Exploring spatial variability in the physical and chemical properties of urban soils using electrical resistivity tomography

1 Introduction

Urban soils constitute a key component of urban ecosystems because their physical, chemical and biological properties underpin numerous ecosystem services, including stormwater regulation, carbon sequestration, biodiversity conservation, vegetation development and climate regulation (1, 2). They also support urban agriculture and green infrastructures, thereby contributing to food security, human well-being and the resilience of cities to environmental change (3, 4). As urbanization continues worldwide, the sustainable management of urban soils has become an important challenge for urban planning and environmental policies (5).

Unlike agricultural or forest soils, urban soils are characterized by an exceptionally high degree of spatial heterogeneity resulting from the superposition of natural pedogenetic processes and successive anthropogenic disturbances. Excavation, backfilling, sealing, compaction, demolition, and the importation of exogenous materials create complex mosaics of soil horizons and anthropogenic deposits over very short distances (6, 7). Consequently, the physical and chemical properties of urban soils often exhibit abrupt lateral and vertical variations in texture, bulk density, porosity, organic matter content, pH, nutrient availability and contaminant concentrations (8–10). Trace metals and organic pollutants inherited from former industrial activities, traffic emissions and construction materials are frequently distributed in localized hotspots, further increasing the spatial complexity of urban soils (11–13). This high degree of heterogeneity challenges conventional soil sampling strategies, which often fail to adequately represent the spatial organization of soil properties and associated environmental constraints.

Over the past two decades, near-surface geophysical methods have become valuable tools for investigating soil spatial variability because they provide rapid, non-invasive and spatially continuous information at resolutions that cannot be achieved through conventional sampling alone (14, 15). Their development has been driven primarily by agricultural and environmental applications, where they have considerably improved the characterization of soil variability for digital soil mapping, precision agriculture, hydrological investigations and the monitoring of soil degradation (15–19). A wide range of complementary techniques—including electromagnetic induction (EMI), ground-penetrating radar (GPR), electrical resistivity tomography (ERT), seismic methods and gamma-ray spectrometry—is now routinely used to investigate soil heterogeneity. These methods respond to different physical properties of the subsurface, such as electrical conductivity, dielectric permittivity, seismic velocity or natural radioactivity, which are themselves controlled by soil texture, water content, porosity, bulk density, clay mineralogy, soil structure and the presence of coarse fragments or anthropogenic materials (14, 20). Although geophysical measurements provide indirect information on soil properties, their integration with conventional pedological observations substantially improves the characterization of soil spatial organization while reducing the intensity of destructive sampling (15, 17). Beyond soil characterization itself, geophysical surveys also provide a spatial framework for designing optimized sampling strategies by identifying homogeneous zones and major subsurface discontinuities before field sampling.

Among these approaches, electrical resistivity tomography (ERT) has received particular attention because it provides high-resolution two- and three-dimensional images of the spatial distribution of subsurface electrical resistivity. The method is based on the injection of electrical current into the ground and the measurement of resulting potential differences, from which the true resistivity distribution is estimated through inversion procedures (21, 22). Soil electrical resistivity is governed by several interacting factors, including water content, pore connectivity, texture, clay mineralogy, porosity, and the abundance of coarse fragments or anthropogenic materials (14, 20). Consequently, resistivity integrates the effects of several key soil attributes and provides a physically meaningful representation of subsurface heterogeneity rather than direct measurements of individual soil properties.

In agricultural and natural soils, ERT has been widely used to characterize soil horizons, investigate soil structure, monitor soil water dynamics and delineate zones of contrasting physical properties (14, 20, 23). These characteristics make ERT particularly attractive for investigating urban soils, where abrupt changes in parent materials, heterogeneous fills and anthropogenic horizons challenge conventional sampling approaches. Despite these advances, the application of ERT to urban soils has remained comparatively limited, and its potential extends beyond the characterization of physical heterogeneity. Because the distribution of anthropogenic materials largely controls both the physical architecture and the geochemical composition of urban soils, electrical resistivity patterns may also reflect the spatial organization of chemical properties and contaminant distributions. To our knowledge, however, few studies have explicitly evaluated whether ERT-derived resistivity patterns can be used to optimize soil sampling and predict the spatial variability of physicochemical and geochemical properties in highly heterogeneous urban soils.

The objective of this study was therefore to evaluate the potential of ERT as a methodological tool for characterizing urban soil heterogeneity. We hypothesized that, in anthropogenic soils where both physical and chemical properties are strongly influenced by the nature and spatial arrangement of human-derived materials, electrical resistivity may serve as a proxy for broader environmental constraints. To test this hypothesis, we combined high-resolution ERT imaging with a geophysically informed sampling strategy in a urban site located in northern Paris (France). Soil samples collected from contrasting resistivity zones were characterized for their physicochemical properties, organic matter composition using mid-infrared spectroscopy (MIRS), and concentrations of major and trace elements using X ray fluorescence spectroscopy (XRF). We then investigated the relationships between electrical resistivity and these variables using multivariate statistical analyses. By integrating geophysical imaging with detailed laboratory characterization, this study assesses whether ERT can improve sampling design and provide a robust framework for interpreting the spatial organization of soil physicochemical and geochemical variability in urban environments.

