Tsunamis are low-frequency but high-impact hazards for coastal societies. The striking tsunami events on 26 December 2004 in the Indian Ocean and on 11 March 2011 in Tohoku raised awareness due to the enormous loss of life and property. The Indian Ocean event in 2004 demonstrated the need for operational early warning systems around the world. However, 7 years later, the 2011 Tohoku event showed the limitations of scientific knowledge concerning tsunami sources, coastal impacts, and mitigation measures. Since then, in the NEAM region (northeast Atlantic, Mediterranean and connected seas) many efforts have been made to understand tsunamigenic sources better and to improve tsunami hazard assessment capabilities. Within the NEAM region, the Gulf of Cadiz is among the most tsunami hazardous areas. The historical reports include events dating back to 60 BC (Mendonça, 1758; Baptista and Miranda, 2009; Kaabouben et al., 2009), but the geological evidence indicates high-energy events as far back as 218 BC (Luque et al., 2001).

The Portuguese coast is highly exposed to tsunami threat from active local and regional tectonic sources. The main tsunamigenic area is the SWIM (Southwest Iberia Margin), with a number of considerable SE dipping inverse faults (Fig. 1b) (Zitellini et al., 2009; Matias et al., 2013). The most severe tsunami occurred on 1 November 1755 and was caused by the Lisbon earthquake, with magnitude of 8.5 as estimated by Martins and Mendes-Victor (1990). This magnitude was more recently re-evaluated by Solares and Arroyo (2004), who estimated it to have been 8.5 ± 0.3. The tsunami hit the entire northern Atlantic basin with a huge impact in Iberia and Morocco (Fig. 1) (Baptista and Miranda, 2009). In the 20th century, the 28 February 1969 earthquake with a magnitude of 7.9 (Fukao, 1973) caused a small tsunami of 0.5 m amplitude in Lagos and Cascais (Fig. 2a) (Baptista et al., 1992; Baptista and Miranda, 2009). The tsunami waves hit the coast at circa 03:00 UTC in low-tide conditions (Baptista et al., 1992), but no significant damage was observed.

The second tsunamigenic zone to be considered is the Gloria Fault (Fig. 1c). The Gloria Fault is a segment of the Eurasia–Nubia plate boundary. It is a large strike–slip fault, located between 24∘ W and 19∘ W, with scarce seismic activity, but it was nonetheless the location of several large events during the 20th century, in particular the 25 November 1941 earthquake, a submarine strike–slip event of a magnitude of 8.3–8.4 (Gutenberg and Richter, 1949) and the event of 26 May 1975 with a magnitude of 7.9 (Lynnes and Ruff, 1985; Grimson and Chen, 1986).

In recent years, a considerable number of tsunami hazard assessment studies were published for the northeast Atlantic area. Most of these studies focus on the tsunami impact in the Gulf of Cadiz using a deterministic approach, namely Lima et al. (2010), Omira et al. (2010, 2011, 2013), Atillah et al. (2011), Baptista et al. (2011a), Renou et al. (2011), Benchekroun et al. (2015), and Lemos et al. (2014). Recently, Omira et al. (2015) published a probabilistic tsunami hazard assessment for the northeast Atlantic.

In this study, we use a deterministic tsunami hazard assessment (DTHA) approach to evaluate the tsunami impact in Sines (Fig. 2). The study area contains the country's most important deep-water port, which is connected to big industrial complexes by fragile infrastructure such as pipelines and conveyor belts (Fig. 2b). In summer the city is a popular tourist destination.

The DTHA approach consists of studying the impact of specific tsunami events – tsunami scenarios – in the study area. The impact is described in terms of maximum wave height (MWH), maximum flow depth (MFD), maximum drawback (MDB), maximum inundation area (MIA), and maximum inland penetration (MIP). We further built the aggregate scenario plotting the MWH and MFD in each cell considering the contribution of the individual scenarios (Tinti et al., 2011).

