust faults are often blind, with long recurrence times, leading to large uncertainties. Offshore, activity is more significant, but epicentral locations are, obviously, even more uncertain. The existence of major canyons, aligned with clearly mapped onshore faults, has led many authors to associate some historical earthquakes with such structures; however, these morphological observations can be misleading (Dinis 2006). In this context, it is not surprising that the location of the tsunamigenic 1755 November 1 earthquake, with magnitude M∼ 8.5–8.8 (Abe 1979; Johnston 1996), is still heavily debated (Johnston 1996; Zitellini et al. 2001; Gutscher et al. 2002; Terrinha et al. 2003; Vilanova et al. 2003; Ribeiro et al. 2006). Similarly, historical information about the large tsunami generated by the 1761 March 31 earthquake, with magnitude 7.5 (Martins & Mendes Víctor 2001), are too sparse to allow a determination of the epicentral region (Baptista et al. 2006).

3 Seismic Intensity Modelling

3.1 Numerical method and source implementation

To model the propagation of seismic waves in a 3-D media, we used the code E3D, an explicit 2-D/3-D elastic finite-difference wave propagation code (Larsen & Schultz 1995), based on the work of Madariaga (1976). The method and computational issues related to the finite-difference scheme are presented in details in the companion paper (Grandin et al. 2007). This numerical method was successfully applied in 3-D by a large number of authors for generating synthetic seismograms in the USA (e.g. Olsen & Schuster 1992; Frankel & Vidale 1992; Olsen & Archuleta 1996; Larsen et al. 1997; Wald & Graves 1998; Stidham et al. 1999; Dreger et al. 2001; Pitarka et al. 2004; Hartzell et al. 2006) and in Japan (e.g. Pitarka et al. 1994, 1996, 1998; Sato et al. 1999, Satoh et al. 2001; Kagawa et al. 2004). However, the size of the region studied in the present paper (∼400 × 400 km) is exceptionally large, and our knowledge of the crustal structure in the region of interest is more limited than in the above studies. Therefore, we expect only the broad features of the wavefield to be reproduced (Grandin et al. 2007). We recall that simulations are only reliable at low frequency, the maximum frequency being proportional to the minimum velocity in the grid, and to the inverse of grid spacing. In this paper, a grid spacing of 1 km is used, thus yielding a maximum frequency of 0.3 Hz.

In the case of the two larger earthquakes studied here (7.5 < M_w < 8.5), the finiteness of the dimension of the fault and of the duration of rupture cannot be ignored. Following the source implementation scheme of E3D, we model these extended sources by superimposing a large number of point sources over a rectangular fault plane that shares the same strike and dip as the individual subevents. The kinematics of rupture, namely rupture propagation direction, is simulated by triggering rupture on each subfault at the right time: rupture can nucleate at a certain location on the fault, and then propagate radially until the fault edge is reached; rupture velocity is assumed to be constant over the fault plane. To prevent high-frequency noise from entering the radiated spectrum, and thus generate a smooth source time function, one has to make sure that rupture on each subfault is initiated before rupture on the previous adjacent subfault has stopped. We chose to use a Brune signal as the elementary source time function for rupture on each subfault (Brune 1970). The displacement rate , as a function of time t, is given by:

1

where τ is the characteristic time of subevent rupture, that sets both the rise time and the duration of the rupture (Beresnev and Atkinson 1997). In the case of a finite source, rupture velocity is fixed to 2.5 km s⁻¹, and the grid spacing is 1 km. By setting τ to 0.4 s, rupture initiates on a subfault when the previous subfault has reached approximately 25 per cent of the total displacement.

We additionally assume a uniform seismic moment distribution over the fault plane. Thus, the slip is not uniform due to the variation of rigidity modulus with depth inside the velocity model. However, in the epicentral distance range considered here (d > 100 km), we have checked that this condition does not induce significant differences from a uniform geometric moment distribution. Furthermore, we set the depth to the top of the fault so that coseismic rupture does not extend beyond the seismic basement. This prevents the occurrence of supershear rupture velocities in the shallow sedimentary cover, but assumes that sediments are not involved in significant seismic wave generation. From a seismological point of view, this hypothesis is valid. Indeed, the total seismic moment is given by:

2

where μ and b are, respectively, the rigidity and slip at an elementary surface dS, located on the fault plane Σ. For large earthquakes, the low rigidity of shallow sedimentary layers (two times lower than average crustal rigidity), and their small thickness (a few per cent of the thickness of the brittle Jurassic oceanic crust) ensure that their contribution to the total seismic excitation is low. This is valid for the case of a uniform coseismic slip, or, inversely, of a uniform seismic moment distribution.

