Imaging mass‐wasting sliding surfaces within complex glacial deposits along coastal cliffs using geophysics

This study presents a multidisciplinary survey combining geological fieldwork and geophysical data to better constrain the parameters influencing the morphology and behaviour of a retreating coastal cliff. Erosion rates are spatially highly variable and hard to predict because of the manifold parameters acting on them. Among these parameters, rock resistance exerts a paramount influence on cliff retreat. Characterizing the rock resistance distribution along a coastal region requires the mapping of several key subsurface properties including the bulk lithology, faulting, fracturing, or weathering. This is a difficult and expensive task because of the high spatial variability of these factors linked to the spatial complexity of the geology. Geophysical methods can be used to tackle this challenge by quickly providing the 3D visualization and distribution of these parameters within the subsurface. A fast‐eroding portion of the Norfolk coast (UK) at West Runton is investigated using a multidisciplinary approach, combining ground‐penetrating radar, electrical resistivity tomography (ERT), cone penetration tests, and outcrop studies. The results allowed us to build a 3D geological and geophysical model of a highly complex area of glacial geology. It forms part of a relict glaciotectonic thrust‐tip moraine and sand basin sequence. The surfaces interpreted on radar data are associated with strong resistivity contrasts on the ERT data. These contrasts have been attributed to petrophysical variations between the lithological units. The base of the sand basin is marked by a low‐permeability clay bed. Its low shear strength is likely to be more susceptible to failure, hereby accelerating the erosion rate of an already fast‐eroding sand basin. The resulting model can be used as input for locally constraining the ground parameters in coastal recession and erosion models.


| INTRODUCTION
In Europe, more than 200 million people live in coastal regions (i.e. regions having either a sea border or having more than half of the population within 50 km of the sea; Collet & Engelbert, 2013). With climate change, global sea-level rise and anthropogenic adjustment, coastal systems and low-lying areas increasingly experience submergence, coastal flooding, and coastal erosion while public authorities are struggling to find solutions (Wong et al., 2014).
Year after year, many buildings are lost to the sea and the population is forced to evacuate in light of increasing coastal flooding and land loss (Burden et al., 2020). Beyond the habitat loss and the increasing natural hazard, erosion can also have a strong environmental impact by exposing coastal landfills that could contaminate waters and disrupt coastal ecosystems (Brand & Spencer, 2019). Worldwide, in the absence of appropriate measures to protect coastal regions, it is estimated that sea-level rise will displace tens of millions of people in the coming decades (McLeman, 2018).
Historically, coastal adaptation measures consisted of the installation of hard defences (e.g. sea walls, revetments, groynes, or artificial reefs) designed to diminish wave attack and slow down the erosion by trapping washed-out sediments. However, these structures showed only short-term and local efficiency. They were not designed to counter an important global sea-level rise, and they led to counterproductive effects with down-drift sediment starvation and erosion (Clayton, 1989;Dickson et al., 2006;Granja & De Carvalho, 2000;Malherbe et al., 2013). Nowadays, more sustainable and climate-resilient measures are being developed and implemented in several countries (Pranzini et al., 2015). To be effective, these management plans must be based on a good understanding of coastal erosion systems and reliable estimates of the shoreline recession (Kettle, 2012).
The main drivers of coastal erosion are wave attack, shore profile, rock properties, and climate (Ashton et al., 2011). Their relative role and combined effects are not yet well understood (Hapke & Plant, 2010) and more work is required to develop reliable predictive models of shoreline recession (Wong et al., 2014). Designing costefficient methods to facilitate data acquisition and site characterization and monitoring is a challenge yet to be tackled.
