1 Background on liquefaction assessment methods

Different approaches to the assessment of the liquefaction potential have been proposed. The most common approach is the “Simplified Procedure”, originally proposed by Seed and Idriss (1971), which is also recommended by Eurocode 8 or EC8 (CEN 2010). According to this procedure, the factor of safety against liquefaction is computed from the ratio between the cyclic resistance ratio (CRR) and the cyclic stress ratio (CSR), as in Eq. 1. The CRR refers to the resisting capacity of the soil to liquefy, while the CSR corresponds to the design seismic action at a specific location in depth.

$$FS_{liq} = \frac{CRR}{CSR}$$
(1)

The liquefaction analysis framework proposed by Boulanger and Idriss (2014) was adopted, which is based on the simplified procedure proposed by Seed and Idriss (1971) and uses the parameters from previous works, namely rd from Idriss (1999), Kσ from Idriss and Boulanger (2004, 2010) and the implementation of the fines content estimates from CPT (Idriss and Boulanger 2004, 2010). In this approach, the resistance values from SPT and CPTu are adjusted to incorporate the effect of fines content. Table 1 presents a summary of the expressions for computation of the governing parameters used in this analysis, as well as the respective references, to obtain the normalized CSR and the respective adjustment parameters.

Table 1 Calculation of CSR and adjustment parameters adopted in the present work
Full size table

On the other hand, the cyclic resistance ratio (CRR) can be estimated from lab and in situ test results. The standard penetration tests (SPT) and cone penetration test (CPT) are particularly convenient, given the extensive worldwide database and past experience. Moreover, the use of the flat dilatometer test (DMT) has been developed in the last two decades, stimulated by the recognised sensitivity of the horizontal stress index KD to a number of factors which are known to increase liquefaction resistance (difficult to sense by other tests), such as stress history, prestraining/aging, cementation, structure, and by its correlation with relative density and state parameter (Monaco et al. 2005). Shear wave velocities also provide a reliable assessment of liquefaction resistance of soils, since both depend on similar factors, namely confining stresses, soil type, void ratio and relative density (Andrus et al. 2004).

In this work, the proposals of Boulanger and Idriss (2014) based on SPT and CPT have been adopted (Eqs. 2 and 3), where (N1)60cs and qc1Ncs correspond to normalised equivalent clean sand values, as suggested by Idriss and Boulanger (2004, 2010). According to these authors, a clean sand is considered to have a fines content (FC) below 5%. It should be noted that the introduction of the FC in these approaches reflects its importance in the liquefaction susceptibility of the soil. However, the estimate of FC based on SPT tests can be ambiguous and may lead to inaccurate results of CRR especially for FC below 25%. Based on Idriss and Boulanger (2004, 2010), a correspondence between soil type and FC has been established, as detailed below (Sect. 4.1).

$$CRR_{7.5} = \exp \left( {\frac{{(N_{1} )_{60cs} }}{14.1} + \left( {\frac{{(N_{1} )_{60cs} }}{126}} \right)^{2} - \left( {\frac{{(N_{1} )_{60cs} }}{23.6}} \right)^{3} + \left( {\frac{{(N_{1} )_{60cs} }}{25.4}} \right)^{4} - 2.8} \right)$$
(2)
$$CRR_{7.5} = \exp \left( {\frac{{q_{c1Ncs} }}{113} + \left( {\frac{{q_{c1Ncs} }}{1000}} \right)^{2} - \left( {\frac{{q_{c1Ncs} }}{140}} \right)^{3} + \left( {\frac{{q_{c1Ncs} }}{137}} \right)^{4} - 2.8} \right)$$
(3)

For DMT-based liquefaction analyses, the Marchetti (2016) CRR-KD curve has been used. Since the effects of higher fines content have not yet been fully investigated and clearly established, all the DMT triggering curves apply to clean sands. Therefore, the CRR is defined by combining the Idriss and Boulanger (2006) CRR-Qcn correlation and the Robertson (2012) average Qcn-KD interrelationship (Eq. 4), where Qcn is the normalized cone resistance. A combined correlation for estimating CRR based on Qcn and KD (Eq. 5) was also obtained by Marchetti (2016), by adopting the geometric average between a first CRR estimate obtained from Qcn (Eq. 4) and a second CRR estimate obtained from KD (introducing KD into Eq. 4).

$$CRR_{7.5} = \exp \left( {\frac{{Q_{cn} }}{540} + \left( {\frac{{Q_{cn} }}{67}} \right)^{2} - \left( {\frac{{Q_{cn} }}{80}} \right)^{3} + \left( {\frac{{Q_{cn} }}{114}} \right)^{4} - 3} \right),\;{\text{where}}\;Q_{cn} = 25 \cdot K_{D}$$
(4)
$${\text{Average}}\,{\text{CRR}} = \left[ {\left( {{\text{CRR}}\,{\text{from}}\,Q_{cn} } \right) \cdot \left( {{\text{CRR}}\,{\text{from}}\,K_{D} } \right)} \right]^{0.5}$$
(5)

