Geotechnical characterization and seismic response of shallow geological formations in downtown Lisbon

The geological and geotechnical characterization of shallow formations is one of the main steps in performing a microzonation study. This paper presents an example of the usefulness of the information compiled in a geological and geotechnical database for the estimation of the seismic response of the shallower formations of the Lisbon downtown area of Baixa. The geotechnical characterization of this area was performed based on the analysis of Standard Penetration Test (SPT) data compiled in the geological and geotechnical database. This database, connected to a geoscientific information system (CGIS), allows, also, the definition of 2D geological profiles used for estimating the thickness of the shallower layers. The shear-wave velocities (VS ) for each layer were estimated from empirical correlations using mean SPT values computed from the statistical evaluation of the compiled data. These VS values were further calibrated with ambient vibration recording analysis. The seismic response of Baixa’s superficial deposits was estimated by applying a 1D equivalent linear method to a set of soil profiles, regularly distributed across the area, and using synthetic accelerograms to simulate input motions associated with probable earthquake occurrences in Lisbon. The results are presented in terms of maps of predominant frequencies, with the corresponding amplification level, as well as spectral amplification factors for 1 Hz and 2.5 Hz. The results show that the fundamental frequency of the Baixa area is between 1.2 Hz and 2 Hz, for the whole central valley, reaching 3 Hz near the edges where anthroprogenic and alluvial deposits have less expression. Amplification factors up to 5 were obtained. These results were achieved regardless of the considered input motion. The similarity of the obtained fundamental frequency with the natural frequency of Baixa’s old building stock increases the probability of resonance effects in future


Introduction
It is recognized that the characteristics of seismic ground motion can be locally modified due to the existence of soft surface layers or basin geometry [e.g., Bard andBouchon 1985, Idriss 1990].Examples may be found from several past earthquakes going back over several decades [e.g., Chavéz-García and Bard 1994, Bouckovalas and Kouretzis 2001, Duval et al. 2001, Giammarinaro et al. 2005, Fritsche et al. 2009, Navarro et al. 2009, Maugeri et al. 2011].Several studies have shown the existence of resonance effects due to the coincidence of the natural period of the shallower soil layers and the fundamental period of buildings built on the soils resulting, often, in unexpectedly higher levels of damage [e.g., Bakir et al. 2002, Gallipoli et al. 2003, Mucciarelli et al. 2004, Gosar et al. 2010].Consequently, the damage distribution observed during an earthquake can be conditioned by this effect, which can increase up to two degrees the observed EMS intensity (European Macroseismic Scale -EMS98; Grünthal [1998]).A paradigmatic example occurred during the 1985 Michoacán earthquake (Mexico).However, other examples can be found for more recent earthquakes occurring worldwide (e.g., Izmit, Turkey, 1999;Chi-Chi, Taiwan, 1999;Mula, Spain, 1999;Al Hoceimas, Morocco, 2004;Abruzzo, Italy, 2009), where the upper, normally consolidated soft deposits were responsible for an increase in the ground motion level for some specific periods [e.g., Navarro et al. 2000, Maugeri et al. 2011].Therefore, the estimation of the seismic behavior of soils for a large town exhibiting moderate to high seismic risk, such as Lisbon, is of great importance for the damage assessment for a future earthquake.
During its history, Lisbon has been affected by several medium to strong earthquakes that caused considerable damage and produced large economic and social impacts.In particular, the very large and well known November 1st, 1755, earthquake (M ≥ 8) caused the