To determine the stability of our results, we have performed a large number of simulations to evaluate the importance of the parameters we have arbitrarily fixed. In the case of the extended rupture, we have verified that dividing or multiplying τ by a factor of 2 does not affect significantly the radiated wavefield. Modifying the value of v_r within reasonable bounds only slightly affects the amplitude of ground shaking, by enhancing or weakening the directivity effect. Finally, we have checked that a complex geometry, involving a slow, shallow rupture on a plane extending inside the sedimentary cover does not affect the results significantly, while the additional number of parameters, and thus the risk of generating spurious results, increases dramatically. On the other hand, variations of focal parameters, fault plane geometry and rupture directivity have strong effects on the resulting radiated wavefield.

In the rest of the paper, the computational domain has a grid spacing of 1 km, and extends to a depth of 70 km. Anelastic attenuation is included (Table 1 in companion paper).

Table 1.

Regressions of Peak Ground Velocity (PGV) measured in m s⁻¹ versus Modified Mercalli Intensity (MMI) discussed here. Parameters a and b are defined in eq. (2).

3.2 Regression of Peak Ground Velocity versus Modified Mercalli Intensity

Four sources of information exist for the seismicity that occurs offshore SW Iberia. First, seismological data is available for recent instrumental earthquakes. Second, macroseismic data has been retrieved from inquiries carried out immediately after major events, or a posteriori, by the study of historical information. Third, several recent studies aimed at imaging the relation between the complex bathymetry and the structure of the sediment cover, using geophysical techniques, to locate active tectonic structures. Fourth, near-field tsunami data has been deduced from available tide-gauges and historical observations. Additional observations have been made in the case of the great 1755 earthquake (far-field tsunami, near and far-field hydrological effects, landslides, seiches and seismic activity triggering), but, since they are not available for any other earthquake, this type of information will not be discussed here.

In the area, all four types of information are reliably available for only one major earthquake, namely the 1969 February 28 Horseshoe Abyssal Plain earthquake. More recent earthquakes have been much smaller, and, as a consequence, were barely felt on land, and probably did not produce surface ruptures, nor a detectable tsunami. Older earthquakes of comparable magnitude are badly constrained from a seismological point of view (excluding the 1964 earthquake, which occurred in the eastern Gulf of Cádiz) and macroseismic observations, as well as tsunami observations, are less reliable. This is obviously the case of historical earthquakes. Our major objective is the computation of synthetic seismic intensity maps, in order to assess seismic risk associated with sources located offshore SW Iberia. In this context, correctly reproducing ground motion distribution for the 1969 earthquake is the first step towards the prediction of isoseismal maps due to any hypothetical source (Levret 1991; Johnston 1996; Paula & Oliveira 1996): the 1969 event is the only event of comparable magnitude for which source parameters are well constrained, and macroseismic information is reliable and dense enough to ensure a safe regression. However, we must bear in mind that seismic intensity heavily depend on local effects, which explains the large standard deviation of observations made in the field; such effects are beyond our level of resolution. Therefore, we will only consider the broad features of the isoseismals, deduced by comparing the ‘smoothed’ intensity values provided by various authors.

Laws relating the intensity of an earthquake (I₀) to epicentral distance in SW Iberia are available from a series of studies (Levret 1991; Johnston 1996; Baptista 1998; López Casado et al. 2000; Martínez-Solares & López-Arroyo 2004); these studies suggest that, in the area, attenuation is very low for large magnitude earthquakes. We use a different approach, where we relate the level of shaking, obtained by a quantitative simulation of wave propagation through the crust and upper mantle, to the macroseismic intensity. Thus, in addition to geometric attenuation, the effects of source finiteness, directivity, and heterogeneities of the propagation medium are also included.

The observed correlation between the level of seismic intensity and the peak value of ground motion has been described at length by Trifunac & Brady (1975), who gave regressions of the peak values of velocity and acceleration, versus MMI, in a study focused on the Western United States. Other authors provided alternative relations for the Western United States (Wald et al. 1999), Taiwan (Wu et al. 2003) or Eastern North America (Kaka & Atkinson 2004). The relations referenced here are of the form:

3

where PGMP is the Peak Ground Motion Parameter (displacement, velocity and acceleration).

The value of a mostly affects the geometric attenuation of intensity, while b is a constant. Wald et al. (1999) and Wu et al. (2003) suggest that, for high intensities (>VII), Peak Ground Velocity (PGV) correlates better with MMI than Peak Ground Displacement (PGD) or Peak Ground Acceleration (PGA). The values of the parameters a and b used by a series of authors are presented in Table 1, with PGV expressed in m s⁻¹. Since they are clearly region-specific, these relations should be used with caution when applied to synthetic ground motion in SW Iberia: they concern different tectonic and geological environments, and were obtained using higher frequencies (>1 Hz) than frequencies available in this study (<0.3 Hz). Thus, they provide bounds to the regression that may apply to the specific case of SW Iberia.

