Geotechnical site characterization using multichannel analysis of surface waves: A case study of an area prone to quick-clay landslides in southwest Sweden

Quick-clay landslides are important geohazards in Sweden, Norway and Canada. While they have been studied using various geotechnical and geophysical methods, only a handful of seismic surveys have been reported for their studies. Here, we repro-cess active-source seismic data from a quick-clay landslide site in southwest Sweden to complement earlier studies of reflection imaging and first-break traveltime tomography with surface-wave dispersion analysis. Results suggest extremely low shear-wave velocities, even as low as 60–100 m/s. From a geotechnical perspective, this implies that the region classifies as a high-risk zone for landslides and construction purposes. High or anomalous values of Poisson’s ratio (or similarly P-and S-wave velocity ratio) depict a zone within the normally consolidated sediments that likely represents a coarse-grained layer, thus confirming earlier results from a number of boreholes drilled in the study area. Overall, the results presented further support to the previous hypothesis that the coarse-grained layer plays a major role in the formation and creation of quick-clay landslides in the study area. Additionally, an attempt to model the distribution of potential quick clays along one of the seismic profiles is performed through a combination of the modelled geophysical properties and soil textures. This study illustrates the potential of seismic methods, and how the integration of multiple geophysical properties and different data handling strategies can help to accurately characterize regions susceptible to quick-clay landslides.

information about the subsurface materials in areas prone to quick-clay landslides.
Multichannel analysis of surface waves (MASW) is one of the most applied non-invasive geophysical methods to obtain near-surface V S models.The combination of V S with other types of geophysical or geotechnical data can allow a more complete site characterization (Boiero et al. 2013;Sauvin et al. 2014;Malehmir et al. 2016;Brodic et al. 2018) compared with using single data sets.V S together with P-wave velocity (V P ) and density can infer other geotechnical parameters, such as shear modulus, bulk modulus, and Poisson's ratio, relevant, for example, for landslide characterization and engineering purposes (Israil and Pachauri 2003;Konstantaki et al. 2013;Petronio et al. 2016;Stokoe et al. 1994;Uhlemann et al. 2016).Normally, measurements of in situ ground stiffness should be performed during the design of a construction site and after the works are completed, in order to estimate, as accurately as possible, any change in the in situ stress-strain behaviour that could indicate permanent deformation or failure.The estimation of the in situ stress field along a 2D profile using seismic methods, without any disturbance caused by direct borehole measurements or other kind, is valuable when large amounts of earthworks are required (Matthews et al. 1997).
MASW is based on the dispersive behaviour of the Rayleigh and Love waves, where the velocity is frequency dependent when propagating in a layered media (Park et al. 1999).Depending on the type of MASW data, active or passive, the depth that can be reached goes from a few tens of metres (active) to a few hundred metres (passive).The frequency content can be broadened by combining active and passive data, through which higher resolution and deeper information can be obtained (Foti et al. 2007;Park et al. 2005).The final results of MASW are sets of 1D shear-wave velocity profiles, which are combined to produce a pseudo-2D velocity profile of the subsurface.
MASW studies do not require specifically designed data acquisition geometries or techniques (Neducza 2007;Konstantaki et al. 2013;Pasquet et al. 2015;Socco et al. 2009), although the investigation depth can be increased by using long spreads and low-frequency geophones (Pasquet et al. 2015).This study is based on applying MASW to a non-optimized data set acquired for active-source reflection seismic processing (Malehmir et al. 2013a, b;Salas-Romero et al. 2019) in an area prone to quick-clay landslides in southwest Sweden.Geotechnical characterization of the site is then achieved by complementing the results from MASW with previous results from reflection seismic (Malehmir et al. 2013a, b;Salas-Romero et al. 2019), full-waveform tomography (Adamczyk et al. 2014), P-wave refraction tomography (Wang et al. 2016), radio-magnetotelluric (RMT) resistivity (Lindgren 2014;Shan et al. 2014;Wang et al. 2016) and borehole data (Salas-Romero et al. 2016).The main objectives of this study are: (i) to obtain 2D V S models; (ii) to estimate V P /V S and Poisson's ratio along the seismic sections with previously obtained V P models; (iii) to compare the new models with the geological data obtained in three boreholes drilled in the study area in order to make an integrated interpretation of the study site for identifying softer subsurface areas; and, finally, (iv) to find relationships between the models that could provide the quickclay distribution in the area.