2 Materials and methods

2.1 Study site

The experiment conducted during this study was carried out on a plot of approximately 1000 m² located on the IRD site in Bondy, within the Seine-Saint-Denis department in the Paris metropolitan area (Figure 1). Prior to the experiment, the area was covered with grass and was infrequently used. According to the land-use plan of the city of Bondy, the site lies over a succession of alluvial deposits: Polygenetic colluvium, scree; Marl with Pholadomya ludensis and gypsum foundation (fourth stratum); Massive and marly gypsum layers. These data are derived from the 1:50,000-scale geological map produced by the BRGM (French Geological Survey). Historically, the city of Bondy has undergone continuous transformation. In the mid-18th century, the area was known for its dense and poorly frequented forest, as shown in the 1756 Cassini map. During the 19th and early 20th centuries, Bondy became a popular destination for Parisians seeking rural retreats. The development of a railway station facilitated population movements and spurred the construction of numerous detached houses. Over time, Bondy continued to expand and urbanize in parallel with the growth of Paris. Very few historical records regarding past land use on the study site were found. However, the site was significantly altered during the construction of the IRD research center in 1947. At that time, many gardens, vegetable plots, and refuse areas were consolidated, and the topography was substantially modified through the import and leveling of various materials, which were subsequently covered with a layer of forest-origin topsoil that was initially uncontaminated. The climate of this region is temperate. Winter and summer temperatures range from 0 to 10 °C and 15 to 25 °C, respectively. Precipitation is relatively stable throughout the year, ranging between 40 and 60 mm per month (data were recorded at Le Bourget meteorological station, located approximately 8 km from the study site).

2.2 Electrical resistivity tomography

The ERT survey was carried out along linear profiles using SYSCAL Pro resistivity (Iris Instruments) that operates quadrupoles sequentially selected among seventy-two electrodes (A and B for electrical current injection; M and N for the measurement of resulting electrical potential at ground surface). The survey was conducted under the prevailing winter field conditions and was not performed after equilibration of the soil at a standardized field-capacity condition. Soil volumetric water content and pore-water electrical conductivity/salinity were not measured simultaneously with the ERT acquisition. Consequently, the measured bulk resistivity represents the combined response of intrinsic soil properties and the hydrological conditions prevailing at the time of the survey. Measurement parameters were chosen as follows: Injection time: 500 ms; potential difference between injection electrodes (V_AB): 100 V; Intensity of electrical current between injection electrodes (I_AB): ~200 mA (depending on soil resistance, following Ohm’s law: U = R·I). Measurements were performed using two electrode arrays namely: reciprocal Wenner-Schlumberger (WSr: M A B N) and Dipole-Dipole (DD: A B M N). The WSr array is well adapted to detect vertical contrasts in electrical resistivity, while the DD array better captures horizontal contrasts (24). WSr is also characterized by a better signal-to-noise ratio and a greater depth of investigation compared to DD (25). The total number of measurements collected was 1,870 for DD and 2,282 for WSr. Short (35x0.75 = 26.25m) transverse profiles (C) were spaced every 6 meters starting from x = 3 m, while long (71x0.75 = 53.25m) longitudinal profiles (L) were spaced every 1.5 meters. The unit electrode spacing was 0.75 meters for both sets of profiles. The ERT survey was conducted during the winter field campaign under relatively moist soil conditions, which favored good electrode contact and signal quality. The survey was designed to characterize spatial variability under these prevailing field conditions and was not intended as a time-lapse or seasonal monitoring experiment. Data interpretation was performed using RES2DINVx64 v4.06.10 and RES3DINVx64 v3.11.55 (Geotomo Software; 26, 27). Spatial variation maps of calculated resistivity as a function of depth were generated using Surfer^® v13 (28). Statistical processing of the vertical resistivity profiles was conducted using Coheris Analytics SPAD v9.2.13 (Coheris) (29), applying Principal Component Analysis (PCA) and Hierarchical Ascending Classification (HAC). This analysis enabled the establishment of a typology of geo-electrical profile types and the subsequent delineation of main soil mapping units. The geoelectrical classification was established independently of the subsequent soil physicochemical measurements. The latter were therefore used as an external assessment of whether the geoelectrical units corresponded to meaningful differences in measured soil characteristics.

2.3 Soil physico-chemical analyses

Soil analyses were conducted on samples collected in each type of the four soil category (n=10, n=9, n=6 and n=5, respectively for C1, C2, C3 and C4) and at four different depth (totalizing 120 samples). The particle size distribution was determined by sieving and sedimentation using Robinson pipette (ISO 11277:2020) after CaCO₃ and organic matter removal (using HCl and oxygen peroxide, respectively) and soil aggregates separation (using sodium hexametaphosphate as a dispersing agent). The pH_Waterand pH_KCl were measured in a 1:5 ratio of MilliQ water (conductivity <0.5 µS cm^-¹) or KCl solution to sample, using a pH meter (Mettler Toledo^®), in accordance with ISO 10390:2021. The organic carbon (C) and nitrogen (N) content of the samples were determined using an elemental analyzer (Flash HT, ThermoFisher, Waltham, MA, USA) after CaCO₃ removal using HCl. Assimilable phosphorus was determined using the modified Olsen method and inductively coupled plasma mass spectrometry (Agilent 7500cx ICP-MS) in accordance with ISO 11263:1994. Cation exchange capacity (CEC) and exchangeable cation (Ca²⁺, Mg²⁺, K⁺ and Na⁺) concentrations were measured using a cobalt (III) hexamine chloride solution and ICP-MS measurements, in accordance with ISO 23470:2018. Total soil concentrations of macroelements (Al, Fe, K, Mg, P) and trace elements (As, C, Cr, Cu, Ni, Pb, Zn) were measured with ICP-MS after perchloric acid (HClO₄) and hydrofluoric acid (HF) soil digestion (ISO/TS 16965:2013).

