Comparison of Multi-Frequency and Multi-Coil Electromagnetic Induction (EMI) for Mapping Properties in Shallow Podsolic Soils.

Electromagnetic induction (EMI) technique is an established method to measure the apparent electrical conductivity (ECa) of soil as a proxy for its physicochemical properties. Multi-frequency (MF) and multi-coil (MC) are the two types of commercially available EMI sensors. Although the working principles are similar, their theoretical and effective depth of investigation and their resolution capacity can vary. Given the recent emphasis on non-invasive mapping of soil properties, the selection of the most appropriate instrument is critical to support robust relationships between ECa and the targeted properties. In this study, we compared the performance of MC and MF sensors by their ability to define relationships between ECa (i.e., MF-ECa and MC-ECa) and shallow soil properties. Field experiments were conducted under wet and dry conditions on a silage-corn field in western Newfoundland, Canada. Relationships between temporally stable properties, such as texture and bulk density, and temporally variable properties, such as soil water content (SWC), cation exchange capacity (CEC) and pore water electrical conductivity (ECw) were investigated. Results revealed significant (p < 0.05) positive correlations of ECa to silt content, SWC and CEC for both sensors under dry conditions, higher correlated for MC-ECa. Under wet conditions, correlation of MF-ECa to temporally variable properties decreased, particularly to SWC, while the correlations to sand and silt increased. We concluded that the MF sensor is more sensitive to changes in SWC which influenced its ability to map temporally variable properties. The performance of the MC sensor was less affected by variable weather conditions, providing overall stronger correlations to both, temporally stable or variable soil properties for the tested Podzol and hence the more suitable sensor toward various precision agricultural practices.


Introduction
Characterization of spatiotemporal variability of relevant physicochemical properties of soil is crucial for precision agriculture and for various environmental sectors [1]. Commonly, soil sampling and laboratory analyses are carried out to understand the spatiotemporal variability of soil properties. However, conventional methods involve invasive soil sampling which is expensive and time consuming and only provide point information. Moreover, soil sampling is often technically not feasible for large-scale and extended temporal monitoring or for areas with restricted accessibility [2][3][4][5]. Mapping of proxy properties, such as apparent electrical conductivity (EC a ) by electromagnetic induction (EMI)

Study Area
The study was conducted at the Pynn's Brook Research Station (PBRS) (49 • 04 23 N, 57 • 33 39 W), located in the Humber Valley, Western Newfoundland, Canada ( Figure 1). The soil texture in the top 0-15 cm soil layer is sandy loam to loamy fine sand, overlain over sandy fluvial and glacio-fluvial deposits [25]. The experimental area covers approximately 0.4 ha of five different silage-corn varieties with different agronomic treatments and an adjacent grassed field [2,26]. A detailed study using EMI instruments was focused on one variety of the silage-corn experiment only, which covered approximately 350 m 2 area. The crop was fully grown and considerably similar for both wet and dry days where EMI data collection was carried out. The mean annual precipitation and temperature obtained from the nearby weather station in Deer Lake, are 1113 mm and 4 • C, respectively, (http://climate.weather.gc.ca/). EMI and soil samples were collected on 18 August 2017 after several consecutive hot and dry days,

Soil Sampling and Analysis
We collected undisturbed and composite soil samples from the selected study plot. The undisturbed soil samples were taken from 0 to 15 cm depth and used for texture (n = 24) and BD (n = 48) analysis. Composite soil samples were collected from 0 to 20 cm to investigate the depth averaged values of SWC, CEC, pH, and ECw. Each composite sample comprised three samples collected in each treatment plot on a diagonal line with 1 m distance and 0.3 m spacing ( Figure 1). All soil samples were analyzed according to the standard protocols (Table 1). For simplification, we assumed a homogenous distribution of the soil texture and BD within the depth of 0-20 cm with no temporal changes throughout the study period.

Soil Sampling and Analysis
We collected undisturbed and composite soil samples from the selected study plot. The undisturbed soil samples were taken from 0 to 15 cm depth and used for texture (n = 24) and BD (n = 48) analysis. Composite soil samples were collected from 0 to 20 cm to investigate the depth averaged values of SWC, CEC, pH, and EC w . Each composite sample comprised three samples collected in each treatment plot on a diagonal line with 1 m distance and 0.3 m spacing ( Figure 1). All soil samples were analyzed according to the standard protocols (Table 1). For simplification, we assumed a homogenous distribution of the soil texture and BD within the depth of 0-20 cm with no temporal changes throughout the study period.

Field Data Collection
The working principle of EMI is based on a two-coil system (transmitter and receiver coil) and has been established for several decades [33,34]. A transmitter coil generates the primary magnetic field which noninvasively induces eddy currents in the soil that in turn generate a secondary magnetic field. From the ratio between both magnetic fields, the bulk EC a (as integral over a certain soil volume) can be derived under low induction number conditions [33]. Meanwhile MC as well as MF sensors are commercially available, allowing for simultaneous recording of EC a from different depth integrals. Two established EMI instruments were used in this study, both build to investigate the shallow soil properties: the CMD-MINIEXPLORER (GF-Instruments, Brno, Czech Republic) operating with a fixed frequency of 30 kHz and three coil separations (0.32 m, 0.71 m and 1.18 m), [11,35], and the GEM-2 (Geophex Ltd., Raleigh, NC, USA) with up to six manually set frequencies and one coil separation (1.67 m plus bucking coil at 1 m) [36,37]. EC a data were recorded in vertical coplanar (VCP) and horizontal coplanar (HCP) coil orientations in both instruments. Instruments were warmed up for >20 min before data recording and held approximately 0.20 m (MC) and 1.0 m (MF) above ground according to the manufacturer instructions. We used a track distance of 1.0 m and instrument orientation parallel to the transects ( Figure 1). No GPS was used. Temperature corrections for the EC a were done using the temperatures values from soil probes [38].

Theoretical Investigation Depth
Although the investigation depths (depth of EC a origin) for MC instruments are widely accepted, the related depth function is based on a theoretical equation, derived for ideal homogeneous material [34]. According to McNeill's approximation [34], the investigation depth is related to the coil separation and its orientation to the surface. The VCP orientation has its highest sensitivity closer to the surface while the HCP orientation reaches deeper depths. The employed CMD-MINIEXPLORER provides six integral depths if both coil orientations are used. If 75% of the cumulative signal were considered, as suggested [34], this would provide EC a from the following investigation depth: VCP-C1 (25 cm), VCP-C2 (50s cm-shallow), VCP-C3 (90 cm), HCP-C1 (50d cm-deep), HCP-C2 (105 cm), and HCP-C3 (180 cm) [11,39]; while the signal origin from the VCP is generally closer to the surface. Field experiments as well as the local sensitivity derived from the theoretical function, however, suggested that the majority of the signal response originates from a shallower depth (<100 cm for all separations and configurations) [35]. Still, given the integral characteristics of EC a and the heterogeneities from natural soils, the actual signal origin in field could be varied in relation to the affecting conditions. The depth of investigation for the MF instruments is theoretically controlled through the employed frequencies, with lower frequencies reaching greater depths (http://geophex.com). Hence, the MF principle and its possibility of changing frequencies was proposed for depth sounding [40]. The GEM-2 employed in this study allows for the simultaneous recording of up to ten frequencies. However, selection of too many frequencies reduces the strength of each frequency signal and consequently lowering the resolution. Based on previous studies, we selected three frequencies, which were also suggested by the manufactures (default settings): 18 kHz, 38 kHz, and 49 kHz. Using again both coil configurations, the GEM-2 provided six sampling depths; hereafter these depths are denoted as VCP-18 kHz, VCP-38 kHz, VCP-49 kHz, HCP-18 kHz, HCP-38 kHz, and HCP-49 kHz ( Figure 2).
Sensors 2020, 20, x FOR PEER R EVIEW 5 of 15 the MF principle and its possibility of changing frequencies was proposed for depth sounding [40]. The GEM-2 employed in this study allows for the simultaneous recording of up to ten frequencies.
However, selection of too many frequencies reduces the strength of each frequency signal an d consequently lowering the resolution. Based on previous studies, we selected three frequencies, which were also suggested by the manufactures (default settings): 18 kHz, 38 kHz, and 49 kHz. Using again both coil configurations, the GEM-2 provided six sampling depths; hereafter these depths are denoted as VCP-18 kHz, VCP-38 kHz, VCP-49 kHz, HCP-18 kHz, HCP-38 kHz, and HCP-49 kHz ( Figure 2).

EMI Data Processing
All ECa were initially quality checked based on their signal-noise ratio. Thereby we considered data to be very noisy if the noise level reached a higher magnitude than the informal content in the variogram analysis. The readings from VCP-C1 and HCP-C1 (MC) as well as VCP-18 kHz and HCP-18 kHz (MF) did not pass the quality check and were consequently ignored from further processing as previously suggested [2,6,42]. ECa data were interpolated by ordinary block kriging using Surfer11 software (Golden Software Inc., Golden, CO, USA) [18]. Point values for the soil sampling locations were derived from interpolated maps and averaged for each treatment plot resulting in 16 points for both dry day and wet day. Simple Pearson's correlation (r) was calculated between soil properties and discrete ECa data for each frequency, coil configuration, and coil separation, using the statistical software Minitab 17 (Minitab Inc., State College, PA, USA).
To assess the practical purpose of the mapping quality from both sensors, we used the provided correlations and predicted the targeted soil properties based on the linear models from each ECa data set. We validated the model accuracies by using leave-one-out validation [11,43]. By comparing the independent prediction vs. the measured values, we displayed the coefficient of determination (R 2 ) as indication for the validated explanatory power of each linear model.
Two assumptions were made in this study: (i) The quadrature component of the secondary field was proportional to ECa under low induction number condition [3,39]; and (ii) soil texture and BD data were assumed to be stable over the monitoring duration and the same values were used for both dry and wet day analyses [10,44].

EMI Data Processing
All EC a were initially quality checked based on their signal-noise ratio. Thereby we considered data to be very noisy if the noise level reached a higher magnitude than the informal content in the variogram analysis. The readings from VCP-C1 and HCP-C1 (MC) as well as VCP-18 kHz and HCP-18 kHz (MF) did not pass the quality check and were consequently ignored from further processing as previously suggested [2,6,42]. EC a data were interpolated by ordinary block kriging using Surfer11 software (Golden Software Inc., Golden, CO, USA) [18]. Point values for the soil sampling locations were derived from interpolated maps and averaged for each treatment plot resulting in 16 points for both dry day and wet day. Simple Pearson's correlation (r) was calculated between soil properties and discrete EC a data for each frequency, coil configuration, and coil separation, using the statistical software Minitab 17 (Minitab Inc., State College, PA, USA).
To assess the practical purpose of the mapping quality from both sensors, we used the provided correlations and predicted the targeted soil properties based on the linear models from each EC a data set. We validated the model accuracies by using leave-one-out validation [11,43]. By comparing the independent prediction vs. the measured values, we displayed the coefficient of determination (R 2 ) as indication for the validated explanatory power of each linear model.
Two assumptions were made in this study: (i) The quadrature component of the secondary field was proportional to EC a under low induction number condition [3,39]; and (ii) soil texture and BD data were assumed to be stable over the monitoring duration and the same values were used for both dry and wet day analyses [10,44].

Descriptive Analysis of Soil Properties
The soil was a loamy sand with high sand content (73.2%), relatively uniformly distributed among the samples (CV = 4.7%) ( Table 2). On the other hand, the variability of silt (20.8%) and clay contents (6.0%) were greater (CV of 15.3% and 13.1%, respectively). The BD of 1.4 g/cm 3 was at the upper end of the range considered ideal for plant growth (www.nrcs.usda.gov) with relative low variation among the samples (CV 5.1%), indicating uniform compaction across the field. Except for EC w , all tested temporally variable soil properties were higher on the wet day compared to the dry day, SWC decreased from 19.7% to 12.3%. We interpret the lower EC w on the wet day to be due to dilution effect leading to lower ionic strength of the soil solution. Moreover, the root uptake of nutrients and leaching from the bottom of the soil column will also result in low EC w . The CEC was relatively stable while the strongly acidic soil (pH 5.4) became moderately acidic (pH 5.7) when wet [45].

Descriptive Analysis for EC a Data
The readings from both EMI instruments (Table 2) show relatively low EC a for their sandy soil, as also reported by several previous studies [2,46]. Both instruments recorded higher EC a on the wet day, most likely as consequence of higher SWC ( Table 2). The MF measured relatively higher values throughout all readings which might be related to the depth of signal origin. The MF-EC a also had higher CVs (up to 58.7% for HCP-38 kHz) than MC-EC a (<13%). The second coil separation (C2) of the MC produced the highest mean EC a , of 4.0 ± 0.3 mS/m) for HCP-C2 (dry) and 6.2 ± 0.8 mS/m for VCP-C2 (wet). Interestingly, the CV was higher for MC on the wet day whereas it was lower for the MF-EC a . EC a measured from VCP-49 kHz was 20.3 ± 0.7 mS/m, the highest value among both instruments and coil orientations, and produced the lowest CV (3.7%). The 38 kHz frequency data of the MF showed high CVs on both days compared to all other EC a values, indicating higher variability of the recordings. On the other hand, the EC a measurements by 49 kHz frequency had a relatively low CV (3.7) and higher mean EC a value, ranging from 7.5 (±0.7) to 20.3 (±0.7) mS/m for both days ( Table 2).
For the dry day, the VCP mode of the MF EMI gives a higher EC a compared to the HCP mode. A similar pattern of high variability on dry day vs. wet day for MF instruments has been reported [4,19]. Overall, for the wet day, the 38 kHz data from MF EMI, and soil properties including silt, clay, SWC, and CEC showed similar variability. Likewise, EC a data measured by 49 kHz frequency showed narrow variability for the same soil properties for the dry day. All MC EMI data showed adequate variability range with the aforementioned soil properties for both days compared to MF EMI sensor.

Correlation to the Targeted Soil Properties
The Pearson correlation coefficient (r) between EC a and soil properties (Table 3) show various significant (p < 0.05) correlations with the stable properties. EC a from both sensors were negatively correlated to sand, while for the MC, almost every coil separation and orientation was significant. As for MF-EC a data, correlations were poor and non-significant. The negative correlations were most likely the results of the high amount of low (electric) conductivity sand content in the soil. In general, larger sand particle sizes were associated with decrease in the EC a [47][48][49].
Silt was clearly and positively correlated to all EC a readings from both sensors. The MC-EC a provided again higher correlations, consistently significant, while the MF-EC a readings provided significant (except HCP-38 kHz), yet lower correlations. The positive correlations between silt and EC a were previously documented [50]. Both correlations, to sand and silt, could be due to a relatively high data range of these properties. Clay content is also generally considered to have a positive relationship to EC a [9,51]. However, in contrast to the majority of previous findings, the correlations to clay in our study remained insignificant for both instruments and partly even negative. One reason therefore could be found in the overall low clay content (6%) and its narrow data range, hindering positive correlations, as reported by Bronson et al. [52]. EC a for both sensors had further negative and insignificant correlations to BD. Although BD is commonly considered as positively correlated to EC a as a result of higher current flow because of greater particle contacts [53][54][55], it was also observed to be negative in soils with higher organic matter content [11]. Additionally, the correlation between EC a and BD is related to mineral content, soil solution, and air phases resulting the interplay between the non-conductive dry mineral part and the liquid conductive phase, while only the liquid phase conducts electrical current through the soil [56]. Therefore, high BD values do not necessarily result in higher EC a . We explained the observed negative correlations of EC a to BD at our test site by the limited amount of conductive (wet) clay minerals and the high non-conductive mineral (sand) parts, acting rather as insulation. The relative homogeneous BD among the tested site ( Table 2) and its corresponding narrow data range additionally hindered to establish proper relationships with EC a . The potential correlation of SWC and EC a are probably the most prominent relationship and build the baseline for various SWC documented mapping studies [56]. During the dry day, SWC correlations with EC a readings were positive and significantly consistent for both sensors ( Table 3). The 38 kHz data of the MF even reached the highest r (0.83). These correlations declined for the wet day, particularly visible for the 38 kHz values recorded by the MF which was opposite to our expectations. This phenomenon could result a lower EC w value in the solute phase as measured during wet conditions. The simultaneous higher correlations to sand and silt on the wet day pointed to an inflecting influence of higher SWC on the MF-EC a . Higher negative correlations to sand might be caused by its insulating effect on the overall higher bulk electrical conductivity, while the higher correlations to silt were probably caused by activating of the conductive surface layers under wet conditions. In contrast, the MC-EC a values were relatively stable for both days, providing moderate to strong and always significant correlations to SWC.
While pH could not be related to any of the EC a sets, CEC on the dry day had significant influence for both instruments. However, CEC significantly correlated only VCP-C3 and HCP-C2 of MC data on the wet day, while other MC and all the MF recordings remained insignificant. We overall explained the weaker correlations on the wet day with lower CEC values and spatial variabilities ( Table 2) limiting its allocation to the EC a . On the dry day, the correlation to EC w were positive and partly significant to the MC values only, while its influence on the MF was negligible. Under the wet condition, all relationships of EC a to EC w were higher for both sensors, but significant only to VCP-49 kHz (MF), VCP-C3, and HCP-C2.
We explained the higher correlation from EC a to EC w during wet conditions by the accompanying higher SWC resulting in a higher saturation percentage and potential dissolution of ions to the soil solution potentially increasing the ionic strength. Since the soil had not reached the saturation of 47.2% on the wet day, based on hydrological simulations [57], there was no chance for ions to be leached out from the soil lowering the EC w . In contrast to the hypothesis and the previous findings [6], the correlations to SWC under wet conditions were lower or similar. This phenomena of higher correlation between EC a and SWC on drier days was also observed in sandy soils [58]. Although higher EC w values were empirically related to deflecting the EC a -SWC correlations [17,53], which was also observed under field conditions [6], very low EC w values demonstrated as narrowing the EC a -SWC relationship, which could lead to lower prediction accuracy [56]. The higher EC a -SWC correlations on the dry day might be the result of the higher EC w . On the other hand, the correlation of EC a to EC w were higher for the wet day, despite the narrow EC w data range, indicating an interplaying effect of SWC and its ionization on the recorded EC a [56]. Regardless of the differences between the sensors, the analysis highlights the complexity of signal response and its interactions between, free water, absorbed water, EC w , and particles content [56]. However, because of the limited data in this study, we are not capable to explain the mixed behavior of EC w under different sensors and coil orientation in total, which needs to be investigated in future studies.
Considering the explanatory power of the EC a correlations in Table 3 and its practical use for field mapping applications, Table 4 shows the accuracies from the respective linear models by means of the R 2 . The R 2 was generated by the predicted vs. the measured soil properties using the leave-one-out validation. Although the r in Table 3 reveled several promising significant correlations, their practical applications under field conditions remain limited. Using these correlations, only the SWC model based on VCP-38 kHz and HCP-C3 on the dry day and VCP-C3 on the wet day, as well as CEC from HCP-C2 and HCP-C3 on the wet day reached prediction accuracies higher up to 50%. Clay, BD, and pH accuracies were zero or negligible for both sensors. Prediction accuracies for sand ranged between 4.4% (VCP-49 kHz-dry day) and 13.8% (VCP-49 kHz-wet day) for the MF and 27.5% to 38% (HCP-C2, VCP-C2-dry day) and 17.1% to 36.3% (HCP-C2, VCP-C3-wet day) for the MC. Similar higher predictions of the MC-based models were observed for silt in both days, reaching up to 29% for the MF (VCP-49 kHz-wet day) while up to 45.3% for the MC (VCP-C3-wet day). However, we like to emphasize that the limited amount of samples restricted the overall prediction quality of all models. To archive sufficient accuracies for practical predictions, higher samples amounts would be required which was however not the focus of this study.
With respect to the overall results, the MC sensor performed better for the tested podzolic soil and selected variables, providing more stable EC a data sets and higher correlations to the targeted soil properties. With the exception of one single data set (SWC vs. VCP-38 kHz on the dry day), all correlations were lower and less significant for the MF sensor. The results further suggested a higher susceptibility of the MF-EC a to SWC variation, which could limit its operation for SWC mapping. However, we need to emphasize that our study considered variables from one soil type and shallow depths only. Bold numbers correspond to significant correlations (*** p < 0.001, ** p < 0.01, * p < 0.05) BD-bulk density; SWC-soil water content (gravimetric); CEC-cation exchange capacity; EC w -pore water electrical conductivity Table 4. Coefficient of determination (R 2 ) of the leave-one-out validation as obtained by the linear models between the soil properties (0-20 cm depth), and temperature corrected EC a data for both wet and dry days as displayed in Table 3 (n = 16).

Conclusions
We performed a comparative study using MF and MC EMI sensors to assess their ability for mapping properties for shallow soils under dry and wet conditions as relevant for agricultural purposes. The results show significant (p < 0.05) positive correlations to silt, SWC and CEC to EC a from both sensors under dry conditions, with higher correlations and significant levels for the MC sensor. The correlations of the MF-EC a to temporal variable properties SWC and CEC became insignificant under wet conditions, except for one frequency (38 kHz) to SWC. However, wet conditions increased the correlation of MF-EC a to sand and silt opposite to our expectations. In contrast, the correlations of the MC-EC a sensor to sand, silt, SWC and CEC remained relatively stable under both, dry and wet conditions. Sand was negatively correlated for both sensors on the dry day, however, only significant for the MC reading for both days. No significant correlations to clay, BD, and pH were found for either instrument, likely because of their limited amount and variability in the tested soil. Likewise, the prediction accuracies based on the linear correlations were lower for the MF sensor. From all considered variables and correlations, only SWC and CEC were reasonably projected (R 2 > 0.5) if EC a -based models were used for independent prediction of its variation, contrary to our hypothesis.
Given the results presented here, we concluded that the MF sensor is more affected by temporally variable soil properties, in particular by variation in SWC, which influenced its ability to establish proper relationships to these targeted variables (except 38 kHz) but also affecting its ability to map stable properties. The performance of the MC sensor was less affected by different weather conditions, providing overall stronger correlations to both, stable and temporally variable soil properties. At the tested loamy sand, the EC a data from the MC sensor were more suitable to investigate the spatiotemporal variability of shallow, agriculturally relevant soil properties compared to the MF sensor. Similar tests for different soil types and management conditions are needed to further verify these findings.