Using Apparent Electrical Conductivity to Delineate Field Variation in an Agroforestry System in the Ozark Highlands

Abstract

Greater adoption and better management of spatially complex, conservation systems such as agroforestry (AF) are dependent on determining methods suitable for delineating in-field variability. However, no work has been conducted using repeated electromagnetic induction (EMI) or apparent electrical conductivity (EC_a) surveys in AF systems within the Ozark Highlands of northwest Arkansas. As a result, objectives were to (i) evaluate spatiotemporal EC_a variability; (ii) identify EC_a-derived soil management zones (SMZs); (iii) establish correlations among EC_a survey data and in situ, soil-sensor volumetric water content, sentential site soil-sample EC, and gravimetric water content and pH; and (iv) determine the optimum frequency at which EC_a surveys could be conducted to capture temporal changes in field variability. Monthly EC_a surveys were conducted between August 2020 and July 2021 at a 4.25 ha AF site in Fayetteville, Arkansas. The overall mean perpendicular geometry (PRP) and horizontal coplanar geometry (HCP) EC_a ranged from 1.8 to 18.0 and 3.1 to 25.8 mS m⁻¹, respectively, and the overall mean HCP EC_a was 67% greater than the mean PRP EC_a. The largest measured EC_a values occurred within the local drainage way or areas of potential groundwater movement, and the smallest measured EC_a values occurred within areas with decreased effective soil depth and increased coarse fragments. The PRP and HCP mean EC_a, standard deviation (SD), and coefficient of variation (CV) were unaffected (p > 0.05) by either the weather or growing/non-growing season. K-means clustering delineated three precision SMZs that were reflective of areas with similar EC_a and EC_a variability. Results from this study provided valuable information regarding the application of EC_a surveys to quantify small-scale changes in soil properties and delineate SMZs in highly variable AF systems.

1. Introduction

Electromagnetic-induction (EMI)-based methods have become a common technique for characterizing site-specific soil property variations. Electromagnetic induction-based methods characterize soil property variability through proximally sensing soil apparent electrical conductivity (EC_a), which is the ability of soil to conduct an electrical current. Field-measured EC_a is the result from complex interactions between soil salinity, base saturation (BS), bulk density (BD), clay content and mineralogy, soil water content (SWC), soil organic matter (SOM), cation exchange capacity (CEC), and soil temperature [1,2,3]. Electromagnetic-induction measurements are non-invasive, mobile, reliable, and relatively inexpensive. Electromagnetic-induction-based methods can collect EC_a measurements at very dense spatial resolutions to accurately delineate spatial changes in below-ground properties [4,5]. With proper ground-truthing, the soil characteristics that can be derived from EC_a surveys include salinity, nutrients, SWC, texture, depth to sand layers or claypans, BD, and many indirect properties and processes (i.e., SOM, CEC, leaching, groundwater recharge, and soil drainage class, among others) [4]. Applications of EC_a measurements largely include precision agriculture [6,7], where the collected data can be used to optimize soil sampling strategies, delineate soil management zones (SMZs), and predict yield variability [7].

Although there has been significant research conducted applying EC_a surveys to assess field variability in different types of land management systems and ecosystems [5], little work has been conducted using EC_a surveys to investigate field variability within agroforestry (AF) systems. Agroforestry is a conservation-oriented, land management practice that is defined as the intentional integration of agriculture and forestry to benefit from the subsequent interactive effects that are produced from growing crops and/or livestock alongside trees and/or shrubs [8]. The benefits of AF practices include soil quality enhancements, opportunities for economic diversity for the producer, and additional ecosystem services [9] such as reduced soil erosion (wind and water), water quality enhancement, increased biodiversity, elevated aesthetic value, carbon (C) sequestration, oxygen (O) production, and mitigation of greenhouse gases [10].

Because AF systems are combinations of agricultural production systems, AF systems are complex and can have more intricate management operations. The combination of trees, shrubs, pastures, crops, and/or livestock results in AF systems having increased soil property variability within smaller spatial distances, making soil property variability more difficult to characterize. Thus, the non-invasive, mobile, and measurement capability of EMI-based methods makes the application of EC_a surveys to delineate soil property variations to be potentially well suited for AF systems. However, EC_a surveys have not been used in AF systems to rapidly evaluate in-field variability and identify the spatiotemporal relationship between repeated EC_a surveys and soil property variations in AF systems, which are needed to better inform management in the Ozark Highlands.