Subject classification:
Site effects, Lisbon, Geotechnical database, Theoretical transfer functions, Ground motion amplification.complete destruction of its downtown area (Baixa), which was reconstructed with the application of the first implemented seismic resistant rules [Pereira de Sousa 1909, Cóias e Silva 2005].Besides this kind of event, generated due to the slow collision of the Euroasiatic and the African tectonic plates, Lisbon can also be affected by earthquakes with moderate magnitudes and with epicenters located nearer to the city, such as the January 26, 1531, earthquake (M 7) generated inland in the Lower Tagus valley seismogenic zone [Moreira 1991].
Knowledge of the seismic response of the surface layers is fundamental for estimating the potential damage due to the occurrence of a medium to strong earthquake.The estimation of the soil fundamental frequencies and, if possible, the amplification of the ground motion during an earthquake, should be the main goal of microzonation studies conducted in urban areas [e.g., Fäh et al. 1997, Ansal et al. 2004, Giammarinaro et al. 2005, Anbazhagan and Sitharam 2008, Ansal et al. 2010].Several methodologies can be applied to perform these kinds of studies but there is no consensus on the most effective procedure for estimating the seismic behavior of soils and the potential site amplification effects [Bard 1999, Mucciarelli andGallipoli 2006].
In recent years, it has become common practice to classify soils into a small number of classes according to the V S30 value (average shear-wave velocity in the upper 30 m of the sub-surface), as presented in Eurocode 8 (EC8) [IPQ 2010], although any statistical test would conclude that this parameter has no (or a very weak) link to seismic amplification [Castellaro et al. 2008, Lee andTrifunac 2010].Several authors proposed soil classifications that are not only based on code recommendations (V S30 values), but also use complementary information, usually F 0 (fundamental frequency) obtained from microtremor analysis [e.g., Luzi et al. 2011, Cadet et al. 2012].However, both geotechnical engineers and seismologists agree that the site conditions can be estimated using the V S profile down to bedrock.
Several geophysical techniques can be used for measuring S-wave velocities and to define the V S profile.Some of these techniques could provide good quality results but they are not often used in microzonation studies because they are not easy to implement in an urban environment.Quick and cheap techniques have been developed using ambient vibration measurements.In particular, Nakamura's technique [Nakamura 1989[Nakamura , 1996[Nakamura , 2000]], based on the interpretation of the horizontal-to-vertical spectral ratio (H/V) computed from ambient vibration records, has been widely used over recent decades in several cities worldwide [Lermo and Chávez-García 1994, Theodulidis and Bard 1995, Duval et al. 2001, Lebrun et al. 2001, Teves-Costa and Senos 2004, Tuladhar et al. 2004, Kamalian et al. 2008, Gosar et al. 2010, Fnais et al. 2014].The theoretical basis of the method is controversial but the use of this technique to estimate site effects has been validated by comparison with both simulations and earthquake recordings [e.g., Bard et al. 1997, Bonnefoy-Claudet et al. 2006a, 2006b, Haghshenas et al. 2008].
Due to its seismic history, Lisbon has been the subject of several studies performed by different authors since the early 1990's, with the objective of estimating plausible seismic scenarios for the city.Oliveira [2008] presented an extensive review of the studies performed, pointing to the uncertainties and proposing the development of further work to reduce them.Also, several hazard studies were performed in the last few years for mainland Portugal and the Iberian Peninsula [Montilla et al. 2002, Sousa and Costa 2009, Vilanova et al. 2012] focusing mainly on the estimation of attenuation laws.
The seismic response of Lisbon's soils was also addressed in previous studies: Mendes-Victor et al. [1994] characterized the seismic behavior of soils by means of impedance contrast to bedrock; Teves- Costa et al. [1995] developed a map of dominant frequencies based on ambient vibration recordings analysis; Teves- Costa et al. [2001] performed a linear 1D theoretical analysis based on data presented in the geological map of Almeida [1986].In the work developed for the Metropolitan Area of Lisbon, in 2001, the soil behavior was estimated by 1D non-linear analysis of specific soil profiles defined from geological and geotechnical logs.However, the work required was very large and the information provided for Lisbon was poorly detailed [Oliveira 2008].
Recently, as part of the Portuguese research project GeoSIS_Lx (http://geosislx.cm-lisboa.pt),new and old information on detailed the geology and geotechnical properties, based on down-hole information, were compiled and implemented in a geographical information system (GIS) [Almeida et al. 2010].The main objective of this project was to allow the use of conventional geological information, as presented in the Lisbon geological map, together with engineering geological data obtained from site investigations, for a wide variety of applications.In particular, it was possible to perform geological and geotechnical 3D modelling of the city [Matildes et al. 2010].
The goal of this paper is to show the applicability of the geological and geotechnical database to microzonation studies, in particular on the geotechnical characterization of the shallower formations, as has been carried out in other countries with similar databases, e.g. in Spain [Cadet et al. 2011], Turkey [Hasancebi andUlusay 2006, Koçkar andAkgün 2007], India [Anbazhagan and Sitharam 2008, Maheswari et al. 2010b, Anbazhagan et al. 2013], Malaysia [Nabilah and Balendra 2012] and Iran [Akbari et al. 2011].With this objective in mind, the geotechnical characterization of Baixa's shallow geological formations, based on Standard Penetration Test (SPT) data analysis, and the estimation of their seismic response, are presented.After providing a general geological and geomorphological framework of Lisbon and its downtown area -Baixa, the paper is structured as follows: (i) a brief description of the geological and geotechnical modeling, obtained in the GeoSIS_Lx project using the database information [Matildes et al. 2010], enabling the identification of the physical properties of any soil profile (lithological description, thickness and depth to bedrock); (ii) the statistical analysis of the Standard Penetration Test data contained in the database (N SPT values), enabling the geotechnical characterization of three distinct zones in Baixa; (iii) estimation of unit weight and V S values for each layer, based on empirical relations using the analyzed NSPT data [e.g., Imai 1977, Lee 1990, Dikmen 2009]; (iv) calibration of the obtained V S values using microtremor records; (v) computation of a 1D soil response for a set of soil profiles regularly spaced using the equivalent linear model [Schnabel et al. 1972] and considering a set of near and far simulated earthquakes.Corresponding transfer functions were computed.
The results are presented in terms of maps of dominant frequencies and spectral amplification factors for the Baixa area.