For the assessment of liquefaction resistance of soils based on shear wave velocities, two methodologies have been adopted, namely those proposed by Andrus and Stokoe (2000) and Kayen et al. (2013). Andrus and Stokoe (2000) follow the same approach of the simplified procedure, with CRR computed from the stress-corrected shear wave velocity in depth (VS1), as follows:

$$CRR = \left[ {0.022 \cdot \left( {\frac{{K_{a1} V_{S1} }}{100}} \right)^{2} + 2.8 \cdot \left( {\frac{1}{{V_{S1}^{*} - K_{a1} V_{S1} }} - \frac{1}{{V_{S1}^{*} }}} \right)} \right] \cdot K_{a2} ,\;{\text{where}}\;V_{S1} = V_{S} \cdot \left( {\frac{{p_{a} }}{{\sigma_{v0}^{{\prime }} }}} \right)^{0.25}$$
(6)

where VS1 is the normalised shear-wave velocity; Ka1 and Ka2 are ageing correction factors on VS1 and CRR, respectively, both corresponding to 1 for uncemented recent soils; \(V_{S1}^{*}\) is the upper boundary value of VS1 for liquefaction occurrence; pa is the reference atmospheric pressure (= 100 kPa) and \(\upsigma_{{\text{v0}}}^{\prime }\) is the initial effective overburden stress.

On the other hand, Kayen et al. (2013) developed probabilistic correlations, based on a vast database of well-documented case histories, for VS-based probabilistic and deterministic assessment of liquefaction susceptibility. In this paper, the deterministic approach has been employed for a liquefaction probability (PL) of 15%, using the equations provided below. The respective factors of safety are computed, as before, as the ratio of the soil capacity to resist liquefaction at PL (15%) and the corresponding seismic demand, CSR.

$$\begin{aligned} & P_{L} = \Phi \left\{ { - \frac{{[(0.0073 \cdot V_{s1} )^{2.8011} - 1.946 \cdot \ln ({\text{CSR}}) - 2.6168 \cdot \ln (M_{w} ) - 0.0099 \cdot \ln (\sigma_{v0}^{{\prime }} ) + 0.0028 \cdot ({\text{FC}})]}}{0.4809}} \right\} \\ & {\text{CRR}} = \exp \left\{ {\frac{{[(0.0073 \cdot V_{s1} )^{2.8011} - 2.6168 \cdot \ln (M_{w} ) - 0.0099 \cdot \ln (\sigma_{v0}^{{\prime }} ) + 0.0028 \cdot {\text{FC}} - 0.4809 \cdot \Phi^{ - 1} (P_{L} )]}}{1.946}} \right\} \\ \end{aligned}$$
(7)

Alternative approaches to the assessment of liquefaction potential have been suggested, mainly focusing on estimates of liquefaction-induced damages, based on quantitative liquefaction risk indexes, namely the Liquefaction Potential Index (LPI) and the Liquefaction Severity Number (LSN). Originally developed by Iwasaki et al. (1978), LPI combines the safety factor with depth, z, down to 20 m. Iwasaki et al. (1982) classification was adopted, as indicated in Table 2, since it is also implemented in CLiq® and the differences with other classifications are minor. The adopted colour code relative to each LPI class is also included in the table.

Table 2 Classification of liquefaction potential based on LPI (after Iwasaki et al. 1982)
Full size table

Tonkin & Taylor (2013) developed another quantitative indicator of the liquefaction-induced damages, the Liquefaction Severity Number (LSN). This index represents the expected damage effects of shallow liquefaction on direct foundations, based on post-liquefaction volumetric deformations, associated with reconsolidation settlements. Using this approach, the liquefaction severity can be classified in terms of expected damage, according to Tonkin & Taylor (2013), as in Table 3, where the adopted colour scheme is also shown.

Table 3 Liquefaction severity and damage based on LSN (Tonkin & Taylor 2013)
Full size table

2 Selection of the pilot site

2.1 Seismicity and liquefaction zonation of Portugal

Portugal’s mainland and its Atlantic coast are located on the western and southern margins of the Iberian Peninsula. The seismicity of the Portuguese territory is heterogeneous and is classified according to regions with distinct seismic behaviour. Seismicity increases in intensity from North to South, with a spatial distribution concentrated in the South and the Atlantic margins. According to existing records, earthquake epicentres are concentrated near the city of Évora, in the Lisbon region, in the Lower Tagus River Valley region, and along the Algarve coast (Ferrão et al. 2016). The greater Lisbon area is probably the zone with greater seismic risk, coincidently where the capital and largest city of Portugal is located. It is affected by the occurrence of large moment magnitude (Mw> 8) distant earthquakes and of medium magnitude (Mw> 6) near earthquakes (Azevedo et al. 2010). An example of a distant event is the 1755 earthquake (Mw > 8.5) generated in the Eurasian-Nubia plate boundary zone. However, local intraplate (Mw≈ 6–7) earthquakes have occurred more frequently, in 1344, 1531 and 1909.

The Portuguese National Annex of the European Standard for Design of structures for earthquake resistance, EN 1998-1, Eurocode 8 or EC8-NA (CEN 2010), established the seismic zonation of continental Portugal, as shown in Fig. 1. This zonation considers two types of seismic actions: Type 1 and Type 2. Type 1 refers to a “distant earthquake” scenario, corresponding to greater magnitude earthquakes at longer distances (with epicentre in the Atlantic region), while Type 2 refers to a “near earthquake” scenario, associated with moderate magnitude earthquakes at close distance (with epicentre in the continental territory). According to EC8, seismic hazard is described in terms of the peak ground acceleration in type A ground (rock), agR. The values of agR for each zone and seismic action type are included in Fig. 1. Following these seismic actions, examples of liquefaction assessment by in situ tests are available in the Algarve (e.g. Rodrigues et al. 2016).