The Ozark Highlands is characterized by a wetter, warmer climate and gentle to steeply sloped hills with well-developed karst topography [11,12]. Many of the soils in the Ozark Highlands overlie the karst topography, exist on steep slopes, are stony (i.e., chert coarse fragments), and shallow to bedrock [11,12]. The combination of the Ozark Highlands’ features results in an increased potential for conservation issues (i.e., surface and groundwater quality degradation) [11,12], which in turn, suggests that the implementation and research of AF systems within the Ozark Highlands is particularly advantageous due to the many potential ecosystem services and benefits that AF systems offer (i.e., water quality enhancement, and reduced soil erosion and runoff). A better understanding of how AF systems affect soil properties and the suitable methods for soil variability delineation in AF systems can potentially lead to the greater adoption and better management of the conservation practice in areas prone to conservation issues.

To increase the information regarding the spatiotemporal relationship between repeated EC_a surveys and soil property variations in AF systems, specifically within regions with similar features to the Ozark Highlands, and to provide spatially resolved soil data for precision soil management, the objectives of this study were to (i) use monthly EC_a surveys to assess the spatiotemporal variation in measured EC_a within a 20-year-old AF system within the Ozark Highlands; (ii) identify clusters of similarly behaving populations for precision SMZ delineation; (iii) identify correlations between EC_a survey data and in situ, soil-sensor-measured volumetric water content (VWC) and EC_a measurements at two depths, as well as benchtop-measured soil sample EC, gravimetric water content (GWC), and soil pH in two depth intervals; and (iv) determine the optimum frequency that monthly EC_a surveys should be conducted to capture temporal changes in field variability within 1 year. It was hypothesized that there would be differences between the EC_a mean, standard deviation (SD), and coefficient of variation (CV) of surveys that were conducted in different weather seasons and in the growing/non-growing season at a 20-year AF system within the Ozark Highlands. Furthermore, it was hypothesized that monthly EC_a surveys can be grouped into similar functional populations and be made into zones for precision soil management. It was also hypothesized that monthly EC_a survey data are correlated with soil-sensor-based VWC and EC_a, and soil-sample-based EC, GWC, and pH. Additionally, it was hypothesized that fewer surveys than monthly could have been conducted in a 1-year period to capture the same amount of overall EC_a variance as the 12 monthly EC_a surveys conducted.

2. Materials and Methods

2.1. Site Description

2.1.1. Mapped Soils and Tree and Forage Establishment

This study was conducted on a 4.25 ha paddock at the University of Arkansas Milo J. Shult Agricultural Research and Extension Center (SAREC) in Fayetteville, AR (36.09°N, 94.19°W). The study site was located within the Ozark Highlands, Major Land Resource Area (MLRA) 116A [13]. The experimental area is mostly mapped as Captina silt loam (fine-silty, siliceous, active, mesic Typic Fragiudults) with some Pickwick silt loam (fine-silty, mixed, semiactive, thermic Typic Paleudults) and small areas containing Johnsburg silt loam (fine-silty, mixed, active, mesic Aquic Fragiudults), Cleora fine sandy loam (Coarse-loamy, mixed, active, thermic Fluventic Hapludolls), and Nixa cherty silt loam (loamy-skeletal, siliceous, active, mesic Glossic Fragiudults; Figure 1) [14]. The paddock also contains an inclusion that is dissimilar, lower in elevation, and consistently wetter than surrounding areas and is not captured in the soil mapping units across the site. The wetter location within the paddock was classified as fine, mixed, active, thermic Typic Endoaqualfs [15]. The elevation and slope percent of the study site ranged from 379 (south-western portions) to 387 m (north-western and -eastern portions), and from 0.0 to 9.1% (Figure 1). The study site has an annual minimum, maximum, and average air temperature of 8.7, 20.3, and 14.6 °C, respectively, and receives an annual average (30-year mean, 1981 to 2010) of 1156 mm of precipitation [16].

In 2000, sixteen rows of eastern black walnut (Juglans nigra L.), Northern red oak (Quercus rubra L.), and pecan (Carya illinoinensis Wangenh. K. Koch) were established and oriented east–west with 15 m of spacing between rows (Figure 1). The site was not ideal for growing eastern black walnut, where they grew adequately on the east end, but struggled to grow in the wet, middle portion and dry Nixa soil on the west end of the rows. As a result, the eastern black walnut trees were replaced in 2014 with rows that consisted of three different fast-growing species: American sycamore (Plantanus occidentalis L.) in the well-drained portion, cottonwood (Populus deltoides W. Bartram ex Marshall) in the poorly drained portion, and pitch/loblolly pine (Pinus rigida x Pinus taeda) in the drought-prone area of the field (Figure 1). Additionally, two forage-species treatments were established in the alleys between tree rows, including a cool-season species (orchardgrass (Dactylis glomerata L., var. Tekapo)) that was seeded in fall 2015 with 17 kg of pure live seed (PLS) ha⁻¹ and a native warm-season mix (8:1:1 big bluestem (Andropogon gerardii Vitman), little bluestem (Schizachyrium scoparium (Michx. Nash) and indiangrass (Sorghastrum nutans L.) seeded in spring 2016 at 10 kg PLS ha⁻¹ (Figure 1). A Haybuster 107C no-tillage drill (DuraTech, Jamestown, ND) was used to plant the alleys. Cornerstone^® Plus (N-[phosphonomethyl] glycine, Winfield Solutions, St. Paul, MN, USA) was applied before establishment to rid the alleys of any existing vegetation at a 2.2 kg ha⁻¹ rate (41% active ingredient (ai)), and alleys were treated with Plateau^TM (ammonium salt of imazapic) after establishment at a 0.28 kg ha⁻¹ rate (23.6% ai) to remove any remaining non-forage vegetation.