2.4 Organic matter composition characterization

The molecular composition if the OM was characterized using mid-infrared spectroscopy (MIRS) with a Fourier transform infrared spectrophotometer (FTIR 660, Agilent Technologies, Santa Clara, California, USA) operating in diffuse reflectance mode. Measurements were made at 2 cm^-1 intervals in the MIR range (4000–400 cm^-1) and converted to absorbance. The raw spectra were subjected to preprocessing, including Savitzky-Golay smoothing polynomial filter with a 13-point window), baseline correction (polynomial fitting with degree 3) and normalization by the sum of absorbance across 4000–400 cm^-1. Then, analyses were focused on five spectral regions corresponding to specific C functional groups: (1) the saturated hydrocarbons region 2750–3025 cm^-1 corresponding to alkyl C; (2) the unsaturated region 1700–1760 cm^-1 and 1576–1618 cm^-1 corresponding to C=O and aromatic C=C bonds respectively; (3) the polysaccharide region 1222–1287 cm^-1 corresponding to O-Alkyl C, and (4) the aromatic CH region 852–898 cm^-1. The selected wavebands were finally used to calculated the hydrophobicity index, the index of oxidation, the aromaticity index, the maturation index, the polysaccharide content, the condensation index and the microbial transformation, as described by Puissant et al. (30).

2.4 Statistical analyses

Statistical analyses were performed using R and SPAD v7 (Cohéris). Variance partitioning was first used to assess the effects of soil classes defined by ERT profiles along depth. As the datasets showed non-Gaussian distributions and abnormal residuals, non-parametric methods were preferred, as they do not assume normality and are less sensitive to outliers. To evaluate significance, values were compared to the median and quartiles of the dataset: very significantly high (x > Q3), significantly high (Q3 − 5% < x ≤ Q3), significantly low (Q1 ≤ x < Q1 + 5%), and very significantly low (x < Q1). Kruskal–Wallis tests followed by post-hoc pairwise comparisons were performed in R. Principal component analyses (PCA) were conducted on physico-chemical properties, major and trace elements, and MIRS spectra, followed by between-class analysis (BCA) using the ade4TkGUI package. Differences between treatments were assessed using a Monte Carlo permutation test (9999 permutations).

3 Results

3.1 Geophysical analyses

Geophysical analyses provided fourteen L-profiles acquired using the Dipole–Dipole (DD) configuration and the Wenner-Schlumberger (WS) configuration (Supplementary Figure S1). Subsurface responses vary significantly depending on the chosen electrode configuration, due to differences in depth of investigation and sensitivity. Apparent resistivity values range from approximately 20 to 90 Ω·m for the WS configuration, and from around 15 to 110 Ω·m for the DD configuration. As previously discussed, these configurations do not provide identical information; therefore, combining both methods is advantageous for generating more comprehensive resistivity maps. Results obtained were then compared with different electrode configurations (WS and DD) and orientations (C and L) (Supplementary Figure S1). Two depth intervals—D1 (0–0.25 m) and D4 (0.9–1.3 m)—were selected to investigate resistivity variations. For the shallow layer D1, once again, notable differences are observed between the WS and DD configurations. The DD configuration yields more pronounced resistivity contrasts, likely due to its higher sensitivity at shallow depths compared to WS. This enhanced sensitivity allows DD to better resolve near-surface variations, which is reflected in the combined resistivity map that integrates both configurations and orientations (WS-DD_L-C). In the deeper layer D4, differences between the two configurations persist but are less pronounced than in D1. The trench located at x = 5, clearly visible across all D1 maps, is only detectable in D4 using the DD configuration. This is attributed to DD’s higher sensitivity, even at depth. In this case, combining both configurations proves useful to mitigate deeper artifacts caused by near-surface sensitivity effects in the DD configuration. Using a single configuration, such as Dipole–Dipole (DD), may be sufficient when the objective is to investigate the most superficial soil layers. However, the Wenner–Schlumberger (WS) configuration remains highly valuable for deeper prospecting, particularly to compensate for near-surface artifacts commonly associated with DD. At both investigated depths and for both configurations, incorporating the C-profiles in addition to the L-profiles does not significantly affect the inversion results. However, model convergence tends to be slower when C-profiles are included, suggesting that the addition of a second orientation does not provide substantial complementary information.

Based on the most comprehensive resistivity model, the subsurface heterogeneity with depth can be represented (Figure 2). These maps reveal the spatial organization of distinct layers characterized by their resistivity values, highlighting significant variability both laterally and vertically. Resistivity maps are displayed using logarithmic scales to better capture contrasts across layers. A linear anomaly observed at x = 5, extending down to approximately 90 cm depth, corresponds to a trench. Additional trench-like features are visible near the center of the site, radiating from the central manhole, though these appear shallower. Vertical changes in resistivity reflect the evolving composition of the soil with depth. At the surface, a first layer with resistivity around 50 Ω·m is observed, corresponding to the topsoil layer distributed across the site. Below this, starting at approximately 25 cm depth, a second layer appears with markedly lower resistivity values, indicative of coarse backfill material. The thickness of this second layer varies across the site: it persists through the third, fourth, and fifth layers along the eastern edge, whereas in other areas—particularly from the third to the sixth layer—high-resistivity materials emerge, with values reaching up to 315 Ω·m. At greater depths, intermediate resistivity zones (~102 Ω·m) are detected in the southern and northeastern parts of the site, possibly corresponding to sandy pockets. These lateral variations in resistivity suggest differences in soil composition and structure, which may influence local processes and dynamics depending on the specific area. It should be noted that these resistivity contrasts were obtained under the prevailing moisture conditions during the survey and should therefore be interpreted as spatial contrasts under the surveyed hydrological state rather than as resistivity values normalized to a common soil water content or field-capacity condition.

