Investigation of Topsoil Salinity and Soil Texture Using the EM38-MK2 and the WET-2 Sensors in Greece

Abstract

The electromagnetic induction (EMI) and frequency domain reflectometry (FDR) sensors, which measure the soil apparent electrical conductivity (ECa) in situ, have emerged as efficient and rapid tools for the indirect assessment of soil salinity, conventionally determined by the electrical conductivity of the saturated soil paste extract (ECe). However, the limitations of applying a single soil sensor and the ECa dependence on multiple soil properties, such as soil moisture and texture, can hinder the interpretation of ECe, whereas selecting the most appropriate set of sensors is challenging. To address these issues, this study explored the prediction ability of a noninvasive EM38-MK2 (EMI) and a capacitance dielectric WET-2 probe (FDR) in assessing topsoil salinity and texture within 0–30 cm depth across diverse soil and land-use conditions in Laconia, Greece. To this aim, multiple linear regression models of laboratory-estimated ECe and soil texture were constructed by the in situ measurements of EM38-MK2 and WET-2, and their performances were individually evaluated using statistical metrics. As was shown, in heterogeneous soils with sufficient wetness and high salinity levels, both sensors produced models with high adjusted coefficients of determination (adj. R² > 0.82) and low root mean square error (RMSE) and mean absolute error (MAE), indicating strong model fit and reliable estimations of topsoil salinity. For the EM38-MK2, model accuracy improved when clay was included in the regression, while for the WET-2, the soil pore water electrical conductivity (ECp) was the most accurate predictor. The drying soil surface was the greatest constraint to both sensors’ predictive performances, whereas in non-saline soils, the silt and sand were moderately assessed by the EM38-MK2 readings (0.49 < adj. R² < 0.51). The results revealed that a complementary use of the contemporary EM38-MK2 and the low-cost WET-2 could provide an enhanced interpretation of the soil properties in the topsoil without the need for additional data acquisition, although more dense soil measurements are recommended.

1. Introduction

Soil salinization has become a major environmental concern in the arid and semi-arid regions of the world, including the Mediterranean coastlands, compromising their agricultural productivity and the sustainability of natural resources. As a matter of fact, in Greece, over 4 million hectares of cropland topsoil are estimated to have been affected by soil salinity and sodicity, of which 3.98 million ha are saline soils and 44,000 ha are sodic soils [1,2]. In addition to high evapotranspiration rates in the region, the more frequent high-temperature extremes and the reduction in precipitation caused by climate change are expected to exacerbate the extent and severity of the salt-affected areas [3]. Considering the imperative need for effective mitigation and management practices, robust tools and methods for high-precision soil salinity monitoring are crucial.

Conventional soil salinity assessment relies on dense soil sampling and extensive laboratory analysis for the soil-saturated paste extract’s electrical conductivity (ECe) determination [4], consisting of a laborious and time-consuming procedure. Nowadays, technological advancements have introduced proximal soil sensors (PSS) [5], along with remote sensors (RS), as rapid and efficient alternatives for capturing the soil salinity variations [6,7,8]. Among the wide variety of commercial PSS [9,10], electromagnetic induction (EMI) sensors are the most commonly applied, especially at field scale [11]. The on-the-go EMI sensors offer, in real time, a large amount of noninvasive, georeferenced measurements of depth-weighted soil apparent electrical conductivity (ECa) that can be related to ECe [12,13] at various soil depths, enabling the quantification and mapping of the soluble salt’s spatial patterns [14,15]. Most recently, affordable capacitance sensors (CS) that provide measurements of ECa by using frequency domain reflectometry (FDR) are being increasingly adopted in the precision irrigation systems for soil moisture monitoring and field soil salinity prediction with satisfactory results [16,17,18]. Through the detection of apparent soil dielectric permittivity (εb), the low-frequency capacitance devices have the advantage of measuring ECa and soil temperature (T), while they simultaneously estimate volumetric soil water content (θ), all within the same soil volume [19]. Additionally, some capacitance probes incorporate within their datalogger or meter empirical models for estimating the soil pore water electrical conductivity (ECp) by the ECa and θ [20,21].

ECa measured in the field by EMI or FDR is a complex quantity, affected by transient and static soil properties, such as soil moisture, salinity, texture, temperature, bulk density, and ECp, as well as by their interrelations [16,22,23], hindering the attribution of ECa values to the examined property. As a consequence of the spatial heterogeneity of soil, the ECa-ECe calibration equations are considered site-specific [24,25,26]. Particularly in soils with low salt concentrations, factors beyond soil salinity have a confounding effect on ECa [27] and therefore need to be determined. In these cases, ECa is mostly used as a surrogate for inferring the more stable components of the soil.

Furthermore, due to the different fundamental principles of the PSS techniques, the accuracy of the soil salinity models produced using a single soil sensor depends on its distinct capabilities and intrinsic characteristics. Overall, these attributes are associated with the depth sensitivity, the soil volume of investigation, the operational frequency, and notably the specific soil properties to which the employed sensor is most responsive [27,28]. In this regard, the combination of multiple PSS (fusion) in situ has been demonstrated to provide complementary information and overcome the limitations that stem from the individual sensor’s outputs, increasing the prediction accuracy of the targeted soil properties [15,29,30,31]. For enhancing the soil salinity interpretation by the EMI readings, several researchers implemented calibration and inversion approaches that integrate data from various EMI devices or other technologies like time domain reflectometry (TDR) and electrical resistivity tomography (ERT), which are used as ground-truth estimates [32,33,34,35,36,37,38,39]. Likewise, a few case studies explored the utilization of FDR and EMI sensors [40,41,42]. Although the existing literature has yielded significant findings concerning the application of multi-source data in the soil salinity appraisal, selecting the optimal set of PSS is, of yet, inconclusive, as this is dependent on a number of factors, including the terrain features, the accessibility and cost of the sensors, and the expertise [43]. For instance, many calibration approaches involve taking EMI measurements at multiple heights above the soil surface for a sensitivity analysis at shallower depths [40] or require more sophisticated data processing (e.g., data filtering [33]) and software. Hence, they might become physically and technically demanding, as well as costly, mainly in large-scale surveys. Additionally, in areas of sloping terrain with ridge planting and expanded tree canopy, which are mostly encountered in Mediterranean agricultural systems [44], the instrument’s elevation may not always be feasible, while the effects of the elevation can produce biased readings, if not properly corrected [45,46].

In an attempt to address these challenges and support the appropriate selection of PSS, this study investigates the applicability of an electromagnetic (EMI) EM38-MK2 (Geonics Ltd., Mississauga, ON, Canada) sensor and a WET-2 (Delta-T Devices Ltd., Cambridge, UK) capacitance probe (CS) for the topsoil (0–30 cm) salinity appraisal at fields in the southeastern region of Greece, Laconia. The contemporary EM38-MK2 sensor, when placed on the ground, provides average ECa measurements from both the deeper (0–1.5 m) and shallower (0–0.37 m) layers of the soil profile, representing a suitable candidate and an attractive option for topsoil salinity monitoring. On the other hand, the low-cost WET-2 probe, which gives instantaneous measurements of ECa along with supplementary data of soil moisture, temperature, and ECp, can facilitate a more detailed interpretation of the soil properties’ fluctuations at the upper soil layer and their contribution to the quantity of ECa and, consequently, to soil salinity.