Our method to determine a regression of the form of eq. (3) is the following. As a preliminary step, we performed a large number of simulations for the 1969 event, and tested the dependence of the results on variations of the source parameters in order to evaluate whether small to moderate uncertainties on fault parameters can result in large uncertainties of the final result. We concluded that, at the regional scale, and in the case of an extended fault, the distribution of PGV is, at the first order, mainly controlled by the magnitude of the earthquake and source location (this result is not surprising, and will enable us, in the final stage of this study, to discriminate between realistic and unrealistic locations for the epicentre of the 1755 earthquake). Rupture directivity (controlled by the rupture initiation location), strike direction and fault dimensions are critical factors for the azimuthal distribution of the maximum amplitude oscillations. In contrast, the computation is little sensitive to other source parameters, such as the dip, the rake, rupture velocity and depth to the top of the fault or the duration and shape of the source time function.

Thus, we concluded that the problem can be partitioned: in a first step, we select an average value for source parameters of the 1969 earthquake, which enables us to do the followings.

• 1

  Fix the value of b, which controls the decay of PGV with epicentral distance.

• 2

  Fix the value of a, which controls the average value of PGV.

Because of the linearity of the equations used for the computation of ground motion in this study, the simulated seismic moment has the same scaling effect as parameter b, and there is a trade-off between the two parameters that can only be resolved by the study of at least two earthquakes. We find that a relation similar to that of Trifunac & Brady (1975) gives good results (Table 1).

In a second step, we discuss the effects of the uncertainties on source parameters, which have second order effects on the distribution of intensities. We especially focus on source directivity effects and basin amplification. This step is necessary to the validation of the values of a and b, since source parameters estimates of the 1969 earthquake made by various authors are somewhat contradictory (see discussion below).

4 The 1969 Earthquake

4.1 Seismic source characterization

The focal mechanism and seismic moment of the 1969 February 28 earthquake has been studied by (among others) López-Arroyo & Udías (1972), McKenzie (1972), Fukao (1973), Buforn et al. (1988) and Grimison & Chen (1988). They concluded that the mechanism is a thrust, with a strike of 40–70°, and a small left-lateral component. Grimison & Chen (1988) suggested that the earthquake was composed of two subevents, the second one having an almost pure strike-slip mechanism. López-Arroyo & Udías (1972) suggested that the SE dipping nodal plane should be the fault plane, but Fukao (1973), and subsequent publications by other authors, favoured the NW dipping plane. López-Arroyo & Udías (1972) calculated a seismic moment of 6.3 × 10¹⁹ Nm, while Fukao (1973) and Grimison & Chen (1988) computed a seismic moment nearly 10 times larger, of 6.0 × 10²⁰ Nm and 8.0 × 10²⁰ Nm, respectively. Fault length was estimated to be 85 km (López-Arroyo & Udías 1972) or 80 km (Fukao 1973), but, due to the 10-fold difference of seismic moment calculations, Buforn et al. (1988) and Grimison & Chen (1988) disagree on the value of fault width, with 20 and 50 km, respectively. Fukao (1973), and Grimison & Chen (1988) both suggested that the rupture propagation was directed towards the NE. The computed epicentral depths are 22 km (López-Arroyo & Udías 1972), 33 km (Fukao 1973) and 32 km (Grimison & Chen 1988).

Macroseismic data for the 1969 earthquake in Iberia has been compiled and published by López-Arroyo & Udías (1972), Mendes (1974), Martínez-Solares et al. (1979), Paula & Oliveira (1996) and Martínez-Solares & López-Arroyo (2004). Using data provided by Paula & Oliveira (1996) for Portugal, and by Mezcua (1982) for Spain, we built a new map of the observed seismic intensities for the 1969 earthquake (Fig. 2). Authors cited above report the largest intensities (VII—VIII) along the southern coast of Portugal, and underline the occurrence of moderate ground motion amplification in the Guadalquivir Basin (V—VI). A N—S elongated region, located on the western coast of Portugal, from the Algarve to the Setúbal area, also had intensity VI.

Figure 2.

Compiled isoseismal map of the 1969 February 28 earthquake, using data provided by Paula & Oliveira (1996) for Portugal (the size and colour of dots varies with seismic intensity), and by Mezcua (1982) for Spain. The location of the epicentre, and the approximate fault dimensions and configuration are also shown.

No anomalously high intensities are explicitly reported in the Lower Tagus Basin, which has lead several authors to claim that no site amplification was observed during the 1969 earthquake in this area (Vilanova et al. 2003). Therefore, one would favour scenarios that produce the same level of shaking in the Neogene and Mesozoic basins. In the rest of the Iberian peninsula, a simpler geometric attenuation seems to have occurred. However, as can be seen in Fig. 2, no data points are available in a wide area located NE of the Tagus Lagoon. The whole zone is covered by thick Neogene alluvial deposits (Carvalho et al. 2006). This ‘hole’ in the data is surprising, and will necessitate further investigations, but does not allow us to discuss whether amplification, or attenuation, occurred in the central part of the Neogene Lower Tagus Basin due to the incoming long-wavelength seismic waves excited by the 1969 earthquake (M_s = 8.0). Similarly, observations made by Pereira de Sousa (1919) in the case of the 1755 earthquake (M∼ 8.5) are bewildering: in the Neogene Lower Tagus Basin, large buildings (parish churches) located at the same epicentral distance and above comparable sedimentary column thicknesses suffered significantly different levels of shaking (for instance, IX—X in Benavente versus VI in Alcochete). However, in the case of the 1969 earthquake, we must admit that, should amplification have occurred, it must have been very localized.