3.2 Geo-electrical profiles typology

The 3D inversion of the dataset enables the generation of vertical resistivity profiles across the site, with a spatial resolution of 37.5 × 37.5 cm². Two combined statistical analyses—Principal Component Analysis (PCA) and Hierarchical Cluster Analysis (HCA)—were used to classify all resistivity profiles by depth into four distinct groups (Figure 3), which served as the basis for the resulting classification map. The selected classification scheme which maximized between-class discrimination while minimizing within-class variability, consists of four classes, each representing a distinct vertical resistivity pattern. While resistivity values are relatively similar across classes in the uppermost layer (0–0.5 m), they become highly discriminant between 0.5 and 2.5 m depth. Types 1, 2, and 4 exhibit peak resistivity values at approximately 1 m depth, but with varying magnitudes: in Class 1, both the maximum and median resistivity values remain below 100 Ω·m; in contrast, Class 4 reaches a maximum resistivity exceeding 300 Ω·m, with median values around 220 Ω·m.; Classes 2 and 3 are intermediate, showing a more pronounced increase in resistivity within the central depth range than Class 1, but significantly lower than Class 4.; Class 3 is notably distinct from the other three, characterized by a near-linear increase in resistivity with depth and high values beyond 2.5 m. Following the soil typology, a 30 auger surveys (Figure 3) were distributed according to a proportionally stratified sampling design, considering the surface area of each mapped resistivity unit (Supplementary Figures S3–S7). Because the geoelectrical classes were defined before soil sampling, the subsequent soil measurements provided an independent assessment of the correspondence between geoelectrical units and measured soil variability.

3.3 Physical and chemical properties

Soil texture was primarily controlled by depth (Table 1), with only a weak effect of soil class (P < 0.1). Classes C3 and C4 were relatively enriched in fine silt and clay. Overall, the soil profiles exhibited a coarse texture, with sand contents of 71% in layer A, 74% in layer B, and 76% in layers C and D (Table 2). Higher proportions of clay and silt were generally observed in the surface layer (A). Significantly coarser textures occurred in layers B and C of soil class C4, which was also characterized by a relatively high sand content and low clay content in the surface layer A (72.1% sand and 13.9% clay). This textural characteristic is consistent with the higher resistivity observed for C4, although the relatively weak class effect on particle-size distribution indicates that texture alone cannot explain the observed resistivity contrasts. In contrast, class C1 in layer D showed a relative enrichment in clay. Class C2 was characterized by a particularly low coarse element content in layer C. Class C3 stood out due to very high proportions of coarse material and fine sand in the deepest layer (D), associated with a low clay fraction. Most of the physico-chemical variability was explained by depth alone (Table 1). High concentrations of organic carbon (Corg) and total nitrogen were observed in layer A, corresponding to the added topsoil (Table 3). Organic matter content then decreased markedly with depth, with approximately tenfold lower organic carbon concentrations in layer D compared to layer A, and no significant differences among soil classes, except between C3 and C4. In contrast, the C/N ratio increased by approximately twofold from layer A to layer C, before decreasing sharply to values below 10 in layer D. A small but significant effect of soil class on organic matter quality was observed, depending on depth, with higher C/N values for classes C1 and C4 in layer C and lower values in layer D. Both pHWater and pHKCl varied only with depth, showing an increase of approximately 0.5 pH units from layer A to layer D. Calcium carbonate (CaCO₃) content remained stable along the soil profile for class C1, increased slightly in layers B and C for classes C2 and C4, and increased markedly in layer D for class C3. Cation exchange capacity (CEC), Olsen phosphorus (POlsen), and most exchangeable cations were mainly driven by depth, positively correlated with SOC but negatively correlated with pH (Supplementary Figure S8), with decreasing values from surface to deeper horizons. Soil classes showed no consistent differences, except for higher values recorded in class C3 in layer D, likely related to its higher clay and carbonate contents. When considering all physical and chemical properties together (Figure 4), significant differences among soil classes were observed at all depths except in layer C, with classes C3 and C4 being the least similar.

Table 1. Proportion (%) of the variance explained by the different classes of soils defined by ERT, the depth of investigation and the interactions between the two latter factors for each parameter (n=10, n=9, n=6 and n=5, respectively for C1, C2, C3 and C4).

Significant effects are represented as follows: “*”: P<0.001; “”: P<0.01; “*”: P<0.05 or “˙”: P<0.1.

Bold values indicate statistically significant factors.

Table 2. Texture of the four soil categories as a function of the depth.

Data are mean of replicates for each class and each depth (n=10, n=9, n=6 and n=5, respectively for C1, C2, C3 and C4).

Table 3. Chemical properties of the four soil categories as a function of the depth.

Data are mean of replicates for each class and each depth (n=10, n=9, n=6 and n=5, respectively for C1, C2, C3 and C4).