In this respect, the main objectives of the research were (i) to evaluate the performances of the WET-2 and the EM38-MK2 in assessing soil salinity at 0–30 cm depth, under varying soil moisture, texture conditions, and salinity levels, without prior inversion of the EMI data, and (ii) to examine whether a complementary use of the sensors could potentially enhance the comprehensive understanding of the soil salinity and texture spatial patterns. The present study aims to provide new insights into the investigation of soil properties in arid and semi-arid regions by introducing a practical and field-applicable approach that lies in the combined operation of the EM38-MK2 and WET-2 sensors directly at the soil surface, without requiring data inversion or instrument elevation.

2. Materials and Methods

2.1. Study Area and Survey Design

Three campaigns of soil sampling and field point measurements were executed using the EM38-MK2 and the WET-2 probe at the ground in four spatially distributed areas with varying land-use and soil conditions in the northern and southern regions of Laconia, Greece. The climate of the studied area is characterized as arid and semi-arid with warm and dry summers and mild winters. The mean annual precipitation is around 500 mm, with approximately 70–80% of it falling during the period from October to March.

The first session was conducted in the early stage of the dry season, in April of 2022, at two non-irrigated field plots: an olive orchard (37°00′28″ N, 22°29′54″ E) and a post-harvest potato crop (37°01′03″ N, 22°29′11″ E). The second one was carried out towards the end of the rainy season in March of 2023 and included measurements from different sites of bare land and non-irrigated farmlands, while the last survey took place in September 2024, at a drip-irrigated orange grove planted on the top of ridges (36°49′28″ N, 22°42′36″ E) (Figure 1).

In the 2022 and 2024 field surveys, the point measurements were recorded in approximate grids of 10 m × 10 m and 20 m × 20 m, respectively. In 2023, a random sampling design was applied to capture the soil salinity variations of the study area, and for the 2024 campaign, measurements were taken along and next to the driplines, near the trees.

All soils of the study are classified as calcaric Fluvisols in terms of the World Reference Base for Soil Resources (WRB) [47,48]. These soils are primarily formed from alluvial deposits of the Evrotas River and present a layering of the sediments rather than pedogenic horizons. They contain low organic matter (~1.8%) in the surface and subsurface layers, relatively high calcium carbonate in the whole soil profile, and exhibit slightly alkaline soil reaction (pH) [48]. The presence of distinct horizons in the soil profile has been reported to contribute substantially to the vertical variations in electrical resistance and conductivity within the topsoil layer (0–30 cm) [49,50].

Different survey areas and dates were pursued to represent the soil properties’ variability occurring under the diverse land-use patterns that extend across the region of Laconia. After thorough research, the selected locations allowed the study to evaluate the sensors’ performances at a wide range of soil types, under dry and wet moisture conditions, and at varying salinity levels and land types. Over the past decade, the combined effect of intensive local irrigation practices, groundwater extraction, and the hydrological environment in the coastal zones of Laconia has led to seawater intrusion and the accumulation of salts in the surface layer, an issue that is likely to worsen in the coming years. In the examined coastal areas of the 2023 and 2024 surveys, soil salinization is mainly associated with sodium chloride, whereas due to the area’s calcium-rich rock formations, the soils of the 2022 survey are mostly affected by irrigation water with high concentrations of calcium salts [2].

2.2. EM38-MK2 Sensor and WET-2 Probe

For the field data collection by EMI and FDR, an electromagnetic EM38-MK2 sensor (Geonics Ltd., Mississauga, ON, Canada) and a WET-2 (Delta-T Devices Ltd., Cambridge, UK) capacitive dielectric probe (CS) were deployed at the examined fields (Figure 2).

The portable, non-destructive EM38-MK2 comprises one transmitter that induces an electromagnetic field in the soil at a 14.5 kHz frequency, and double receiver coils, located at 1 m and 0.5 m distances from the transmitter. Based on McNeill’s approximation [51] for homogeneous material, the dual-fixed spacing between the coils allows the EM38-MK2 to obtain simultaneous measurements of ECa (mSm⁻¹) at 4 average depth ranges, using its dipole mode. More specifically, the sensor’s effective depths of measurement at the 0.5 m and 1 m coil separations are approximately 0.37 m (H0.5) and 0.75 m (H1) when the sensor is positioned horizontally on the soil surface and 0.75 m (V0.5) and 1.5 m (V1), respectively, in the vertical orientation. Contrary to the vertical mode, in which the sensor indicates a greater relative sensitivity with increasing depth, the majority of the signal in the horizontal mode originates from beneath or near the soil surface and decreases with depth [45,52].

Moreover, the device is equipped with a built-in GPS for the ECa data traction and an enhanced temperature compensation circuitry that automatically eliminates the temperature-related drifts during surveying [45].

The capacitive WET-2 is a lightweight, three-rod probe that detects the dielectric properties of the soil and measures soil water content, electrical conductivity, and temperature. The metal rods are 6.8 cm long with a 3 mm diameter and spaced 1.5 cm apart. When inserting into the soil, the probe applies a 20 MHz signal to the central rod and an electromagnetic field is generated. Then, through an HH2 moisture meter, which is connected to the sensor, the capacitance (C) and conductance (G) of the soil are measured, and the apparent dielectric permittivity εb and ECa (mSm⁻¹) are determined by embedded calibration files. Finally, volumetric soil water content (θ) (%) and soil pore water electrical conductivity (ECp) (mSm⁻¹) are computed, and alongside temperature (T) (°C), all are displayed on the screen of HH2 [53].

θ is calculated by the dielectric probe using a simple calibration equation that relates the measured εb to θ as follows [54]:

θ = (εb^0.5 − b₀)/b₁

(1)

where b₀ and b₁ are empirical coefficients that depend mainly on the type of the porous medium, and they can be easily determined for enhanced precision. The linear θ − εb^0.5 relationship has been reported to be valid for a range of mineral soils [55], for which a set of coefficient values are incorporated in the manual. In the present study, the manufacturer’s proposed values for inorganic soils (b₀ = 1.8 and b₁ = 10.1) [53] are utilized.

ECp has been documented to reflect a more precise indicator of soil salinity than ECa since it is related to the actual salinity stress experienced by the plants [56,57]. The values of ECp provided by the WET-2 are estimated by the linear model of Hilhorst (2000) [58], described as follows:

ECp = (εp ECa)/εb − 4.1

(2)

where εp is the dielectric constant of soil solution, and its value is commonly equal to that of water (= 80.4), while 4.1 is the value of εb when ECa = 0. The model (Equation (2)) is applicable for cases where θ > 0.10 m³ m⁻³. Hereafter, the bulk ECa measured by the WET-2 is noted as WET-ECa.

2.3. Field Measurements Using the EM38-MK2 and the WET-2 Sensors

At each site, the handheld EM38-MK2, set at the manual recording mode, was placed on the ground initially in horizontal and then in vertical orientation, and the ECa readings of H0.5 (ECa in horizontal mode at 0.5 m coil spacing), H1 (ECa in horizontal mode at 1 m coil spacing), V0.5 (ECa in vertical mode at 0.5 m coil spacing), and V1 (ECa in vertical mode at 1 m coil spacing) were acquired. The H0.5 mode of EM38-MK2 with a sensitivity depth of about 0.37 m aligns most closely with the targeted 0–30 cm topsoil, while the H1, V0.5, and V1 modes with effective depths reaching to approximately 0.75 and 1.5 m are additionally employed to assess the electrical conductivity distribution within the soil profile and capture the subsoil salinity variations that may influence the topsoil.