Geological and geomorphological setting
Lisbon grew around an historical center which includes the downtown area (Baixa) (Figure 1).This area has been occupied since prehistorical times due to its strategic geographical location in relation to fluvial, estuarine and marine environments.Archaeological remains show a prehistory of early occupation in small settlements and a long, evolving urban occupation, since the mid-first millennium BC to the present.
In 1755, when Lisbon was struck by a major earthquake, Baixa was the most damaged area, not only due to the site response to the strong ground motion but also due to the tsunami that followed and the fire triggered by the earthquake.The almost total destruction of this area required major reconstruction leading to a new urban plan, with a geometric design.The new town was built over the ruins and, as a consequence of the great volume of debris, a thick layer of man-made (anthropogenic) materials, locally buried in the soft alluvial deposits, covered the creek area.Due to this process, the local coastline was artificially moved closer to its present-day location [Almeida et al. 2008].
The geological setting of Lisbon (Figure 1) is characterized by two geological environments.The SW area landscaped in Mesozoic formation materials in-cludes Cretaceous marls and limestones and neo-Cretaceous basalts.The E and NE area have Cenozoic formations, mainly Palaeocene and Miocene sedimentary series, associated with the genesis and evolution of the Tagus river basin.During the Miocene, an open connection with the sea allowed the deposition of a complete estuarine sequence, with alternating marine and continental facies.The thickness of the complete sequence can be as great as approximately 300 m.As the Miocene units form a monocline dipping east, the sequence becomes thicker eastwards.The Pliocene and Pleistocene sedimentation represents new conditions in the basin with a predominantly sandy sequence.The Holocene fluvial deposits are characterized by a sequence of sandy and clayey lenticular beds with lateral and vertical facies variations.
The Baixa area, located in the northern estuarine margin of the Tagus River, corresponds to the fluvial outlet of a 6.2 km 2 elongated basin cut in the Miocene bedrock [Almeida et al. 2009].The valley is filled by a thick layer of alluvial sediments (normally consolidated silty sands and organic silty clays).
The Miocene formations present in the area correspond to the base of the Lisbon sequence defined by Cotter [1956] and characterized by alternating units with marine and continental influence and large vertical and lateral variations of facies.In the area, the Miocene sequence includes overconsolidated soils and soft rocks, gently tilted south and southeast with local undulations, giving rise to the geomorphologic setting with incised valleys bounding gentle hills.In this sequence, the different units, although very variable, are characterized by the main lithologies including (Figure 2): silty clayey soils and calcarenites (M Pr and M FT ); fine micaceous sandy and silty sandy soils (M Es , M QB and M pm ); limestones, calcarenites and coquinites (M EC , M CV and M Mu ).
Located at the northern Tagus' margin, the Baixa area is morphologically marked by a depressed area between gentle hills (Figure 2).Affected by tides since historical times, the natural alluvial infill is almost completely covered by anthropogenic deposits.

Geological and geotechnical modelling
As already mentioned, downtown Lisbon suffered a number of changes due to human interventions during its different phases of occupation and, also, severe damage caused by earthquakes.In recent decades, a large number of engineering works, especially for urban, subway and infrastructural development, were carried out.The amount of geological and geotechnical information produced constitutes an important contribution to the knowledge of the geology and the geotechnical characteristics of the area.As part of the GeoSIS_Lx research project, a geotechnical and geological database has been developed to include in situ investigation data (borehole interpretation, sampling and geotechnical measurements) and laboratory test results.The database design follows a hierarchical data structure keeping subjective data interpretation to a minimum.Due to the lack of standard practices, it was also necessary to simplify and group the properties of some parameters to minimize their variability.The implementation of the GeoSIS_Lx geoscientific information system (CGIS) [Turner 2003] interactive with the database, allowed the storage of the information collected and made the data available for several purposes such as urban planning and engineering design [Almeida et al. 2010].Currently, the database contains 924 site investigation reports, 5976 borehole logs and 38,560 N SPT data.This geotechnical information is a secondary but important input for the 3D geological model and, once implemented, is also useful to cross validate the model.
The geological modeling was carried out based on the interpretation of the geological 1:10,000 scale map [Almeida 1986], the retrieval of database information and the automatic processing of such information with Matlab scripts.Considering the large volume of information and the uncertainty associated with existing data it was necessary, in a first iteration, to select the most trustworthy records purging more obvious errors.
Using a small Matlab script on the spreadsheets it was possible to debug some information to produce a more consistent data set [Matildes et al. 2010].The information included in the collected borehole profiles was considered in this processing, as well as the correspondence between stratigraphy, lithology and in situ tests (SPT).The registered data points were interpolated for the whole downtown Lisbon area, through a kriging algorithm, to determine the surfaces representing the lower boundary of each formation of interest.
Taking into account that anthropogenic deposits only appear at the top of the sequence, the thickness of surface materials was obtained by considering the digital terrain model surface and the base of the anthropogenic and alluvial deposits (Figure 3).
The data regarding anthropogenic deposits in all boreholes were used to model their thickness.A total of 666 borehole logs were used to interpolate the corresponding lower boundary surface.Considering the amount of urban development in the study area, it was assumed that anthropogenic deposits are constantly present although with variable thickness and significance (Figure 3).
For the alluvial deposits modeling, the logs of all boreholes that did not reach bedrock were discarded as they provided limited information on the thickness of the alluvium [Matildes et al. 2011].The alluvium deposits basal surface was created by interpolation data LISBON DOWNTOWN SHALLOW FORMATIONS CHARACTERIZATION from 442 borehole logs.The interpolation algorithm takes only into account the depth of the surface at each point and the output is a continuous surface in space, regardless of other parameters of influence on the spatial distribution of the materials.As the deposition of the alluvial materials is morphologically constrained by the relief, it was necessary to carefully analyze the slope and curvature of the area in order to limit the geographical extension of this primarily continuous surface (Figure 3).