3.4 Soil organic matter composition

Most organic matter (OM) quality indicators were primarily driven by depth, with the exception of the AROCH/C=C and SAT/POLY ratios (Table 4). Only a limited fraction of the variability was attributable to soil class, notably for indices related to hydrophobicity (SAT/SOC), microbial transformation (SAT/POLY), and condensation (AROCH/C=C). Multivariate analyses based on the full mid-infrared spectroscopy (MIRS) spectra (400–4000 cm^-¹) revealed significant differences among soil classes at all depths except in layer C, with class C4 being the least similar to the others (Figure 5). Overall, soil class C4 was characterized by higher relative contributions of unsaturated hydrocarbons and polysaccharides, whereas class C1 exhibited higher proportions of saturated hydrocarbons. Hydrophobicity (SAT/SOC), oxidation (C=O/SOC), and aromaticity (C=C/SOC) indices, as well as polysaccharide-related bands (POLY/SOC), increased markedly in the deepest layer (D). In contrast, the maturation index (C=C/UNSAT) showed a decreasing trend with depth. Within layer D, class C1 displayed significantly higher hydrophobicity (SAT/SOC) values than the other soil classes, while class C3 exhibited higher maturation (C=C/UNSAT) and microbial transformation (SAT/POLY) indices. Altogether, these results indicate that the main differences in OM quality among soil classes occur in the deeper horizons, with more limited differentiation in the surface layer. Across all samples, hydrophobicity (SAT/SOC), oxidation (C=O/SOC), aromaticity (C=C/SOC), and polysaccharide content (POLY/SOC) were negatively correlated with soil organic carbon (SOC) concentration (Supplementary Figure S8), whereas the maturation index (C=C/UNSAT) was positively correlated.

3.5 Major and trace elements

The elemental composition determined by ICP–MS showed a strong and significant effect of depth for most elements, with the exception of Cr (Tables 5, 6). Total concentrations of Al and K increased progressively with depth, whereas P concentrations decreased markedly in the deepest layer (D). Trends for the other major elements were less consistent. Iron (Fe), potassium (K), and magnesium (Mg) concentrations were strongly correlated with one another (Supplementary Figure S8). Phosphorus (P) was highly and positively correlated with soil organic carbon (SOC), while K showed a negative correlation with SOC. With the exception of Cr, trace element concentrations remained relatively stable throughout the soil profiles and decreased only in layer D. Although less pronounced than depth-related effects, significant differences among soil classes were also observed for several major elements, including Al, Fe, Mg, and P, as well as for trace elements such as Cr and Zn, and marginally for Ni. The greatest differentiation among soil classes occurred in the deepest layer (D), as indicated by the multivariate analyses of elemental composition (Figure 6). At this depth, soil class C3 was clearly enriched in major elements (Fe, Al, and Mg) and in the trace element Ni compared to classes C2 and C4. Across all samples, trace element concentrations were strongly intercorrelated (Supplementary Figure S8). In addition, all trace elements, with the exception of Cr and As, were positively related to soil organic matter content.

4 Discussion

4.1 Ability of ERT to resolve fine-scale spatial variability in urban soil

Urban soils are generally expected to exhibit pronounced spatial heterogeneity owing to successive phases of excavation, filling, levelling and landscaping that disrupt natural pedogenesis and generate complex assemblages of natural and anthropogenic materials (2, 7). Based on the documented history of the study site, we therefore hypothesized that Electrical Resistivity Tomography (ERT) would reveal marked subsurface variability associated with excavation during the construction of adjacent buildings, backfilling with excavated materials, and the addition of a natural topsoil layer during landscaping in the late 1950s. Contrary to this expectation, however, both the geophysical survey and the pedological observations revealed a relatively homogeneous soil profile, suggesting that these historical earthworks were either implemented relatively uniformly across the site or that subsequent pedogenic processes have progressively attenuated the contrasts initially generated by anthropogenic disturbances. This finding highlights an important aspect of urban soil evolution: although urban soils are often considered intrinsically heterogeneous, the degree of anthropization and resulting spatial variability can differ substantially among sites depending on their management history and post-disturbance evolution (2, 7).

Despite the relatively limited variability observed at the site scale, the high-resolution ERT survey successfully resolved subtle lateral and vertical variations in electrical resistivity down to approximately 2 m depth. These variations closely matched the stratigraphic organization observed in the soil pits, where most horizons developed from the same marly gypsum parent material. The correspondence between geophysical and pedological observations demonstrates that ERT remains sufficiently sensitive to detect relatively small changes in subsurface organization even when lithological contrasts are weak. Rather than identifying discrete anthropogenic deposits, the resistivity images primarily reflected variations in the physical organization of the soil profile, illustrating the ability of ERT to characterize continuous rather than merely abrupt transitions.

This observation is consistent with the conceptual basis of electrical resistivity measurements. Bulk soil resistivity is not controlled by a single soil property but represents the integrated response of several interacting factors, including particle-size distribution, clay mineralogy, soil moisture, porosity, bulk density, pore connectivity and pore-water electrical conductivity (14, 20, 31). Consequently, similar resistivity values may arise from different combinations of soil characteristics, while relatively small variations in several properties may together generate detectable geophysical contrasts. ERT should therefore not be interpreted as a direct proxy for any individual soil attribute but rather as an integrative descriptor of soil architecture and functioning. This integrative nature is particularly advantageous in urban environments, where soil variability often results from multiple superimposed processes that cannot easily be disentangled using conventional point measurements alone.