Before each survey, all the metal objects were removed, and the instrument was nulled and calibrated according to the manufacturer’s instructions. Additionally, to minimize potential instability from the sensor’s drifts in ECa, the calibration was repeated during the field surveys [45].

Following the EMI operation, the dielectric WET-2 probe was used. After inserting the probe into the soil surface, at the same measurement points and below the center of EM38-MK2, the data on WET-ECa, εb, T, θ, and ECp were obtained for a depth up to approximately 10 cm. For both sensors, two measurements were taken at each point, and the mean value was recorded.

Since EM38-MK2 compensates for the effects of temperature, and the soil temperature difference in all surveys did not exceed 10 °C as recorded by the WET-2, the raw ECa measurements were used for the data analysis. Furthermore, for each survey, the measurements and soil sampling were completed within one day to limit any temperature fluctuations.

From all study areas and periods, a total of 81 point-measurements were performed using the WET-2 and the EM38-MK2.

2.4. Soil Sampling and Laboratory Analysis

After completing the EM38-MK2 and WET-2 measurements for every session, disturbed single soil samples from 0 to 30 cm depth (topsoil) were collected from each measurement plot using a hand auger for the determination of soil properties. Each sample was taken and sealed in plastic bags immediately after each measurement to minimize temporal variability and to ensure consistency between field data and soil sample conditions. The total of 81 soil samples, cleared from surface litter and crop residues, were transferred to the laboratory, air dried, sieved at 2 mm, and then analyzed for soil texture, pH, organic matter (OM) content, ECe, and CaCO₃. The particle size fractions of clay (%), silt (%), and sand (%) were estimated using the Bouyoucos hydrometer method [59] and soil texture classes were delineated according to the USDA classification system. For the CaCO₃ equivalent percentage, the Bernard calcimeter method was employed by measuring the eluted CO₂ after the addition of HCl. Soil pH was measured using standard glass/calomel electrodes in 1:1 w/v soil–water suspension and the soil organic matter (OM) (%) was determined by K₂Cr₂O₇ oxidation method [60]. Finally, ECe (dSm⁻¹) was measured using a conductivity meter (WTW, Cond 315i) on solutions extracts that were acquired from soil-saturated pastes after applying vacuum, in accordance with the standard USDA protocols [4]. Based on the soil salinity classification criteria of USDA [4], the soils are defined as non-saline (ECe < 2 dSm⁻¹), slightly saline (ECe = 2–4 dSm⁻¹), moderately saline (ECe = 4–8 dSm⁻¹), strongly saline (ECe = 8–16 dSm⁻¹), and extremely saline (ECe ≥ 16 dSm⁻¹).

2.5. Statistical Analysis and Predictive Modeling

All EM38-MK2 and WET-2 measurements were initially quality examined, and the negative values, as indicative of interference, were removed from the datasets. The field data corresponding to volumetric soil water content (θ) values lower than 0.10 m³ m⁻³ were also excluded from the analysis, due to the fact that below this threshold, ECa measurements become unreliable for soil salinity estimation [57]. Data distribution of the studied variables was explored with the Shapiro–Wilk test and graphical methods (Q–Q plots) to assess normality.

Thereafter, to identify the strength of the relationship among the estimated ECe, the soil properties of clay, silt, sand, CaCO₃, and the field measurements obtained by the ΕΜ38-ΜΚ2 and WET-2 field measurements, the Spearman’s rank coefficient correlation (r_s) was used. The Spearman’s coefficients are nonparametric measures ranging from −1 to +1 and are computed from the ranks of the data values rather than directly from the values, like Pearson’s coefficients. Therefore, they are less sensitive to outliers, and more appropriate for small and not normally distributed datasets. A principal component analysis (PCA) was further applied to pooled data (all parameters and samples) in order to reduce the dimensionality of the data, and facilitate the investigation of the relationships between the variables.

Subsequently, multiple linear regression (MLR) was used to construct predictive models between ECe and the EM38-MK2 and WET-2 measurements for all study sites. MLR models were developed utilizing the ECe values as dependent variables, and the readings of H0.5, V0.5, H1, V1, and WET-ECa, as explanatory variables for EM38-MK2 and WET-2, respectively. For the capacitance probe, instead of WET-ECa, the sensor-calculated ECp was incorporated in the regression as an independent variable and examined for the ECe prediction. Furthermore, to account for the influence of soil properties on the sensors’ performances in assessing ECe, the determined textural fractions of clay, sand, and silt, and CaCO₃ were introduced as covariates in the models. Based on the strongly linear relationship between θ and εb^0.5 for a variety of soils [19,55], the εb^0.5 was used as a surrogate of soil moisture, θ, and was added in the MLR as a covariate for the case of the WET-2 probe. In addition, under the low moisture and non-saline soil conditions, the soil texture (clay, silt, and sand) prediction by the EM38-MK2 was also explored.

The evaluation of the developed models for each survey and sensor was made with selected accuracy metrics, including adjusted R² (adj.R²), root mean square error (RMSE), and mean absolute error (MAE), as expressed in the following equations:

$$\mathrm{a}\mathrm{d}\mathrm{j}.{\mathrm{R}}^2 = 1 - \left({\displaystyle \frac{{\sum }_{i=1}^n{\left( y_i-{\widehat{y}}_i\right)}^2∕\left(n-P-1\right)}{{\Sigma }_{i = 1}^n{ \left({ y}_i- {\overline{y}}_i\right)}^2∕\left(n-1\right)}}\right)$$

(3)

$$\mathrm{R}\mathrm{M}\mathrm{S}\mathrm{E}=\sqrt{{\displaystyle \frac{1}{n}}{\Sigma }_{\mathrm{i}=1}^n{\left({\mathrm{y}}_{\mathrm{i}}-{\widehat{y}}_{\mathrm{i}}\right)}^2}$$

(4)

$$\mathrm{M}\mathrm{A}\mathrm{E}={\displaystyle \frac{1}{n}}\sum _{\mathrm{i}=1}^n\left|y_{\mathrm{i}}-{\widehat{y}}_{\mathrm{i}}\right|$$

(5)

where n is the sample size, $y_{\mathrm{i}}$ the measured value, ${\widehat{y}}_{\mathrm{i}}$ the predicted value, and ${\overline{y}}_i$ is the average value. Adj. R² values closer to 1 indicate higher model-fitting accuracy, whereas for RMSE and MAE, values closer to 0 show stronger prediction ability. Moreover, the ratio of the performance to interquartile distance (RPIQ) was applied to amplify the reliability beyond the errors. RPIQ is calculated as PRIQ = (Q3 − Q1)/RMSE, where Q3 − Q1 is the difference between the 75th and 25th percentile values of the sample [61]. Finally, the soils were classified according to the salinity severity classes by FAO [62].

The statistical analysis and regression models were performed with the statistical software Statgraphics Centurion 17 (Statgraphics Technologies, Inc., Warrenton, VA, USA).

3. Results and Discussion

3.1. Descriptive Statistics of the Soil Properties Determined in the Laboratory

Table 1. Descriptive summary of the laboratory-determined soil properties of soil salinity (ECe) (dSm⁻¹), soil texture (clay, silt, and sand) (%) and CaCO₃ (%) at 0–30 cm depth for each study site.