Heinrich et al. (1994), Guesmia et al. (1996), Miranda et al. (1996) and Gjevick et al. (1997) conducted studies of the tsunami associated with the 1969 event, and concluded that Fukao's (1973) source parameters allow the tsunami observations (traveltimes and amplitudes) in southern Iberia to be reproduced relatively well. However, the synthetic mareograms do not succeed in reproducing the traveltime and polarity of the primary wave observed in Casablanca, Morocco. Gjevick et al. (1997) suggested that a slightly rotated strike of the source (from 55° to 73°) provides a better agreement with the observations. These conclusions support the hypothesis of a complex 1969 tsunami source.

4.2 Results and discussion

The main disagreement between authors about the 1969 earthquake lies in the determination of the seismic moment, and therefore, of fault width and coseismic slip. We chose to test the two sources proposed by Fukao (1973), and Buforn et al. (1988), with fault widths of 20 and 50 km, respectively. We also tested a third source, with fault dimensions and focal parameters fixed at intermediary values. For each of the three sources, the fault plane was rotated by 10°, clockwise and anticlockwise, in order to account for the uncertainties on the strike. A total of nine tests were performed (Table 2).

Table 2.

Source parameters used for the computation of synthetic isoseismal maps of the 1969 earthquake. For each scenario, fault dimensions and dip angle are different, and we test three values of the strike. The value of the rake is adjusted so as to maintain approximately the direction of the axis of maximum compression. Source A2 and C2 correspond to the models of Buforn et al. (1988) and Fukao (1973), respectively.

Rupture is nucleated SW of the fault centre (at a distance of 20 km along strike), at two-thirds of the fault width (distance measured along dip, from fault top), and propagates radially at a velocity of 2.5 km s⁻¹. We have checked that a bilateral rupture gives similar results, with slightly lower intensities, while a rupture directed towards SW results in small intensities on land (maximum of VI), incompatible with observations (maximum of VIII).

Due to the difficulty in estimating the depth to the top of the fault, this parameter is not assessed by any author. In our study, we noticed that the results were little dependent on this depth, provided that the top of the fault is located below the seismic basement. Therefore, we fixed this parameter to 8 km. The middle of the top of the fault is located at (36.0°N; –10.45°E). The value of the seismic moment was adjusted for each hypothetical source in order to obtain a similar level of synthetic seismic intensity for all cases.

The resulting synthetic intensity maps are shown in Fig. 3. The tested values of the seismic moments are 3.6 × 10²⁰, 6.0 × 10²⁰ and 1.0 × 10²¹ Nm for models A, B and C, respectively (Table 2). These values are in good agreement with the magnitudes provided by Fukao (1973) and Grimison & Chen (1988). On the other hand, we tested a value of the seismic moment of 6.3 × 10¹⁹ Nm, following the proposition of López-Arroyo & Udías (1972), and we obtained much lower intensities (except by choosing an unrealistically high value of parameter a), suggesting that the seismic moment is significantly underestimated. López-Arroyo & Udías (1972) had already expressed doubts about the compatibility of this value with the high stress drop of the 1969 earthquake, inferred from the surprisingly small number of aftershocks.

Figure 3.

Synthetic isoseismal maps of the 1969 February 28 earthquake, using source parameters summed in Table 2 (see text for discussion). The focal mechanism of each source is shown.

All 9 models succeed in reproducing the ‘L-shaped’ macroseismic field, characterized by the same level of intensities along the southwestern coast of Portugal and Spain, with a decrease when moving inland. The largest intensities are found either in the eastern Algarve (southern Portugal), or in the Lower Tagus Valley Basin. For model A, because of the smaller fault width, directivity is stronger, and small variations of the fault strike induce appreciable differences in the distribution of intensities. On the other hand, model C is less sensitive to such variations. In all cases, solutions with a fault plane rotated anticlockwise give larger intensities in the Lower Tagus Valley Basin, and thus can be excluded.

Model A2, extrapolated from López-Arroyo & Udías (1972) and Buforn et al. (1988), reproduces well the observations, with the highest intensities in the Algarve (VIII) and a large ‘L-shaped’ region of lower intensities (VI) extending from the western Guadalquivir Basin to the Lisbon Area. The rotated mechanism (model A3) yields similar features. On the other hand, model C2, which corresponds to the solution proposed by Fukao (1973) gives much higher intensities in the Lower Tagus Valley Basin (IX) than in the Algarve (VII). A rotation of the fault plane (model C3) does not improve this result significantly. However, a reduction of the fault width (B2) increases the directivity effect, and re-establishes a balance between intensities calculated in the Algarve, in the Guadalquivir Basin and the Lower Tagus Valley Basin.