This non-uniqueness implies that interpreting a resistivity anomaly on the basis of a single soil property may lead to ambiguous or erroneous conclusions. For example, a decrease in resistivity may result from increased water content, higher pore-water conductivity, greater pore connectivity, or an increase in fine particles, while simultaneous changes in several of these factors may reinforce or compensate for each other. Conversely, the absence of a resistivity contrast does not necessarily imply the absence of spatial variability in an individual soil property. Resistivity should therefore be interpreted as an integrated response rather than as a one-to-one indicator of a specific soil attribute. In the present study, this limitation was addressed by combining complementary electrode configurations and by classifying complete vertical resistivity profiles using multivariate statistical analyses. This approach reduces the dependence of the spatial interpretation on any single resistivity value or individual soil property.

4.2 Using ERT to design a balanced stratified soil sampling scheme

One of the most significant outcomes of this study is that the ERT survey provided an objective framework for defining four geophysically coherent soil classes, which subsequently guided the sampling strategy. Instead of relying on systematic grids or random sampling, the delineation of geophysical units enabled sampling to be targeted toward representative zones characterized by distinct subsurface conditions. Such an approach increases the likelihood of capturing the major sources of spatial variability while reducing redundant sampling within relatively homogeneous areas. Similar strategies are increasingly advocated in precision agriculture and digital soil mapping, where proximal geophysical measurements are considered exhaustive environmental covariates that improve the delineation of soil mapping units and optimize sampling effort (16, 20, 32, 33). Although relatively few studies have applied this concept to urban soils, the present work demonstrates that ERT can provide valuable support for designing sampling strategies even when the overall degree of anthropogenic heterogeneity is lower than expected.

Importantly, the usefulness of ERT should not be viewed solely in terms of its ability to identify highly disturbed sites. Electrical resistivity imaging has been widely used to characterize former industrial sites, mining areas, landfills and brownfields, where strong resistivity contrasts often reveal buried foundations, heterogeneous backfill materials, contaminant plumes or technogenic deposits (34, 35). Our results highlight a complementary application that is less frequently emphasized: ERT can also demonstrate the relative continuity of soil horizons and confirm the absence of major subsurface discontinuities. In this sense, geophysical surveys reduce uncertainty irrespective of whether they reveal strong heterogeneity or relative homogeneity. Such information is equally valuable for environmental assessments because it provides an objective basis for selecting representative sampling locations and evaluating the spatial representativeness of soil observations.

Beyond its conventional use for subsurface characterization, ERT can also serve as a practical decision-support tool for objectively delineating homogeneous soil units and optimizing balanced stratified sampling in highly heterogeneous environments. To our knowledge, relatively few studies have combined ERT-derived resistivity profiles with multivariate statistical analyses for this specific operational purpose. Such an approach has broad potential applications, including the optimization of soil quality and contamination monitoring programs, as well as the design of field experiments by accounting for pre-existing soil heterogeneity and thereby reducing experimental bias (36). Looking forward, recent developments in artificial intelligence, fuzzy decision-making, and machine learning offer promising opportunities to further enhance the integration and interpretation of geophysical and environmental datasets (37, 38). Although these approaches were beyond the scope of the present study because they generally require extensive training datasets, they represent valuable directions for future research on ERT-based soil characterization.

Although the present study does not constitute a predictive cross-validation of individual soil properties, the sequential design provides an independent reality check of the ERT-derived spatial stratification. The geoelectrical classes were established before soil sampling, and the subsequent physicochemical measurements were used to assess whether these classes corresponded to systematic differences in soil characteristics. The significant differences observed when the physicochemical variables were considered jointly support the relevance of the spatial classification, while the absence of a single property explaining all resistivity patterns confirms the multivariate nature of the observed relationships.

4.3 Influence of soil physical properties on ERT

The effectiveness of ERT for soil characterization relies on the fundamental relationships between electrical resistivity and soil physical properties. In most soils, electrical current is transmitted predominantly through the pore water rather than the solid mineral phase. Consequently, bulk electrical resistivity is controlled by the amount and composition of the pore water, pore connectivity, and the geometry of the pore network, all of which are strongly influenced by soil texture, structure, coarse fragment content, compaction, and horizon development (39, 40). Because resistivity integrates the combined effects of these properties, it generally spans several orders of magnitude, providing a much greater dynamic range than individual physical parameters such as water content or porosity alone. This high sensitivity makes ERT particularly effective for identifying vertical and lateral changes in soil horizons and anthropogenic materials, even before direct soil observations or sampling are undertaken. In the present study, ERT was therefore used not as a direct indicator of soil quality, but as a rapid, non-destructive tool to objectively characterize site heterogeneity and delineate homogeneous soil units suitable for balanced stratified sampling.

The differences observed among the ERT-derived soil classes provide further insight into the factors controlling electrical resistivity in urban soils. Although significant differences were detected for numerous soil properties, not all variables contributed equally to the geophysical classification, illustrating the integrative nature of the resistivity signal. Rather than reflecting individual soil characteristics, electrical resistivity results from the combined influence of multiple interacting physical and chemical properties that together govern the movement of electrical current through the soil matrix (14, 20). Consequently, interpreting ERT images requires considering the relative contribution of these different factors rather than seeking one-to-one relationships with individual soil parameters.