Survey Date	Study Site	Variable	Mean	Maximum	Minimum	SD *	CV (%) *	Skewness
2022	Field 1./olive orchard N = 25	ECe (dSm−1)	1.01	1.96	0.67	0.34	33.70	1.38
		Clay (%)	33.12	38.40	26.60	2.99	9.02	0.25
		Silt (%)	39.62	45.70	32.60	3.14	7.93	0.22
		Sand (%)	27.05	31.00	19.00	3.21	11.88	−2.41
		CaCO3 (%)	22.75	38.30	16.81	5.68	24.98	2.71
	Field 2./potato crop N = 23	ECe (dSm−1)	0.92	1.28	0.64	0.24	25.67	2.06
		Clay (%)	35.59	47.00	28.40	4.70	13.22	1.53
		Silt (%)	38.19	43.40	34.00	2.43	6.38	0.80
		Sand (%)	26.22	33.00	17.00	4.41	16.82	−0.98
		CaCO3 (%)	9.14	13.94	6.56	2.08	22.74	0.90
2023	Bare land and farmlands N = 13	ECe (dSm−1)	9.69	20.40	1.58	7.69	79.42	0.57
		Clay (%)	23.99	44.00	10.70	10.57	44.08	1.07
		Silt (%)	37.01	58.00	23.30	12.43	33.58	0.89
		Sand (%)	39.00	66.00	10.60	20.61	52.84	−0.42
		CaCO3 (%)	21.21	31.30	9.63	9.59	45.22	−0.35
2024	Orange orchard N = 20	ECe (dSm−1)	0.90	1.32	0.62	0.17	19.03	1.88
		Clay (%)	16.43	20.20	12.20	2.20	13.21	−0.26
		Silt (%)	44.78	52.00	38.60	3.50	7.82	−0.04
		Sand (%)	39.37	47.20	29.20	4.50	11.67	0.27
		CaCO3 (%)	10.60	15.99	8.20	2.08	19.63	0.97

* SD: standard deviation, CV: coefficient of variation

presents the descriptive statistics of ECe (dSm⁻¹), clay, silt, and sand particles (%), and CaCO₃ (%) at the topsoil (0–30 cm) as they were determined in the laboratory for each study area. Topsoil pH values across the examined sites ranged from 7.3 to 8.3, indicating neutral to moderately alkaline conditions, while organic matter was low, ranging from 0.66% to 1.56%.

The soils of the olive orchard in 2022 (field 1) are clay loam with mean values of clay, silt, and sand of 33.12%, 39.62%, and 27.05%, respectively. The clay and silt contents reflect homogeneous textural composition (CV= 9.02% and CV = 7.93%, respectively) across the topsoil, whereas sand shows slightly higher variability (CV = 11.08%). Moreover, the increased CaCO₃ content, with a mean value of 22.75%, is consistent with the calcareous soils of the wider area, in which olive trees are commonly cultivated due to their high tolerance and adaptation. The soils of the potato crop (field 2) are also classified as clay loam, exhibiting a moderate variability in sand (CV = 16.8%) and clay (CV = 13.22%) and lower concentrations of CaCO_3. The higher CV values of the estimated clay (CV = 44.08%), silt (CV = 33.58%), sand (CV = 52.04%), and CaCO₃ (CV = 45.22%) in the district of the 2023 survey reflect the diverse land patterns and soil textures under which the experiment was conducted, ranging from finer-textured (silty clay, silty clay loam, and sandy clay loam) to coarse-textured (silty loam, loam, and sandy loam) soils. In contrast, the loam soils of the orange orchard (2024) have more stable concentrations in CaCO₃ (CV = 19.63%) and moderate variations in clay and sand, whereas silt is distributed more uniformly (CV = 7.82%).

Concerning soil salinity at the topsoil (0–30 cm), ECe shows moderate variability across both fields of 2022, with CV values of 33.7% and 25.67% for the olive orchard and potato crop, respectively. However, at the potato crop field, ECe data are positively skewed (skewness = 2.06). For the orange orchard (2024), ECe exhibits a narrower range (CV = 19.03%), indicating a relatively more uniform distribution of topsoil salinity across the field. For all the cultivated fields of the 2022 and 2024 surveys, ECe values do not exceed 2 dSm⁻¹ at 0–30 cm depth. Meanwhile, ECe mean values are 1.01 dSm⁻¹ and 0.92 dSm⁻¹ for the olive orchard and the potato crop, respectively, and 0.90 dSm⁻¹ for the orange orchard in 2024. On the contrary, in the district area of 2023, ECe has a substantial variability (CV = 79.42%), with values varying from 1.58 to 20.40 dSm⁻¹ and a mean of 9.69 dSm⁻¹.

The 2022 and 2024, soil samples are classified as non-saline, indicating the topsoil of the fields is not salt-affected, whereas for the study area of 2023, 15.38% of the samples are extremely saline, 30.77% strongly saline, 30.77% moderately saline, 15.38% slightly saline, and 7.69% non-saline.

3.2. Descriptive Analysis of the EM38-MK2 and WET-2 Measurements

As follows, the descriptive statistical characteristics of all EM38-MK2 and WET-2 measurements are shown in

Table 2. Descriptive summary of the ECa and the soil properties measured by the EM38-MK2 (H0.5, V0.5, H1, and V1) and the WET-2 probe (WET-ECa, ECp, θ, εb, and T) at the experimental sites.

Survey Date	Study Site	Variable *	Mean	Maximum	Minimum	SD **	CV (%) **	Skewness
2022	Field 1./olive orchard N = 25	H0.5 (dSm−1)	0.19	0.26	0.08	0.05	24.30	−1.23
		V0.5 (dSm−1)	0.20	0.32	0.04	0.06	31.95	−0.96
		H1 (dSm−1)	0.29	0.39	0.18	0.06	20.77	−0.92
		V1 (dSm−1)	0.34	0.51	0.13	0.10	29.84	−1.57
		WET-ECa (dSm−1)	0.11	0.20	0.07	0.04	34.88	1.43
		ECp (dSm−1)	1.51	2.00	1.14	0.17	11.26	1.96
		θ (%)	14.32	33.63	10.18	5.31	37.07	2.69
		εb	10.87	27.04	8.01	4.15	38.15	3.07
		Τ (°C)	26.98	28.00	25.50	1.23	4.39	−0.97
	Field 2./potato crop N = 23	H0.5 (dSm−1)	0.11	0.15	0.09	0.02	21.02	0.24
		V0.5 (dSm−1)	0.14	0.17	0.12	0.02	12.81	−0.04
		H1 (dSm−1)	0.20	0.24	0.17	0.02	10.75	0.14
		V1 (dSm−1)	0.25	0.32	0.18	0.04	16.92	0.39
		WET-ECa (dSm−1)	0.08	0.12	0.06	0.02	27.60	0.14
		ECp (dSm−1)	1.40	2.12	1.14	0.31	22.41	1.72
		θ (%)	11.87	15.50	10.10	1.70	14.29	0.88
		εb	9.01	11.29	8.01	1.08	12.04	1.05
		Τ (°C)	16.34	18.00	14.90	0.99	6.06	−0.07
2023	Bare land and farmlands N = 13	H0.5 (dSm−1)	1.02	2.57	0.35	0.84	82.73	1.60
		V0.5 (dSm−1)	1.24	2.81	0.36	0.76	61.39	1.42
		H1 (dSm−1)	1.22	2.88	0.02	0.88	72.03	1.00
		V1 (dSm−1)	1.38	2.33	0.17	0.72	51.73	−0.67
		WET-ECa (dSm−1)	1.94	3.78	0.42	1.23	63.78	0.03
		ECp (dSm−1)	6.88	13.99	1.69	4.77	69.31	0.26
		θ (%)	34.69	44.50	26.10	6.69	19.27	0.37
		εb	23.57	43.30	7.08	9.61	40.77	0.58
		Τ (°C)	19.73	21.60	19.30	1.15	3.81	0.85
2024	Orange orchard N = 20	H0.5 (dSm−1)	0.73	1.41	0.33	0.27	36.99	0.95
		V0.5 (dSm−1)	1.10	1.87	0.54	0.33	30.02	0.41
		H1 (dSm−1)	0.75	1.22	0.39	0.22	28.99	0.39
		V1 (dSm−1)	1.45	2.24	0.87	0.43	29.76	0.25
		WET-ECa (dSm−1)	0.34	0.49	0.22	0.06	18.32	0.52
		ECp (dSm−1)	0.93	1.14	0.83	0.09	9.58	1.93
		θ (%)	39.43	46.40	31.80	3.69	9.35	−1.36
		εb	33.18	41.99	25.20	3.31	9.97	1.33
		Τ (°C)	22.64	23.3	22.3	0.78	3.59	0.35