Fig. 1
figure 1

Seismic zonation of Portugal mainland: a Action Type 1; b Action Type 2 (adapted from EC8)

Full size image

Earthquake Induced Liquefaction Disasters (EILDs) are responsible for significant additional structural damage and casualties, particularly in zones where specific geologic, geomorphological, hydrological and geotechnical characteristics indicate liquefaction potential of soils below structures (LIQUEFACT 2017). The presence of thick profiles of recent alluvial sandy deposits in a high seismicity area is a good example of the combination of the necessary liquefaction triggering conditions.

Information regarding seismic activity in Portugal only started being collected after the 1755 earthquake. For older events, the available data only include the testimonials of people experiencing large earthquakes. Since these are mostly subjective descriptions of ordinary people, it has been hard to assess the level of reliability of this information with reference to liquefaction; this means that doubts arise in several circumstances as to whether the phenomenon actually occurred. For this reason, as discussed by Jorge (1993), data in the catalogue are classified in terms of quality of information and localization of the source. In particular, the categories are ‘certain’, ‘doubtful’, ‘very doubtful’ and ‘credible’ liquefaction. The first three categories refer to descriptions directly related to liquefaction, with more or less certainty. The ‘credible liquefaction’ category provides information, not directly describing but potentially related to the liquefaction phenomenon. Following this approach, Jorge and Vieira (1997) identified in the map shown in Fig. 2, the locations of historical liquefaction events coupled with a reliability classification. This is considered the most reliable source of information on the evidences of the liquefaction phenomenon in Portugal. From the earthquake catalogue, Jorge and Vieira (1997) identified six earthquake events associated with liquefaction, as indicated in Fig. 2: 26/01/1531 (M = 7.1); 01/11/1755 (M = 8.5); 31/03/1761 (M = 7.5); 12/01/1856 (M = 6.0); 11/11/1858 (M = 7.2) and 23/04/1909 (M = 6.6). The details of these events are listed in Portuguese catalogues, including the magnitude, macroseismic intensity and coordinates of the epicentre. The locations where liquefaction occurred as well as the epicentral distances were not reported, but were assumed, according to the site where liquefaction was observed, even considering the large degree of uncertainty. This uncertainty was reflected in the calculation of the estimated epicentral distances, however the error made in this computation was taken into account.

Fig. 2
figure 2

Location of liquefaction events associated with historical earthquakes (adapted from Jorge 1993). Note: “Very doubtful” occurences have been removed from the original map. Permission granted by the author

Full size image

A liquefaction potential zonation map of Continental Portugal was developed by Jorge (1993) and further discussed by Jorge and Vieira (1997). This zonation map was derived from the superposition and generalization of two basic maps: the liquefaction ‘opportunity’ map and the liquefaction susceptibility map. For the greater Lisbon area, a more detailed representation was produced, which evidenced the high liquefaction potential of that region, as illustrated in Fig. 3.

Fig. 3
figure 3

Liquefaction zonation map (Jorge 1993; Jorge and Vieira 1997). Permission granted by the author

Full size image

After the identification of the ‘high to very high’ liquefaction susceptibility areas in Fig. 3, mostly along the Lower Tagus Valley, the collection and analysis of existing geotechnical data in that region was carried out, mainly covering the municipalities of Vila Franca de Xira, Benavente, Montijo and Barreiro.

2.2 Collection and analysis of existing information

For the selection of the location of the pilot site, the investigation was initiated with the collection of existing geological and geotechnical information in the metropolitan region of Lisbon along the Lower Tagus River Valley area. With the collaboration of numerous public institutions, governmental agencies, private companies, contractors and design offices, a considerable volume of geotechnical data was assembled. After careful inspection, 95 geotechnical reports were selected for analysis, in a total of more than 350 test results. The majority of these tests, about 72%, corresponded to SPT and borehole logging, in a total of 257 test results, but also included 70 CPT(u), 12 DMT and 17 VS measurements (from SCPT, Cross-Hole or seismic refraction). Information on the position of the groundwater level at the time of testing was also available in most test reports.

These reports refer only to the North-East to South part of the Lower Tagus Valley in the Greater Lisbon, where quaternary sand deposits are expected, involving the municipalities of Vila Franca de Xira, Azambuja, Salvaterra de Magos, Benavente, Alcochete, Montijo and Barreiro, mostly located along the left bank of the Tagus river and estuary. Important works associated to the construction of a major highway (A10), including a 12 km extension bridge and viaduct crossing the river Tagus and agricultural plains, have provided a wealth of information from extensive geological and geotechnical site characterisation tests, which were collected and analysed for the present research.

For the assessment of liquefaction susceptibility in this region, the peak ground acceleration amax was computed according to EC8-NA (CEN 2010), as summarised in Table 4.