Geotechnical characterization
The irregular spatial distribution of the borehole information together with the geological variability and the lateral and vertical variations in composition and geotechnical properties of the shallower layers requires a complex geotechnical model.The geotechnical characterization was performed based on the results of N SPT data and lithology.A wide range of factors, which include the irregular spatial distribution and the lack of standardization in national practice, restrict the efficient use of these data.In the data analysis, despite the careful exclusion of clearly abnormal values, it has always been considered that the results are estimations that have associated uncertainties.
A total of 376 boreholes were selected for analysis in the Baixa area, which included 1398 N SPT data values.Since part of the borehole information is incomplete, invalidating the application of the corrections to the SPT blow counts, it was decided to use the uncorrected value considering always the N SPT value as the number of blows corresponding to 300 mm of penetration.In cases where, according to the practice adopted in Portugal, the test was suspended at 60 blows before total penetration, the value of N SPT was extrapolated to 300 mm.This solution enabled the analysis of certain situations including the effect of the degradation of properties due to superficial weathering of the overconsolidated Miocene hard soils and soft rocks.
Considering the geological genesis and evolution of alluvial sedimentation and the different phases of deposition of the anthropogenic materials, it was expected that the geotechnical properties would have wide spatial variations.To check those differences, the area was divided and analyzed in 3 zones: northern, central and southern (Figure 2).3.2.1.Anthropogenic deposits Anthropogenic deposits, including the debris from the 1755 earthquake, have a heterogeneous lithological composition, depending on their genetic context.In most cases they are sandy.In the fluvial channel, the anthropogenic materials can be intercalated in the soft alluvial deposits making interpretation a difficult task.
The N SPT values show the presence of normally consolidated soils, with 52% of the tests having values up to 10. Values higher than 50 (10%), corresponding to the presence of cobbles or larger fragments included in the softer matrix, are not representative for the behavior of these materials and were ignored (Figure 4).The statistical analysis of the N SPT values shows mean values that are affected by a small number of high values.To overcome this problem, the median was used.Separate analyses are presented for tests performed at depths less than and greater than 5 m (Table 1).
The distribution of the N SPT values with depth is very irregular but shows a slight trend to increasing values with increasing depths, due to the increasing over- burden pressure (see Figure 4).
The lower values observed at greater depths in the southern zone may result from the settlement of these materials within the alluvial deposits.
In terms of the geographical distribution of the anthropogenic deposits, the analyzed values show a slight tendency towards the presence of denser materials in the upper 5 m in the central and southern zones.These results may be a consequence of the increase in densification, due to the contribution of the local dense network of short wooden piles used for foundations in the 18th-19th centuries.

Alluvium
The alluvial deposits are characterized by the presence of lenticular bodies and significant lateral and vertical facies variations.The main lithological facies include soft to hard silt and clay, loose to dense sands, and a range of transitional lithologies.A gravelly layer is frequently present at the bottom of the alluvial sequence.
The N SPT values indicate the presence of normally and slightly overconsolidated soils (Figure 5).The lower values correspond to soft, silty and clayey soils, while the higher values correspond to sandy soils and overconsolidated intercalations resulting from the complex depositional and diagenetic history of the alluvial sequence, taking into account the effect of the water level variation and erosion episodes.Values higher than 50, probably associated with the presence of coarse particles or fragments of anthropogenic materials that have settled into the alluvial sediments, were interpreted as isolated occurrences (Figure 5).These values should not influence the overall behavior of the soils and, therefore, were ignored in the analysis.In the righthand graphs of Figure 5, beyond the projection of the points corresponding to the tests, the lines of the values of the quartiles (Q1, Q2 and Q3) and percentiles (P10 and P90) are also shown.Given the uneven distribution of sandy and clayey layers, due to the lenticular character of the alluvial deposits, as well as the discrimination of the major lithologies, in the statistical analysis the overall results at different depths were considered (Table 2).
The very irregular distribution of the N SPT values with depth can be interpreted as a result of the vertical and lateral lithological variation within the lenticular structure which also contributes to the uneven distribution of the overconsolidated layers (Figure 5).

Miocene bedrock
The Miocene bedrock is characterized by a sequence of sands, clays, marls, calcarenites, coquinites (a limestone conglomerate composed mainly of shell fragments) and limestones, with important vertical and lateral facies variations.The bedrock was analyzed as a single unit and not considered separately for each of the three selected zones.
The N SPT values indicated the presence of overconsolidated hard soils and soft rocks, with 49% of the tests with extrapolated N SPT values higher than 60 (Figure 6).The large range of values is a consequence of the heterogeneity in lithology and of the superficial degradation of the mechanical properties of the overconsolidated Miocene materials.The erosion of several hundred meters of sediments, exhumation and corresponding pedogenetic processes, have led to an increase in porosity and destruction of cementation bonds.The tests performed on the unweathered hard soils systematically produce very high resistance, with reduced sampler penetration, and very high extrapolated N SPT values, while N SPT values less than 30 were registered in the weathered materials, even at depths greater than 20 m (Figure 6 and Table 3).Although it is not very clear in Figure 6, due to the high number of tests and to the plotting of results of different study areas, the N SPT distribution with depth shows the superficial degradation of the mechanical properties.

Estimation of physical properties (V S and c)
Shear wave velocity (V S ) can be obtained directly from seismic field experiments (cross-hole, down-hole, etc.).However, these methods are expensive and they are seldom performed in the urban environment.In this case, if no further data are available or detailed investigations are not possible, the very common SPT is often used.Since the 1970s, many authors have developed empirical correlations between N SPT values and shear wave velocity [e.g., Imai 1977, Imai andTonouchi 1982].Many of the existing correlations have been derived for specific soil types and geological contexts  (1) The authors proposed values of a and b, for different materials (types of soils and stratigraphical context).Some examples are presented in Table 4.
Taking into consideration the lithology of each surface formation, shear wave velocities for anthropogenic deposits and for muddy and sandy alluvium were first estimated using the empirical relationships proposed by Imai [1977].Similar expressions, derived by Lee [1990], were used to estimate the shear wave velocities of the Miocene materials (Table 4).
The N SPT values associated with the different surface materials were also used to estimate the unit weight, c (kN/m 3 ), according to the following expressions [Bowles 1982]: c = 2 ln (N SPT ) + 12.1 (for alluvium) (2) c = 2.1 ln (N SPT ) + 11 (for anthropogenic deposits) (3) Due to the diversity of the geotechnical properties of surface materials, the range of estimated values showed large variation.Taking the median of the N SPT values for the identified surface materials, V S values ranging from 160 m/s to 188 m/s for anthropogenic deposits, 152 m/s to 187 m/s for silty and clayey alluvium, and 132 m/s to 216 m/s for sandy alluvium were obtained, with the lower values for shallower deposits and the higher values for the deeper deposits.The unit weight varied from 15.6 kN/m 3 , for the most superficial anthropogenic deposits (down to 2 m), to 17.9 kN/m 3 , for the deeper alluvial deposits.
For the weathered upper levels of the Miocene hard soils and soft rocks, the V S values varied from 324 m/s to 585 m/s, according to the depth and the lithological composition.The estimated unit weight was 20 kN/m 3  for weathered overconsolidated Miocene material and 22 kN/m 3 for bedrock.These values are in accordance with measured data obtained for the same types of geological formations by Almeida [1991].

Procedure
To constrain the shear wave velocity of the shallower formations, spectral ratios H/V computed from ambient vibration recordings were used.The H/V curves, obtained at specific sites, were compared with theoretical transfer functions computed for the soil profile, which were obtained independent of the geotechnical borehole data.The procedure consisted of five steps (Figure 7): (1) computing the H/V curve from ambient vibration records acquired at selected sites close to geotechnical boreholes; (2) identifying the 1D soil profile from the geotechnical information (thickness of each surface layer and the depth to bedrock); (3) estimating the shear wave velocity for each layer from N SPT values collected from several geotechnical boreholes, using empirical relationships appropriate to the geological setting (see Table 4); (4) computing transfer functions for these soil profiles using synthetic accelerograms; (5) selecting the most appropriate empirical relationship by fitting the fundamental frequency (F 0 ) of the theoretical transfer functions to the peak frequency of the experimental H/V curve.This fitting took into account potential non-linear behavior of the surface deposits.
The main steps of this procedure are presented and discussed below.

H/V curves obtained from ambient vibration measurements and soil profile definition
Taking into consideration the spatial distribution of boreholes with satisfactory information, 13 sites were selected to perform ambient vibration measurements in the study area (Figure 8).The locations of the sites were chosen to sample alluvium and anthropogenic deposits with different thicknesses, and also over different Miocene formations.The measurements were carefully performed according to equipment specifications, field conditions and guidelines from the European SESAME project results [SESAME WP12 2004, Chatelain et al. 2008, Guillier et al. 2008].Ambient vibrations were recorded for 20 to 30 minutes at each site, using a Cityshark digitizer coupled with a 3D Lennartz seismometer of 5 second period.The measurements were taken in good weather conditions (no rain and no wind), mostly during the night, and the sensor was installed, whenever possible, directly on concrete or asphalt; in some sites, the sensor was on the sidewalk.
The H/V curves were computed using Geopsy LISBON DOWNTOWN SHALLOW FORMATIONS CHARACTERIZATION  4. software (http://www.geopsy.org/index.html),following the procedure described in detail by Chatelain et al. [2008].In each record, stable windows with lengths between 20 s and 40 s duration were selected using an anti-triggering system [SESAME WP12 2004, Chatelain et al. 2008].Only records with at least 20 windows were processed.Fourier spectra were computed for each window of each component (V = vertical component; NS and EW = horizontal components).After smoothing and merging the horizontal components with a quadratic mean, the H/V curve was computed for each window.Besides all individual H/V curves, the results present the averaged H/V curve and its standard deviation computed on a decimal logarithmic scale.Standard deviation for F 0 (H/V peak frequency) was also computed from the F 0 values obtained for each individual window.The H/V curves are displayed in Figure 9 and the corresponding F 0 and A 0 values are presented in Table 5.All computed H/V curves satisfy the reliability criteria defined in Sesame guidelines [SESAME WP12 2004], but 5 of these curves did not satisfy the clear peak criteria (sites M3, M4, M6, M7 and M8).Sites M3 and M4 are located on the Colina do Castelo foothills, over a Miocene bedrock formation with no soft surface deposit; so, it is not surprising that the H/V peaks are not very pronounced.The failure of the clear peak criteria for sites M6 and M8 is due to the existence of two very close peaks (see Figure 9).For site M7 the failure of these criteria is due to the "high" values of the H/V curve presented at lower frequencies (compared with the peak amplitude).However, it was noticed by Haghshenas et al. [2008] that the clear peak criteria recommended in the SESAME guidelines are too strict.This is particularly true when the fundamental frequency is low, as many criteria allow a variation within a small percentage of F 0 (for instance 5 to 10%).As F 0 in Baixa varies between 1.1 and 1.4 (see Table 5), the acceptable interval is very small and so it was decided to consider all H/V curves in the following discussion.
From the analysis of the H/V curves displayed in Figure 9 and the values presented in Table 5, it is possible to observe that the average fundamental frequency of Baixa lies between 1.2 Hz and 1.4 Hz.Only the most northern site (M1), already outside the main area of Baixa, has a fundamental frequency of 2.4 Hz.With the exception of the sites over the Miocene bedrock formations (M2, M3 and M4), the amplitude of these frequency peaks varies from 2.9 to 5.2.As expected, the sites over the Miocene bedrock formations show a very poorly-defined peak.
To compute the transfer functions and perform V S estimation (next step), three soil profiles (one for each zone) were defined from three geotechnical boreholes selected close to sites where ambient vibration measurements were performed (Figure 8): M9 (for the northern part; borehole S5), M8 (for the central part; borehole S9) and M11 (for the southern part; borehole S32).From the borehole information the lithology and thickness of each layer, as well as depth to bedrock, were defined.The shear wave velocity and unit weight were computed applying the empirical expressions presented above to the respective N SPT (median values) for each layer.For bedrock (composed of unweathered Miocene rocks), shear wave velocity was set at 800 m/s and the unit weight equal to 22 kN/m 3 .Figure 10 presents the three test soil profiles.

Computing transfer functions and estimating V S
Transfer functions can be defined as the ratio between the Fourier amplitude spectra of the seismic motion at the surface and the seismic motion at at the bedrock surface [Kramer 1996].To define the bedrock input motion Lisbon's historical seismicity, as well as the seismic code for mainland Portugal [Costa et al. 2008], have been taken into account.Two different types of input motion corresponding to relevant scenarios were selected: (i) a distant earthquake (d = 200 km) with large magnitude (M = 7.9) and (ii) a nearby earthquake (d = 25 km) with smaller magnitude (M = 6.0).These two scenarios, similar to the ones proposed by other authors [Carvalho et al. 2008, Oliveira 2008], can be associated with several past earthquakes that affected Lisbon, for instance, the nearby January 26, 1531, (M = 7) and April 23, 1909, (Mw = 6.1) earthquakes, from Lower Tagus Valley sources, and the distant larger magnitude earthquakes of November 1st, 1755 (M = 8.5), and February 28, 1969 (M = 7.9), from offshore sources.Due to the non-existence of strong motion records on the Portugal mainland, within these magnitude and distance ranges [Vilanova et al. 2009], synthetic accelerograms were computed to simulate the strong motion associated with each scenario.
The synthetic accelerograms, generated through a physical simulation of the fault rupture and travel path mechanisms using stochastic methods [Estêvão and Oliveira 2012], were computed specifically for this study.With this method, it was possible to select the inland and offshore sources associated with the close and distant earthquake scenarios.The corresponding seismic motions are in accordance with the two types of seismic action defined in EC8 for mainland Portugal [Costa et al. 2008].For each source ten synthetic accelerograms were computed.Figure 11 presents the response spectra of each seismic motion, for the two scenarios, as well as the Portuguese seismic code response spectra for a 475 year return period [IPQ 2010].Mean PGA are 1.58 m/s 2 (0.16 g) for the near-field and 1.22 m/s 2 (0.12 g) for the far-field simulated seismic motions.These values are smaller than the ones specified in the Portuguese code [IPQ 2010]; however they are in agreement with the attenuation law derived by Ambraseys et al. [1996] using real European strong motion records.
Several methods could be used to compute the 1D soil response.Most authors agree that soft soils subjected to strong ground motions present non-linear behavior [Hartzell et al. 2004].Simple approaches are commonly used, such as the equivalent linear method, which approximates the soil behavior using an iterative procedure [Schnabel et al. 1972], and the frequency dependent linear methods, which were proposed to improve the accuracy of the approximate solution and better simulate a wide range of shear strains [Yoshida et al. 2002].However, as shown by Kwak et al. [2008] these latter methods may lead to different results depending on the frequency dependent algorithms used and do not always improve the accuracy of the solution.
It is also recognized that the 1D equivalent linear analysis provides a reasonable estimate of ground vibration under a seismic event [Idriss 1990].In general, this approach is conservative compared with the results obtained by other methods using recorded or artificial accelerograms [Castellaro and Mulargia 2014].Being aware of the analytical limitations and considering that there was insufficient information about the behavior of Baixa soils, it was decided to use the equivalent linear model to compute the 1D site response analysis, as it is implemented in the ProShake (n.d.) software [Schnabel et al. 1972].To take into account the soil behavior, modulus ratio and damping ratio curves were selected from the literature according to the lithology present in the different layers [Seed andIdriss 1970, Sun et al. 1988].
The computed transfer functions for each soil profile were compared with the corresponding H/V curves obtained from the ambient vibration analysis at the same site, to check for the similarity between the fundamental frequencies of the transfer functions (F 0 ) and the H/V frequency peaks.In spite of the large difference in the strains involved on ambient vibrations and simulated seismic input motions, several authors agree that the peak frequency derived from the H/V curves provides a very satisfactory estimate of the natural frequency of the soil deposit [Bard 1999].Haghshenas et al. [2008] compared the fundamental frequency obtained from spectral ratios to a reference site using earthquake records and the H/V frequency peak obtained with ambient vibrations.They examined several sites with different site conditions, and they found a generally good agreement between the fundamental frequencies obtained with both techniques for most of the analyzed sites.However, since strong seismic motions and soft soils are involved, the occurrence of nonlinear effects must be considered.
To illustrate the procedure, Figure 12  curves.It can be seen that F 0 did not fit the H/V peaks as expected (plots on the left-hand side).Considering that the main parameters affecting the soil response are the thickness and the shear-wave velocity of each layer, it is necessary to change one of these two parameters to have a better compatibility between the transfer function and the H/V curve peaks.As the geometry of the soil profile cannot be changed (it is constrained by the geotechnical borehole information), the disagreement is likely to be due to a wrong estimation of the V S values.So, the V S values were adjusted, using different empirical relationships [e.g., Rodrigues 1979, Imai and Tonouchi 1982, Iyisan 1996, Jafari et al. 2002].Also, taking into consideration that the non-linear behavior can decrease the amplitude of the frequency peak and shift it to lower frequencies [Régnier et al. 2013], final relationships that provided transfer functions with fundamental frequency closer to the H/V frequency peaks were selected.The comparison between the final transfer functions and the H/V curves is displayed on the right-hand side of Figure 12; the empirical relationships selected were derived by Dikmen [2009] (see Table 4).Final V S values vary from 131 m/s to 158 m/s for anthropogenic deposits, from 132 m/s to 170 m/s for silty and clayey alluvium (Baixa southern zone), and from 105 m/s to 170 m/s for sandy alluvium (Baixa northern and central zones).The unit weight was set equal to 16 kN/m 3 for anthropogenic deposits and to 18 kN/m 3 for alluvial deposits, which are in the range of measured values for the same types of materials [Almeida 1991].For the overconsolidated Miocene material, V S velocities of 400 m/s for sand and sandstone, and 450 m/s for clay and limestone, were assumed; the unit weight was kept equal to 20 kN/m 3 .The bedrock properties were not changed.Table 6 presents the initial and final V S values.
These values should be checked against experimental measurements.In the Baixa area, only one direct measurement was performed (cross-hole test) giving values that are slightly higher, between 150 m/s and 240 m/s [LNEC 1998].Geotechnical studies for site characterization before the construction of the Vasco da Gama Bridge, which crosses the Tagus estuary north of Lisbon, were performed in the Tagus alluvium about 9 km from Baixa [Oliveira et al. 1997].This alluvium was divided into 6 sub-units, 4 of them with composition and geotechnical properties similar to the Baixa alluvium.Seismic cross-hole experiments gave shear wave velocities for these 4 sub-units varying from 51 m/s to 348 m/s.The usual range of V S values for the  et al. [2005] obtained a value of 110 m/s for the shear wave velocity of an alluvial deposit 7 m thick, also close to the Tagus River, a little further north of Lisbon, from spectral analysis of surface waves (SASW).These experimental values are also summarized in Table 6.
Considering the wide range of measured and calculated values from in situ experiments, the values estimated for downtown Lisbon, using the mentioned empirical relationship, seem realistic.