The final results are presented in integrated hazard maps for all the scenarios considered and for the aggregate scenario. Each integrated hazard map consists of MWH, MFD, MDB, MIA, and MIP of the corresponding scenario. The static effect of the tide is analysed for three different tidal stages: mean lower low water (MLLW), mean sea level (MSL), and mean higher high water (MHHW). Further, we present the contribution of each scenario to the aggregate tsunami impact in MSL conditions.

Numerical modelling of tsunamis is commonly divided into three stages: generation, propagation, and inundation. We use a benchmarked numerical code NSWING (Non-linear Shallow Water model wIth Nested Grids) (Miranda et al., 2014) developed in-house to model the tsunami. The model supposes an instant seabed deformation that has been rendered using the half-space elastic theory (Okada, 1985) embedded in Mirone suite (Luis, 2007). The vertical sea bottom deformation is assumed to be equal to the free-surface deformation and transferred to the ocean surface.

The code solves linear and non-linear approximations of shallow-water equations (SWEs) to calculate tsunami propagation and inundation in a Cartesian or spherical reference system. In the deep ocean, non-linear convective inertia forces are of secondary order as waves travel with amplitudes much smaller than water depths. When the tsunami enters shallow coastal areas, the non-linear convective inertia force and bottom friction become increasingly important. We applied non-linear SWEs approximations in all instances, for deep-ocean, near-shore, and onshore propagation.

NSWING employs a dynamically coupled system of nested grids and solves SWEs using an explicit staggered finite leapfrog numerical scheme for linear terms and an upwind scheme for non-linear terms. In NSWING the incorporation of the system-coupled nested grids is mainly based on the code COMCOT (Cornell Multi-grid Coupled Tsunami Model; Liu et al., 1998). The code further applies a radiating boundary condition, allowing wave motion to pass from one domain to another, through boundaries with very small reflections. A moving boundary algorithm (Liu et al., 1995), based on “wet” and “dry” cells, is adopted to track shoreline movement during inundation.

Propagation and behaviour of tsunamis change because of varying bathymetry when they enter coastal areas. To model the impact in Sines, we implement a dynamically coupled system of nested grids. We interpolated the half-minute North Atlantic grid (GEBCO, 2014) to 640 m resolution for the parent grid. Using four layers and applying a refinement factor of 4, we achieved a 10 m final resolution in the DEM.

The amplitude of the tide in southwest Portugal is above 2 m and must be taken into account in Sines (Baptista et al., 2011a). To study the tide effect, the tidal variation in the last 3 years was considered. We used the values of mean high water (MHW) and mean low water (MLW) from 2012 to 2014 (Antunes, 2014) and calculated the mean to obtain the MHHW and MLLW respectively and referenced them to MSL. The MHHW is 1.22 m above MSL, and the MLLW is 0.88 m below MSL. These values have been subtracted and added respectively to the established DEM. For each scenario designed, we ran the model in MHHW, MSL, and MLLW conditions to study the static influence of the tide; the result is presented in respect to the aggregate scenario.

To design the tsunami scenarios we use the main seismogenic source zones and the associated maximum credible earthquake (MCE) (Miranda et al., 2008; Omira et al., 2009). We used the typical faults (TFs) presented in Omira et al. (2009) except for the Portimao Bank fault (PBF) because it does not direct enough energy to the western Portuguese coast.

The seismogenic sources used here are SWIM and Gloria. For this study we considered four TFs in the SWIM area and their MCE scenarios to reproduce initial conditions for tsunami propagation; these TFs are the Cadiz Wedge fault (CWF), the Gorringe Bank fault (GBF), the Horseshoe Fault (HSF), and the Marques de Pombal fault (MPF) (Fig. 1b). Additionally, we use a seismogenic scenario consisting of a composite rupture of HSF and MPF (HSMPF), proposed by Ribeiro et al. (2006) for the source of the 1 November 1755 earthquake. This source is also stated in Matias et al. (2013), with a maximum magnitude estimation of 8.75. This magnitude value coincides with the upper limit of the magnitude estimate for the 1755 earthquake of Solares and Arroyo (2004): 8.5 ± 0.3. As this source has been proposed, we cannot evaluate a worst-case scenario impact without considering it.