Table 4 Calculation of amax for Vila Franca de Xira and Benavente, according to EC8-NA (CEN 2010)
Full size table

The analysis of the collected reports was carried out, according to the type of test, based on the previously described approaches to the assessment of liquefaction susceptibility. The classification of the liquefaction susceptibility of each soil profile was made, according to two criteria: (a) minimum factor of safety of 1.00; (b) minimum thickness of the liquefiable soil layer of 3 m. Consequently, three classes have been considered: low, moderate and high. For the purpose of geographical referencing and future microzonation, each test point was geographically located and colour-coded, according to the adopted colour scheme, introduced in Table 5. On a first approach, geo-referencing was made by introducing all coordinates on Google Earth®. In order to aid visual identification of liquefiable areas, the same colour code was associated with paddle icons for SPT data, diamond paddle icons for CPT data and target circles for CH (cross-hole) data, as schematically shown in Table 5.

Table 5 Susceptibility colour code used for existing data points, based on the factor of safety to liquefaction (FSliq)
Full size table

This colour classification of SPT, CPT and CH data points has been superimposed on the liquefaction zonation map in Fig. 3 (from Jorge 1993), as illustrated in Fig. 4.

Fig. 4
figure 4

Location of the geotechnical reports collected in the greater Lisbon area, superimposed on the existing liquefaction zonation map (from Jorge 1993)

Full size image

Despite some variability regarding liquefaction susceptibility, there is a substantial agreement between the general zonation map and the analysed data points. In effect, the red points in Fig. 4 are predominantly located in the area previously identified as having high to very high liquefaction susceptibility, mainly involving the municipalities of Vila Franca de Xira and Benavente.

2.3 Location of the pilot site

The area in the agricultural plains of the “Lezíria Grande de Vila Franca de Xira” was found to have the ideal geological, hydrogeological and geotechnical, as well as operational conditions, for constituting a research pilot site on liquefiable soils. The area of the pilot site was divided into zones, named Site Investigation (SI) points, identified by the respective number. Table 6 summarises the number, type and location of the tests performed at the pilot site and in each SI, and Fig. 5 indicates the testing locations in a map. The location of each type of tests was selected based on a geological and geomorphological interpretation of the site, described in detail in Viana da Fonseca et al. (2017) and Saldanha et al. (2018). The position of the groundwater level was measured in each testing location, which is particularly relevant for liquefaction analyses. An extensive series of microtremor measurements was also performed, complementary to these investigations, for the purpose of the liquefaction microzonation of the region, which will not be addressed in this paper.

Table 6 Tests performed in the pilot site
Full size table
Fig. 5
figure 5

Location of the site investigation (SI) points and of the main tests at the pilot site

Full size image

For the purpose of liquefaction susceptibility assessment from penetration tests, the analysis will focus on SPT, CPTu and DMT data. For Vs-based liquefaction analysis, direct measurements of SDMT and estimated values based on SPT, CPT and DMT results will be considered, since CH results were found to be unreliable due to equipment malfunctioning. On the other hand, surface geophysics results were applied for complementing the geological and geotechnical characterisation of the site, namely for layer detection, by effectively covering large areas. The predictions of shear wave velocities from the geotechnical tests were included, given its valuable contribution to liquefaction analyses, as detailed in Ferreira et al. (2018).

3 Characterisation of the pilot site

3.1 SPT results and preliminary liquefaction assessment

Two SPT tests were carried out in SI1 and SI7, respectively. High quality samples were collected in an adjacent borehole, using the Mazier sampler, for complementary laboratory studies. The SPT test results in the two locations in terms of (N1)60,cs are presented in Fig. 6, together with a simplified soil profile defined from the SPT results, as well as a preliminary analysis for liquefaction susceptibility using the simplified procedure, considering Type 1 and 2 seismic actions. The resulting factors of safety against liquefaction refer only to the sandy layers.

Fig. 6
figure 6

SPT-based assessment of liquefaction potential at the pilot site: a SI1, b SI7

Full size image

For clearer perception of the evolution of the factor of safety, FSliq, with depth, 3 coloured zones have been added, corresponding to values below 1.00 (red), between 1.00 and 1.25 (yellow) and above 1.25 (green). The value of 1.00 is conventionally, as previously stated, the minimum factor of safety; however, EC8 is more conservative, proposing a minimum FSliq value of 1.25, hence the transition area in yellow.

In the illustrated cases of SI1 and SI7 in Fig. 6, it is clear that thick sandy layers exhibit high to very high liquefaction susceptibility, except for a medium dense sand layer at 5–8 m in SI1. Based on these SPT results, a preliminary liquefaction analysis of each location can be made. At SI1, a non-liquefiable clayey crust of about 2 m is followed by a 20 m thick liquefiable sandy layer, interbedded by a medium-dense sand layer between 5 and 8 m, after which a silty clay non-liquefiable layer was found. On the other hand, at SI7, the non-liquefiable clayey crust is 6 m thick and the liquefiable sandy layer is about 11 m thick, located between 6 and 17 m, followed by a clay layer. This analysis will be further discussed by comparison with other geotechnical data.

3.2 CPTu testing

In this pilot site, ten piezocone tests (CPTu) were performed. The tests were performed according to the ISO 22476-1.2012 (ISO 2012) and the normative procedures proposed by the TC16. The results were treated using the methodology of Boulanger and Idriss (2014) for soil liquefaction analysis, as previously introduced. The groundwater level was measured in each in situ test location, varying from 0.3 to 2.0 m. The in situ measured values were used in the calculations. Figure 7 shows an example of the CPTu results in three plots: (a) cone resistance (qc) and pore pressure (u2); (b) soil behaviour type index (Ic) and simplified soil profile; (c) liquefaction factor of safety (FSliq).