Spectral response of downtown Lisbon
To produce a map of the spectral response of the Baixa area, a set of 13 cross-sections parallel to the river bank and regularly spaced at 100 m intervals was produced.For each cross-section, a set of soil profiles spaced at 50 m intervals were defined, using data retrieved from the subsurface model.A total of 256 soil profiles were uniquely identified (Figure 13A).The depth and thickness of the anthropogenic deposits and of the alluvial deposits were defined using the geological 3D model, as well as the identification of the bedrock geological formation (Figure 13B).
Two sets of 10 transfer functions, corresponding to the near-field seismic motions and to the far-field seismic motions, were computed for each soil profile.For each soil profile the transfer functions for the two types of seismic motion were not very different (see Figure 13C,D).So, it was decided to consider the mean transfer function, for each soil profile and for each type of seismic motion (near-field and far-field), to present and discuss the results.
Figure 14 shows the obtained fundamental frequencies, as well as the corresponding amplification factors, for each type of seismic motion.For both scenarios it can be seen that the fundamental frequency of the Baixa area lies between 1.2 Hz and 2 Hz in the middle of valley, reaching 3 Hz near the edges where the anthropogenic and alluvial deposits have less expression.Higher frequencies, up to 10 Hz, are observed outside the valley over the Miocene formations.This result is in accordance with the predominant frequencies of the first microzonation map obtained by ambient vibration analysis for the Lisbon town by Teves- Costa et al. [1995].The amplification factors for these funda-LISBON DOWNTOWN SHALLOW FORMATIONS CHARACTERIZATION It was expected that some correlation between the alluvium thickness and the fundamental frequency of the soil would be observed.However this is not very pronounced.In detail, it is possible to observe that the fundamental frequencies for the near-field motion seem to better represent the thickness of the alluvium deposits (Figure 14A); however, the variability of the thickness of the anthropogenic deposits can mask the expected effect.Comparing Figure 14C and D it is possible to observe that the amplification is a little lower in the middle of the valley for the far-field seismic motion and a little higher in the southern zone for the near-field motion.However, as mentioned, these differences are very small (less than or equal to 0.4).As a whole, this is a very consistent result.
The spectral responses at 1 Hz and 2.5 Hz are displayed in Figure 15.It is clear that these responses are very similar for both seismic motions.The spectral amplitude at 1 Hz (Figure 15A,B) reaches a maximum of 4 in the middle of the valley, and it seems to correlate well with the thickness of the alluvial deposits.The spectral amplitude at 2.5 Hz (Figure 15C,D) is close to 1 in the central zone of the valley, reaching 3.5 in the northern zone, and it exceeds 5 in some areas in the southern zone of the valley.This effect could be associated with a local increase in the thickness of the superficial deposits, in the southeastern areas, whereas in the southwestern zone this effect is related to the response at a single point that has an abnormal thickness of anthropogenic deposits and it should be further investigated.
The analysis of the spectral amplitude at 2.5 Hz, which is similar both for near-and far-field seismic mo- tions, is very important because this frequency is within the range of the natural frequency of the Baixa building stock which varies from 2.3 Hz to 3 Hz [Oliveira 2004].