* H0.5 and H1 represent the EM38-MK2 ECa measurements in horizontal mode at 0.5 and 1 m coil separation, respectively. V0.5, V1: the EM38-MK2 ECa measurements in vertical mode at 0.5 and 1 m coil separation, respectively. WET-ECa: ECa measured by the WET-2, ECp: soil pore water electrical conductivity measured by the WET-2. ** SD: standard deviation, CV: coefficient of variation

.

The highest electrical conductivity values recorded by both the EM38-MK2 (H0.5, V0.5, H1, and V1) and WET-2 (WET-ECa and ECp) appear in the study site of the 2023 survey, which is anticipated given the high soil salinity (ECe) levels and the relatively increased soil water content (θ) measured by the WET-2. All readings show increased CV values, ranging from 51.73% in the V1 up to 81.12% in the H0.5, the latter of which indicates a greater variability of electrical conductivity in the upper soil layer (0–0.37 cm). The EM38-MK2 measurements ranged from 0.02 (H1) to 2.88 dSm⁻¹ (H1). The WET-2 provided consistently high ECp values throughout the survey, featuring a mean value of 6.88 dSm⁻¹. Moreover, the mean volumetric soil water content, θ (mean = 34.69%), with a coefficient of variation of 19.27%, shows the topsoil was adequately wetted during the survey. The higher mean WET-ECa value compared to those of EM38-MK2 could be explained by the smaller measurement volume of the WET-2, which reveals that variations in soil salinity occur at the localized, small-scale areas and which EM38-MK2 tends to smooth. A similar pattern has been reported between the older version of EM38-MK2, EM38, and TDR [33,63] and between EM38 and WET-2 [41], after inversion of the EMI data. It is mentioned that EM38 is equipped with one receiver coil at 0.1 m, providing measurements up to 0.75 and 1.5 m depth, while further differences and applications of these two models are discussed in the recent literature [22,52].

In the fields of the 2022 and 2024 surveys, EM38-MK2 and WET-2 recorded overall lower measurements than 2023. More specifically, in the case study of 2022, EM38-MK2 at the olive orchard has mean values varying from 0.19 dSm⁻¹ in H0.5 to 0.34 dSm⁻¹ in V1, with CVs from 24.30% to 31.95%, whereas for the potato crop, the mean and CV values are slightly lower, from 0.11 dSm⁻¹ (H0.5) to 0.25 dSm⁻¹ (V1) and CVs between 10.75% and 21%. In both fields, the horizontal and vertical readings at both coil spacing are of similar magnitude, exhibiting a relative uniformity of electrical conductivity. The ECa by the WET-2 probe is also decreased across both fields with mean values of 0.11 dSm⁻¹ and 0.08 dSm⁻¹ for the olive orchard and the potato crop, respectively, with a moderate variability (CV = 34.88% and CV = 27.60%). It is noteworthy that, despite the low WET-ECa values, ECp shows the highest mean among the measurements in both fields of 2022, reflecting the ionic concentration within the soil solution. In the estimation by the WET-2, θ varied between 10.18% and l3.63% for the olive grove and 10.10% and 15.50% for the potato crop, indicating restricted wetness conditions on the upper soil layer.

At the orange orchard in the 2024 survey, however, θ displays considerably higher levels (31.80%–46.40%), with an average value of 39.43% and narrow variability (CV = 9.35%), suggesting homogenous and sufficient soil moisture across the topsoil. The larger mean values of the vertical readings (V0.5 = 1.10 dSm⁻¹, V1 = 1.45 dSm⁻¹) from the corresponding horizontal (H0.5 = 0.73 dSm⁻¹, H1 = 0.75 dSm⁻¹) support that the electrical conductivity is more increased in the deeper soil layers. Nevertheless, the highest CV of H0.5 (CV = 36.99%) underlies stronger variability near the soil surface. The variability reduction through the subsoil layers is attributed to the fact that with increasing depth, the spatial and temporal fluctuations of soil salinity and moisture tend to dampen, leading to an averaging process [7]. The WET-2 ECa and ECp have a mean of 0.34 dSm⁻¹ and 0.93 dSm⁻¹, respectively, with ECp values being more stable (CV = 9.58%) than WET-ECa (CV = 18.32%) across the plot.

For all surveys, in the measurement by the WET-2, temperature (T) remained within a stable and acceptable range (CV < 6.06%), minimizing its impact on the reliability of soil electrical conductivity measurements.

As has already been addressed, the surface topography of many cultivated systems in the Mediterranean region can impede the surveying process when using EMI and FDR. This was the case in the orange orchard (2024), where the collection of ECa measurements on the top of the ridges and next to the driplines, as is suggested for obtaining reliable data [64,65], was rather challenging due to the extent of the crop canopy.

3.3. Correlation of ECe and Soil Properties with the Measurements of EM38-MK2 and WET-2

Spearman’s correlation coefficients (r_s) among the estimated ECe, soil properties, and the measurements by EM38-MK2 and WET-2 illustrate various significant correlations (p < 0.05, p < 0.001) for all study sites, as summarized in

Table 3. Spearman’s correlation coefficient (r_s), among the determined laboratory soil properties (ECe, clay, silt, sand, and CaCO₃) for 0–30 cm depth and the measured EM38-MK2 (H0.5, V0.5, H1, and V1) and the WET-2 (WET-ECa, εb^0.5, T, and ECp) data, for each survey.