S T U DY A R E A
The quick-clay survey site is located adjacent to the Göta River valley, in southwest Sweden, north of the municipality of Lilla Edet, in an area called Fråstad (Fig. 1).The area on both sides of the river is characterized by a gently sloping landscape of glacial, postglacial and fluvial deposits (clay, silt and sand), with some exposed granite to granodiorite bedrock observed to the south.Several landslide scars are observed on both sides of the shorelines (Fig. 1).A prominent fracture zone follows the river channel and multiple morphologically and geologically significant lineaments cover the area (Fig. 1).
This area has been subjected to numerous geophysical, geotechnical and hydrological surveys (Fig. 1; Löfroth et al. 2011;Adamczyk et al. 2013Adamczyk et al. , 2014;;Dahlin et al. 2013;Malehmir et al. 2013a, b;Lundberg et al. 2014;Shan et al. 2014Shan et al. , 2016;;Salas-Romero et al. 2016, 2019;Wang et al. 2016;Comina et al. 2017).The main target of these different investigations was mostly quick-clay mapping.The aim of this project was to study quick-clay landslides using different geophysical methods and to interpret and characterize the subsurface down to bedrock using an integrated analysis of several types of data sets.The studies using reflection seismic (Malehmir et al. 2013a, b;Salas-Romero et al. 2019), electrical resistivity tomography, cone penetration test with resistivity, and geochemical sampling (Löfroth et al. 2011), core sampling and downhole property measurements (Salas-Romero et al. 2016), and P-wave refraction tomography and RMT resistivity (Wang et al. 2016) revealed the subsurface structures and provided information on a coarse-grained layer underlying potential quick clays in some locations.

DATA AC QU I S I T I O N
The acquisition set-up of the reflection seismic data collected in the study area was not optimized for recording surface-wave data.Therefore, some aspects of the data acquisition, such as the geophone natural frequency, receiver spacing, seismic source and receiver spread, restricted the frequency content of the surface-wave data, the signal penetration depth and resolution.Malehmir et al. (2013a, b) and Salas-Romero et al. (2019) described the reflection seismic data acquisition and processing for the seven seismic lines collected in 2011 and 2013 (this information is summarized in Supporting Information Table S1).The most important information about these surveys is that: (i) the natural frequency of the cabled geophones was 28 Hz, except in the northern part of line 5-5b where wireless stations were deployed, alternating single-component (1C) vertical geophones of 10 Hz and three-component (3C) broad-band digital MEMs (micro-electro mechanical systems) sensors; (ii) receiver spacing for the cabled geophones was 2-4 m and wireless stations were separated by 10 m distance; (iii) dynamite, accelerated weight drop and sledgehammer were used as seismic sources; and (iv) maximum receiver-source offsets were large, reaching up to approximately 2300 m in line 5-5b.At source locations where dynamite was not used, several shot records were generated in order to increase the signal-to-noise ratio by vertical stacking of the repeated records.
Two examples of raw data for shots in the middle of lines 5-5b (merged lines 5 and 5b collected in 2011 and 2013) and 7 are shown in Figure 2(a) and Supporting Information Figure S1(a), respectively.Line 5-5b is divided in two parts by the Göta River, being the wireless stations deployed on the northern side and the cabled geophones on the southern side of the river.Dynamite was used as a seismic source in line 5-5b, while a sledgehammer and dynamite (in the extremes of the line) were used in line 7.

Data preparation
For the reasons previously outlined, the reflection seismic data required some preparations prior to applying MASW. Figure 2 and Supporting Information Figure S1 show the different steps in data preparation for lines 5-5b and 7, respectively.The data preparation required, first of all, sufficient record length for recording surface waves.A length of 4 s was chosen for producing the data along all the lines (Fig. 2a and Supporting Information Fig. S1a).In a second step (Fig. 2b and Supporting Information Fig. S1b), a 'conservative' mute function was designed to remove the direct body waves and noise above the surface-wave cone in the seismic gathers.In a somewhat similar approach, Ivanov et al. (2005) used muting in shot gathers to separate fundamental from higher modes.The reason behind this is that even though the surface waves are the strongest energy recorded in a shot gather, at far offsets and high frequencies the body waves and high-frequency surface waves dominate and the low-frequency surface waves are attenuated (Park et al. 1999), which may complicate picking of the fundamental-mode dispersion curves (Ivanov et al. 2005).
Improvement on fundamental and higher modes imaging can also be achieved using longer receiver spreads, which increase the spectral resolution (Ivanov et al. 2005).In the following step (Fig. 2c and Supporting Information Fig. S1c), data were separated in positive and negative offsets, and a certain number of near offsets (30 m), larger than half of λ max (Stokoe et al. 1994), were removed in order to meet the horizontal travelling plane-wave criteria (Park et al. 1999;Foti et al. 2018) and allow the development of surface waves.Finally, a division of the seismic data in smaller subsets of 100 m each (minimum 25 receivers per subset) was required by RadExPro (MASW software) to be able to import the large amount of data (Fig. 3b and Supporting Information Fig. S1d).