Particle-size distribution is generally regarded as one of the principal factors controlling soil electrical resistivity because it directly influences pore-size distribution, water retention and surface electrical conduction. Fine-textured soils, particularly those enriched in clay minerals, typically exhibit lower resistivity than coarse-textured materials due to their greater specific surface area and higher capacity to retain conductive pore water (14, 31). In the present study, however, only relatively limited textural differences were observed among the ERT-derived soil classes. Significant contrasts were mainly associated with localized enrichment in coarse sand within the upper horizons and increased clay contents in the deeper horizons of Class 3. These observations suggest that particle-size distribution alone cannot explain the spatial organization of resistivity across the site. Instead, the measured resistivity patterns most likely reflect the combined effects of texture, bulk density, pore geometry and water distribution, all of which influence the continuity of conductive pathways within the soil (17, 20). Because soil water content was not monitored during the survey, the respective contributions of intrinsic soil properties and transient hydrological conditions cannot be fully separated. Repeated measurements under contrasting moisture conditions would therefore help distinguish structural heterogeneity from temporal variations in soil water distribution, an approach that has proven particularly valuable in hydrogeophysical studies (20). The particularly high resistivity values observed for Class C4 are therefore consistent with its relatively coarse texture, especially its high sand and low clay contents in the upper layer. However, this correspondence should not be interpreted as a direct causal relationship between sand content and resistivity, because the statistical analysis showed that depth accounted for substantially more variance in particle-size distribution than soil class. The higher resistivity of C4 is thus more appropriately interpreted as the combined response of its textural characteristics and associated pore structure and water distribution.

Electrical resistivity is inherently influenced by temporal variations in soil water content, and consequently the delineation of geoelectrical units may vary slightly with changing hydrological conditions. Time-lapse ERT studies have demonstrated the sensitivity of resistivity to infiltration processes and seasonal soil moisture dynamics (41–43). Nevertheless, the principal geological and pedological structures controlling resistivity, such as soil horizons, anthropogenic materials, and soil compaction, remain unchanged over time. In the present study, measurements were acquired under favorable winter moisture conditions to ensure good electrode contact and a high signal-to-noise ratio. Furthermore, soil samples were collected in the central part of each geoelectrical unit rather than near class boundaries, thereby minimizing the potential impact of slight seasonal shifts in boundary location. Repeated ERT surveys under contrasting hydrological conditions would nevertheless be valuable for assessing the temporal stability of the proposed geoelectrical classification. Accordingly, the present study addresses spatial variability under a defined set of field conditions and does not assess the temporal stability of the proposed geoelectrical classification. Repeated ERT surveys under contrasting hydrological conditions would be required to quantify the temporal component of the observed resistivity variability. However, hydraulic properties such as saturated hydraulic conductivity (Ksat), water retention characteristics, field capacity and permanent wilting point were not measured in the present study. Consequently, the present dataset does not allow us to determine whether the ERT-derived classes differ in hydraulic functioning or to establish direct relationships between resistivity and hydraulic parameters. Such measurements would provide an important complementary assessment in future studies by helping to distinguish the effects of soil structure, water retention and transient soil moisture on the observed resistivity patterns. However, hydraulic properties such as saturated hydraulic conductivity (Ksat), water retention characteristics, field capacity and permanent wilting point were not measured in the present study. Consequently, the present dataset does not allow us to determine whether the ERT-derived classes differ in hydraulic functioning or to establish direct relationships between resistivity and hydraulic parameters. Such measurements would provide an important complementary assessment in future studies by helping to distinguish the effects of soil structure, water retention and transient soil moisture on the observed resistivity patterns.

4.4 Relationship between soil organic matter and ERT

The relationship between electrical resistivity and soil organic matter appeared even more complex. Total organic carbon contents did not differ significantly among the ERT-derived soil classes, indicating that bulk carbon stocks were not a major driver of the observed resistivity contrasts. In contrast, Fourier Transform Infrared (FTIR) analyses revealed clear differences in organic matter composition, particularly for hydrophobic functional groups and polysaccharide-related compounds. These results suggest that ERT may be more sensitive to qualitative differences in organic matter than to its total abundance. Such an interpretation is physically plausible because organic matter exerts a strong indirect control on soil electrical properties through its influence on aggregate stability, pore architecture, wettability and water retention rather than through its intrinsic electrical conductivity (44–46). Hydrophobic organic compounds may reduce soil wettability and modify infiltration patterns, whereas microbial polysaccharides promote aggregate formation and enhance pore connectivity, thereby influencing the spatial distribution of soil water and, consequently, the propagation of electrical current. These mechanisms may explain why differences in organic matter quality were better associated with the ERT classification than total organic carbon concentrations.

These observations also open promising perspectives for investigating soil carbon dynamics using geophysical methods. Because electrical resistivity is highly sensitive to water distribution, and because organic matter composition strongly influences soil hydraulic behavior, repeated ERT measurements could potentially be used to monitor temporal changes associated with biological activity, seasonal wetting and drying cycles, or carbon turnover. Coupling time-lapse ERT with measurements of soil respiration, CO₂ fluxes and microbial activity may therefore provide new opportunities to investigate the links between soil structure, hydrological functioning and carbon cycling in urban ecosystems (20). Although such applications remain largely unexplored, they represent an interesting direction for future research.