2022
Field 1./Olive Orchard	ΕCe (dSm−1)	Clay (%)	Silt (%)	Sand (%)	CaCO3 (%)
ΕCe (dSm−1)	1.00	−0.26	−0.07	0.23	−0.47 *
H0.5 (dSm−1)	−0.12	−0.26	0.18	0.08	0.00
V0.5 (dSm−1)	0.05	−0.09	0.22	−0.01	0.02
H1 (dSm−1)	0.24	0.00	0.17	−0.12	0.06
V1 (dSm−1)	0.40 *	−0.14	0.34	−0.07	−0.01
WET-ECa (dSm−1)	0.33	0.05	0.31	−0.44 *	0.47 *
ECp (dSm−1)	0.55 *	−0.08	0.16	−0.04	−0.11
εb0.5	−0.17	0.00	0.23	−0.33	0.44 *
Τ (°C)	0.17	−0.07	−0.24	0.18	−0.18
Field 2./Potato Crop	ΕCe (dSm−1)	Clay (%)	Silt (%)	Sand (%)	CaCO3 (%)
ΕCe (dSm−1)	1.00	−0.04	−0.04	0.05	0.14
H0.5 (dSm−1)	−0.22	0.00	−0.11	0.08	−0.09
V0.5 (dSm−1)	0.12	0.05	0.25	−0.15	0.05
H1 (dSm−1)	0.04	0.45 *	−0.15	−0.33	−0.50 *
V1 (dSm−1)	0.05	0.06	0.04	−0.10	−0.27
WET-ECa (dSm−1)	0.15	−0.09	−0.12	0.21	0.21
ECp (dSm−1)	0.10	−0.21	−0.21	0.38	0.25
εb0.5	0.18	−0.01	0.03	0.00	0.23
Τ (°C)	−0.18	−0.17	0.26	−0.01	0.23
2023
Bare Land and Farmlands	ΕCe (dSm−1)	Clay (%)	Silt (%)	Sand (%)	CaCO3 (%)
ΕCe (dSm−1)	1.00	−0.04	−0.04	0.05	0.14
H0.5 (dSm−1)	0.60 *	−0.48 *	−0.30	0.41 *	0.50
V0.5 (dSm−1)	0.81 *	−0.31	−0.16	0.29	0.36
H1 (dSm−1)	0.57	−0.60 *	−0.36	0.63 *	0.52
V1 (dSm−1)	0.64	−0.45	−0.02	0.38	0.14
WET-ECa (dSm−1)	0.74 *	0.26	0.19	−0.29	0.38
ECp (dSm−1)	0.92 **	−0.30	−0.40	0.34	0.80 *
εb0.5	−0.35	0.74 *	0.77 *	−0.83 *	−0.67
Τ (°C)	0.42	−0.67	−0.59	0.69	0.62
2024
Orange Orchard	ΕCe (dSm−1)	Clay (%)	Silt (%)	Sand (%)	CaCO3 (%)
ΕCe (dSm−1)	1.00	−0.05	−0.09	0.05	−0.02
H0.5 (dSm−1)	−0.16	0.37	0.48 *	−0.58 *	−0.34
V0.5 (dSm−1)	−0.15	0.35	0.42 *	−0.58 *	−0.40
H1 (dSm−1)	0.35	0.09	0.07	−0.08	−0.22
V1 (dSm−1)	0.17	0.06	0.09	−0.17	−0.37
WET-ECa (dSm−1)	0.32	0.54 *	0.27	−0.43 *	0.09
ECp (dSm−1)	0.44	0.49 *	0.01	−0.21	0.27
εb0.5	−0.06	0.34	0.31	−0.38	−0.11
Τ (°C)	0.16	−0.04	0.38	−0.38 *	−0.23

Significant correlation at * p < 0.05, ** p < 0.001 (2-tailed).

. The r_s values between ECe and the particles of clay, silt, sand, and CaCO₃ equivalent percentages are also shown.

For both fields of 2022, the results revealed low or negligible correlation between the readings of EM38-MK2 and WET-2 with ECe, reflecting the influence of low soil moisture in the ECa surveys. Particularly, at the potato crop (field 2), apart from the H1 of EM38-MK2, which is correlated with the more static properties of soil, including clay (r_s = 0.45) and CaCO₃ (r_s = −0.50), no statistically significant relationships were developed between the sensors’ variables and the determined soil properties. By documenting a correlation between clay content and EMI-ECa (r = 0.35) in extreme low soil moisture conditions, Pedrera-Parilla et al. [44] suggested that an EMI-based survey can be applied even in soil water-limited systems to delineate the spatial distribution of more stable soil properties such as soil texture, albeit with higher correlations that are to be found in well-moisturized soil profiles. Furthermore, the non-significant correlation between ECe and the measurements of the EM38-MK2 and WET-2 probe is consistent to the substantially low soil water content θ levels that did not exceed 15.87% (mean θ = 11.87%) (Table 2) across the field, indicating dry and non-conductive upper soil layers.

In the case of the olive orchard (field 1), however, where the WET-2’s estimated θ reaches up to 33.63% (mean θ = 14.32%) on the soil surface (Table 2), a moderate correlation is observed between ECe and the EM38-MK2 reading of V1 (r_s = 0.40). This correlation reflects that ECe (0–30 cm) is captured by the averaged electrical conductivity measured by EM38-MK2 in the deeper soil layers. Furthermore, in contrast to the non-significant relationship with WET-ECa, ECe is moderately correlated with ECp (r_s = 0.55). This is relevant to the fact that ECp represents the ion concentration dissolved in the soil solution; hence, it is more associated with the soil salinity indicated by ECe.

In comparison to the 2022 survey, at the experimental area of 2023, where the soil moisture conditions on the soil surface were adequate, both sensors’ readings were related to ECe at 0–30 cm. In detail, among the EM38-MK2 measurements, V0.5 was found to have the strongest correlation with ECe (r_s = 0.81), followed by H0.5 (r_s = 0.60), demonstrating the sensor’s response was more affected by the variations of soil salinity occurring in the deeper soil layers (V0.5) and less by those beneath the soil surface (H0.5). In general, at the 0.5 coil separation, the relative sensitivity of the sensor in the vertical mode becomes maximum at about 20 cm depth [52,66]. As regards the WET-2 probe, ECe is highly correlated with ECp (r_s = 0.92, p < 0.01), while a lower, yet strong, correlation of r_s = 0.74 is shown between ECe and bulk WET-ECa.

Opposite to the majority of findings reporting a positive relationship of clay with the EMI-ECa [67,68,69,70], a negative correlation was observed between clay and the H0.5 and H1 at the district of 2023. This negative correlation may be attributed to the highly heterogeneous soil surface and the clay mineralogy that can affect the near-surface ECa measurements and will be investigated in future studies.

At the orange orchard (2024), due to the low salinity levels at the topsoil (ECe < 1.32 dSm⁻¹) (Table 1), and the narrow range of the obtained ECa measurements up to about 2 dSm⁻¹ (Table 2), both devices present weak and non-significant correlations with ECe. This absence of correlation comes to agreement with the hypothesis that under adequately wetted (mean θ = 39.43%), yet non-saline soils, ECa cannot be directly related to soil salinity, since further soil factors might govern its magnitude. Indeed, the H0.5 and V0.5 derived by EM38-MK2 show a moderate positive correlation with silt (r_s = 0.48 and r_s = 0.42, respectively) and a stronger negative correlation with sand (r_s = −0.58), highlighting the contribution of soil texture to the sensor’s ECa readings. The WET-ECa provides moderate correlations with clay (r_s = 0.54) and sand (r_s = −0.43), whereas ECp is correlated only with clay (r_s = 0.49), validating the high sensitivity of the apparent dielectric permittivity in the variations of soil texture [19] at small scales.

Interestingly, H0.5 and V0.5 yield weak and statistically insignificant correlations with clay content (r_s = 0.37 and r_s = 0.35, respectively), which could be explained by the loamy-textured soils of the studied area (Table 1) with lower concentrations of clay and potentially less conductive clay mineralogy. Weak correlation between clay and EMI-ECa has also been documented in recent publications for sandy loam soils [65,71].