Coastal cliffs (albeit comprising cliffs made of 'soft' rocks) are characterized by dynamically linked cliff retreat and shore platform erosion, notably through sudden mass-wasting processes (Collins & Sitar, 2008;Prémaillon et al., 2018;Rosser et al., 2013). Many coastal cliff recession models focus on the influence of marine drivers (Ashton et al., 2011;Dickson et al., 2006;Fitton et al., 2016;Inman et al., 2002;Limber et al., 2018), but some studies have also demonstrated that rates of cliff retreat vary greatly according to lithology (Del Río et al., 2009;Hapke & Plant, 2010), moisture content (Dietze et al., 2020), and discontinuities or décollement layers that favour mass-wasting processes (Le Cossec et al., 2011). The compilation of erosion studies at a global scale presented by Prémaillon et al. (2018) suggests that cliff retreat is most clearly governed by the lithological nature of the cliffs. They suggest that cliffs' resistance to erosion can be expressed by a simplified Hoek and Brown rock mass strength criterion (i.e. weak, medium, strong) that is merging lithological description and fracturing/weathering state of the rock (Hoek & Brown, 1997). However, due to insufficient knowledge about rock strength distribution, 3D rock models are often oversimplified (Limber et al., 2018). There is a need for a cost-efficient strategy to acquire rock properties data so these can be more systematically included in the cliff recession models.
The majority of subsurface investigations rely on in-situ testing such as cone penetration tests (CPTs) and the collection of rock samples for laboratory testing (Coe et al., 2018). They provide direct and quantitative measurements of the rock mechanical properties, but only sample local points and provide discrete data. Consequently, the broad distribution of the geotechnical properties must be approximated via geo-statistics based on these local measurements. The reliability of geostatistical models is limited when the subsurface geology is complex, because of the high disparity of these properties. Further constraints to guide the statistical modelling in 3D are necessary.
In this study, we aim to characterize the cliff strength by acquiring volumetric geophysical data. Geophysical methods are nondestructive tools able to quickly measure geophysical properties within the subsurface. Constraining the geophysical data with groundtruth knowledge allows building 3D geological models of the subsurface. Furthermore, petrophysical property relationships allow translating geophysical data into 3D models of geotechnical and hydrological properties (Coe et al., 2018;Mayne et al., 2002). Previous studies have already used geophysical data to constrain the cliff properties, notably seismic methods to identify the basement rock (Inman et al., 2002), Bouguer gravity methods to identify aquifers on coastal cliffs (Jacob et al., 2018;Sajinkumar et al., 2017;), aerial electromagnetic methods to obtain large-scale ground resistivity distribution that can be correlated to unstable landslide areas (Nakazato & Konishi, 2005), or ground penetrating radar (GPR) methods to study paleo-events related to coastal hazards (Switzer et al., 2020). Here, we utilize high-resolution data to map the geology and water content of the subsurface by combining 3D electrical resistivity tomography (ERT) with multi-frequency GPR measurements, CPTs, and geological mapping along the shoreline. On the one hand, electrical resistivity can be used to determine in-situ rock properties and map the top of the water table and the soil salinity (Coe et al., 2018). On the other hand, GPR methods are sensitive to the electromagnetic properties of the soil. They provide complementary images, at a higher resolution than ERT, which help determine the stratigraphy in the shallow subsurface, locating fault zones, mapping the water table and the bedrock depth and saline interfaces (Jol & Bristow, 2003;Neal, 2004). Both methods are sensitive to the moisture and clay content (Goldman & Kafri, 2006;Huisman et al., 2003). An increased moisture or clay content leads to a reduction in the rock resistance, facilitating the triggering of landslides and more generally, coastal retreat (Dafalla, 2013;Yalcin, 2007). Laboratory studies have shown that reduced resistivity of granular material can be indicative of areas of low shear strength (Saarenketo & Scullion, 1996).
By combining 3D GPR and ERT data, this study aims to build a representative ground model that provides information regarding the soft sediment distribution, the presence of faults or décollement surfaces, and the relative moisture content distribution within the subsurface. The resulting ground model can be used as input to constrain the cliff strength spatial distribution. The study area is representative of many other coastal sites and the findings from this site can help understand coastal erosion at sites with similar geological settings. Furthermore, the methodological approach developed in this paper is applicable to other coastal sites of different geological settings where geophysical contrasts and properties allow the successful use of ERT and GPR.

| STUDY SITE LOCATION AND GEOLOGICAL SETTINGS
The study site is situated between West Runton and East Runton, northeast Norfolk coast, UK ( Figure 1). It extends above a large glacitectonic sand basin that retreats at a higher rate than the adjacent till-dominated cliffs. In this area, coastal erosion of the undefended cliffs can be averaged to À0.9 m/yr over the last 20 years (Environment Agency, 2013). The subsurface geology is clearly observable in the cliff section, providing a unique opportunity to link geophysical measurements with direct geological observations ( Figure 2). It is a representative example of many glacitectonic sand basins, which are common along this part of the coastline.
F I G U R E 1 Location of the study site compared with the indicative erosion zones and extent of the electrical resistivity tomography and ground penetrating radar surveys. The survey covers a sand basin well exposed on the shore's cliff to calibrate the geophysical models. The current models assume a relatively linear erosion trend parallel to the coast, which does not illustrate the locally higher erosion at the sand lens (modified from AECOM Limited, 2010). Imagery map from Esri, DigitalGlobe, GeoEye, Earthstar Geographics, CNES/Airbus DS, USDA, USGS, AeroGRID, IGN, and the GIS User Community (https://services.arcgisonline.com/ArcGIS/rest/services/World_Imagery/MapServer) [Color figure can be viewed at wileyonlinelibrary.com] The regional Quaternary geology was most recently defined by Lee et al. (2017) as a complex glacigenic sequence comprising diamicton, glaciolacustrine sands and muds, and glaciofluvial outwash deposits.
They record the successive episodes of ice marginal advance and retreat across the region that occurred during the Anglian Glaciation, 450 000 years ago ( Figure 3). In total, six major ice advances (A1 to A6) have been described in East Anglia (Lee et al., 2017). In the study area, four glaciogenic units have been recognized (Phillips et al., 2008;Carr et al., 2016). From the base upwards, they comprise the Happisburgh Diamicton Member (HTM; associated with the A1 ice advance), the Bacton Green Diamicton Member (BGTM; associated with the A4 ice advance), the West Runton Mélange Member (WRM; associated with the A5 ice advance), and the Runton Sand and Gravel Member (RSGM; associated with the A5 ice advance). They overlie Upper Cretaceous Chalk bedrock, which crops out on the foreshore beneath the highwater mark, and the overlying Early to Middle Pleistocene shallow marine and coastal deposits of the Wroxham Crag Formation (WCF).
At the study site and extending westwards, these materials have been remobilized and largely intermixed, forming a glacitectonic mélange (the WRM). The mélange incorporates large rafts of remobilized chalk bedrock and the development of large glacitectonic sand basins (infilled by the RSGM) that formed contemporaneously along structurally defined thrust-tip ridges interpreted as thrust moraines (Phillips et al., 2008). This deformation occurred during the A5 advance (Lee et al., 2017).
Previous investigations along the North Norfolk coast concluded that erosion occurred largely through mass wasting (Dickson et al., 2006;Lee, 2008), when the strength of the cliff materials cannot support the stresses imposed by gravity and by the waves. In their numerical model, Dickson et al. (2006) attributed a single value of material strength to represent all areas of till cliff, thereby neglecting the variability in material strength. The Quaternary glacial deposits that compose the Norfolk cliffs are, however, very heterogeneous in nature and probably play an important role in the erosion variability along the coast. They consist of layers of permeable sand, gravels, low-permeability clays, and diamicton that locally result in zones of low shear strength prone to erosion and failure (Frew, 2009). The ponding of percolating water along low-permeability geological structures, where the porewater pressure builds up, further reduces the shear strength and creates slip surfaces prone to mass wasting (Lee et al., 2011).
The mapping of the lithology and of these slip surfaces is therefore a valuable input for coastal erosion models.
So far, studies in North Norfolk have been mainly limited to 2D cross-sections along the outcrops, primarily along the shore.
From those results, understanding the character and processes by which these sequences form and how the geological features (notably possible décollement surfaces) are evolving inland is difficult because information is available in one direction only (Lee, 2008).
F I G U R E 2 Photograph of the cliff outcrop taken from the beach at three different times between 2014 and 2017. The limits of the sand lens are outlined with white dashes. Erosion affects mainly the softer sandy sediments within the lens, while other glacial tills are more resistant. By comparing the differences between July 2014 and April 2017, and between April 2017 and July 2017, one can see that the coastal retreat rate in 2017 was more important than within the previous three years [Color figure can be viewed at wileyonlinelibrary.com]

| MATERIALS AND METHODS
Previous studies have proven that GPR is an efficient method for mapping 3D glacitectonic structures and for investigating the lithology and depositional settings of glacial sediments (Bakker & Van Der Meer, 2003;Busby & Merritt, 1999;Fallon, 2012;Lønne et al., 2001;Overgaard & Jakobsen, 2001;Sadura et al., 2006;Wells, 2007). GPR was successfully used to image the inland extension of glacitectonic structures in North Norfolk between Sheringham and West Runton, but mainly using 2D common offset GPR (Carr et al., 2016). The maximum depth of penetration ranged between $24 and 6 m with 25 and 100 MHz antennas, respectively. Results demonstrated that the combination of three survey frequencies (25, 50, 100 MHz antennas) was required to enable identification and interpretation of glacial deposits.
Isolated 2D GPR profiles may be contaminated by out-of-plane reflections and may not be sufficient for imaging highly heterogeneous structures (McClymont et al., 2008). In contrast, 3D GPR data can provide high-resolution images of structures in any direction with approximately equal fidelity, facilitating the prediction of sediment distribution. 3D geophysical data are therefore considered more appropriate for understanding the processes that control the distribution of the glacial deposits in the North Norfolk coast.
3D ERT and topographic data from an unmanned aerial vehicle (UAV) were acquired in 2015. The 3D ERT survey comprised 11 parallel acquisition lines, with 8 m line spacing and 4 m electrode spacing along those lines. Data were acquired using a dipole-dipole array, which has been shown to provide superior resolution of subsurface features compared to Wenner or Schlumberger arrays (Chambers et al., 2002). The survey comprised measurements with dipole length (a) of 4, 8, 12, 16, 20, and 24 m, and dipole spacings (n) of 1 to 8a, and a complete set of reciprocal measurements for error analysis and data weighting in the inversion (Tso et al., 2017). The measured resistances were inverted using E4D (Johnson et al., 2010) with an L2 norm on the model misfit to gain a 3D model of the subsurface resistivity distribution. Overlapping photography from the UAV was processed using the structure from motion (SfM) technique to derive orthographic imagery, 3D point clouds, and ultimately a digital surface model. The photography was geocoded using a series of ground control points that were referenced using real-time kinematic (RTK) GPS.
The UAV digital surface model was filtered and smoothed to produce the 3D topography of the studied cliff section that was then used to build the surface and subsurface discretization for the ERT inversion.
Prior to inversion, data were filtered based on their reciprocal errors (removing all data with reciprocal errors above 10%). Measured reciprocal errors were used as data weights in the inversion under the condition that they were above a lower limit of 10% of the average reciprocal error level, which was calculated based on the errors between the 10th and 90th percentiles . A forward modelling error of 2% of the measured resistance was added to the reciprocal error. The inversion converged to a chi-squared misfit criterion of one, meaning that the model fits well the data within their errors. After convergence, the root-mean-squared misfit is equal to 2.4%, which is interpreted to represent the data noise.
100 MHz 3D GPR data and low-frequency (50 and 25 MHz) 2D data were acquired in 2017. The survey parameters were based on Annan (2003) and adjusted based on the field conditions to avoid aliasing and ensure a dense inline sampling. The survey parameters are summarized in Table 1 Common offset GPR survey data were acquired with GPS control using a Sensors & Software pulseEKKO Pro system. The antennas were mounted on a rough terrain smart-cart, with a GPS antenna positioned at the centre between the two GPR antennas, 0.88 m above the ground. Each GPR trace was associated with a GPR record by the central unit, with a GPS record on average every eight traces (40 cm). In addition to the GPR points recorded during the acquisition, RTK GPS points were also measured using a Leica Viva GS15 antenna at a coarser interval to calibrate the results of the 3D grid and position the 2D lines.
Three different source frequencies were used to obtain data of different resolution and signal penetration: the lower frequencies aimed at imaging the base of the sand lens and the general architecture, while the higher frequencies were recorded to image the internal stratigraphy of the sand basin at higher resolution. Additionally, common midpoint (CMP) lines were acquired in both the inline and crossline direction to estimate and monitor the wave velocity, with values ranging between 0.1 and 0.12 m/ns. Based on this velocity, the maximum penetration depths were estimated to be around 16, 10, and 2 m for the 25, 50, and 100 MHz antennas, respectively.
Datasets were processed using ReflexW (v.8.1) and Matlab. GPS data were processed to correct for mis-positioning, altitude correction, and coordinate system conversion (see Figure S1 in the online Supplementary Information). To correct for the altitude anomalies of the GPS points and to generate a regular positioning grid, a topographic surface was interpolated based on the RTK GPS records. The GPR traces were then assigned to the new set of regular GPS points.
The surveys of the different frequencies were processed following a common processing workflow designed to remove noise and bad traces, correct for acquisition settings, and compensate for signal attenuation before migration and depth conversion (see Figure S1).
Velocity models used for migration and depth conversion were based on the velocity analysis of CMP surveys and diffraction hyperbola analysis. CMP data could not be acquired at perfectly horizontal dipping layers and have been recorded only at three different points.
To calibrate the velocity, shallow CPT using a Panda dynamic cone penetrometer have been performed to test soils' cone resistance. It could only reach a depth of 4.5 m from the surface, thus the calibration could not be performed in the deeper part of the profiles ( Figure 4). The cone resistance can be related to the undrained shear strength and thus can be an indicator of the subsurface lithology (Langton, 1999). In this study, we approximate the relation between cone resistance and lithology without corrections. The measured data were filtered to remove anomalous measurements. The rapid increase in cone resistance from $7-10 MPa to $15-20 MPa is interpreted as the boundary between the soft sand and the hard tills. The low cone resistance at the intersection between the two is interpreted as a weak shear strength local plane, which is evidenced on the radar profile as the boundary surface between the two lithological units.
An average interval velocity of 0.13 m/ns for the sand lens and 0.1 m/ns for the till have been retained. The best migration results were obtained using a finite-difference scheme with a 15 approximation method. It should be noted that this migration scheme has

| Outcrop study
Three distinctive lithofacies assemblages were identified on the cliff outcrop ( Figure 5): • The WCF, consisting of grey interbedded gravels, sands, silts, and clays, with a weathered black bed at its lower part.
• Overlying the WCF, the WRM is a heterogeneous mixture of inter- Hence, the exact location of the lower boundary of the sand basin cannot be accurately imaged.

| GPR
The interpretation of GPR data followed the scheme defined by Neal (2004) to describe radar surfaces, radar facies, and radar packages. Radar packages, facies, and surfaces are defined based on the differences in patterns, continuity, and strength of reflections on the 100 and 50 MHz datasets. Three radar packages, four radar facies, and three radar surfaces have been defined ( Figure 5). The interpreted radar surfaces match well with the formation boundaries observed in the cliff.
The radar facies and the corresponding formations are summarized in Table 2. From the oldest to the youngest, they comprise: • f0, a reflection-free facies, with low-resistivity material (below 50 Ωm). It fits with the position of the WRM on the side of the basin, where it is predominantly clayey (hence the higher attenuation and lower resistivity), reducing the depth of penetration (Jol & Bristow, 2003). It forms radar package RP-1, associated with the deformed rock hosting the sand basin.
• f1, a low-amplitude, low-continuity reflections facies, with numerous overlapping diffractions. The boundary G-s0 is a horizontal low-amplitude, non-continuous, large-wavelength reflector dipping towards the west at an angle of 23 . Radar facies f1 is truncated at the top by the radar surface G-s1, which is marked by a convex strong-amplitude reflection and matches well with the isoresistivity surface at 1000 Ωm ( Figure 6). This facies displays varying resistivity decreasing downward from 1000 to 100 Ωm. It is assimilated to the ridge at the centre of the basin, which appears to be made of WRM and WCF on the outcrop, underlain by either WCF or chalk. It forms a complex, highly deformed radar package RP-2 found beneath the ridge. • f3, a continuous, concave parallel, high-amplitude, high-wavelength reflectors facies, onlapping the underlying radar surface G-s2. It comprises two units, separated by an unconformity that shows a variable thickness. f3 is related to medium to high resistivity between 200 and 1000 Ωm.
The radar facies identified are comparable to the facies described within different glacial deposits with comparable GPR methods (Jakobsen & Overgaard, 2002;Overgaard & Jakobsen, 2001). By comparison with these studies, radar facies f0 and f1 correspond to   Concerning the ERT, inversion results are inherently non-unique, and an infinite number of solutions exist that explain the data equally well (Olayinka & Yaramanci, 2000). The sensitivity of geoelectrical measurements also decreases with depth, thus lowering the resolving capability of an array in the inversion of the data (Szalai et al., 2007).
In our case, data were acquired along 200 m-long parallel transects, with a depth of investigation of approximately 30 m (Oldenburg & Li, 1999). This depth is close to the depth of the lower boundary of the sand lens, and hence interpreting the lower boundary of the sand lens is challenging. Finally, the smoothness constraints will cause even a sharp interface to be smooth. These factors make it difficult to determine an accurate depth of the sand lens from the ERT data.
By comparing the results with the outcrop, the depth estimated by the GPR data is considered more reliable than that estimated by the ERT. However, the ERT provides more insights on the geology beneath the sand lens (particularly at the shallower sections), where the GPR signal is strongly attenuated. Joint inversion of the ERT data using the GPR surfaces could help to characterize the geological features of interest and improve the 3D resistivity model of the investigated site. Studies by Doetsch et al. (2012) or Merz et al. (2015), for example, used GPR to image the 3D shape of the bedrock topography, and employ it as a 2D structural constraint for tomographic inversions of seismic and geoelectric data. A similar approach could also aid here to better define the depth of this deeper interface. including several steep-sided ice contact hills that occur along the North Norfolk coast (Lee et al., 2017).

| Geological model
In a previous study, Burke et al. (2009)

| Constraining erosion rates in space through geophysical methods
The lithology is likely to be the main driver of the high erosion rate of this sand body. Results from the CPT show that the RSGM is characterized by a low cone resistance, related to the relatively weak shear strength of the sand. Our results show that these sand basins prone to erosion can be easily mapped in 3D using ERT methods, where the sand is characterized by a high resistivity (above 1000 Ωm). But the high rate of erosion is also linked to the presence of low-permeability surfaces that dip towards the beach, highlighted by the radar surfaces G-s1 and Gs-2 on GPR results. These surfaces are outlined by a strong low-frequency reflector before a sudden loss of GPR signal, particularly at the centre of the basin, and a high-resistivity contrast around 1000 Ωm ( Figure 5).
The strong attenuation of the radar signal along these surfaces could be related to a high clay or moisture content or a strong signal scattering. The presence of lacustrine marls at the centre of some of the sand basins has been reported by Phillips et al. (2008), thus favouring the presence of clay-rich low-permeability surfaces.
Such impermeable beds could cause ponding of the percolating water along these surfaces. The low-resistivity anomaly observed in the upper RSGM seems to confirm this hypothesis. The resistivity of the upper RSGM goes down to 100 Ωm, far below the expected resistivity for dry gravel and sands, which usually range between 500 and 10 000 Ωm (Palacky, 1987). This low anomaly can be linked to a high moisture content within the upper RSGM. An increased moisture content generates a higher pore pressure, which in turn increases the shear stress along these surfaces (Souisa et al., 2015).
A high shear stress can facilitate slope failure processes. Even though the high erosion rate of these sand basins is primarily linked to their weak shear strength, the mass-wasting process, notably after heavy rains, could accelerate the erosion of these sand basins.
Using geophysical methods, these low-permeability surfaces could be regionally mapped to better constrain erosion trends along the coast.
It should be noted that the current coastal protection plan intends to let these sand lenses erode naturally (AECOM Limited, 2010). This is because the eroded sand delivered from these basins to the beach is a finite sediment supply feeding the adjacent beaches. Consequently, their erosion seems more beneficial to coastal protection than their preservation.

| Limitations and perspectives
This study has tested the efficiency of ERT and GPR methods to investigate subsurface properties that can influence erosion.
ERT was able to clearly define the outline of the sand basin and its internal structure. It also provided information on the corresponding lithologies and likely water saturation. However, due to the inherent limitations of this technique, accurate imaging of the true depth of the features remains challenging. This limitation was addressed using GPR data, which imaged the interfaces at significantly higher resolution.
The use of 100 MHz antennas for investigating features of this type and anticipated geometry is not favoured. In this study, 100 MHz was chosen for both practical reasons (it was not possible to mount the 50 Hz antennas on the cart) and because of an incorrectly interpreted depth of the base of the basin during preliminary testing. The 3D GPR 100 MHz data failed in imaging the entirety of the sand basin. They provided the best resolution, but the weak depth of penetration only enabled the imaging of the uppermost units. The 3D GPR 100 MHz data could potentially reach the bottom of the shallow subbasin made of facies f3 (the surface G-s2), which was clearly visible on depth-slices in the 3D volume, but the centre of the shallow basin could not be imaged due to the high attenuation. It also suffered from a low signal-to-noise ratio, which was difficult to compensate for during processing. The geometry of the sand basin was essentially defined with the 50 MHz data, but the resolution of this dataset is low (above 1 m), and only five 50 MHz GPR 2D sections were available. Isolated 2D GPR profiles may be contaminated by out-of-plane reflections and may not be sufficient for imaging highly heterogeneous structures (McClymont et al., 2008). Therefore, even though the results match well with ERT data, uncertainties remain regarding the exact geometry of the basin and its structural settings. Despite these drawbacks, 2D GPR sections have proven very useful for imaging the internal structures related to the basin's depositional history.
They offer a resolution in depth which could not be reached with the ERT method.
For future investigations, acquiring 50 MHz 3D data or a dense 50 MHz 2D data grid could allow an improved assessment of the geometry of the basin and how it evolves over time during ongoing coastal erosion. Including the collection of CMP data throughout the study area, with an offset of at least 1.5 times the desired depth of investigation, is crucial to better constrain the migration and the timeto-depth conversion.

CONCLUSIONS
A representative cliff on the North Norfolk coast, UK, where erosion is controlled by the distribution of soft sediments and mass-wasting events, has been studied using a multidisciplinary approach combining 3D ERT, multi-frequency 3D and 2D GPR, geotechnical tests (CPT), and an outcrop study. Based on this example, we show that this approach allows building a detailed ground model of retreating coastal areas that can contribute to a better prediction of coastal recession.
ERT proved to be an efficient method to quickly delineate the basin and estimate the moisture content distribution within the sub- Interpretation of GPR and ERT data showed that the three identified radar surfaces (G-s0-G-s2) are characterized by an abrupt change in resistivity and a sudden high attenuation of the GPR-emitted waves that are associated with an increasing clay and/or water content. These surfaces appear impermeable enough to retain water, hence locally building up shear stress and reducing the shear strength, further increasing the risk of sand slides along these planes. GPR methods can be used to map these surfaces prone to mass events. The radar surfaces can also be used to structurally constrain the inversion of the ERT data and improve the accuracy of the estimated sand basin volumes.