Application of electrical resistivity tomography for geotechnical assessment near the Vistula flood protection infrastructure

Flood embankments and sheet pile walls protect urban areas from river floods (Knox et al., 2022). How they behave under load and water pressure depends strongly on the soil around them (Foster et al., 2000). Warsaw is a city where housing estates are built close to the Vistula floodplain, so knowing the ground conditions along the flood defenses is a practical necessity and not only of academic interest.

Drilling is still the standard way to obtain such information. Boreholes give detailed data, but only at one point. When soils change quickly in the lateral direction, which is typical for alluvial plains, borehole data can miss local features that matter for embankment stability. Non-invasive methods that image the subsurface continuously are therefore of growing interest for this kind of site.

ERT is one of the most popular geoelectrical methods for shallow investigations (Loke et al., 2013). The method measures how the ground resists electric current. Resistivity changes with grain size, porosity, water saturation and ion content in pore water. By placing many electrodes on a line and running a numerical inversion, a 2D section of true resistivity is obtained (Loke & Barker, 1996). ERT has been used for landslide studies, dams, levees, foundations and detection of underground voids or pollution (Chambers et al., 2006; Sjödahl et al., 2008).

The choice of measurement protocol affects both the depth of investigation and the resolution. The gradient array gives a good balance between lateral resolution and signal strength, which is why it is often used for general surveys. The dipole-dipole array gives better lateral resolution but is more sensitive to noise. Urban sites with metal structures and buried cables are a typical problem for this array (Dahlin & Zhou, 2004).

Most published ERT work on embankments and levees comes from rural or semi-rural sites, where electromagnetic noise is limited and the study is usually focused on seepage or internal erosion (Sjödahl et al., 2008). Reports from dense urban flood-plains are less frequent. Direct comparison of two protocols along the same line is also rare. The present work is different in three ways: First, the site lies in a built-up part of the Vistula floodplain, with buildings, buried utilities and metallic sheet piles close to the survey line; Second, gradient xl and dipole-dipole were applied on the same profile, so the two protocols can be compared under the same field conditions; Third, the interpretation is geotechnical rather than purely hydrogeophysical, and the results are meant to support the design and verification of flood protection structures.

The paper presents an ERT survey made along the Wał Zawadowski embankment in Wilanów, close to the Larsen sheet pile walls. The aim was to describe the ground-water and lithological conditions down to about 25 m – 30 m and to compare the two protocols in this urban setting.

The fieldwork was carried out on 5 November 2025. Two locations were surveyed along Wał Zawadowski, in the Zawady housing estate in Wilanów, Warsaw (Fig. 1). The area lies in the southern part of the city, next to the Vistula flood embankment, where Larsen sheet pile walls form part of the flood protection system. Each profile was 120 m long and was oriented along the embankment.

Fig. 1.

Location map of the ERT profiles near the Vistula flood protection infrastructure (own research)

Data was collected with an ABEM Terrameter LS, a multichannel resistivity and IP meter. The instrument was connected to four multicore cables with 21 take-outs every 2 m. Each profile was laid out in a straight line, giving a total length of 120 m. The controller chose the quadrupoles automatically, according to the selected protocol. An acquisition time of 1.0 s was used, with stacking of 3 readings per measurement. The maximum output voltage was set to 200 V and the current range was 1 mA – 1000 mA. The measured apparent resistivities were then inverted in 2D to obtain the true resistivity distribution.

Two protocols were applied. At Location 1, the line was surveyed twice: once with gradient xl and once with dipole-dipole. The gradient xl protocol reached a depth of about 24 m – 26 m (close to 1/5 of the profile length). Dipole-dipole reached about 30 m and yielded a higher number of datum points (957, compared with 820 for gradient xl), but with a weaker signal. At Location 2, only gradient xl was used. Topographic corrections were applied with a digital elevation model from the Geodesy and Cartography State Institution (GUGiK). Processing and inversion were done in Res2Dinv.

Data quality was controlled in two stages. In the first stage, raw values with negative apparent resistivity were removed. In the second stage, the repeatability of the measurement was checked.