The major tsunami event in the SWIM is the one associated with the 1 November 1755 earthquake, and the exact source remains unknown. Numerous studies and campaigns have been carried out in order to identify the source of the 1 November 1755 tsunami. Data (multi-channel reflection seismic, refraction seismic, multibeam swath bathymetry) have been gathered to reveal more accurate information on the tectonics in the SWIM. These investigations in the SWIM revealed much geological evidence for the TFs used in this study. Several authors proposed different sources for this event.

Johnston (1996) suggested the GBF as a possible candidate source of the 1 November 1755 event through scale comparison of isoseismal maps with the 28 February 1969 event. The source for the 1755 tsunami suggested by Baptista et al. (1998) is closer to the Portuguese coast compared to the GBF. Zitellini et al. (1999) found an active thrust fault, the MPF, through the interpretation of multi-channel seismic data. Gutscher et al. (2002) identified an active subduction, the CWF, and concluded that it is a candidate source for the 1755 event. Therefore it must be considered in natural hazard assessments. Further neotectonic structures and deformed seafloor sediments at the HSF also show clusters with shallow seismicity (Gràcia et al., 2003). Some authors suggest considering multiple fault rupture scenarios to explain the high magnitude observed in 1755 (Zitellini et al., 2001; Gràcia et al., 2003; Ribeiro et al., 2006). As this overview shows, the exact source of the 1755 Lisbon event is still a matter of discussion. These uncertainties related to the 1755 source and the tectonic processes acting in the SWIM confirm the need of systematic tsunami hazard assessment in surrounding areas.

The Gloria zone is our regional source area. The Gloria Fault is a transform fault running from 24∘ W to 19∘ W (Laughton and Witmarsh, 1974). Three strong magnitude earthquakes occurred in the last 130 years: 22 December 1884 (Moreira, 1984), 25 November 1941 – magnitude 8.3 (Gutenberg and Richter, 1949; Moreira, 1984), and 26 May 1975 – magnitude 7.9 (Lynnes and Ruff, 1985; Grimson and Chen, 1986). 25 November 1941 and 26 May 1975 produced small tsunamis recorded at the tide stations in the northeast Atlantic basin (Debrach, 1946; Moreira, 1984; Baptista et al., 1992; Baptista and Miranda, 2009). The 25 November 1941 epicentre location and the focal mechanism are presented in Baptista et al. (2011b). We used these parameters to draw a 1941-like scenario for the Gloria source zone. The TF parameters are presented in Table 1, and the fault is presented in Fig. 1.

We ran a total of 18 simulations. For each typical fault we considered three tide conditions: MLLW, MSL, and MHHW. The results are presented in the form of integrated hazard maps showing MWH, MFD, MDB, MIA, and MIP (Fig. 3a–e). For the scenarios at MSL, we additionally present synthetic waveforms (Figs. 4 and 5) at chosen positions (Fig. 2b).

In Fig. 6 we present the aggregate scenario, considering all calculated models. Figure 7 shows the inundation and the drawback limits, considering the aggregate scenarios at the three tide conditions.