For each model, we can calculate the average displacement using the eq. (2), with a uniform rigidity modulus. Given the high rigidity modulus at the fault location (Table 2), the calculated average displacements (∼4 m) for models B and C are in good agreement with empirical regression moment magnitude versus average displacement (Wells & Coppersmith 1994). On the other hand, for models A and B, the deduced value of the moment magnitude (M_w = 7.7–7.8) differs slightly from the surface wave magnitude (M_s = 8.0, according to USGS), confirming the high stress drop character of the 1969 earthquake (Kanamori & Anderson 1975). We conclude that our preferred model is model B.

In this part, we confirmed that the distortion of isoseismals, due to the variety of geological settings found in the region, could be correctly matched by using a relatively simple 3-D velocity model and appropriate source parameters. From the macroseismic study of the 1969 earthquake, we can draw three conclusions. First, a clockwise rotation of the mechanisms can be excluded, because it would ‘transfer’ radiated energy from the Guadalquivir region to the Lisbon area, and generate significantly higher intensities in the Lower Tagus Valley Basin. Second, the seismic moment calculated by López-Arroyo & Udías (1972) is significantly lower than the seismic moment deduced from this study, which is, on the other hand, in good agreement with the value provided by Fukao (1973). Third, a fault width of 35 km gives better results than the fault widths of 20 and 50 km proposed by Buforn et al. (1988) and Fukao (1973), respectively.

However, two main discrepancies between observed and synthetic isoseismal maps persist. First, intensities at the NE border of the model are higher in the synthetics. This could be explained by the poor correlation between the lowest intensities and computed PGV in the range of frequencies considered here. Second, synthetic intensities in the Lusitanian Basin are much smaller that observed intensities. This can be improved by a better modelling of the geological structure in the area, especially at shallow depth.

5 The 1755 Earthquake

5.1 Source characterization

The 1755 November 1 earthquake was the strongest earthquake ever reported in Europe, and was extremely destructive throughout Portugal (Pereira de Sousa 1919), Spain (Martínez-Solares et al. 1979) and Morocco (Levret 1991) (Fig. 4); the shock was even felt in Northern Germany, the Azores and the Cape Verde Islands (Reid 1914). The large size of the earthquake is further attested by the observation of seiches in southern England, in Holland and as far as Finland (Reid 1914). The large tsunami-waves generated by the earthquake also provoked extensive damage along the coasts of Portugal, southern Spain and Morocco, and were even detected in the Lesser Antilles and southwestern England. Extensive geological evidence of tsunami deposits associated with the 1755 earthquakes have been reported in southern Portugal (Andrade 1992; Dawson et al. 1995), southern England (Foster et al. 1993), on the Spanish coast (Luque et al. 2001), at the Tagus Estuary mouth (Abrantes et al. 2005) and in the Cape da Roca-Cascais area west of Lisbon (Scheffers and Kelletat 2005).

Figure 4.

Compiled isoseismal map of the 1755 November 1 earthquake, using data provided by Pereira de Sousa (1919) and Moreira (1984) for Portugal, Martínez-Solares et al. (1979) for Spain, and Levret (1991) for Morocco. The three tested fault models at Gorringe Bank (Johnston 1996), along the Marquês de Pombal - Pereira de Sousa Fault Zone (Zitellini et al. 2001; Terrinha et al. 2003) and in the Gulf of Cádiz (Gutscher et al. 2002; Gutscher et al. 2006; Thiebot & Gutscher 2006) are shown (Table 3).

The problem of epicentral location has been addressed by various early studies (Reid 1914; Pereira de Sousa 1919), and, since the beginning of the instrumental period, a consensus attributed the origin of the earthquake to a structure located between the Gorringe Bank and the Coral Patch Ridge (Machado 1966; Moreira 1985; Johnston 1996).

In the most recent hypothesis among several scenarios, Johnston (1996) suggested that a NE—SW trending fault, possibly outcropping at the base of the NW flank of Gorringe Bank, could be responsible for the 1755 earthquake. Using isoseismal regression analysis, he deduced a seismic moment of 1.26 × 10²² Nm, and a fault area of 16 000 km² compatible with an average slip of 12 m. The SE dipping fault plane suggests that the Eurasian oceanic plate is being overridden by the African oceanic plate, at least a shallow depth.

However, the recent re-evaluation of historical information available for the 1755 earthquake and tsunami (Baptista et al. 1995, 1998a; Vilanova et al. 2003), as well as results from new seismic surveys offshore SW Iberia (Zitellini et al. 2001; Gutscher et al. 2002), has led a series of authors to propose alternative scenarios.