Fig. 7
figure 7

CPTu results in the pilot site at SI6

Full size image

The first plot (Fig. 7a) provides the basic information of the soil profile, allowing to distinguish the depths at which the soil layer is granular (higher cone resistance and pore pressure coincident with the hydrostatic line) or fine-grained (lower cone resistance and excess pore pressure). The Ic plot (Fig. 7b) illustrates a preliminary soil profile, based on the proposal of Robertson and Wride (1997); in addition, a simplified soil profile has been defined, by approximating the original Ic by constant values, where similar behaviour is expected. As proposed by Cubrinovski et al. (2017), the simplified soil profile considers: gravel and coarse sand (Ic ≤ 1.3); clean sand (1.3 ≤ Ic ≤ 1.8); sands with low fines content (1.8 ≤ Ic ≤ 2.1); silty sand, sandy silt and non-plastic silt (2.1 ≤ Ic ≤ 2.6); and, non-liquefiable silt or clay (Ic ≥ 2.6). This soil classification is different from the original classification proposal from Robertson (1990), updated by Robertson (2009), as it is focused on soil response with respect to earthquake-induced liquefaction. From this point of view, there is no distinction between silts, clays and organic or sensitive soils; instead, these soil types have been grouped together as non-liquefiable soils. On the other hand, sands have been sub-divided to account for different fines content: from clean sand to low FC sands, to silty sands, since liquefaction case histories suggest that small variations in fines content strongly influence liquefaction susceptibility. Finally, Fig. 7c illustrates the variation of the factor of safety against liquefaction, FSliq, in depth. Again, coloured zones have been included to ease identification of the critical layers: red for values below 1.00, yellow between 1.00 and 1.25 and green for values above 1.25.

In the case of SI6, shown in Fig. 7, the simplified Ic plot shows distinct soil layers, which can be clearly identified and summarised as follows: a top non-liquefiable layer about 7 m thick, followed by a 5 m thick clean sand layer down to 12 m, then a non-liquefiable layer down to 22 m and a deeper soil layer, consisting of sands with low fines content, again with high liquefaction susceptibility. It should be noted that, below 20 m, liquefaction evaluation is less reliable and should be analysed by means of specific site response analyses, since the uncertainty in some of the computation factors becomes larger (Boulanger and Idriss 2014).

A general overview of 6 CPTu at different locations within the pilot site are plotted in Fig. 8. Thick liquefiable layers can be identified in all of these profiles, despite the significant variability in depth among the different testing locations.

Fig. 8
figure 8

Selection of CPTu results in the pilot site at SI1, SI3, SI4, SI5, SI7 and SI12

Full size image

3.3 SDMT testing

In this pilot site, four Seismic Flat Dilatometer tests (SDMT) were performed in the first stage, according to Eurocode 7-Part 3 recommendations and ISO/TS 22476-11. However, at SI1, operational problems were experienced, having reached a depth of only 4 m. The seismic dilatometer is an extension of the traditional DMT, introduced by Marchetti (1980) with a seismic module implemented above the steel blade (Marchetti et al. 2008). The seismic module consists of an instrumented rod connected between the DMT blade and the rods, equipped with two horizontal geophones spaced 0.50 m, for measuring shear wave velocities, VS. The presented DMT results were obtained directly from the usual DMT interpretation formulae according to Marchetti (1980) and Marchetti et al. (2001). In this respect, Fig. 9 shows the profiles of the material index ID (indicating soil type) and of the horizontal stress index KD (related to the stress history) together to the corresponding liquefaction safety factor FSliq at the investigation sites, namely SI7, SI8 and SI9. At each of the sites, FSliq was calculated using the Marchetti (2016) CRR-KD correlation (DMT data only), while at SI7 DMT and CPT results were combined, according to the Marchetti CRR-KD-Qcn formulation.

Fig. 9
figure 9

SDMT results in the pilot site: a SI7, b SI8, c SI9

Full size image

Comparing with CPT results, DMT liquefaction assessment also detects a non-liquefiable silty-clayey crust of 3 to 6 m thickness, depending on the site investigation location, before encountering the sandy and silty-sandy deposits that provide most of the liquefaction down to 14–16 m depth. The combined use of CPT and DMT in SI7 follows the same DMT tendency, even though the liquefaction susceptibility appears to be much lower, probably due to the presence of interbedded layers that do not allow a correct coupling of DMT and CPT data at certain depths.