Dispersion curve estimation
Dispersion curves of the Rayleigh waves in the frequencyphase velocity spectrum were obtained along all the seismic lines.Quality of the dispersion curves was comparable in all the lines, as the data acquisition was similar in all of them.Dispersion curve analysis was performed for every midpoint location defined at the centre of the receiver spread of every subset for each shot location, both on positive and negative offset data.Picking these curves at the maximum amplitudes was an important step that was carefully assessed.In our study, dispersion curves were picked for fundamental modes (all the seismic lines) and some first higher modes (all the lines except line 3).The fundamental mode usually had the strongest amplitude and was therefore easier to pick, compared with the higher modes that were not always visible.Picking higher modes increases penetration depth and achieves higher resolution models, which may help resolve e.g., low-velocity materials (Foti et al. 2018).
Following Park et al. (1999), receiver spacing and receiver spread influence the shallowest resolvable depth of investigation (Z min ) and the maximum penetration depth (Z max ), which are estimated to be about half of the minimum wavelength λ min (minimum velocity divided by maximum frequency) and half of the maximum wavelength λ max (maximum velocity divided by minimum frequency), respectively.Figure 3(a) shows an example of a dispersion curve picked in the frequencyphase velocity domain (including the limits of λ max and λ min ) and Figure 3(b) shows the subset corresponding to that curve (midpoint in 819 m and source in 1068 m).This shot from line 5-5b is the same as shown in Figure 2. The curve can reliably be picked within the limits of λ max and λ min .Picks at frequencies above the aliasing limit (λ min ) should only be accepted with caution (Cornou et al. 2006).Figure 4(a) shows a typical example of all the fundamental-mode dispersion curves picked along line 5-5b (see position in Fig. 1).This figure provides an overview of the frequency range (3-25 Hz) and phase velocity values (between 60 m/s and 150 m/s) along this line.In Figure 4(b), the maximum depth of investigation obtained from these dispersion curves is represented.

Inversion of dispersion curves
Inversion of the dispersion curves allows the recovery of 1D V S models and estimated values for V S 30 (the averaged V S values for the top 30 m used for site classification; Foti et al. 2018).A scheme of the processing and inversion workflow is shown in Figure 5.The initial subsurface model used for the inversion was a horizontally layered, linear elastic and isotropic medium.The model parameterization was similar in all the lines.Constant values for density (1.7 g/cm 3 ) and Poisson's ratio (0.48) were estimated from the data collected in previous studies (Polom et al. 2013;Salas-Romero et al. 2016, 2019).Half-space depth was defined using the information from the extracted dispersion curves (0.5 λ max ), resulting in an averaged value estimated for all the lines (26 m).Through different inversion tests and the a priori information avail- able from reflection seismic, resistivity, borehole and P-wave refraction/full-waveform tomography data, the number of layers with progressive thickness was defined to be 7.
As our approach is based on a local-search method, the chosen starting model influenced the resulting 1D profiles.In order to obtain optimal results, different starting models were tested (Foti et al. 2018) and the most adequate one was chosen by visual examination of the inversion results (root-meansquare -RMS -error of the fit of the inverted to the picked curve was below 2% and convergence was reached after 2-3 iterations out of 6-8 allowed iterations).If higher-mode dispersion curves were available, the fundamental mode was first inverted and the result was used as a starting model to invert both modes.

Run inversion
Figure 5 Processing and inversion workflow.
The picked dispersion curves are inverted using a leastsquares algorithm (Xia et al. 1999) in an automated iterative scheme, for obtaining the variation of V S with depth at a midpoint position (1D model).After a first inversion run, curves with an RMS fit above 2% were checked visually.A small number of picks were considered as outliers and either corrected or removed.The inversion was run again afterwards.Pseudo-2D velocity models were then derived using linear interpolation of the 1D models.A horizontal moving average (with window size equal to two times the receiver spacing) was applied in order to obtain more coherent results between neighbouring 1D V S profiles.This step introduced negligible lateral smoothing (the 1D V S profiles represent the average velocity over the spread length that is 100 m), preserving the 1D structure of the results.Supporting Information Figures S2  and S3 show the starting, 1D inversion and pseudo-2D inversion V S models for lines 5-5b and 7, respectively.The starting model in line 5-5b is different on each side of the river, because the inversion of surface waves was done separately for each part, as the receiver type and interval were different.Observing more in detail the pseudo-2D inversion V S models, the results for both lines (Supporting Information Figs S2c and  S3c) are still patchy with vertical smearing, especially in the southern side of the river in line 5-5b.
Other types of data, such as reflection seismic, resistivity or borehole data were used only after the inversion for evaluation of and comparison with the V S models.In order to compute V P /V S and Poisson's ratio models, the previously obtained V P models were resampled to match the grid discretization of the V S models.No distinction was made between V P models from P-wave refraction and full-waveform tomography data.

R E S U LT S A N D I N T E R P R E TAT I O N S
In order to provide a more coherent description of the results and their interpretations, lines with similar features and models are explained within the same section.All the models were cut to show the same investigation depth appropriate for the V S models, even though some of these may provide good resolution at greater depth (P-wave refraction tomography data have a lateral resolution between 27-35 m down to 80 m of depth; full-waveform tomography data have a lateral resolution of 10 m in the first 20-30 m of depth, and about 20 m down to 80 m of depth; RMT resistivity data can reliably reach down to 20-30 m of depth).