3.4. Principal Components Analysis

Principal component analysis (PCA) was applied to 12 variables, which were measured from 81 soil samples collected from the four experimental areas. PCA was employed to reduce the dimensionality of the data and to detect the most important causes of variability, since a high correlation between the variables was noticed. By this means, the variables that are most capable of distinguishing the soil samples are identified. The 12 variables included ECe, soil parameters of WET-ECa, ECp, and εb^0.5 by the dielectric WET-2 probe, the H0.5, V0.5, H1, and V1 readings as recorded by the ΕΜ38-ΜΚ2, the particle size fractions of clay, silt, sand, and the CaCO₃ equivalent percentage for the selected soil samples. PCA resulted in three principal components, which accounted for 83.45% of the total variability. In Figure 3, the aforementioned variables and soil samples can be seen as a function of both first and second principal components.

The first principal component (PC1) explained 50.42% of the total variability and was defined by the parameters H0.5, V0.5, H1, V1, and εb^0.5. They were placed away from the axis origin, suggesting that they were well reflected by PC1 and clustered together, showing a strong positive correlation. The first PC could be recognized as a proxy of the ΕΜ38-ΜΚ2 measurements. The second principal component (PC2) explained another 19.56% of the total variability and was associated with ECe, WET-ECa, ECp, and CaCO₃. ECe, WET-ECa and ECp were located close together on the positive side of PC2, indicating they are strongly and positively correlated. The second PC could be regarded as a representative of the estimated soil salinity (ECe) and the parameters of WET-ECa and ECp, provided by the WET-2.

The third principal component, which explained another 13.47% of the total variability, was driven by the percentages of clay, silt, and sand and appears to interpret the grain size distribution of the soil samples.

Moreover, samples selected from fields 1 and 2 (2022) were clustered near clay, reflecting a higher concentration than the samples from the orange orchard (2024), which were clustered on the positive side of PC1, near the H0.5, V0.5, H1, and V1 measurements and εb^0.5. Samples from this site (4), as can be seen, had higher concentrations of silt and sand. On the other hand, samples from the wider district of the 2023 survey (3) were placed near ECe, WET-ECa, and ECp, having the highest soil salinity values. The two data points of the 2023 survey, which are observed at the far right of the bi-plot, and are distinctly separated from the rest of the variables, reflect the soil samples with the highest soil salinity values (ECe) within the dataset (19.9 and 20.4 dSm⁻¹).

3.5. Prediction of Soil Salinity and Texture Using the EM38-MK2 and WET-2 Measurements

Multiple linear regression models of topsoil salinity (0–30 cm) were developed using the readings from the EM38-MK2 (H0.5, V0.5, H1, and V1) and the WET-2 probe (WET-ECa, ECp, εb^0.5, and T) along with the determined soil properties as independent variables, and ECe values as dependent for each survey. In the non-saline soils and fields with limited soil moisture (2022, 2024), the prediction of soil texture (clay, silt, and sand) by the EM38-MK2 measurements was additionally explored.

The models generated for each sensor using stepwise selection were significant at p < 0.001 and met the basic assumptions (normality, linearity, homoscedasticity, and independence of residuals) of regression. Subsequently, the performance of the models was evaluated, and the results are compiled in

Table 4. Multiple linear regression models of ECe (0–30 cm) and soil texture (clay, silt, and sand) using the measurements of EM38-MK2 (H0.5, V0.5, H1, and V1) and WET-2 (WET-ECa, ECp, T, and εb^0.5) sensors for the study areas. ECe, H0.5 (horizontal ECa at 0.5 coil separation), V0.5 (vertical ECa at 0.5 coil separation), H1 (horizontal ECa at 1 m coil separation), V1 (vertical ECa at 1 m coil separation), WET-ECa (ECa measured by the WET-2), and ECp (soil pore water electrical conductivity measured by the WET-2) are expressed in dSm⁻¹ and clay, silt, and sand in %.

Survey Date	Sample Size (N)	Sensor	Model *	adj. R2	RMSE	MAE	RPIQ **
2022	25	WET-2	ECe = 1.038 ECp–0.193 εb0.5	0.95	0.20	0.18	2.75
2023	13	WET-2	ECe = 1.447 ECp	0.96	2.08	3.99	5.67
			ECe = 4.963 WET-ECa	0.82	4.51	4.00	2.61
		EM38-MK2	ECe = 11.738 H0.5 + 0.523 Clay–14.85	0.93	2.24	1.93	5.25
			ECe = 8.372 V0.5–0.67	0.88	4.10	3.55	2.87
2024	20	EM38-MK2	Silt = 3.67 V0.5 + 39.1	0.49	3.18	2.50	2.24
			Silt = 9.30 V0.5–3.31 H1 +38	0.51	3.74	2.88	2.15
			Sand = 29.33 V0.5–30.5 H0.5 + 27.5	0.52	4.87	3.97	1.97

* All regression coefficients: p < 0.001, ** RPIQ < 1.7: poor model, 1.7 ≤ RPIQ < 2.2: good model, RPIQ ≥ 2.2: excellent model

.

It is worthy to mention that in the potato crop field (field 2) of the 2022 survey, no ECe or soil texture model could be established with the measurements of EM38-MK2 and WET-2, indicating the high influence of soil moisture content in ECa. Consistent with the correlation results (Table 3), it was shown that under limited soil water content (θ) conditions of approximately 10–15%, soil properties of topsoil cannot be reliably determined by FDR probes or the non-destructive EMI sensors due to the discontinuity of conductance pathways in the soil solution that disrupts and reduces the penetration of the ECa signal on the upper soil layers [72].

Furthermore, despite the moderate correlation observed between H1 (EM38-MK2) and clay content, the regression analysis could not produce a statistically significant model for predicting clay. While the characterization of the soil texture by EMI-ECa has been documented in the literature for non-salinized areas and dry soils, this was not achieved in the present study, most likely due to the limited number of measurements and soil samples of the survey.

As shown in Table 4, in the olive orchard of 2022, the topsoil ECe was predicted by the readings of the WET-2 probe, including ECp and εb^0.5 with high accuracy (adj. R² = 0.95) and low RMSE, and MAE values of 0.20 dSm⁻¹ and 0.18 dSm⁻¹, respectively. In addition to the high RPIQ value of 2.75, which reflects the strong predictive performance of the variables ECp and εb^0.5, the model’s form comes in agreement with the findings by Visconti et al. [41] and Kargas et al. [16]. Contrary to the aforementioned studies, though, the temperature did not enhance the accuracy of the prediction. The developed model highlights that WET-2, through the estimated ECp and εb^0.5, which is used as a surrogate of soil water content (θ), provides accurate and reliable estimations of the low soil salinity levels (0.67–1.96 dSm⁻¹), even in limited soil water conditions, with a mean field θ value of about 14%. Regarding EM38-MK2, the generated models did not satisfy the modelling requirements, which could be attributed to the larger volume of soil sensed by the instrument compared to the extracted soil core samples (0–30 cm).

In the heterogeneous soil conditions of the experimental site in 2023, with a broad range of ECe levels (1.58–20.40 dSm⁻¹) and adequate soil moisture, both EM38-MK2 and WET-2 provided good predictions of soil salinity at 0–30 cm depth.