Baptista et al. (1995, 1998a) presented a review of historical records of tsunami parameters that have been used as the basis for a series of studies that aimed at determining the epicentral area location by backward ray tracing of the tsunami (Baptista et al. 1996, 1998b). A source located in Gorringe Bank, with a fault length of 120 km and a dip directed towards NE, was tested by Baptista et al. (1996, 1998b), and led to simulated traveltimes systematically longer than observations in Iberia; the authors concluded that ‘the Gorringe Bank is a very unlikely source for the 1755 event’. Several subsequent publications aimed at testing the coherence of various hypothetical sources in terms of tsunami traveltimes (Baptista et al. 1996, 1998b, 2003; Gutscher et al. 2006). These studies concluded that a source compatible with an incipient subduction along the western Portuguese margin (e.g. Ribeiro 2002) provides a better fit between inferred and modelled tsunami traveltimes than a source located at Gorringe Banks.

At the same time, Zitellini et al. (2001) reported active thrusting at the base of a large submarine hill dubbed ‘Marquês de Pombal’, located 80 km west of Cape São Vicente. This inferred thrust zone is associated with seismic activity, may have a listric geometry, and might be connected to a backthrust fault via a décollement surface located at a depth of 16–18 km. However, the surface of the inferred fault is too small (7 000 km²) to generate a M_w∼ 8.5 earthquake. Terrinha et al. (2003) subsequently proposed that the Marquês de Pombal Thrust should be extended further north, where, instead of cutting directly to the surface, it would connect to a shallower décollement level (depth of ∼8–10 km), and finally reach the base of the continental slope at the eastern border of the Tagus Abyssal Plain. The complex geometry of the second segment would be responsible for the inversion of the normal faults of the Pereira de Sousa Fault Zone, and would yield a much larger surface (19 000 km²), compatible with the magnitudes of the earthquake and associated tsunami. Additionally, Gràcia et al. (2003b) showed that half of the NNE—SSW striking scarp created by this ESE dipping fault, is buried below submarine landslide deposits, suggesting that slope failure processes could have contributed to tsunami generation.

On the other hand, based on interpretation of other seismic profiles, and gravimetry and thermal modelling, Gutscher et al. (2002; 2006) and Thiebot & Gutscher (2006) propose that the olistostrome units of the western Gulf of Cádiz were accumulated as a result of the ongoing westward displacement of a shallow east dipping fault plane, in a suduction context. This fault plane, currently locked, would have been responsible for the 1755 earthquake, at least as a secondary source associated with a primary source further west. Platt & Houseman (2003) and Fonseca (2005) objected that the absence of a Wadati—Benioff zone does not allow a straightforward confirmation of the activity of a hypothetic subduction plane, neither in the Gulf of Cádiz, nor along the western Portuguese margin. Another objection comes from preliminary GPS data, which suggests that the western Betic-Rif belt and NW Morocco behave rigidly (Fadil et al. 2006; Stich et al. 2006), thus bringing into question the connection of the southern ramp of the presumed subduction plane, which lies in a region of low seismic activity (Fig. 1), with unmapped active structures onshore.

Another complex scenario has been proposed by Vilanova et al. (2003), who argued that the multiple shocks felt all over Iberia, and the high seismic intensities observed in the Lisbon area demonstrate the presence of a secondary source in the Lower Tagus Valley. The rupture of the inferred NNE—SSW striking, WNW dipping thrust plane outcropping on the western bank of the Tagus river, at the frontier between the Mesozoic and Neogene basins (Fonseca et al. 2000, 2001), would be due to stress triggering. Such a scenario would imply that strong motion in the Lisbon area is dominated by the existence of major local sources, rather than offshore sources, as the epicentral location(s) of the 1755 earthquake(s) is (are) critical to our understanding of the seismic hazard in the whole SW Iberia region (Montilla et al. 2002). However, as pointed by Matias et al. (2005), should clear evidence of the validity of this hypothesis be found, it is likely that the available historical information will never allow a magnitude to be assigned to this secondary event. Consequently, we are unable to discuss the scenario proposed by Vilanova et al. (2003) in this study.

Seismic intensity modelling was performed by various authors, using probabilistic and stochastic methods (Mendes-Victor et al. 1999; Baptista et al. 2003), or site-effect functions and attenuation relations (Gutscher et al. 2006). These methods suffer important limitations, because they do not take into account physical considerations attached to the complexity of the seismic source, on one hand, and the propagation medium, on the other hand. Hence, it is not surprising that the conclusions of the studies cited above are contradictory: Mendes-Victor et al. (1999) back the hypothesis of the two NNW—SSE striking sources (‘N160N135’) described by Baptista et al. (1996, 1998b); Baptista et al. (2003) favour a ‘L-shaped’ rupture (‘MPTF/GB’) on two segments located below Marquês de Pombal and Guadalquivir Bank, respectively; Gutscher et al. (2006) suggest that a source located at (−8.5°W; 36.0°N), possibly an asperity at the northwestern corner of the subduction plane proposed by Gutscher et al. (2002), would explain most of the observed macroseismic intensities in Iberia and Morocco.