3.4 Geophysical investigations

Seismic wave velocities were measured in the pilot site by means of geophysical surface wave methods, namely via seismic refraction (SR), spectral analysis of surface waves (SASW), as well as in borehole tests, such as the seismic dilatometer (SDMT) and cross-hole (CH) tests. For the purpose of liquefaction assessment, the results of seismic refraction tests were also considered, despite being better suited for profiling and layer detection, by identifying changes in seismic wave velocities in depth. However, borehole seismic tests are considered more reliable and detailed and were analysed, based on direct measurements of VS, as well as its prediction from penetration tests. In effect, from the variety of in situ penetration tests performed at the pilot site, it was possible to obtain predictions of VS from correlations with SPT, CPTu and DMT test results. For the SPT-Vs correlations, the proposals of Wair et al. (2012) for different soil types were used, which also take into account the effective vertical stress at each depth of the soil profile. For CPT-VS correlations, the proposals of Hegazy and Mayne (1995), Mayne (2006), Andrus et al. (2007), Robertson (2009) and Monaco et al. (2005) were analysed. As detailed in Ferreira et al. (2018), the prediction proposed by Robertson (2009) was found to be the most appropriate for these soils. For VS predictions based on DMT, the proposal of Marchetti et al. (2008) was adopted. Amoroso (2014) demonstrated that the DMT-based predictions are more consistent than those based on the CPT. For this analysis, Fig. 10 presents the results obtained at SI1 and SI7, in terms of measured VS via SR and SDMT, as well as estimated VS profiles based on:

Fig. 10
figure 10

Measured and estimated VS results and respective FSliq: a SI1, b SI7

Full size image

  • Wair et al. (2012): SPT (W 2012)

  • Robertson (2009): CPT (R 2009)

  • Marchetti et al. (2008): DMT (M 2008)

Figure 10 also includes the computed factors of safety against liquefaction using the two distinct approaches: Andrus and Stokoe (2000) and Kayen et al. (2013) for the two seismic actions (T1 and T2), taking into account the estimated fines content.

In both locations, the results show significant approximation between measured and predicted VS values. As expected, seismic refraction provides simplified profiles, assuming a stiffness increase with depth, which is not always the case in SI7, as shown in the SDMT profile. DMT-based predictions are remarkably similar with SDMT measurements, which demonstrates the good performance of Marchetti et al. (2008) proposal. As evidenced by Amoroso (2014), DMT-based predictions appear to be more consistent than those based on the CPT considering that DMT-VS correlations include the horizontal stress index KD, noticeably reactive to stress history, prestraining/aging and structure, scarcely detected by cone tip resistance qc from CPT. On the other hand, CPT-VS predictions are subjected to the additional uncertainty arising from the selection of which one of the numerous existing correlations is adopted, depending on geological age, cementation, effective stress state. With regard to the liquefaction susceptibility assessment, the obtained FSliq values are indicative of very thick liquefiable soils at both locations. However, in SI7, there are significant discrepancies in the results, which are likely linked to the soil type consideration and estimate of fines content, based on FC, IC and ID, respectively.

4 Analysis and discussion

4.1 Combining field and laboratory data

For comparing the results of these field tests, especially in terms of liquefaction susceptibility assessment, two site investigation locations were selected: SI1 and SI7. In order to specifically address the impact of soil type, especially fines content, the laboratory results of grain size distribution and plasticity, obtained on SPT samples, have been integrated in the SPT-based liquefaction assessment. Figure 11a shows the first 20 m of the simplified soil profile in SI1, and Fig. 11b presents the comparison between the SPT-estimated and laboratory measured fines content and plasticity index. The SPT-estimated FC were defined, considering the proposal by Idriss and Boulanger (2004, 2010), and based on the lithological description of the SPT log (below 5% for clean sand; 5–10% for sand with fines; 10–30% for silty sand; above 30% for fine non-liquefiable soils). In addition, the soil type parameter IC from CPTu, with a cut-off at 2.35 (average value between 2.1 and 2.6) corresponding to the midpoint between silty sands and non-liquefiable soils, is provided in Fig. 11c. The combination of field and laboratory data enabled to redefine the soil profile, by identifying the sandy layers, potentially susceptible to liquefaction, as illustrated in Fig. 11d.

Fig. 11
figure 11

SI1 results: a SPT simplified soil profile based on lithology; b SPT-estimated and lab-measured fines content; c simplified IC for liquefaction; d revised soil profile. Note: where PI is not specified means non-plastic (NP) soil

Full size image

The most striking observation, at first glance, is that the revised soil profile is more complex and stratified than the simplified profile derived from the lithological description of the SPT. This is due to the laboratory measurement of fines content, which provides a very different outline of the soil type, as shown in Fig. 11b. In this figure, the plasticity indexes at different depths are also included, which are relevant in liquefaction analyses (Boulanger and Idriss 2014). It is clear that the SPT test alone fails to identify the existence of clay/silt layers interbedded with the sand deposits, which have a very significant impact in the liquefaction response of the profile, so the use of complementary information, especially from the laboratory analysis of the collected SPT samples, is highly beneficial.

Based on this revised soil profile and using the laboratory-measured fines content information, the factors of safety against liquefaction obtained from SPT, as well as from the estimated VS-SPT and VS-CPT profiles (Kayen et al. 2013 approach) have been recalculated, as indicated in Fig. 12, from which the critical layers can be easily identified. In addition, the CPTu profile has also been revised, by removing FSliq values for IC above 2.35 (midpoint between silty sands and non-liquefiable soils). For clarity, the results from seismic refraction tests were not included in this comparison.

Fig. 12
figure 12

Identification of critical layers in SI1 taking FC into account: a revised soil profile, b SPT FSliq, c revised Ic, d CPTu FSliq, e VS FSliq

Full size image

In contrast with the FSliq profiles in Figs. 6a and 8 (SI1), the consideration of the adjustments in fines content enabled a clearer distinction between layers, particularly useful in the identification of the critical ones. In this case, a layer of moderate to low liquefaction susceptibility was also detected. Despite the larger scatter in the VS-based FSliq profiles, the same critical layers can be recognised, mainly for T1 seismic action. For the lower magnitude seismic demand (T2), the VS-FSliq profiles are substantially higher, suggesting that the computed DWF (Distance Weighting Factor, similar to MSF) may need further adjustments.

In sum, in this location, three highly liquefiable layers have been identified, between 2 and 5 m, then at 10–12 m, and then from 13 to 17 m. A very thin deep liquefiable layer was also found nearly at 20 m, which effect at the surface is expected to be negligible. Since the SPT and CPT tests were performed very close to each other, the discrepancies in the results can only be attributed to the nature and specificities of the in situ test, as it is necessarily the same soil profile. Since the CPT measurements are nearly continuous (every 1 cm), while the SPT was performed at every 1 m in depth, the observed differences are a reflection of the many intercalations of fine layers, which often are not visible in the SPT results. In fact, the CPT results show some points where the FS is high, as well as the SPT results. What is apparent from this comparative analysis is that the greater detail of the CPT is fundamental to identify these heterogeneous soil profiles, while the SPT may lead to a different perception of the soil profile.

For the second site at SI7, a similar analysis was performed, as outlined in Fig. 13, with simplified SPT soil profile of the first 20 m (Fig. 13a), the SPT-estimated (from the lithological description of the SPT log) and lab-measured fines content (Fig. 13b), CPTu soil type profile from IC ((Fig. 13c). Combining this information, a revised soil profile has been produced (Fig. 13d).

Fig. 13
figure 13

SI7 results: a SPT simplified soil profile based on lithology, b SPT-estimated and lab-measured fines content, c simplified IC for liquefaction, d revised soil profile. Note: where PI is not specified means non-plastic (NP) soil (for FC < 50%)

Full size image

In this case, the original soil profile has been converted into a simpler three-layered profile, despite the existence of thin interbedded layers of finer soil, as noted in the soil type description. The comparison between SPT-estimated and laboratory-measured fines content reveals clear differences, as before, particularly near the interface of the layers. The integration of this information in the revised computation of the factors of safety is illustrated in Fig. 14, which also includes the identification of the critical layers in terms of liquefaction susceptibility.

Fig. 14
figure 14

Identification of critical layers in SI7 taking FC into account: a revised soil profile, b SPT FSliq, c revised Ic, d CPTu and DMT FSliq, e VS FSliq

Full size image

In this location, the simplified soil profiles from SPT and CPT are relatively similar, with two clayey layers at the crust and below about 16 m, and a central critical zone. However, the estimate of the thickness of the sandy layers slightly differs: the SPT results identified about 10 m of liquefiable sands (between 5 and 15 m), while the CPT indicates about 7 m of sandy soils (from 7 to 14 m), with a few interbedded layers of fine soil. In turn, DMT results suggest that the liquefiable layer is about 9 m thick, located from 5 to 14 m in depth. As highlighted in the figure, the combination of these results suggests that it is reasonable to consider a thick liquefiable layer, approximately between 6 and 15 m. With regard to VS-based FSliq results, a good agreement with the previous plot is evident, especially after the FC adjustment obtained from the laboratory measurements (by comparison with the VS-FSliq profile in Fig. 10b). It is again discernible that the inclusion of soil type information, such as from laboratory analyses, is vital to obtain a reliable VS-based assessment of liquefaction susceptibility, clearly improving its capability for identifying liquefiable and non-liquefiable soil layers.

4.2 Overview of the liquefaction response of the pilot site

As discussed in the introduction, the use of alternative and quantitative liquefaction indexes is advocated, providing relevant information in terms of the damage induced by soil liquefaction. For this purpose, LPI and LSN values have been computed, from the field penetration test data, namely SPT, CPT and DMT. At first, it is worth comparing all the results obtained at the pilot site from CPT data, as presented in Fig. 15. In this figure, LPI and LSN have been calculated considering the two types of seismic actions and a coloured background shading has been included, based on the classification of Tables 1 and 2.

Fig. 15
figure 15

Severity damage based on LPI and LSN from CPTu at the pilot sites

Full size image

As shown in Fig. 15, LPI values fall on the high or very high liquefaction severity, except for SI2 and SI13, where the LPI is low. SI4 appears to be the location with the highest liquefaction susceptibility, in terms of LPI, but SI5 and SI12 are also classified as highly liquefiable. In sum, from LPI results, it can be concluded that the majority of testing points exhibit high (50%) to very high (30%) liquefaction severity. In turn, based on the LSN results in Fig. 15, greater surficial liquefaction-induced damages are expected in SI4 and SI5, however the values fall within the moderate to severe class, that is, below 40. In terms of the variability of LSN values, there is greater scatter in its classification, with about 20% of the testing points in each class. Since LPI and LSN are liquefaction severity indicators, some authors have proposed a parallelism between them, namely Wotherspoon et al. (2015), who made use of the observed superficial manifestations after the Christchurch series of earthquakes to establish the comparison. The proposed classification relationship is provided in Table 7.