Conclusions
In this paper, the use of the Lisbon geological and geotechnical database, developed in the project Geo-SIS_Lx (http://geosislx.cm-lisboa.pt), for site characterization and microzonation purposes has been tested.The identification of some inconsistencies and some shortcomings in data organization were identified during data processing (only using the database with different objectives enables the identification of its weaknesses).On the one hand, this limited the accuracy of the results and identified the need to use more efficient approaches for data processing.However, on the other hand, this contributes to the improvement of the database that has been continuously updated.As expected, the results show spatial variability conditioned by the irregular ge-ographical distribution of the data.
Aware of these limitations, the geotechnical characterization and the definition of 1D soil profiles were performed making use of the information included in the database.Shear-wave velocities for the different layers were estimated using empirical correlations between N SPT and V S from the literature.The selection of the most suitable correlation was performed with the help of ambient vibration analysis (H/V curves).Dikmen [2009] relationships were found to be the most suitable for Baixa surface materials and the final V S values are in the range of those obtained during experimental tests performed at different sites in the Tagus estuary for the same types of materials.All these values should be checked with experimental measurements or complementary techniques performed in situ.In spite of the constraints, this methodology allowed the estimation of the physical properties of the shallower formations, and it seems to be applicable in re- gions where no geophysical data exist but geotechnical data are commonly available.
Simple 1D equivalent linear analysis was undertaken, despite the possible existence of non-linear and/or 2D effects.The lack of knowledge about the non-linear dynamic parameters of Baixa's soils made the choice of this simple approach necessary.The numerical simulations were performed using synthetic accelerograms due to the fact that there are no real strong motion records in the Lisbon area.The results show that the fundamental frequency of Lisbon downtown shallow formations lies between 1.2 Hz and 2 Hz, reaching 3 Hz at the edges of the valley, with amplification factors of 3.8 to 4 in the northern and central zones of the valley and reaching 5 in the southern zone.These results are in agreement with the natural frequencies derived from ambient vibration records [Teves-Costa et al. 1995] and with the microzonation study of Lisbon performed by Teves- Costa et al. [2001] using numerical simulations: natural frequency for the Baixa area was about 2 Hz (or less) and the amplification factors can reach 5.However, these previous studies were done for the whole of Lisbon and only two to four simulated/sampled points were in Baixa itself.The detail of this study was much higher, enabling identification of differences in the Baixa area.
Spectral responses at 1 Hz and 2.5 Hz are similar for both near-field and far-field seismic motions.This can be conditioned by the seismic input motions used in the computations.The use of real accelerograms and the introduction of non-linear models could change the results and improve the simulations.However, the main goal of the paper was to show the applicability of the geological and geotechnical Lisbon database on microzonation studies and, for this, it was not imperative to compute an accurate response for the Baixa's shallower formations.
Taking into consideration that most of the local building stock has natural frequencies between 2.3 Hz and 3 Hz, and although a simple approach was used, resonant effects during a future earthquake that could produce an increase in the estimated level of damage can be expected.Whatever the seismic source considered, it is possible to predict that the natural frequency of Lisbon downtown (Baixa) will play an important role during the next earthquake in Lisbon.

Figure 2 .
Figure 2. Left: Surface geology of the Baixa area (adapted from Pais et al. [2006]).The central part is filled by alluvial sediments covering Miocene formations (see lithological composition in the text).Right: Digital terrain model (DTM) obtained from a 1:1000 survey scale.Dashed black lines separate the three defined zones (north, central and south).

Figure 4 .
Figure 4. Distribution of N SPT values in the anthropogenic deposits for the three zones.Left: histograms of N SPT for depths less and greater than 5 m.Right: distribution of N SPT (< 50) with depth.

Figure 5 .Figure 6 .
Figure 5. Distribution of N SPT values in the alluvial deposits for the three zones.Left: histograms of N SPT for different depths.Right: distribution of N SPT (< 50) with depth for clayey and sandy materials.Percentiles lines (in black) and quartiles lines (in red) are also presented.

Figure 7 .
Figure 7. Synthesis of the procedure for estimating the Vs values for each layer.The empirical relationships used are presented in Table4.

Figure 8 .
Figure 8. Location of the ambient vibration measurements (M) and the closest geotechnical boreholes (S).Boreholes used to calibrate the V S values are identified.The dashed black lines separate the three defined zones (northern, central and southern).

Figure 9 .
Figure 9. H/V curves computed for each site.

Figure 11 .
Figure 11.Response spectra for the 10 seismic motions computed for each source: (left) near-field seismic motion; (right) far-field seismic motion.The mean of the 10 spectra are presented in bold (black curve).Bold red curve: Portuguese seismic code response spectra for 475 year return period.

Figure 12 .
Figure 12.Comparison between the computed transfer functions and the H/V curves obtained from ambient vibration analysis.Left: transfer functions computed with the first estimated V S values.Right: transfer functions computed with the final estimated V S values.

Figure 13 .
Figure 13.Location of the 256 soil profiles for which the transfer functions were computed (A).Example of one cross section performed to define the soil profiles (B).Transfer functions computed for 10 near-field (C) and 10 far-field (D) seismic motions for an arbitrary profile.The black bold curves are the mean transfer functions for each type of motion.

Figure 14 .
Figure 14.Fundamental frequencies (top) and the corresponding amplifications (bottom) for a near-field seismic motion (left) and a far-field seismic motion (right).

Figure 15 .
Figure 15.Spectral amplitude at 1 Hz (top) and 2.5 Hz (bottom) for a near-field seismic motion (left) and a far-field seismic motion (right).

Table 4 .
a and b values proposed by several authors for different types of soils and stratigraphical contexts.

Table 3 .
Median of N SPT values of the different lithologies in the Miocene formations.

Table 5 .
Peak frequency (F 0 ) and corresponding amplitude (A 0 ) of the H/V curve obtained at each site.Corresponding standard deviations are also presented.

Table 6 .
Freitas et al. [2014] values obtained from N SPT relationships (three first rows) and V S values obtained from field experiments (last row).All V S values are in m/s.(*)At depths greater than 30 m, the Miocene formations were considered as bedrock.(**)Singleexperimentperformedfor a 7 m thick alluvium layerTagus estuary alluvium lies between 100 m/s and 230 m/s, often with higher values for the deeper layers[Lopes 2005].Freitas et al. [2014]obtained values from 140 m/s to 290 m/s from cross-hole experiments performed in Tagus alluvium at a site in the Alcântara Valley, located approximately 3.4 km west of Baixa.Lopes