Generally, all authors cited above agree on a primary source located offshore SW Portugal.

5.2 Results and discussion

In this study, we test, by forward modelling, three previously published sources for the 1755 earthquake that can be considered as end-members of the set of proposed offshore seismic sources (Fig. 4). First (model G), a steep fault, dipping towards SE, and extending down to a depth of 50 km below Gorringe Bank is tested (Johnston 1996). A second source (model M) is implemented, consisting of listric faults rooted in the middle crust, below the Marquês de Pombal - Pereira de Sousa fault zone, and distant of ∼50 km from the SW coast of Portugal, along the Alentejo margin (Zitellini et al. 2001; Terrinha et al. 2003). Finally (model C), we simulate a rupture along a shallow dipping subduction-related thrust in the Gulf of Cádiz (Gutscher et al. 2002; Gutscher et al. 2006; Thiebot & Gutscher 2006).

Source parameters are shown in Table 3. The seismic moment for each model is deduced from eq. (2) using the value of slip and fault area proposed by the various authors, and the average rigidity at the depth. For each source, fault plane geometry (dimensions, strike, dip and depth) is set using available published information, but many other parameters (rake, rupture velocity and directivity) are, obviously, not constrained by historical information, and must be extrapolated from the study of similar instrumental earthquakes.

Table 3.

Source parameters used for the computation of synthetic isoseismal maps of the 1755 earthquake. Latitude, longitude and depth are specified at the centre of the top of the fault.

We assume that all fault segments act with a pure thrust mechanism; this hypothesis is supported by the observation of source mechanisms of recent great (M_w > 8.0) shallow dip-slip earthquakes (Harvard CMT Catalogue).

Rupture velocity was set to 2.5 km s⁻¹ for the shallow rupture models B and C, which is a typical value for dip-slip crustal earthquakes. For model A, which involves faulting at larger depths, a slightly higher rupture velocity of 2.7 km s⁻¹ was used.

As emphasized by McGuire et al. (2002), the majority of large (M > 7) earthquakes exhibit a preferential direction of rupture, and the 1755 earthquake may be expected to have behaved in the same way. Hence, for sources A and B, in addition to a bilateral rupture scenario, the two unilateral cases are tested. However, the low aspect ratio of source C is not typical of great tsunamigenic earthquakes, and an along-dip directed rupture cannot be excluded. Therefore, for this earthquake, two additional nucleation points are considered, in addition to the three along-strike cases (Fig. 5).

Figure 5.

Synthetic isoseismal maps of the 1755 November 1 earthquake, using source parameters summed in Table 3. Stars denote epicentral location (see text for discussion). The first and second rows correspond to model C, the third row corresponds to model M, and the fourth row corresponds to model G.

Model C, with a fault located a few tens of kilometres away from the coasts of the eastern Gulf of Cádiz, systematically produces a similar level of ground shaking in southern Spain, southern Portugal and northern Morocco. Compared to the historical isoseismal map, the highest intensities (X—XI) are shifted from the southern coast of Portugal to the southwestern coast Spain. In the most favourable case (a rupture directed towards WSW), simulated intensities are still much higher than intensities deduced from historical studies carried out in Morocco (IX—X versus VII—VIII) and in the Internal Betics (VIII versus V) by Levret (1991) and Martínez-Solares & López-Arroyo (2004), respectively. Overall, this model fails to reproduce a key feature of the 1755 November 1 earthquake, that is the observation of higher intensities in Portugal than in Spain and Morocco.

In the case of model M, we observe high intensities in the Lisbon area (IX—X), in the Lusitanian Basin, in the Lower Tagus Valley Basin, and in the Arrabida chain, that are compatible with observations. These intensities are conditioned by rupture directivity, because the area lies in the strike direction of the fault. Intensities in Algarve (IX) are also correctly reproduced, mainly because of the proximity of the source. However, intensities computed in the western direction (VIII) are much higher than observations (VI—VII). In the synthetics, the decrease of intensities with distance is very low, with intensities VIII distributed over a 300-km wide zone. The relation between PGV and MMI is not the cause of this phenomenon, since, in order to remove this effect, an unrealistically high value of the parameter a, in eq. (3), would be required. The radiation pattern induced by the style of faulting can explain these discrepancies: in the most remote areas, surface waves generated by the roughly E—W coseismic slip have a maximum amplitude in the direction of the Algarve, the Guadalquivir Basin and the Betic-Rif Belt, but also in the areas further north, in eastern Portugal and western Spain. This effect might be reduced if a smaller fault width was used, but would require a smaller seismic moment, and a lower tsunamigenic efficiency.