We ran a total of 18 scenarios to study the tsunami impact at Sines. Our results show that all SWIM scenarios cause severe inundation and drawback. To complement the integrated hazard maps of MWH, MFD, MDB, MIA, and MIP, we recorded synthetic waveforms at chosen points (see Fig. 2). The signals of the waveforms are diverse. This fact may be explained due to differences in the parameters of the TFs. Waveforms from HSF, MPF, GBF, and HSMPF are comparable in terms of period and arrival time (Fig. 4a–e). These TFs are dextral reverse with SW–NE trending and the hanging block in the SE. They are all located in the SWIM area. They are distinguishable by their dimensions and slip. Other parameters like the fault strike, dip, and rake are similar. Among the single-fault scenarios, the GBF is the largest fault, producing an MWH of 17.1 m in the study area (see Table 2). The HSF, although smaller, produces a similar inundation and wave height pattern as the slip is 5 m higher compared to the GBF (see Table 1). The MPF is smaller in terms of dimensions and has a slip of 8 m and therefore produces the weakest tsunami in Sines among the SWIM scenarios (Table 1 and Fig. 3e). Nevertheless, MWHs are above 10 m and MPF is the nearest fault to the test site that produces a short tsunami travel time (22 min in Fig. 4e). The composite scenario HSMPF is the worst-case scenario and combines the effects of both faults: the first wave arrives 22 min after the earthquake, and the tsunami triggered by HSMPF causes the worst inundation and drawback in Sines. The CWF is a subduction slab and has different fault parameters compared to the other TFs in the SWIM. The shallow, east-dipping slab has dimensions of 170 × 200 km and a slip of 20 m (see Table 1). The analysis of the waveforms shows that wave periods generated by the CWF are larger than those generated by the other TFs in the SWIM. This fact may be explained by the larger co-seismic deformed area in this scenario. Cape St. Vincent (Fig. 1b), in the southwest of Portugal, might act as an obstacle to the tsunami leading to a reduced impact. The CWF has a higher impact in the southern part of the study area but with decreasing inundation and amplitudes towards the north. Nevertheless, wave amplitudes of 5 m cause considerable inundation in the northern part of the port. The waves produced by CWF reach Sines 38 min after the earthquake (Table 2). The Gloria Fault scenario located at 37∘ N between 14∘ W and 24∘ W produces the smallest inundation in the study area. It is a transform fault triggering slight vertical movement because of a 160∘ rake with a slip of 11 m (Table 2). The scenario produces amplitudes between 0.3 and 0.4 m with approximately a 10 min period (Fig. 5b). The earthquake in 1941 generated similar waveforms, showing weak attenuation with amplitudes around 0.4 m in Cascais (Baptista et al., 1992; Baptista and Miranda, 2009). Site effects, observed in some few coastal locations (Fig. 5a), caused MWH over 1 m with some smaller inundation in uninhabited area between the container terminal and the intake and restitution points of the EDP power plant. Because of the larger distance to the Portuguese coast, the tsunami travel time is approximately 85 min (Table 2).

We also calculated flow velocities for the composite scenario HSMPF at MSL for different stages of tsunami propagation. The median values are about 10 m s-1 in the inundation area at all terminals in the port. Some extremes of about 20 m s-1 or higher occur close to the breakwaters, jetties, and in the inundation area when the flow depth values are small depending on the considered propagation instant. In general, we find that flow velocities increase with lower flow depth values in the inundation area.

Considering the HSMPF scenario in MSL conditions, the pipelines at the liquid-bulk and petrochemical terminal are entirely inundated, with up to 5 m flow depth values. These structures are subject to flow velocities of about 10 m s-1 at first wave impact. At the 17 m topographic contour, the pipelines behind the liquid-bulk and petrochemical terminal are not affected by the tsunami (Fig. 3d). We find similar flow velocity values at the multipurpose terminal where the pipelines of the liquefied natural gas storage tanks pass. Here the maximum flow velocity values are slightly above 10 m s-1 at wave impact, and MFDs are between 5 and 10 m. The conveyor belt and the stockpiles at the multipurpose terminal are nearly entirely inundated up to a water level of 5 m and show flow velocities of 10 m s-1 at first wave impact. The pipelines at the liquid-bulk, petrochemical, and multipurpose terminal are inundated in all scenarios in the SWIM. These quantitative DTHA results indicate a high risk of potential damage in the case of tsunami impact. However, the topic of building vulnerability is beyond the scope of this study.

The tide has an important influence on tsunami impact in Sines. The tidal regime is semi-diurnal with an amplitude of about 2 m. As expected, the aggregate scenario in MHHW conditions caused larger inundation areas and higher MFD values. On the other hand, the aggregate scenario at MLLW produced larger drawback areas. A tsunami impact at low tide does not exclude the risk of heavy inundation and increases MDB by 16 % compared to MHHW (Fig. 7).

Our results are compatible with the PTHA (probabilistic tsunami hazard assessment) results for the northeast Atlantic, recently published by Omira et al. (2015). This study shows that wave heights exceeding 5 m have a probability of 45 % of occurrence in 500 years at Sines. Only the scenarios of the SWIM area have the capacity to produce such a high tsunami impact along the Portuguese west coast. Moreover, our results are comparable with the unique historical report showing that the tsunami did not reach the city (Falcão, 1987).

We computed a map showing the contribution of the individual scenarios to the aggregate scenario at MSL (Fig. 8). Four scenarios contribute to the aggregate scenario, namely CWF, GBF, HSF, and HSMPF. The main factor in the aggregate model is the HSMPF scenario that contributes more than 60 % in terms of MWH and MFD area, independent of the tidal amplitude (Table 3). The scenarios CWF, GBF, and HSF contribute about 12 ± 4 % to the aggregate model. The MPF and Gloria do not contribute to the aggregate scenario (Table 3 and Fig. 8).

Finally, we simulate the extreme scenario corresponding to the combination of the worst seismic scenario (HSMPF) and the highest annual tide (2 m above MSL). The annual frequency of occurrence of the HSMPF seismic scenario – magnitude 8.75 is 8 × 10-4 (Omira et al., 2015) – and the annual frequency of the extreme tide is 1, resulting in a very unlikely event but corresponding to the worst tsunami impact at Sines. On the other hand, the probability of having a tsunami in MHHW conditions is much higher than the probability of having a tsunami in extreme tide conditions because the MHHW tide level occurs several times a year. In Fig. 9, we show a comparison between the MIPs for HSMPF in MHHW conditions, HSMPF in extreme annual tide conditions, and the aggregate scenario. The results show that the inundation area for the extreme scenario is 5 % higher than the one in MHHW conditions. However, the comparison between the extreme scenario and the aggregate results in a 0.5 % greater inundated area. Therefore, we recommend considering the 5 % as an additional buffer regarding inland penetration when applying mitigation measures.

In conclusion, we find that all SWIM scenarios (CWF, GBF, HSF, MPF, and HSMPF) demonstrate a high impact at Sines test site. Nevertheless, the weakest source, the MPF, still causes considerable inundation and an MWH above 10 m. The proximity of the faults within the SWIM results in short tsunami travel times. For the scenarios MPF and HSMPF, we calculated 22 min propagation time from the source to the Sines test site. This closeness to possible tsunami sources creates the need for an efficient early warning system and meticulously planned evacuation for the port and other coastal areas. Also, coastal societies need to be educated about and prepared for possible tsunami impact.

The Gloria Fault scenario differs from the other scenarios and produces an MWH of approximately 1 m in certain areas, 1 order of magnitude less than the scenarios in the SWIM.

The aggregate scenario allows us to consider a set of faults to produce a synthesis of different scenarios. We further state the importance of this tool as an important indicator for evacuation and city planners. We showed with the contribution map that different sources have varying degrees of importance in our study area. Although the worst-case scenario may contribute more to the aggregate scenario than the other faults considered, the other faults may still have a more significant impact on other parts of the test site. The aggregate scenario is a valuable tool for quantitative presentation of tsunami impact from multiple sources. Especially in areas exposed to threat from near-field sources, the aggregate scenario maps help to establish accurate evacuation plans and thus to allow efficient and faster reaction to tsunami warning.