Another promising application concerns the detection of organic contaminants in urban soils. Several studies have shown that petroleum hydrocarbons, diesel and other non-aqueous phase liquids modify soil electrical properties by altering pore-fluid conductivity, microbial activity and moisture distribution (47, 48). In contaminated environments, biodegradation processes frequently generate secondary geochemical gradients that are readily detectable using electrical methods. Consequently, while ERT cannot directly quantify hydrocarbon concentrations, it may provide an efficient, non-invasive tool for identifying contaminated zones and guiding targeted soil investigations in urban brownfields or former industrial sites. Integrating ERT with chemical analyses could therefore considerably improve the efficiency of contamination surveys while reducing the number of samples required for laboratory analyses.

4.5 Potential of ERT for detecting trace elements contamination

Elemental composition exhibited similarly complex relationships with the geophysical classification. Significant differences among ERT-derived soil classes were observed for several major elements, including Al, Fe, Mg and P, as well as for trace elements such as Cr and Zn, with weaker differences for Ni, Pb and As. However, these relationships should not be interpreted as evidence that elemental concentrations directly determine electrical resistivity. Instead, both resistivity and elemental distributions are likely controlled by common environmental factors, including parent material, mineralogical composition, weathering intensity, soil structure and hydrological conditions. For example, Fe and Al oxides influence aggregation processes and reactive surface area, while soil moisture and redox conditions govern both electrical conductivity and trace element mobility. This interpretation agrees with previous studies showing that electrical resistivity generally correlates poorly with total contaminant concentrations but effectively identifies zones where contaminant transport, accumulation or transformation are controlled by subsurface hydrological processes (34, 35).

From an environmental perspective, these findings suggest that ERT may be more useful for assessing contaminant behavior than for directly mapping contamination levels. In urban soils, trace element mobility and bioavailability are governed not only by total concentrations but also by soil moisture, redox conditions, pore connectivity, mineral surfaces and organic matter composition. By integrating several of these controlling factors, ERT could contribute to identifying functional zones where contaminants are more likely to accumulate, migrate or become bioavailable. Future research should therefore investigate relationships between geophysical signatures, trace element speciation, pore-water chemistry and plant uptake in order to better assess environmental risks associated with urban soil contamination. Such integrated approaches would considerably strengthen the role of geophysics in urban soil assessment, moving beyond simple site characterization toward process-based evaluation of soil functioning.

5 Conclusions

This study demonstrates the value of Electrical Resistivity Tomography (ERT) as a non-invasive tool for characterizing the spatial organization of urban soils and supporting the design of representative soil sampling strategies. Although the study site was expected to exhibit pronounced heterogeneity owing to its history of excavation, backfilling and landscaping, the geophysical survey revealed a relatively continuous soil profile with only moderate spatial variability. This finding highlights that urban soils are not systematically heterogeneous and emphasizes the importance of objectively assessing subsurface variability rather than inferring it solely from historical land use. The high-resolution ERT survey delineated four geophysically coherent soil classes that were subsequently validated through pedological observations and physicochemical analyses. While no single soil property fully explained the observed resistivity patterns, the ERT classification captured meaningful differences in soil texture, major and trace element distributions, and, notably, in the molecular composition of soil organic matter. These results confirm that electrical resistivity should be regarded as an integrative indicator reflecting the combined influence of soil architecture, moisture distribution, mineralogical composition and pore connectivity, rather than as a direct proxy for individual soil properties. The main methodological contribution of this study is the integration of high-resolution ERT imaging with multivariate classification to objectively delineate geoelectrically coherent soil units and translate these units into a balanced stratified soil sampling design. The approach therefore extends the use of ERT beyond conventional subsurface characterization by using continuous geophysical information as an operational decision-support tool for subsequent soil sampling. The subsequent independent physicochemical characterization provided an external assessment of the relevance of the ERT-derived units: although no single soil property explained the resistivity distribution, several physical, chemical, elemental and organic-matter characteristics differed among the geoelectrical classes. Importantly, ERT proved valuable not only for identifying spatial discontinuities but also for confirming the relative homogeneity of the site, thereby reducing uncertainty in soil characterization regardless of the degree of anthropogenic disturbance. The complementary use of Dipole–Dipole and Wenner–Schlumberger arrays further highlighted the importance of adapting acquisition strategies to survey objectives. Future studies should investigate the potential of time-lapse ERT, combined with complementary techniques such as electromagnetic induction, ground-penetrating radar and digital soil mapping, to better characterize soil functioning, contaminant mobility and temporal changes in urban environments. Overall, this study shows that ERT is more than a tool for mapping subsurface heterogeneity. By objectively delineating geophysically coherent soil domains prior to field investigations, it bridges non-invasive geophysical imaging and pedological sampling, providing an efficient framework for optimizing soil characterization and advancing process-based investigations in complex urban environments.

Table 4. Infrared based indices of SOM composition of the four soil categories as a function of the depth.

Data are mean of replicates for each class and each depth (n=10, n=9, n=6 and n=5, respectively for C1, C2, C3 and C4).

Table 5. Total concentration of major elements of the four soil categories as a function of the depth.

Data are mean of replicates for each class and each depth (n=10, n=9, n=6 and n=5, respectively for C1, C2, C3 and C4).

Table 6. Total concentration of trace elements of the four soil categories as a function of the depth.

Data are mean of replicates for each class and each depth (n=10, n=9, n=6 and n=5, respectively for C1, C2, C3 and C4).

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fsoil.2026.1924576/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: José A. González-Pérez, Spanish National Research Council (CSIC), Spain

Reviewed by: Shahrokh Rahbari, Tarbiat Modares University, Iran

Khalid Essa, Cairo University, Egypt