Inversion was run with the smoothness-constrained least-squares scheme (L2 norm) available in Res2Dinv (Loke & Barker, 1996; Loke et al., 2013). The cell width was set to 1 m, which is half of the electrode spacing. The cell depth was increased with depth using the default scheme of the software. Topography from the GUGiK model was included directly in the finite-difference mesh. A damping factor of 0.15 was used, together with a vertical-to-horizontal flatness ratio equal to 1.0. Iterations were stopped when the relative change in RMS misfit was smaller than ~15 %. In practice, the solution was stable after 7 to 10 iterations in each profile. Areas with low sensitivity, based on the Jacobian coverage, were excluded from the geological interpretation.

The two protocols are complementary, not equivalent. The gradient xl array uses remote current electrodes and several simultaneous potential pairs (Dahlin & Zhou, 2006). This gives a strong signal and fairly uniform coverage along the line. In a noisy urban environment the array is robust, but vertical resolution in the deeper part of the section is limited. The dipole-dipole array uses short current and potential dipoles, systematically shifted along the line. It reaches deeper for the same profile length and gives better lateral resolution. On the other hand, the signal becomes weak at large n-factors, and the response to near-surface metallic objects and buried cables is stronger (Dahlin & Zhou, 2004). In our survey, the consequences of these differences are visible both in the RMS misfit (Table 1) and in the character of the sections, which is discussed in Section 3. It is also worth noting that the time required to conduct a geophysical survey using the dipole-dipole method is significantly longer (3–4 times) than in the case of the gradient array.

The interpretation of the sections was based on resistivity contrasts that can be linked to changes in lithology, water saturation and content of fine-grained or organic material. Low-resistivity zones were read as saturated or clay-rich material. High-resistivity zones were read as unsaturated, sandy or loose deposits. The quality of each inversion was expressed by the RMS misfit between measured and calculated apparent resistivities. Table 1 summaries the survey parameters.

All three profiles produced resistivity sections of acceptable quality. For the two gradient xl profiles (1a and 2) the RMS misfit was below 5 %, which indicates good agreement between the model and the data. For the dipole-dipole profile (1b) the misfit was about 15 %. This higher value is consistent with the stronger sensitivity of the dipole-dipole array to noise from buried utilities present in the survey area.

The sections show a three-layer structure, the same in all profiles (Fig. 2). The interpretation is given in Table 2 and described below.

Fig. 2.

Inverted resistivity sections: a) Profile 1a, Location 1 (gradient xl); b) Profile 1b, Location 1 (dipole-dipole); c) Profile 2, Location 2 (gradient xl) (own research)

Layer I is the uppermost and reaches a few meters below the ground surface. Resistivity values in this layer are between 185 Ωm and 700 Ωm. The layer was interpreted as mixed-grain sands with a high proportion of fine and silty sand. The high resistivity points to an unsaturated and loose material. This is typical for near-surface fill and reworked alluvial deposits that often occur near embankments. Full lithological identification still requires confirmation from reference boreholes.

Layer II has an average thickness of about 10 m and resistivity values between 10 Ωm and 185 Ωm. It was interpreted as saturated sandy deposits, possibly with clay admixtures or thin clayey sand interbeds. The drop in resistivity compared with Layer I results from two effects: the presence of groundwater and the higher content of fine particles. Both effects lower the resistivity of the medium.

Layer III starts at about 15 m depth and continues below the prospection limit. Resistivity is very low, below 10 Ωm. The layer was interpreted as saturated alluvial muds, most likely silty clays with some organic matter and with high plasticity. The very low resistivity values result from full saturation, from a high share of clay minerals, and from decomposed organic material.

The comparison of Profiles 1a and 1b, which were recorded at the same place, shows that the general layer structure is the same for both protocols. The dipole-dipole section, however, contains more localised anomalies. These anomalies are linked to metallic objects and utility lines in the shallow subsurface. The dipole-dipole array produces a denser set of measurement points and is more sensitive to such objects, so this result is not surprising. The dipole-dipole array is highly effective for resolving lateral resistivity variations, but this sensitivity also makes it more vulnerable to external disturbances.