Lines 1 and 4
Figures 6 and 7 show V P (Adamczyk et al. 2014), V S , V P /V S and Poisson's ratio obtained in this study using MASW, as well as RMT resistivity models (Lindgren 2014;Shan et al. 2014) for the processed seismic lines 1 and 4 collected in 2011 (see position of the lines in Fig. 1; Malehmir et al. 2013a, b).From previous studies, two interfaces, S1 and S2, were identified in both seismic lines and interpreted to belong to coarse-grained layers underlying quick clays.The bedrock interface B1 is only delineated in line 1 and has the shape of a small basin, with bedrock peaks on both sides of the line.
Line 1 (Fig. 6) shows two horizontal high-velocity layers at around 15 m and 0 m (V P is around 2000 m/s and V S is higher than 120 m/s).The models at the centre of the line do not reach the top of the bedrock; however, on the sides, at CDPs 100 and 350, V P (Adamczyk et al. 2014) and, in less degree, V S models show high values, which seem to be related to the presence of bedrock at these positions.In the V P model, the bedrock interface B1 seems to be well delineated, although there are high values above it, which do not seem to fit with the assumed coarse-grained layers.The V P model also shows a low velocity area (V P is below 1500 m/s) between the S1 and S2 interfaces.This low-velocity area is less clear in the V S model and difficult to delineate (Fig. 6b), for example, in the centre and the south-western side V S is lower on average showing few signs of increases with depth, while in the north-eastern side the values are mostly higher.The V S model has higher resolution in the top 10 m, where velocities are lower than 80 m/s.In the derived V P /V S and Poisson's ratio models, (Fig. 6c-d), S1 and S2 seem to delineate a zone with low values for both parameters imbedded between materials with higher values above and below.Poisson's ratio values seem to indicate the presence of mostly saturated deposits (saturated clays and sands; Bowles 1988;Konstantaki et al. 2013;Uhlemann et al. 2016).The groundwater level according to the borehole data (Salas-Romero et al. 2016) was at 2 m depth in this profile, thus all materials below it were considered fully saturated.Deposits with higher clay content may have V P /V S equal to 8-9 (Bailey et al. 2013).However, the V P /V S values are extremely high in the edges of the pro- file with values above 20, which seem unrealistic and not representative for saturated clays and sands.The high values of V P /V S in these positions may be related to a lower resolution of the V S and V P models in these areas and/or due to extremely low V S (studies in the same area, Polom et al. 2013 andComina et al. 2017, obtained similar values).Bailey et al. (2013) associate high V P /V S to an increased porosity, saturation and clay content, all of which are conditions present in this study area.The same observations can be made in all the lines.The interpreted interfaces S1 and S2 agree with the high-velocity layers, corresponding to silt and sandy silt materials according to the borehole data (Salas-Romero et al. 2016).The RMT resistivity model (Fig. 6e; Lindgren 2014) also agrees with the interpreted seismic section.The resistivity values corresponding to the low-velocity zone are between 1 m and 10 m (unleached clays; Solberg et al. 2012), and above and below this area the resistivity values are between 10 m and 100 m, which may suggest the presence of leached clay deposits and coarse-grained materials (Solberg et al. 2012).
The V S model for line 4 (Fig. 7b) also shows two subhorizontal high-velocity layers at around 10 m and 0 m (V S is higher than 120 m/s), but these layers are barely visible in the V P model (Fig. 7a).The comparison with the soil textures obtained from the borehole data (Salas-Romero et al. 2016) suggests that the identified sand-sandy silt layer in BH2 is deeper than these high-velocity layers.The first of these layers seems to be from silty materials according to the borehole data.The origin of the second layer is unclear, but could be related to the presence of a thin silt layer not visible in the simplified borehole data (see Salas-Romero et al. 2016 for more detailed information).V S decreases between these two layers down to 100 m/s, and at the landslide scar V S is very high.In the V P /V S and Poisson's ratio models (Fig. 7c-d), the high-velocity layers lack continuity and are not delineated as clearly as in line 1.When comparing with the RMT resistivity model (Fig. 7e; Shan et al. 2014), the resistive layer (<100 m) at a depth of around 0-10 m is roughly equivalent to the low-velocity layer sandwiched between the two high-velocity layers in the V S model from CDPs 150 to 390 approximately.It is worth noting that in the reflection seismic data and the V P model, this layering was not identified, making more valuable the V S model obtained from MASW using a non-optimized active reflection seismic data set.