The best-fit models with the highest RPIQ were produced with the ECp measurements for the WET-2 probe (adj. R² = 0.96, RPIQ = 5.67) and the variables of H0.5 and the clay for the EM38-MK2 sensor (adj. R² = 0.93, RPIQ = 5.25). Despite the moderate prediction errors of RMSE = 2.08 dSm⁻¹ and MAE = 3.99 dSm⁻¹ for the WET-2 model and RMSE = 2.58 dSm⁻¹ and MAE = 2.01 dSm⁻¹ for EM38-MK2, both models exhibit good predictive ability, with RPIQ reflecting substantial evidence and smaller errors relative to ECe variability. In contrast, the models using the WET-ECa (WET-2) and the V0.5 (EM38-MK2) variables featured slightly higher RMSE and MAE values and lower accuracy in predicting the ECe variations at topsoil (Table 4).

The inclusion of clay in the EM38-MK2 model underlies the significance of taking into account additional soil properties when conducting salinity surveys at sites with high textural variability. In such heterogeneous soil conditions, which are prominent in the shallower soil layers, the textural components influence the soil’s water-holding capacity, hence complicating the ECa measurements, particularly when these fall within a range of 1–2 dSm⁻¹, as was observed for the H0.5 values (Table 2). Visconti et al. [41], using inverted bulk soil ECa data from the EM38, documented that although all EM38-based ECe models had a poor performance (R² < 0.50), the incorporation of clay and temperature provided the most reliable ECe predictions (R² = 0.36) for the sensor. Meanwhile, the high predictive performance of the WET-2’s ECp model found in our study indicates that when the soil water content is sufficient, ECe can be estimated across a wide range of high and low salinity levels, through a simple linear model using only ECp data, as was similarly stated in [16].

As anticipated from the Spearman’s correlation results, in the non-saline soils of the orange orchard (2024), no ECe model reached statistical significance. A potential reason for this could be the fact that the measurement taken by both sensors of ECa, with values below 2 dSm⁻¹, is affected by soil factors that exhibit higher spatial variability than soil salinity in the topsoil. In fact, despite sufficient levels, soil water content θ has a low variability (CV = 9%), thus hindering the WET-2 from detecting the differences of soil salinity through ECp, which also exhibits a low range (CV = 9%). In this case, the EM38-MK2 readings predicted silt and sand mass fractions, indicating soil texture could be the main driver of ECa. The developed models of silt content yielded moderate accuracy (0.49 < adj. R² < 0.51), with RMSE and MAE values between 3.18 dSm⁻¹ to 3.74 dSm⁻¹ and 2.50 dSm⁻¹ to 2.88 dSm⁻¹, respectively, while the RPIQ values of 2.14 and 2.25 demonstrated fair model performances. For sand, the model showed moderate predictive ability (adj.R² = 0.50), with higher RMSE (4.87 dSm⁻¹) and MAE (3.97 dSm⁻¹) values and a lower RPIQ (1.97).

The relationship between the estimated ECe and the prediction by each model is given in Figure 4.

At all study sites, instead of applying the gravimetric method to estimate θ, which consists of a destructive and more tedious procedure, the soil water content θ was estimated in situ by the capacitance WET-2 probe, using Equation (1). In this way, the moisture status of the topsoil could be directly quantified, allowing us to evaluate the interactions of ECa with soil salinity and texture under various soil wetting conditions. Since EM38-MK2 does not directly measure soil water content, the WET-2 was employed in a complementary manner, eliminating the time and additional data that would be required for local calibration of EM38-MK2 for each soil type [73]. Similarly, the effects of soil temperature measured by the dielectric probe during the surveys were accordingly examined. It is worth noting that in the current study, θ and ECp estimations by the WET-2 relied on default factory calibrations, which could result in limited accuracy. Applying soil-specific calibrations may substantially improve the accuracy and reliability of both θ [74] and ECp [75]. The produced models may be applicable to soils exhibiting texture, salinity, and moisture conditions similar to those in this study. Nevertheless, additional experiments across diverse soil types, cropping systems, and a larger dataset of sensor measurements are recommended to further assess their robustness. Furthermore, although organic matter content in the topsoil was relatively low across the examined fields, its potential influence on ECa measurements and soil salinity interpretation, particularly in non-saline soils, should be explored in future research.

4. Conclusions

In the present study, EMI and FDR field surveys were conducted across the region of Laconia in Greece for the estimation of soil salinity and texture at 0–30 cm depth. To this aim, on-ground measurements were obtained from an electromagnetic EM38-MK2 sensor and a WET-2 capacitive dielectric probe under varying soil conditions and diverse land uses, and were used to develop site-specific models of topsoil (0–30 cm) salinity (ECe) and soil texture through multiple linear regression. The performances of the sensors were individually assessed and evaluated to identify potential benefits of their complementary application in enhancing the soil properties interpretation.

According to the findings, reliable ECe models were constructed from the readings of both sensors in saline soils across a broad range of texture and soil water content (20–43%), as was estimated by the WET-2. In these soils, the depth-weighted ECa by the EM38-MK2 in horizontal mode (H0.5) along with clay and the calculated-ECp by the WET-2 probe provided the most accurate estimations of topsoil salinity. On the other hand, in all non-saline soils exhibiting low and relatively uniform soil moisture, EM38-MK2 and WET-2 were unable to predict ECe at the topsoil, highlighting the limitations of ECa surveying in dry soil surfaces. A notable exception was observed for the WET-2, which, using only ECp and the εb^0.5 as a proxy of soil moisture, could estimate soil salinity when the field’s volumetric water content was approximately 14%. Additionally, under these soil conditions, prediction models of silt and sand content were developed by the EM38-MK2 readings with moderate accuracy for the loamy soils.

The research demonstrated that the modern EM38-MK2, offering rapid and noninvasive measurements of the shallower soil layers up to 0.37 and 0.75 m, could be a suitable tool for assessing soil salinity at 0–30 cm depth in heterogeneous soils at large scales, without prior data inversion or instrument elevation adjustments. However, additional edaphic factors should be considered to improve the predictions. To facilitate the investigation of the soil properties’ impact on ECa, the affordable WET-2 probe, which measures soil temperature and estimates the volumetric soil water content in real time, could be complementarily used in situ, reducing extensive soil sampling and calibration requirements. Moreover, the WET-2, based solely on the ECp and εb measurements, could be effectively employed for the topsoil salinity appraisal, even in relatively restricted soil water conditions, although proper contact of the probe with the soil is demanded. Caution is also advised when applying the proposed protocol due to the discrepancy between the sensing depth of the WET-2 probe (0–10 cm) and the depth of soil samples at 0–30 cm, which is a limitation that can become critical in soils with strong moisture and salinity stratification and in crop systems with ridges and inter-rows. For a more enhanced evaluation of the sensors’ prediction ability, soil-specific calibrations of θ and ECp are recommended for the WET-2, and a higher density of measurements is suggested for the EM38-MK2. Furthermore, the calculation of ECp, which reflects a promising indicator of soil salinity, necessitates further examination, especially at larger and regional scales.

Future research should focus more on using such cost-efficient and user-friendly FDR devices with EMI or different technologies in situ to leverage prediction accuracy and overcome their applicability constraints. The complementary sensor approach presented in this study may support farmers in extracting meaningful insights on the soil properties’ complex interactions, particularly in soils with low salt concentrations. A comprehensive understanding of the soil salinity dynamics may also be achieved in the shallow soil layers, where the salt accumulation through the evaporation and cultivation practices can become detrimental for agricultural productivity.