In an urban setting, the presence of metallic objects and underground utility networks may introduce artificial anomalies, increase noise levels, and locally distort the measured apparent resistivity values. Consequently, the dipole-dipole results obtained in this study should be treated with caution, as some of the observed anomalies may reflect anthropogenic interference rather than true subsurface heterogeneity. Overall, interpretational reliability of the method in areas affected by dense underground infrastructure is one of the main limitations, alongside high-conductivity soils and the requirement of good ground coupling (Hu et al., 2026).

The three-layer model fits what is known about the Vistula floodplain in Warsaw. The shallow sands match Holocene alluvial deposits that are common in the river terrace zone. The deep low-resistivity layer matches fine-grained sediments that accumulated in the old channel and floodplain environment (Sarnacka, 1992).

Working with ERT near sheet pile walls and flood embankments has two sides. On one side, the method gives a continuous 2D image of the subsurface, which bore-holes alone cannot provide. The continuous image helps to identify lateral changes in lithology or saturation that may form seepage paths along or under flood protection structures (Sjödahl et al., 2008). On the other side, metallic sheet piles and buried cables can distort the data, especially when a protocol with high lateral sensitivity, such as dipole-dipole, is used (Dahlin & Zhou, 2004).

The difference in RMS misfit between the two protocols in our data is a clear example of this trade-off. The gradient protocol worked well and gave low misfit. The dipole-dipole protocol reached greater depth but was strongly influenced by electromagnetic noise from nearby infrastructure. For similar urban conditions, a practical conclusion is simple: start with the gradient protocol; use dipole-dipole only where extra depth is needed and where the noise level allows it. This infrastructural sensitivity may also serve as a tool for identifying and pre-surveying sites where technical records are incomplete or precise mapping is lacking.

From the geotechnical point of view, the layer structure has practical meaning for the Larsen sheet pile walls. The saturated zone starts at relatively shallow depth and is thick, which implies that the sheet piles should reach well into the intermediate or deep layer to provide sufficient stability and to control seepage. The very low resistivity and the high plasticity of the deepest layer indicate weak bearing capacity and high compressibility. The ERT profiling results are coherent with archival lithological profiles – boreholes No. M54-20 and M54-24 (Fig. 3), located near ERT cross-section No. 1 (Fig. 1). They confirm the presence of Holocene fluvial non-cohesive formations at depths of a dozen meters below ground level, in agreement with the Warsaw Engineering-geological Atlas (Frankowski & Wysokiński, 2000).

Fig. 3.

Geotechnical borehole profiles No. M54-20 (1982) and No. M54-24 (1962) from the study area (own research)

Using non-invasive geophysics as the first step of a site investigation also matches the general trend towards more sustainable ground studies. If the main subsurface features are mapped before drilling, the number of boreholes and their positions can be optimised. This lowers cost and reduces environmental impact (Chambers et al., 2006).

Near flood protection structures, where unnecessary ground disturbance is undesirable, this approach fits the idea of sustainable construction practice. The presented approach can support and optimise decision-making processes in the design, monitoring and maintenance of hydraulic and geotechnical structures (Su et al., 2024).

The ERT survey along the Wał Zawadowski embankment in Warsaw demonstrated that this non-invasive method is useful for a first-stage geotechnical characterisation of the subsurface near flood protection infrastructure. The main findings of the study can be stated as follows.

A consistent three-layer model was found in all profiles. It consists of near-surface unsaturated sands (with a high proportion of fine and silty sand), an intermediate saturated sand layer with some silt and clay, and a deep layer of saturated alluvial muds with organic matter and high plasticity. This sequence is typical for the Vistula floodplain and has direct meaning for the design and verification of sheet pile wall foundations.

The gradient xl protocol was the more reliable choice for this urban site, with RMS misfit below 5 %. The dipole-dipole protocol was strongly affected by noise from underground infrastructure and gave about a 15 % misfit. A multi-protocol approach is therefore recommended: gradient xl for the main characterisation and dipole-dipole for selected deeper prospection. The direct comparison of both protocols on the same line, which is the main methodological element of this work, made possible a realistic evaluation of their performance under the same field conditions.

The results also provide a spatial framework for the planning of targeted bore-holes and support the use of non-invasive geophysics in geotechnical workflows for flood protection projects in urban floodplains. Further work should include a comparison with new borehole data and repeated monitoring surveys to strengthen the interpretation and to extend its applicability.