As for model G, most features of the observed macroseismic field seem to be correctly reproduced with a rupture directed towards SW: high intensities in the Algarve (IX—X) and in the Lisbon area (IX), associated with a moderate amplification in the Guadalquivir Basin (VIII). We also observe a rapid decrease of intensities from south to north, at 8°W, and from west to east, at 39°N. The shape of the limit between intensities IX and X is in very good agreement with observations, and the highest level of shaking is located at Cape São Vicente. On the other hand, simulated intensities at the SE corner of the computational grid (6.5°N; 39.5°N) are larger than intensities reported by Martínez-Solares & López-Arroyo (2004) in the area (V—VI). However, it can be noted that this overestimation is also observed for models C (VII) and M (VII—VIII), but the error is much smaller for model G (VI).

Magnitude estimates made from isoseismal area regression methods (Johnston 1996) or tsunami strength (Abe 1979) were obtained assuming an epicentral location close to Gorringe Bank, and constrained using mostly near field data, as the quality of macroseismic and tsunami observations dramatically decreases with their magnitude. An eastward displacement of the source would obviously reduce the inferred moment magnitude. In this study, for both models M and C, modelled intensities in the innermost areas of Iberia (VII—VIII) are significantly higher than observed intensities (VI), even when an average slip of 10 m is used. This incompatibility between the proposed fault locations of models M and C, and the observed macroseismic field, suggest that the seismic moment should be reduced, either by scaling down the average slip, or by reducing fault dimensions. However, Baptista et al. (1998b, 2003) and Gutscher et al. (2006) all admit that an average slip of ∼20 m is needed to reproduce tsunami wave amplitudes and run-up values comparable with observation, in the case of a rupture located close to the coast. Such large average coseismic slips have only been observed for exceptional great (M_w > 9.0) mega-thrust earthquakes (1957 Aleutian earthquake, 1960 Chile earthquake, 1964 Alaska earthquake and 2004 Sumatra earthquake), with rupture lengths an order of magnitude larger than values considered here. Therefore, for models M and C, a second incompatibility appears between proposed fault dimensions and the average slip needed to generate the large tsunami wave amplitude.

We conclude that a fault located below Gorringe Bank, with a rupture directed towards SW, reproduces better the overall pattern of macroseismic observations than a fault aligned along the Marquês de Pombal - Pereira de Sousa Fault zone, or a subduction-related thrust fault in the Gulf of Cádiz.

6 Conclusion

We generated synthetic isoseismal maps for a large (M_s = 8.0) earthquake that occurred 1969 in the region. The fault parameters provided by several previous studies were tested, and we obtained a very good fit of simulated and real intensity distributions for a focal mechanism (strike = 233°; dip = 49.5°; rake = 63.5°) in good agreement with existing solutions. Fault dimensions were also estimated (fault length = 82.5 km; fault width = 35 km), as well as the seismic moment (M₀ = 6 × 10²⁰ Nm), and average slip (approximately 4 m). Finally, using the 1969 earthquake, we calibrated a relationship between PGV and MMI. Using this relationship, the case of the great 1755 earthquake (M_w = 8.5–8.7) was examined, and we tested three end-member sources proposed by different authors. We concluded that, using the methodology presented in this paper, a primary source located at Gorringe Bank is the most realistic hypothesis to fit the observed isoseismal pattern.

In this study, we have shown that unknowns lie in a realistic description of the seismic source in the area of study, but that only a few parameters are crucial when trying to simulate ground motion at low frequency. Obviously, the moment magnitude, focal mechanism and source location influence the distribution of intensities in the near field. For large and great earthquakes, other details of the seismic source can alter dramatically this distribution. Rupture directivity plays a particularly important role, and is conditioned by various parameters, such as fault dimensions, epicentral location with respect to the fault centre, rupture velocity, and fault strike.

1

We would like to acknowledge the occurrence of a M_w = 6.1 earthquake on the 12th of February, 2007, while the present paper was under review. A special issue of the EMSC Newsletter is dedicated to this event.

Acknowledgments

We are grateful to Pascal Podvin who made this work possible, Paul Tapponnier for fruitful discussions, and Geoffrey King whose remarks and continuous support significantly improved the quality of our work. The authors acknowledge helpful reviews of two anonymous reviewers. We thank Shawn Larsen for supplying the code E3D. Some figures were made using the GMT software written by P. Wessel and W. H. F. Smith.

This work has been developed with the support of the Fundação para a Ciência e a Tecnologia, through project POCTI/CTE-GIN/59994/2004: ‘Finite source modelization and strong motion prediction in Portugal’, and POCTI/CTE-GIN/59750/2004: ‘Seismic tomography of the continental lithosphere in Algarve (Portugal)’ and was largely carried out at Centro de Geofísica de Évora (Portugal). This is also IPGP contribution 2269.

Author notes

*

Now at: Institut de Physique du Globe de Paris (IPGP), Laboratoire de Tectonique et Mécanique de la Lithosphère, 4 place Jussieu, 75252 Paris, France. E-mail: