Multifrequency electromagnetic induction soil moisture characterization under different land uses in western Newfoundland

Abstract Identifying and characterizing the spatial patterns in soil moisture variability under different land use conditions is crucial for agriculture, forestry, and civil and environmental engineering. Yet employing multifrequency (MF) electromagnetic induction (EMI) techniques to carry out this task is under-represented in boreal podzolic soils. This study (i) compared four frequencies (∼2.8–80 kHz) for shallow mapping of soil moisture measured with a time–domain reflectometry at 0–20 cm soil depth under three different land use conditions (agricultural land, field road, and a recently cleared natural forest), (ii) developed a relationship between apparent electrical conductivity (ECa) measured using multifrequency EMI (GEM-2) and soil moisture, and (iii) assessed the effectiveness of ECa as an auxiliary variable in predicting soil moisture variations under different land use conditions. The means of ECa measurements were calculated for the exact sampling location (ground truth data) in each land use condition at a research site, Pasadena, NL, Canada. Soil moisture–ECa linear regression models for the three land use conditions were only statistically significant for 38.3 kHz frequency and were further analyzed. Further statistical analysis revealed that ECa was primarily controlled by soil moisture for the three land use conditions, with the natural forest possessing the highest mean ECa and soil moisture. Geostatistical analysis revealed that cokriging ECa with less densely collected soil moisture improved the characterization accuracy of soil moisture variability across the different land use conditions. These results reveal the effectiveness of the georeferenced MF–EMI technique to rapidly assess intrafield soil moisture variability under different land uses.


Introduction
Agricultural production in Newfoundland and Labrador (NL) is limited by the poor inherent fertility of the predominantly coarse-to medium-textured podzols.These soils are acidic and have low nutrient status (Enakiev et al. 2018).In 2017, the provincial government outlined plans to convert substantial areas of boreal forests into agricultural land as part of the "The Way Forward on Agriculture Initiative" to increase food production (Government of Newfoundland and Labrador 2017).The conversion of natural forest land to agricultural fields can introduce spatial and temporal variability in soil properties (Niu et al. 2015;Atwell and Wuddivira 2019).Soil properties naturally vary with land use (Guo et al. 2023), temperature and precipitation (Frazen 2018), as well as soil type (Wu et al. 2023).Properties, such as electrical conductivity (EC) (Tiwari et al. 2019), soil organic matter (Atwell and Wuddivira 2019), and water holding capacity (Bai et al. 2019) have been reported to respond to land use changes.Monitoring such changes is critical for efficient management, particularly in regions with low inherent production capacity, such as those in Newfoundland.
Soil moisture is a vital parameter for optimal site-specific management and ecosystem sustainability (Yu et al. 2018).Understanding spatial variability in soil moisture is key for effective management of hydrology and water resources (Guo et al. 2020), agriculture and irrigation (Funes et al. 2019), and rock and soil mechanics (Susha et al. 2014).Additionally, EC is important for monitoring the fertility and the chemical status of soils.Conventional methods of sampling and analyzing soil moisture or EC (such as gravimetric sampling, neutron probes, and gypsum blocks) can be destructive, laborintensive, time-consuming, and provide point information (Calamita et al. 2015;Badewa et al. 2018).Alternatively, geophysical methods, such as electromagnetic induction (EMI) and time-domain reflectometry (TDR) can be used to characterize soil EC or soil moisture variability more rapidly, noninvasively, and less laboriously (Altdroff et al. 2018;Badewa et al. 2018;Sadatcharam 2019).EMI and TDR employ electromagnetic waves to measure apparent electrical conductivity (EC a ) and dielectric permittivity, respectively.EC a is defined as the depth weighted average of the bulk soil EC within a subsurface soil volume and is considered a proxy of soil properties, including soil moisture (Badewa et al. 2018;Altdorff et al. 2020).Geophysical techniques present an expeditious way for monitoring the spatio-temporal variability of soil moisture.
EMI sensors operate by inducing an alternating current (EM waves in kHz) within the soil via the primary electromagnetic field generated from the transmitting coil and measuring the resultant secondary field via the receiving coil (Von Hebel et al. 2019).The amplitude, phase differences, and intercoil spacing between the primary and resultant fields are then used to determine an "apparent" value for soil EC (Simon et al. 2020).There are two types of EMI sensors that are commercially available, multicoil (MC) and multifrequency (MF) EMI sensors.The MC sensor operates at a fixed frequency and has multiple coil separations (Altdorff et al. 2020).The depth of investigation for this type of sensor depends on the coil separation; thus, the higher the intercoil spacing between the transmitter and receiver coils, the higher the depth of investigation (Deidda et al. 2022;Sadatcharam et al. 2020).The bucking coil cuts off the primary influence at the receiver coil (Calamita et al. 2015).The MF-EMI instruments with slingram geometry (dipole-dipole) can be used for conventional soil characterization (Lück et al. 2022) as well as simultaneous mapping of soil EC a and magnetic susceptibility (Sadatcharam et al. 2020).Altdorff et al. (2020) compared the performance of MF and MC EMI sensors in shallow podzols and concluded superior performance from the MF sensor.Similarly, Sadatcharam (2019) found the MF-EMI sensor to be more reliable for EC a variability in wet soils than in dry soils.
Over the years, geostatistical techniques have been used as interpolation tools to effectively estimate and provide unbiased predictions at unsampled locations (Srivastava et al. 2019).Kriging and cokriging are widely used interpolation techniques (Belkhiri et al. 2020); the former is used when a variable displays spatial dependence, while the latter is employed when other properties have been densely sampled in comparison to the target variable (Rostami et al. 2020).The extensively sampled variable (covariate) is usually measured more cheaply and quickly than the target variable.EC a measured using EMI sensors can be used as a covariate for mapping of soil moisture (Tarr et al. 2005;Naimi et al. 2021).Spatial variability in soil moisture can be predicted by ordinary kriging from the limited soil samples and by cokriging when a strong statistically significant relationship exists between soil moisture and an easily measured auxiliary variable.Previous studies suggest superior prediction accuracy of cokriging compared to ordinary kriging (Belkhiri et al. 2020).Altdorff et al. (2018) used ordinary kriging with EC a data to investigate the performance of EMI sensors in forest ecosystems and reported strong similarities between spatial EC a and soil moisture patterns.Similarly, Atwell and Wuddivira (2019) reported variable effectiveness of EC a data from EMI sensors for characterizing soil moisture in forests, agricultural land, grasslands, quarries, and residential areas.Lastly, Carrière et al. (2021) reported a stronger relationship between EC a and estimated soil depth in forest systems compared to managed systems (cedar plantation).
Converted lands for agricultural purposes usually combine varying agronomic practices, which tend to significantly affect the EC a -soil moisture relation making it necessary to have an accurate site-specific soil moisture prediction.This research aims to apply an MF-EMI sensor to map soil moisture variability under natural forests and managed agricultural land in western Newfoundland.The GEM-2 MF sensor employed in this study produces sensitive and thermally stable measurements (Briggs et al. 2019).Specifically, this work seeks to (i) compare the different frequencies in the 2.8-80 kHz range for shallow mapping of volumetric soil moisture under the different land use conditions, (ii) evaluate the relationship between EC a measured using MF-EMI and soil moisture, and (iii) assess the effectiveness of EC a as an auxiliary variable to predict soil moisture variations under different land use conditions.Although previous studies using EMI techniques to characterize soil moisture have been done in agricultural land in Newfoundland (Altdorff et al. 2018(Altdorff et al. , 2020;;Badewa et al. 2018;Sadatcharam et al. 2020), no study has investigated the use of EMI sensors under different land use types concurrently.It is worthwhile to test the technique under different conditions to assess its usefulness to support various academic, industrial, and government projects being carried out in the province and other subregions.Also, the use of EC a as a covariate in cokriging is yet to be done in the province.To our knowledge, very few studies have used geostatistical techniques with geophysical data as covariates for predicting soil moisture.

Study area
This study was conducted on three different land use conditions: an agricultural field, a recently cleared natural forest, and a field road at the Western Agriculture Center and Research Station, Pasadena (49.0130 • N, 57.5894 • W), NL, Canada (Fig. 1).The study site is managed by the Department of Fisheries, Forestry and Agriculture, Government of Newfoundland, Canada.The soil in the study site is generally classified as a reddish-brown-to-brown Podzol developed on a gravelly sandy fluvial deposit with >100 cm depth to bedrock and a 2%-5% slope (Croquet 2016).The area of the agricultural land was 924 m 2 and included a rotation of corn, canola, faba bean, wheat, and oats/peas.The field road was adjacent to the agricultural land and was 240 m 2 in area, serving as access for equipment, vehicles, and people to other parts of the field.A 50 m 2 area was selected from a recently cleared natural forest to compare its EC a and soil moisture data with other land use conditions of interest (agricultural land and field road) (Fig. 1).Before the commencement of EC a and soil moisture measurements, soil texture, soil bulk density, and soil organic matter for different land uses were determined from ground truthing (Table 1).Based on 3-month data (1 September-30 November 2021) from the nearby Deer Lake weather station A (http: //climate.weather.gc.ca), the area received a total rainfall of 409 mm and had an average mean temperature of 8.18 • C (Fig. 2).EMI survey EC a data were recorded from the three land use conditions: agricultural land, field road, and natural forest using EMI surveys.For each land use condition, a GEM-2 MF-EMI sensor (Geophex, Ltd., Raleigh, NC, USA) was used to simultaneously collect EC a data at four different frequencies (2.8, 18.3, 38.3, and 80.0 kHz).The selected frequencies are suitable for shallow subsurface investigations (Won et al. 1996).During each survey, a 1 m line spacing was maintained.A global positioning system was attached to the EMI sensor to enable the production of georeferenced maps.Before each survey, the EMI sensor was warmed up for at least 30 min to prevent data drift and ensure high-quality data as proposed by Robinson et al. (2004).
During each survey, the MF-EMI sensor was positioned such that the transmitter coil was always ahead of the receiver coil and was carried with the supplied shoulder strap at an average height of about 1 m above the ground.Both the vertical coplanar (VCP) and horizontal coplanar (HCP) coil orientation modes were used when collecting data on agricultural land, field road, and natural forest as done by previous researchers in the same area (Altdorff et al. 2018;Badewa et al. 2018;Sadatcharam et al. 2020).The MF surveys were carried out in a bidirectional order over the three land use conditions.
Soil temperature was measured at 0-20 cm depth for all three land use conditions using a soil temperature probe.Among the several models suggested for the correction of the temperature effect on the mobility of dissolved ions, the "corrected Sheets and Hendrickx model" was adopted in this study as displayed in eq. 1 (Sheets and Hendrickx 1995).According to Ma et al. (2011), this model was adopted because it performs better in most situations and for a wide range of temperatures.
EC t is the EC a data collected at measured soil temperature, t ( • C), and EC 25 is the temperature corrected EC a .Negative values were considered noise and subsequently eliminated.
The surveys were carried out during a wet period (20 October and 8 November 2021; Fig. 2) based on the total rainfall data available at the nearby Deer Lake weather station A (http://climate.weather.gc.ca).The surveys were carried out in the wet period because MF-EMI is more reliable for EC a variability in wet soils than in dry soils (Sadatcharam 2019).
Soil moisture content data recording and time-domain calibration Soil moisture was measured at nine sampling points at 0-20 cm depth in the agricultural land, field road, and natural forest volumetrically using a hand-held TDR (FieldScout TDR 350, Spectrum Technologies, Aurora, IL, USA) with a probe length of 20 cm and gravimetrically via oven drying.Soil moisture data measured gravimetrically were converted to volumetric soil moisture using the bulk density and correlated with the TDR measured volumetric soil moisture using Lin's concordance correlation coefficient (LCCC) and root mean square error (RMSE) performance criteria to obtain the field scale accuracy of the TDR measured data.Lin's Concordance Correlation Coefficient prediction matrix measures the accuracy of the prediction along a 1:1 line to evaluate the agreement between paired readings (Lawrence and Lin 1989) and is of the form where x * and y * are sample means for populations X and Y, and xi and yi are paired ith values from populations X and Y.The value of the index is scaled between −1 and 1, with a value of 1 representing complete agreement between all paired sites.
The 0-20 cm depth was considered because it is an active interface between the soil, vegetation, atmosphere, and human activities mostly affected by precipitation, infiltration, and canopy cover (Choi and Jacobs 2007).

Descriptive statistics
Descriptive statistics (boxplot and coefficient of variation (CV)) were conducted to evaluate the variability of EC a and soil moisture data.Across all the study sites, the relationship between EC a and soil moisture was not statistically significant (p > 0.05) for 2.85, 18.33, and 80.01 kHz frequencies in both HCP and VCP modes, hence was not further analyzed.However, the relationships were significant (p < 0.05) for the soil EC a measured with a frequency of 38.31 kHz in HCP mode across all study sites on both survey days, thus only the 38.31 kHz in HCP mode was considered for further analysis.A similar result was seen by Altdorff et al. (2018), Badewa et al. (2018), andSadatcharam et al. (2020) on a nearby field at the Western Agriculture Center and Research Station.Kolmogorov-Smirnov frequency distribution analysis revealed that all soil properties investigated in this study were normally distributed on both survey days.

Analysis of variance
One-way Analysis of variance (ANOVA) was used to examine the differences in soil EC a and soil moisture across the different land use conditions.The normality of residuals was assessed and least square means were compared using the Fisher's LSD test at 5% significance.

Correlation and regression
The strength of the relationships between EC a and soil moisture for each land use was assessed using Pearson's correlation analysis.A strong relationship between the auxiliary variable and target variable has been seen as an efficacy of cokriging (Tarr et al. 2005).Linear regression was used to characterize the relationship between soil moisture and EC a and to generate predictive models of soil moisture under each land use using EC a from the first survey day.The models were validated by fitting them to the EC a data from the second sampling day to predict soil moisture for the second sampling day.The model performances were evaluated using the level of agreement (accuracy) between the predicted and measured soil moisture for the second day using LCCC prediction matrix and RMSE.All statistical analyses were performed with Minitab 17.

Geostatistical analysis
Ordinary kriging was used to prepare interpolated maps of EC a .The spatial prediction of the unmeasured points was given by the line sum of observed values (eq.3).
where Z * (x o ) is the predicted value at the unmeasured position x o , Z(xi) is the measured value at position x i , λ i is the weighting coefficient from the measured position to x o , and n is the number of positions within the neighborhood searching.
A total of 4202, 1303, and 1062 EC a data were collected (observed values) from the agricultural land, field road, and natural forest, respectively.Soil moisture was less sampled (nine data points) compared to EC a in this study, hence in addition to ordinary kriging, cokriging was used for soil moisture mapping.Cokriging is the multivariate equivalent of ordinary kriging.The main difference is that cokriging has a secondary variable (covariate) (eq.4).
where u and v are the target and covariate variables, respectively.The two variates u and v are cross-correlated, and the covariate contributes to the estimation of the target variate.The EC a data collected on the agricultural land, field road, and natural forest, respectively were used as covariates for soil moisture predictions with cokriging.To assess the effectiveness of EC a as an auxiliary variable for improving soil moisture mapping predictions, ordinary kriging was compared to cokriging.
Several variogram models (linear, exponential, circular, Gaussian, spherical, and power model) were considered for creating EC a and soil moisture maps using ordinary kriging or cokriging.The variogram models with the lowest RMSE based on the cross-validation results were selected (Sówka et al. 2020;Zhou et al. 2020).Cross-validation works by singly removing each point in the sampling scheme and predicting its value based on kriging the remaining data.All variograms were assumed to be isotropic.The ordinary kriging and cokriging under each land use were compared using Surfer 24.Interpolated maps were then created using Surfer 24.

Results and discussion
The development of policies in recent times to maintain water resources requires evidence-based information on the spatial distribution of soil moisture under different land uses (Gebrehiwot et al. 2021).This approach to measuring soil moisture in different land use conditions has the potential to improve land unit mapping, agricultural planning, and afforestation activities (Stavi 2019).

Calibration of TDR
There was a significant strong positive correlation (LCCC = 0.9) between the measured volumetric moisture content with TDR and the calculated volumetric moisture content using the gravimetric analysis and bulk density for 0-20 cm depth with a low RMSE (1.09%) as seen in Fig. 3.This result is similar to that is reported by Topp et al. (1980) and Badewa et al. (2018).

Descriptive statistics
Several factors, including but not limited to land use, can influence the variation in soil properties (Atwell and Wuddivira 2019).All other factors, besides land use, were kept constant during soil data collection, therefore the variations in soil EC a and soil moisture were used as a measure of sensitivity to land use.The CV of the soil EC a and soil moisture within land uses was used as a reflection of the spatial sensitivity of EC a and soil moisture.According to the classification of Warrick (1998), the CV of EC a for agricultural land and field road was moderate (15% < CV < 35%), while that of the natural forest was low (CV <15%) for the first survey day (Fig. 4).For the second survey day (Fig. 4), the CVs for EC a for the three different land use conditions were low based on the classification of Warrick (1998).The agricultural land had the highest CV in EC a among the three land use conditions, indicating this land use was most sensitive to EC a changes.This was primarily attributed to fertilizer inputs in this land use as discussed in Atwell and Wuddivira (2019).The CV for the soil moisture was low for all three land use conditions for both survey days (CV <15%) (Fig. 5).Natural forest showed the lowest CV in soil moisture, indicating a more stable spatial pattern in the soil moisture.Agricultural land and field road showed a higher CV in soil moisture, indicating that factors, such as human disturbance, can increase the spatial heterogeneity of soil moisture as also discussed in Guo et al. (2020).The range of EC a was most extensive in the natural forest relative to the agricultural land and field road for both survey days.The higher EC a range may be ascribed to higher soil organic matter found in the natural forest relative to other land use conditions.This result is similar to the observation by Atwell and Wuddivira (2019).The range of EC a values in the agricultural land was greater than that in the field road, and this may be due to fertilizer application, which potentially leads to increased soil ionization (Kaufmann et al. 2020).

ANOVA
The mean EC a for the natural forest was higher than agricultural land and field road for the first survey day (20 October 2021) and for the second survey day (8 November 2021) (Table 2).Generally, soils with high clay contents possess continuous moisture-filled pores that easily conduct electricity relative to sandy soils (Rhoades et al. 1989).Natural forest had a higher clay and soil organic matter content relative to other land use conditions as shown in Table 1, leading to a higher water holding capacity (Atwell and Wuddivira 2019).The lowest EC a mean was recorded in the agricultural land.This was related to the fact that EC a surveys on agricultural land were carried out during the harvesting period, hence the crops on this land could have utilized most of the applied fertilizer leading to low EC a readings.The frequent use of the field road as an access route could have increased Note: Means that do not share a common letter are significantly different according to Fisher's least significant difference test.
Fig. 6.The site-specific relationship between EC a and volumetric soil moisture with Pearson's correlation coefficient (r) and coefficients of determination (R 2 ) under each land use.
the soils compaction, hence leading to a relatively higher EC a than in the agricultural land (Machado et al. 2015).
The soil moisture mean values were higher on the first survey day than on the second survey day (Table 2), which potentially could be due to a higher total rainfall (mm) on the first survey day (84.2 mm) relative to the second survey day (19.3 mm) as seen in Fig. 2. The total rainfall was calculated for 10 days before each survey date.The mean soil moisture values were higher in the natural forest relative to the agricultural land and field road (Table 2).This was ascribed to the deep litter layer (relatively high soil organic matter) found in this land use.The deep litter can potentially reduce surface evaporation and can improve vertical infiltration effectively by preventing rainfall splashing as well as surface runoff.By improving vertical infiltration, the presence of the litter facilitates the formation of an aerobic environment that accelerates the decomposition of root systems and improves soil porosity by providing a network of continuous root channels.This finding agrees with that of Morris (2004).The availability of high soil organic matter in coarse-textured soils, such as podzolic soils, also tends to expand the volume of soil pores that retain water against gravitational drainage, increasing the water holding capacity in the natural forest as observed by Morris (2004).The low soil moisture in agricultural land and field road was ascribed to the higher rate of evaporation that existed in the surface horizon of the agricultural land and field road than on the natural forest due to factors, such as ploughing, which reduces soil water evaporation by creating a capillary barrier at the surface horizon (Gao et al. 2011).In addition, the agricultural land and field road had relatively higher surface runoff due to surface crusting and compaction, causing them to have a lower infiltration than in the natural forest.
One-way ANOVAs revealed that EC a data were not significantly different between the agricultural land and field road, possibly due to the proximity and similar soil characteristics between these land use conditions; however, EC a in natural land was significantly different from agricultural land and field road as seen in Table 2.In contrast, soil moisture was found to be significantly different among the three land use conditions (p < 0.005) for both survey days as shown in Table 2.These statistics proved that soil EC a and soil moisture differences existed between the land use conditions as indicated by Xu et al. (2021) and Gao et al. (2011).Although soil EC a is dynamic, the results indicate that soil moisture is more sensitive to land use conditions relative to soil EC a in this study area.In the Aripo savannas, soil moisture has also been seen to be more sensitive to land use conditions relative to soil EC a (Atwell and Wuddivira 2019).

Pearson's correlation analysis
There were significant positive correlations between EC a and soil moisture for all three land use conditions (Fig. 6).There was a strong relationship between EC a and soil moisture in the field road and the natural forest, while the agricultural land showed a relatively weaker relationship between EC a and soil moisture on both survey days as shown in Fig. 6.The correlation results show that EC a values in the study area increase with water content and potential ions retained in soil solution.The positive correlation between EC a and soil moisture corroborates with previous studies (Tang et al. 2020;Wang et al. 2020) in agricultural lands in other subregions.

Regression analysis
Regression analysis showed EC a explained more than 70% of the variation in soil moisture in the field road and the natural forest.On the other hand, EC a explained relatively lower (59%) variation in soil moisture in the agricultural land (Fig. 6).A possible explanation of the lower R 2 in the agricultural land could be because the agricultural land is a managed ecosystem with the presence of fertilizer resulting in an increase in the concentration of dissolved ions, and consequently, in the pore water electrical conductivity (EC w ) contributing to a higher variability in EC a (Wuddivira et al. 2012;Hobley et al. 2020).Also, due to heavy compaction and shallow water depth in some areas of the agricultural land, it resulted in poor drainage in certain zones of the field.These factors could have affected EC a -soil moisture relation (Archie 1942;Corwin and Lesch 2005;Altdorff et al. 2018).The regression results across the three land use conditions on both survey days indicate that EC a is primarily controlled by soil moisture for these study sites.
Validation of the generated regression models using LCCC and RMSE obtained by comparing the predicted volumetric soil moisture using EC a data with linear regression models to the measured volumetric soil moisture with TDR revealed that predictions of soil moisture from site-specific linear regression models exhibited much certainty on the land use conditions with higher soil moisture (natural forest) compared to the land uses with lower soil moisture (agricultural land and field road) as seen in Fig. 7.The fitted linear regression developed between EC a and soil moisture in the natural forest (LCCC = 0.7, RMSE = 3.87%) provided the highest prediction accuracy relative to other land use conditions.These findings indicate that moist soils are more favorable for EC a surveys, as also suggested by Brevik et al. (2006) and Sadatcharam et al. (2020).Although several regression models were developed to predict soil moisture during data processing, the site-specific models produced the most accurate soil moisture predictions as seen by Drummond et al. (2003) and Altdorff et al. (2018).
Systematic deviations in soil moisture (Fig. 7) have been reported by Bogena et al. (2007) to be magnified in soils with lower EC a (∼0.06 dS m −1 ).These EC a values reported by Bogena et al. (2007) are similar to those recorded in the agricultural land and field road in our study (Table 2).A further explanation for the underestimation and overestimation is the disparity of sampling depths between TDR and EMI sensor (Calamita et al. 2015;Altdorff et al. 2018).TDR probes were installed vertically at 0-20 cm depth, and the EMI sensors have an effective integral depth of 0-250 cm.Another factor could be the effect of soil temperature on the sensor's electronics as also mentioned by Bogena et al. (2007).However, the effect of temperature on the instrument, and therefore measured EC a of the soil in our study area has not yet been studied, and its influence on the results of the experiment is beyond the scope of this paper.

Interpolated maps
Visual differences generally existed between the maps of soil moisture obtained from kriging and those from cokriging.The maps obtained from cokriging were less smooth and showed more local detail in their representation of the soil moisture variability.However, the trends in variability of soil moisture in both methods were similar.These findings agree with that of Tarr et al. (2005).The parameters used in fitting the variograms for each interpolated map are displayed in Table 3.The various variograms are displayed in Appendix Figs.A1-A6.

Agricultural land
The interpolated spatial maps for EC a (38.3 kHz) for the agricultural land for the two survey dates are displayed in Figs.8a and 9a.EC a values for this land use ranged from 0.05 to 0.2 dS m −1 .In this land use, spherical (Fig. A1a) and Gaussian models (Fig. A2a) displayed the best-fitting models for the spatio-temporal representation of EC a in ordinary kriging for the first and second day, respectively.The variograms Table 3. Geostatistical parameters for ordinary kriging and cokriging analysis and cross-validation results (root mean square error) of apparent conductivity and soil moisture content for 20 October and 8 November 2021 survey.provided a clear insight of EC a spatial structure of both sampling days.Positive nugget values were found for the EC a variograms on both survey days (Table 3).This could be due to the variation in EC a associated with short-range variability of soil properties, such as soil moisture content, ionic composition, and topography (Moral et al. 2010;Narjary et al. 2021).The lower range (3 m) on the second day compared to the first day (19.2 m) (Table 3) suggests that the spatial autocorrelation of EC a increased on the second day (Liu et al. 2017).The increase in RMSE between the first day (RMSE = 0.05 dS m −1 ) and sec-  ond day (RMSE = 0.09 dS m −1 ) obtained from cross-validation (Table 3) indicates a reduction in prediction accuracy, possibly due to an increase in spatial heterogeneity of EC a as displayed in the interpolated map (Fig. 9a).The nugget:sill ratio was used to classify the spatial dependence of EC a .Ratio values lower than 25% and higher than 75% corresponded to strong and weak spatial dependence, respectively, while the ratio values between 25% and 75% corresponded to moderate spatial dependence (Chang et al. 1998).EC a in the agricultural land use exhibited a strong spatial dependence on the first day and a moderate spatial dependence on the second day (Table 3).This land use, generally, had low EC a scattered across the map with the highest EC a region in the agricultural land observed on the lower left region of the map for the first survey day.The interpolated EC a maps for the second survey date, generally, revealed lower EC a values relative to the first survey date, which is consistent with the trend of the measured EC a data.
The spherical variogram model was the best fit for the spatio-temporal representation of soil moisture in both cokriging (Figs.A3a and A4a) and ordinary kriging (Figs.A5a  and A6a) for both survey days.The presence of nugget effect in the ordinary kriged soil moisture maps could be attributed to measurement errors (Kathuria et al. 2019).The maps obtained from cokriging (Figs.10a and 12a) revealed less smoothness in their depiction of soil moisture variation compared to ordinary kriged maps (Figs.11a and 13a).The range values varied slightly between both interpolation techniques on both survey days, indicating a slight variation in the spatial autocorrelation of soil moisture (Table 3).
The land use was generally dry with the highest soil moisture observed as a small pocket on the northern section of the map.The accuracies of the soil moisture map obtained from cross-validation were higher with cokriging (RMSE = 0.45% and 0.47% for day 1 and day 2, respectively) relative to ordinary kriging prediction (RMSE = 0.62% and 1.72% for day 1 and day 2, respectively) (Table 3) revealing the effectiveness of using EC a as a covariate in creating more accurate soil moisture maps relative to ordinary kriging in this land use.

Field road
The interpolated spatial maps of EC a for the field road ranged between 0.05 and 0.2 dS m −1 for the two survey dates (Figs. 8b and 9b).In this land use, spherical models produced the best fit for the spatio-temporal representation of EC a in ordinary kriging for both days (Figs.A1b and A2b).The model had a nugget effect of 0.2 and a range of 12 m on both survey days (Table 3).The lower nugget value on this land use compared to the agricultural land could be attributed to the local scale decrease in EC a variation in the  field road (Narjary et al. 2021).The RMSE obtained from cross-validation did not change between the first and second day (RMSE = 0.04 dS m −1 ) (Table 3).The land use exhibited a strong spatial dependence on both survey days (nugget:sill = 8%) (Table 3).The field road generally had low EC a scattered across the map with the highest EC a found in the upper region of the map on both days (Figs.8b and 9b).These interpolated EC a maps for the second survey date generally revealed lower EC a values relative to the first survey date.A similar pattern was observed in the agricultural land as described above.
For soil moisture, the spherical variogram model presented the best fit in both cokriging (Figs.A3b and A4b) and ordinary kriging (Figs.A5b and A6b) for the first and second survey days.There were no nugget effects in both approaches (Table 3).Maps obtained from cokriging (Figs.10b and 12b) displayed less smoothness in their depiction of soil moisture variation compared to maps from ordinary kriging (Figs.11b and 13b).The ranges in soil moisture values obtained from both interpolation techniques did not differ between the two survey days (2 m) (Table 3).This suggests that the spatial autocorrelation of soil moisture was not significantly affected by either interpolation technique or time.The accuracies of the soil moisture map obtained from cross-validation were higher in the cokriging prediction relative to ordinary kriging prediction for the first and second survey days (RMSE = 0.03% and 0.45% versus RMSE = 0.49% and 0.82%, respectively) (Table 3).This revealed the effectiveness of using EC a as a covariate in creating more accurate soil moisture maps relative to ordinary kriging in this land use.This result corroborates findings from Tarr et al. (2005), who reported superior effectiveness of cokriging relative to ordinary kriging for soil properties, such as soil moisture and organic matter.

Natural forest
The power variogram model (Figs.A1c and A2c) displayed the best fit for the spatial representation of EC a in ordinary kriging for the two survey days (Figs.8c and 9c).The variogram parameters did not change between survey days in this land use (Table 3).The model had a nugget effect of 0.2 and a range of 1 m in this land use (Table 3), suggesting the presence of spatial autocorrelation, possibly due measurement errors when collecting EC a measurements (Haining 2009;Liu et al. 2017).The land use exhibited moderate spatial dependence on both survey days (nugget:sill = 40%) (Table 3).The RMSE obtained from cross-validation was lower on the first day (RMSE = 0.14 dS m −1 ) compared to the second day (RMSE = 0.24 dS m −1 ) (Table 3), indicating a reduction in prediction accuracy.This was possibly due to the higher soil moisture content on the first survey day relative to the second survey day (Table 2).Prediction accuracy of EC a is expected to increase with soil moisture (Sadatcharam et al. 2020).The natural forest generally had the highest EC a readings compared to other land use conditions.This is consistent with the higher soil organic matter in the natural forest compared to other land use types.Soil organic matter improves nutrient retention, cation exchange capacity (Wulanningtyas et al. 2021), moisture retention (Chalise et al. 2019), and EC (Pouladi et al. 2019).Similar to the agricultural land and field road, the interpolated EC a maps for the second survey date showed lower EC a values than those for the first survey date.
For the first survey day, the spherical model produced the best fit for the spatial representation of soil moisture in cokriging (Fig. A3c) and ordinary kriging (Fig. A5c).On the other hand, the Gaussian model was the best-fitting model for the cokriging (Fig. A4c) and ordinary kriging (Fig. A6c) approaches for the second day.There were no nugget effects in both interpolation methods (Table 3).This could mean there were minimal errors associated with data collection (Lui et al. 2017).The maps obtained from cokriging (Figs.10c and 12c) revealed less smoothness and more local detail in their depiction of soil moisture variation compared to the ordinary kriged maps (Figs.11c and 13c).This is consistent with what was observed in other land uses.The range values did not change (2 m) between interpolation methods on the first day (Table 3).A similar finding was seen on the second day (2.5 m), indicating that the spatial autocorrelation of soil moisture did not vary with interpolation technique but varied with time.This land use was wetter compared to other land use conditions.The accuracies of the soil moisture map obtained from cross-validation were higher in the cokriging prediction (RMSE = 0.02% and 0.18% for day 1 and day 2, respectively) than in the ordinary kriging prediction (RMSE = 0.13% and 0.25% for day 1 and day 2, respectively) (Table 3), revealing the effectiveness of using EC a as a covariate in creating more accurate soil moisture maps relative to ordinary kriging in this land use.

Conclusion
This study demonstrates that proximal surveys of EC a using GEM-2 could be a helpful surrogate for assessing intrafield variability of soil moisture.This was achieved by analyzing the relationships between EC a and soil moisture measurements from three different land use conditions (agricultural land, recently cleared natural forest, and field road) on a boreal podzolic soil.The strong influence of soil moisture on EC a under different land use conditions from the generated linear regression was indicative that soil moisture was a major driver of EC a in the study area.Mapping of soil moisture using cokriging with EC a as a covariate produced more local detail than maps produced using ordinary kriging.There were improvements in the prediction accuracies of soil moisture maps when the cokriging technique was applied compared to the ordinary kriging.This suggests that EC a obtained using EMI has the potential as a robust auxiliary variable for accurately predicting soil moisture in boreal podzolic soils.The best prediction was found in the natural forest, the land use type that had the strongest correlation between soil moisture and EC a .These results suggest that cokriging of soil moisture with densely sampled EC a as covariates improves the characterization accuracy of soil moisture variability in the study area.This study reveals the effectiveness of the georeferenced MF-EMI technique to rapidly assess intrafield variability under different land uses.Such surveys may rapidly map the spatial variability of intrafield soil moisture under different land use conditions.It is recommended that more studies should be carried out on other subregions for further validation.

Fig. 1 .
Fig. 1.The location of the different land use conditions in Western Agriculture Center and Research Station, Pasadena, NL, Canada (49.0130 • N, 57.5894 • W).

Fig. 2 .
Fig. 2. Daily total rainfall and mean temperature from August to November 2021 for study area from Deer Lake weather station A.

Fig. 3 .
Fig.3.Relationship between the measured volumetric moisture content with the TDR and the calculated volumetric moisture content using the gravimetric analysis and bulk density for 0-20 cm depth.

Fig. 4 .
Fig. 4. Spatial variability of apparent electrical conductivity (EC a ) data ranges by box and whisker plots ( * --outlier value) with coefficient of variation for the different land use conditions on two survey days.

Fig. 5 .
Fig. 5. Spatial variability of soil moisture data ranges by box and whisker plots ( * --outlier value) with coefficient of variation for the different land use conditions on two survey days.

Fig. 7 .
Fig. 7. (a) Relationship between predicted volumetric soil moisture from linear regression models and measured volumetric soil moisture with TDR under agricultural land, (b) field road, and (c) natural forest with Lin's concordance correlation coefficient and root mean square error of predictions.

Fig. 8 .
Fig. 8. (a) Spatial variability maps of apparent electrical conductivity (EC a ) for agricultural land, (b) field road, and (c) natural forest obtained from ordinary kriging for first survey day.

Fig. 9 .
Fig. 9. (a) Spatial variability maps of apparent electrical conductivity (EC a ) for agricultural land, (b) field road, and (c) natural forest obtained from ordinary kriging for second survey day.

Fig. 10 .
Fig. 10.(a) Spatial variability maps of soil moisture content for agricultural land, (b) field road, and (c) natural forest obtained from cokriging for first survey day with apparent electrical conductivity (EC a ) as covariate.

Fig. 11 .
Fig. 11.(a) Spatial variability maps of soil moisture content for agricultural land, (b) field road, and (c) natural forest obtained from ordinary kriging for first survey day.

Fig. 12 .
Fig. 12.(a) Spatial variability maps of soil moisture content for agricultural land, (b) field road, and (c) natural forest obtained from cokriging for second survey day with apparent electrical conductivity (EC a ) as covariate.

Fig. 13 .
Fig. 13.(a) Spatial variability maps of soil moisture content for agricultural land, (b) field road, and (c) natural forest obtained from ordinary kriging for second survey day.

Fig. A1 .
Fig. A1.(a) Variogram models of apparent electrical conductivity (EC a ) obtained from ordinary kriging for agricultural land, (b) field road, and (c) natural forest for first survey day.

Fig. A2 .
Fig. A2.(a) Variogram models of apparent electrical conductivity (EC a ) obtained from ordinary kriging for agricultural land, (b) field road, and (c) natural forest for second survey day.

Fig. A3 .
Fig. A3.(a) Variogram models of soil moisture obtained from cokriging for agricultural land, (b) field road, and (c) natural forest for first survey day.

Fig. A4 .
Fig. A4.(a) Variogram models of soil moisture obtained from cokriging for agricultural land, (b) field road, and (c) natural forest for second survey day.

Fig. A5 .
Fig. A5.(a) Variogram models of soil moisture obtained from ordinary kriging for agricultural land, (b) field road, and (c) natural forest for first survey day.

Fig. A6 .
Fig. A6.(a) Variogram models of soil moisture obtained from ordinary kriging for agricultural land, (b) field road, and (c) natural forest for second survey day.

Table 1 .
Basic soil properties of agricultural land, natural forest, and field road obtained from ground truthing and laboratory analysis (n = 9).

Table 2 .
ANOVA of apparent electrical conductivity (EC a ) and soil moisture content measurement between agricultural land, natural forest, and the field road at 95% confidence.