Table 7 Classification of liquefaction severity and damage based on LPI and LSN
Full size table

However, the results in Fig. 15 do not fit well within the relationship between LPI and LSN proposed in Table 7, mainly because the LSN values are relatively low, classifying liquefaction severity at all testing locations as minor to moderate, in relation to the relative LPI, which indicates most testing locations as severely affected by liquefaction. Based on the available information, it is not yet possible to state which severity index is being poorly estimated at the pilot site, though it appears that LPI is over-conservative and LSN is possibly unconservative. This poor correspondence, also observed by Wotherspoon et al. (2015) and Cubrinovski et al. (2017), suggests that further studies are required, not only in terms of the liquefaction assessment procedures from which these indices are computed, but also to account for the configuration of the soil profile, namely the thickness of the crust, the depth and thickness of the liquefiable layers, as well as the relative distribution of liquefiable layers and interbedding with fine non-liquefiable layers (Millen et al. 2019). An adjustment based on these test results to the LPI versus LSN classification is also included in Table 7.

It is also interesting to compare these CPT-derived indexes with those from SPT, DMT, DMT combined with CPT tests, as well as from direct VS measurements, summarised in Table 8 for LPI and LSN. The results show considerable differences between the absolute values of LPI and LSN, according to the type of test from which these have been computed.

Table 8 Comparison of LPI and LSN values from SPT, CPTU and SDMT at SI1 and SI7
Full size table

These results suggest that the use of SPT data and VS measurements for LPI or LSN estimates may lead to significant deviation from realistic values, especially in the presence of interbedded layers of sands and silty clays, as in the present case. Both the original and revised values of SPT-FSliq have been included (SPT and SPT_lab FC, respectively) to demonstrate the positive impact of the use of laboratory analyses of SPT samples in the improvement of SPT-derived parameters. From a qualitative perspective, the values from SPT and CPTu indicate similar trends, with higher values at SI1. On the other hand, the values of LPI and LSN obtained from DMT and CPT predictions appear reasonably similar, while the combined use of DMT and CPT provides lower indexes, probably due to the abovementioned interbedded layers that does not allow the correct coupling of DMT and CPT data at each soil depth. Given the inadequacy of VS to distinguish between sandy and clayey soils, the use of VS-based liquefaction indexes should only be used when specific soil type information (grain size distribution and index properties from laboratory analyses or IC from CPTu) are available, otherwise these can be largely overestimated. The combination of VS results with other geotechnical data on soil type proved to be a reasonable alternative solution to overcome this limitation. However, the corresponding LPI values are still overestimated in comparison with those from CPTu.

5 Conclusions

A new pilot site in liquefiable soils has been setup in the Greater Lisbon area, which has provided a wealth of geological, geophysical and geotechnical data to be explored and analysed, mainly in terms of liquefaction assessment protocols. The selection of its location is discussed in detail, based on the collection and analysis of existing geological and geotechnical reports. The conventional approach to liquefaction susceptibility assessment, based on the simplified procedure applied to SPT, CPT, DMT and VS measurements, has been implemented, in terms of the factors of safety against liquefaction (FSliq). The investigated area is constituted by very heterogeneous soil profiles, with interbedded sand-silt–clay layers. In some locations, more homogeneous layers of sand were found and some critical layers were identified, at different depths. However, the profiles are generally very heterogeneous, which is why the use of different in situ tests is even more relevant. In both SI1 and SI7, thick potentially liquefiable layers were found, as well as in many others (see Fig. 8) so it can be concluded that the pilot site area is prone to liquefaction.

Due to the presence of interbedded layers of sand and clayey soils, some discrepancies were observed in the results, particularly from direct interpretation of SPT and VS results. This is a consequence of the lack of specific information on soil type, namely fines content, from these test results, which has a strong impact in the assessment of liquefaction susceptibility. To overcome these limitations, laboratory data from physical identification and grain size distribution obtained on SPT samples, were combined with field data, which considerably improved the convergence and the consistency of different test results. In effect, after the inclusion of laboratory measured fines content, it was possible to clearly identify the critical, highly liquefiable layers from the different tests. The analysis was complemented with alternative quantitative measures of the superficial damage induced by liquefaction, such as the Liquefaction Potential Index (LPI) and the Liquefaction Severity Number (LSN).

The main conclusion of this paper is that the use of different methodologies for the assessment of liquefaction susceptibility by means of in situ tests is beneficial, particularly if complemented with simple laboratory analyses of grain size distribution and consistency limits. This approach enabled to overcome the limitations of some of the approaches, particularly from SPT and VS measurements. For the case study of this paper, which involved sensitive loose granular soils, often interbedded with finer soil layers, the laboratory information proved to be of great value to eliminate some discrepancies obtained by the conventional method on SPT data and VS measurements. However, some discrepancies have not been resolved, evidenced by the LPI and LSN values, since the results from SPT_labFC are still considerably different from CPT results. The presence of many interbedded sand-silt–clay layers was found to compromise an accurate SPT evaluation of the liquefaction potential of the profiles, since the discrete 1-m data points of the SPT are often not representative. In short, the combination of these criteria enabled to identify the areas potentially most affected by liquefaction. Subsequent investigation campaigns are being carried out to refine the database and the results are currently being transferred to geo-statistical modelling software for the microzonation of the pilot site. Complementary information can be found in Viana da Fonseca et al. (2018), Ferreira et al. (2018), Saldanha et al. (2018) and Millen et al. (2019).