Lines 2-2b and 5-5b
Figures 8 and 9 show the V P (Wang et al. 2016), V S , V P /V S , Poisson's ratio obtained in this study using MASW and RMT In line 2-2b, the bedrock interface B1 can be delineated between CDPs 400 and 600 within all models.In comparison, the coarse-grained layer interface S1 is not adequately defined by the velocity models in the south-eastern portion of the line, except at around 20 m between CDPs 100 and 150, and 350 and 400.In contrast, V P /V S and Poisson's ratio models show a well-delineated undulating layer between CDPs 150 and 350.In light of these results, the interpretation of the reflection seismic data for the interface S1 between CDPs 150 and 350 should be corrected following the path of the undulating layer observed in Fig. 8(c and d).The silt-sandy silt layer in the borehole data from BH1 (Salas-Romero et al. 2016) coincides with S1.The RMT resistivity model (Fig. 8e; Wang et al. 2016) does not cover the entire south-eastern portion of the line; nevertheless, a resistive layer (>100 m) appears to coincide with the S1 position between CDPs 350 and 450.
The interpreted interfaces S1 and B1 in line 5-5b are found outside the model limit.In the southern part, the V S model (Fig. 9b when bedrock is close to the surface around CDP 500.V P , V P /V S and Poisson's ratio show high values between CDPs 150 and 450, and 550 and 650, in the first 5 m.This is probably due to the influence of the V P model, which may have worse resolution in these locations.The RMT resistivity model (Fig. 9e; Wang et al. 2016) shows a resistive layer (>100 m) between CDPs 300 and 700 interbedded within other materials with lower resistivity values.In the northern side of the river, the S1 interface at 10 m between CDPs 880 and 1200 is within the velocity model limits and seems well delineated by all the models.

Lines 3, 6 and 7
Figures 10, 11 and 12 show the V P (Wang et al. 2016), V S , V P /V S , Poisson's ratio obtained and calculated in this study using MASW and RMT resistivity models (Shan et al. 2014;Wang et al. 2016) for the processed seismic lines 3 (Malehmir Models for line 3 (Fig. 10) define relatively well the interfaces S1 and B1 in the south-eastern portion of the line.Below the S1 interface, the V S model (Fig. 10b) does not have enough resolution for defining the interface S2.In the north-western portion of the line, V P /V S and Poisson's ratio (Fig. 10c-d) cannot delineate S1 clearly, showing patchy areas in both models.This is probably related to the vertical smearing observed in the V S model (Fig. 10b).Below S1 V P /V S and Poisson's ratio seem to decrease and start to increase again before reaching the interface S2 between CDPs 180 and 290.Without borehole data, the nature and thickness of the S1 interface in this line is unclear and may have an influence in the lower V P /V S values compared with other lines.The RMT resistivity data (Fig. 10e; Shan et al. 2014) also show a resistive layer (10-50 m) at the S1 interface and low-resistivity values around 10 m below it.
Lines 6 and 7 show very similar models, particularly in the eastern portion of both lines (Figs 11 and 12).The interface S1 was not interpreted in line 6, and in line 7 it was interpreted at around −20 m, deeper than the maximum depth of the models.The V S models (Figs 11b and 12b) in both lines show high-velocity values (>120 m/s) at shallower depths (top 10 m) in the western portion of the lines.In the eastern portion, the V S models suggest the presence of a low-velocity layer (around 100 m/s or lower) between 5 and 15 m on top of the bedrock with a westward continuation in line 7. V P /V S , Poisson's ratio and RMT resistivity models (Figs 11c-d and 12c-e;Wang et al. 2016) show very similar features and high values where the low velocity layer is identified.Wang et al. (2016) interpreted this layer as the coarse-grained layer, which was identified neither in the reflection seismic nor in the V P data.The correlation between V S and RMT resistivity anomalies may indicate, as observed before in Fig. 7 (line 4), the presence of silty materials or a coarse-grained layer (S1).The interpretation of the reflection seismic data for the interface S1 should be improved using the models in lines 6 and 7 (Figs 11 and 12).
Figure 13 presents the 3D models for V P , V S , V P /V S and Poisson's ratio.In general, the correlation between the models in different seismic lines is good and similar structures are found among them.The amplitude of the anomalies is similar across all the lines.

D I S C U S S I O N
Shear-wave velocity models obtained using MASW show a good resolution down to 10-15 m of depth, allowing the detection of a low-velocity layer sandwiched between two highvelocity layers in one of the lines (Fig. 7b).When contrasting these results with the data from the boreholes (Salas-Romero et al. 2016), V S appears to be sensitive to silty materials.The low-velocity layer in line 4 (Fig. 7b) coincides with a relatively high resistive layer (around 30-80 m), which suggests the presence of fine-grained till, leached marine clays, silt or potential quick clays (Solberg et al. 2012).The V S model for line 5-5b (Fig. 9b) shows high values between CDPs 400 and 500, and below the landslide scar, similar to line 4 (Figs 7b and 13).The reflection seismic section for line 5-5b lacks continuity at S1 between CDPs 400 and 500 compared with other segments of the line.Salas-Romero et al. (2019) suggest that either a lower fold in the seismic data, possible fractures in the bedrock, or bedrock movement may be causing this pattern.Moreover, below the landslide scar, the sediments appear to have shifted towards the river.The V P model (Fig. 9a) shows low velocity anomalies both between CDPs 400 and 500, and below the landslide scar.The possible non-exclusive interpretations for this are: (1) the presence of fluids that decrease V P but not V S ; (2) low quality seismic data at these positions; and (3) the compression of the sediments below the landslide scar during the sliding.
In general, V S models offer valuable information leading towards a more complete interpretation of other parameters previously available at the site and introducing new, important information by themselves and when combined with previous studies.This is particularly true as no new data acquisition was conducted.The entire study was based on optimizing an available data set acquired for reflection seismic and P-wave traveltime tomography, showing the importance of new data handling techniques.Additionally, other techniques, such as windowing and dispersion stacking, could further improve the resolution and the data quality (Neducza 2007;Socco et al. 2009).As the target depth is not deep (around 30-50 m), the investigation depth of the V S models may be sufficient for site characterization, and the resolution is higher at shallower depths where high-frequency information is more abundant.This is in accordance with other specifically designed activesource MASW data sets (e.g., Foti et al. 2018).At depths down to 26 m, the subsurface structures are mostly sub-horizontal (except at the bedrock outcrop places); therefore, significant lateral variations of the elastic properties are not expected (Malehmir et al. 2013a, b;Salas-Romero et al. 2019).Limitations of the MASW software such as limited model parameterization, no custom inversion workflow and lack of control of regularization, convergence criteria, and model uncertainties should be considered when evaluating the V S models.
The use of different seismic sources can affect the signalto-noise ratio, penetration depth and frequency content of the data.For example, in lines 4 and 5-5b, dynamite was used as seismic source, in lines 1, 2-2b and 3 both an accelerated weight drop and a sledgehammer were used, while in lines 6 and 7 the main source employed was the sledgehammer.It is possible that the sledgehammer data from lines 6 and 7 (Figs.10b and 11b) were of a slightly lower quality on the western portions of the lines, but it is difficult to quantify this further without a comparison of all three sources at these positions, which is unfortunately unavailable and beyond the scope of this study.
The primary goal of the project where this work is included was to map quick clays in the study area using geophysical methods.However, quick clays can only be identified using geotechnical site and laboratory investigations.In this way, a combination of geophysical methods is the best approach for obtaining the location of potential quick-clay areas or other related materials such as a coarse-grained layer that, in this specific area, usually underlies the quick clays.Following this approach, the next step is to use the geophysical data to model the distribution of quick clay in the area.Figure 14 shows 3D cross-plots of the different physical properties (V P , V S , V P /V S , Poisson's ratio and resistivity) for line 5-5b.The data points in the cross-plots were selected from within a 20 m window centred around the borehole position BH3 (Fig. 1).Each data point was colour-coded according to the soil textures identified in the borehole (Salas-Romero et al. 2016).By visual inspection, the cross-plots reveal that the data cluster in at least two groups of different point density and opposite trends (Fig. 14a-c).According to the soil textures, the cluster to the right includes mostly silty clay and silt, and the denser cluster to the left includes sandy silt and clay.The cross-plots show similar data distributions, although V P and resistivity appear to have the larger influence on the clustering (i.e., there is no clustering along the V S axis).Considering that quick clays were visually identified above the coarse-grained layer in BH3 (Salas-Romero et al. 2016), the sandy silt-clay cluster was assumed to contain the potential quick-clay data points.A quick-clay model was thus defined with V P less than 1200 m/s, resistivity below 32 m, V P /V S less than 14 and Poisson's ratio below 0.498 (Fig. 14d-f).Applying this model to all data points along line 5-5b, a dis-tribution of potential quick-clay areas is obtained (Fig. 14g).The southern side of the river shows more abundancy of potential quick clays, while the northern side shows just few small areas.The thicker quick-clay packages are immediately above the interface S1, which is the coarse-grained layer, and thinner packages are found at shallower depths.Earlier studies by Löfroth et al. (2011) confirm the presence of potential quick clays at some spots shown in the model between CDPs 600 and 700.This simple model could be improved with deeper and higher resolution data, which can help to establish more reliable property values, and thereby a more accurate model for mapping quick clays.
Estimates of the V S 30 value at the midpoint positions were also provided.V S 30 is a valuable parameter for geotechnical site characterization.In general, the produced values are low for most of the lines, for example in line 5-5b, V S 30 ranges mostly between 70 m/s and 130 m/s, reaching values above 200 m/s in just a few locations.Based on the soil © 2021 The Authors.Near Surface Geophysics published by John Wiley & Sons Ltd on behalf of European Association of Geoscientists and Engineers., Near Surface Geophysics, 1-17 classification shown by Kanlı et al. (2006), made according to the Uniform Building Code and Eurocode 8, these soils classify mostly as soft soils and soft clays/silts with high plasticity and high-water content indices, or maybe even deposits of liquefiable soils or sensitive clays.Such results flag this area as a high landslide risk site for quick clays.This finding is so far the most striking observation in terms of quick-clay landslide risk in the study area, further suggesting the potential of MASW for these types of studies.

C O N C L U S I O N S
Multichannel analysis of surface waves has been performed on seven active-source seismic profiles, specifically for reflection imaging and P-wave traveltime tomography at a known quick-clay landslide site in southwest Sweden.Raw data from original shot records had to be re-evaluated to increase the record length to 4 s.This extension was deemed necessary since the V S values were estimated to be extremely low, in the range of 60-120 m/s, thus requiring long records for an adequate analysis.Fortunately, the original data were recorded over 6-10 s, which can be recommended as a minimum recording time for similar, future studies.V P /V S and Poisson's ratio, derived from the obtained V S models and previously available P-wave traveltime tomography data, correlate well with coarse-grained layers that underlie quick clays.Previous reflection seismic interpretations (Salas-Romero et al. 2019) should be adjusted in light of these new models.The discovered negative correlation between V S and RMT resistivity models improves the interpretation of the shallowest layers with respect to previous studies.A model that shows the distribution of quick clays in the area was obtained by combination of the modelled geophysical properties and soil textures, indicating large amounts of quick clays along the longest 2D seismic profile collected in the project.
Although V S is claimed to be insensitive to quick clays, our MASW results delineated associated materials, such as the underlying coarse-grained layer, and allowed us to infer geotechnical properties that may help to characterize the study area.Estimated V S 30 values indicate the presence of soft soils and liquefiable deposits.Therefore, our results confirm that the study area is susceptible to quick-clay landslides, but we emphasize that triggering conditions and mechanisms should carefully be studied and understood.Although the study area is highly prone to quick-clay landslides, their occurrence, in terms of their timing, remains unclear at this stage.

AC K N OW L E D G M E N T
The Geoscientists Without Borders program of the Society of Exploration Geophysicists (SEG) and Uppsala University sponsored this project.The geophysical data were collected with the help of PhD and MSc students from Uppsala University, and staff from SGU, in particular S. Ohlsson.This study was initiated as part of a joint research collaboration among Uppsala University, SGU, Leibniz Institute for Applied Geophysics, University of Cologne, Syiah Kuala University, Polish Academy of Sciences, Norwegian Seismic Array, Norwegian Geotechnical Institute, Institute for Geosciences at the University of Oslo, Geotechnical Group at the Norwegian University of Science and Technology, and Geological Survey of Norway.Partial funding from Trust2.2-Geoinfra project (http://trust-geoinfra.se/) supported this work (252-2012-1907).We are thankful to SGU for providing the geological data.We would like to thank S. Wang, A. Adamczyk, C. Shan, A. Lindgren and M. Bastani for providing their resistivity and tomography modelling results.RadExPro Seismic Software was used for the multichannel analysis of surface waves, and MATLAB was used for analysing the results.Figures were prepared using Generic Mapping Tools (http://gmt.soest.hawaii.edu/),Inkscape (https://inkscape.org/) and MATLAB.We thank two anonymous reviewers and the Associate Editor, Sebastian Uhlemann, for providing their critical reviews and constructive comments.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.Four of the land reflection seismic lines are available with restricted access (permission from A. Malehmir is required for accessing the data) at the Swedish National Data Service (https://snd.gu.se/en/catalogue/study/snd1079).Geological data are available from SGU following registration.Borehole, resistivity and tomography modelling data are properly cited and referred to in the reference list.

S U P P O RT I N G I N F O R M AT I O N
Additional supporting information may be found online in the Supporting Information section at the end of the article.

Figure 1
Figure 1 Geological map of the study area (copyright Geological Survey of Sweden-SGU, modified from Salas-Romero et al. 2019) located in Fråstad in southwest Sweden.Geological structures and materials, along with information about some of the investigations carried out by Uppsala University, are shown in the legend at the bottom and on the right-hand side of the map.

Figure 2 Figure 3
Figure 2 Data preparation for MASW for a shot recorded in the middle of line 5-5b.(a) Raw data, (b) after applying a mute function and (c) after separating positive and negative offsets and removing near offsets (30 m).Note the position of the Göta River in the negative offset part.In the southern side of the river cabled geophones were used, while in the northern side wireless stations were deployed.

Figure 4
Figure 4 (a) All the fundamental-mode dispersion curves picked along line 5-5b.(b) The dispersion curves are displayed in the distancewavelength/2 domain.The gap between 1300 and 1600 m corresponds to the Göta River position where no receiver and shot placement was possible.

Figure 6
Figure 6 Reflection seismic section of line 1 (Malehmir et al. 2013a) with superimposed (a) full-waveform tomography results for V P (Adamczyk et al. 2014), (b) V S model obtained from MASW, (c) V P /V S distribution, (d) Poisson's ratio distribution and (e) RMT resistivity model (Lindgren 2014).The simplified soil texture results for BH1 obtained in Salas-Romero et al. (2016) are superimposed on (a)-(e).The legend at the bottom of the figure shows the soil textures present in the borehole data.The seismic interpretation from previous studies is also included (dashed lines and labelled reflections).

Figure 7
Figure 7 Reflection seismic section of line 4 (Malehmir et al. 2013b) with superimposed (a) full-waveform tomography results for V P (Adamczyk et al. 2014), (b) V S model obtained from MASW, (c) V P /V S distribution, (d) Poisson's ratio distribution and (e) RMT resistivity model (Shan et al. 2014).The simplified soil texture results for BH2 obtained in Salas-Romero et al. (2016) are superimposed on (a)-(e).The legend at the bottom of the figure shows the soil textures present in the borehole data.The seismic interpretation from previous studies is also included (dashed lines and labelled reflections).

Figure 8
Figure 8 Reflection seismic section of line 2-2b (Salas-Romero et al. 2019) with superimposed (a) P-wave refraction tomography model (Wang et al. 2016), (b) V S model obtained from MASW, (c) V P /V S distribution, (d) Poisson's ratio distribution and (e) RMT resistivity model (Wang et al. 2016).The simplified soil texture results for BH1 obtained in Salas-Romero et al. (2016) are superimposed on (a)-(e).The legend at the right side of the figure shows the soil textures present in the borehole data.The seismic interpretation from previous studies is also included (dashed lines and labelled reflections).

Figure 9
Figure 9 Reflection seismic section of line 5-5b (Salas-Romero et al. 2019) with superimposed (a) P-wave refraction tomography model (Wang et al. 2016), (b) V S model obtained from MASW, (c) V P /V S distribution, (d) Poisson's ratio distribution and (e) RMT resistivity model (Wang et al. 2016).The simplified soil texture results for BH3 obtained in Salas-Romero et al. (2016) are superimposed on (a)-(e).The legend at the bottom of the figure shows the soil textures present in the borehole data.The seismic interpretation from previous studies is also included (dashed lines and labelled reflections).resistivity models (Wang et al. 2016) for the processed seismic lines 2-2b (Salas-Romero et al. 2019) and 5-5b (Salas-Romero et al. 2019).Both lines show undulating top of bedrock, with a peak close to the surface, and thereafter dipping down towards the river.The coarse-grained layer interface S1 is interpreted along both lines, except next to the river in line 2-2b.Line 5-5b also presents the data from the other side of the river, where again S1 can be interpreted.In line 2-2b, the bedrock interface B1 can be delineated between CDPs 400 and 600 within all models.In comparison, the coarse-grained layer interface S1 is not adequately defined by the velocity models in the south-eastern portion of the line, except at around 20 m between CDPs 100 and 150, and 350 and 400.In contrast, V P /V S and Poisson's ratio models show a well-delineated undulating layer between CDPs 150 and 350.In light of these results, the interpretation of the reflection seismic data for the interface S1 between CDPs

Figure 10
Figure 10 Reflection seismic section of line 3 (Malehmir et al. 2013b) with superimposed (a) P-wave refraction tomography model, (b) V S model obtained from MASW, (c) V P /V S distribution, (d) Poisson's ratio distribution and (e) RMT resistivity model(Shan et al. 2014).The seismic interpretation from previous studies is also included (dashed lines and labelled reflections).

Figure 11
Figure 11 Reflection seismic section of line 6 (Salas-Romero et al. 2019) with superimposed (a) P-wave refraction tomography model (Wang et al. 2016), (b) V S model obtained from MASW, (c) V P /V S distribution and (d) Poisson's ratio distribution.The seismic interpretation from previous studies is also included (dashed lines and labelled reflections).

Figure 12
Figure 12 Reflection seismic section of line 7 (Salas-Romero et al. 2019) with superimposed (a) P-wave refraction tomography model (Wang et al. 2016), (b) V S model obtained from MASW, (c) V P /V S distribution, (d) Poisson's ratio distribution and (e) RMT resistivity model(Wang et al. 2016).The seismic interpretation from previous studies is also included (dashed lines and labelled reflections).

Figure 13
Figure 13 3D views showing (a) V P , (b) V S , (c) V P /V S and (d) Poisson's ratio models.The ranges of values for every property are on the rightupper corner of every 3D view.The identified anomalies are continuous across several lines.Note that the northern portion of line 5-5b (where the wireless stations were deployed) has been excluded from the 3D views, to focus better on the property changes and their consistency along different lines.The Poisson's ratio model shown in Fig. 11 (line 6) has a different colour scale compared with this figure.

Figure 14
Figure 14 3D cross-plots of different properties from BH3 colour-coded according to the type of soil texture obtained in Salas-Romero et al. (2016).See legend on the right-hand side of the figure.(a) Log(resistivity) versus V P and V S , (b) log(resistivity) versus V P and V P /V S , (c) log(resistivity) versus V P and Poisson's ratio, (d), (e) and (f) cluster representation of the same data sets as in (a), (b) and (c), respectively, and (g) model of the quick-clay distribution obtained along line 5-5b.

Figure S1 :
Data preparation for MASW for a shot recorded in the middle of line 7.

Table S1 :
Main acquisition parameters of the seismic lines recorded in September 2011 and March 2013 (for more details see Malehmir et al. 2013a, b and Salas-Romero et al. 2019).