2.1.2. Fertilizer Applications

During every spring between 2001 and 2007, except for 2005, 3.9 to 6.7 Mg of poultry litter (PL) ha⁻¹ was distributed via broadcast application over the eastern half of the AF site, and 50 to 76 kg nitrogen (N) ha⁻¹—as the ammonium nitrate (NH₄NO₃) fertilizer—was broadcast-applied over the western half of the AF site [17]. In 2005, PL and NH₄NO₃ were applied at a rate of 8.9 Mg PL ha⁻¹ and 123 kg N ha⁻¹ in spring and fall, respectively [17]. Starting in June 2004, additional fertilizer was surface-applied to the surrounding ground near each tree as an annual application of Osmocote (The Scotts Miracle-Grow Co., Marysville, OH, USA), a slow-release fertilizer, that contained 5.6, 2.4, and 4.6 g of N, phosphorus (P), and potassium (K), respectively [17]. In 2005, both PL and NH₄NO₃ applications were made in the spring and fall to evaluate the impacts of nutrient source on soil physiochemical properties.

In March 2017 and 2018 and April 2019, orchardgrass and native grass treatments in the alleys between tree rows received 84 kg N ha⁻¹ (4.94 Mg ha⁻¹, fresh weight basis) of locally sourced PL (Figure 1). The chemical composition of the 2017 PL application was 2.7, 0.7, and 1.1% N, P, and K, respectively, and had a pH of 6.1. The 2018 PL had 2.0, 0.6, and 1.0% N, P, and K, respectively, and had a pH of 6.2. The 2019 PL had 2.5, 0.7, and 0.9 N, P, and K, respectively, and had a pH of 5.2. Additionally, the site was grazed by heifers (Bos taurus L.) at a density of 1.9 animal units (AU) ha⁻¹ from May to June 2017, 2.2 AU ha⁻¹ from May to July 2018, and 2.4 AU ha⁻¹ from May to July 2019 [15,18,19]. Additionally, N, in the form of urea (46-0-0), was applied at a rate of 67.3 kg N ha⁻¹ or 146 kg urea ha⁻¹ on 30 March 2020 and 31 March 2021. Urea was applied to all alleys with a fertilizer spreader attached to a 3-point hookup with a spreading width of 15.2 m.

Between 2016 and 2021, the trees were annually individually fertilized in the spring/summer with variable rates and combinations of NH₄NO₃ or a 32-0-0 fertilizer (Oakley’s Carefully Blended 32-0-0 Fertilizer, Bruce Oakley, Inc., Little Rock, AR, USA), a 13-13-13 fertilizer (Greenkeeper’s Secret 13-13-13 Premium Fertilizer, T&M, Inc., Foristell, MO), and gypsum (CaSO₄) The fertilizer combinations applied to the trees were based on fertility treatment and the tree fertilizer application rates were based on the nutrient concentrations of 2016 soil samples. Specific details on the 2016 to 2021 tree fertilizer amounts and application areas per tree species are presented on

Table 1. Summary of the fertilizer types (i.e., NH₄NO₃, 13-13-13, gypsum (CaSO₄)), amounts, and application areas of the fertilizers used to fertilize the trees (i.e., Northern red oak (Quercus rubra L.), pecan (Carya illinoinensis Wangenh. K. Koch), American sycamore (Plantanus occidentalis L.), cottonwood (Populus deltoides W. Bartram ex Marshall), and pitch/loblolly pine (Pinus rigida x Pinus taeda)) of an agroforestry site in Fayetteville, AR between 2016 to 2021.

		2016–2021 Tree Fertilizer Application Rates per Tree Species
		Fertilizer Amount *					Fertilizer Application Area
Year	Fertilizer	Oak	Pec	Syc	Cot	Pine	Oak	Pec	Syc	Cot	Pine
		__________ kg tree−1 ___________
2016	NH4NO3	0.52–1.0	2.3	0.20	0.20	0.20	2.4–4.9 × 6.11 m area	9.1 m diameter area	2.3 × 2.4 m area	2.3 × 2.4 m area	2.3 × 2.4 m area
	13-13-13 †	1.3–2.6	5.7	0.48	0.48	0.48
	Gypsum	0.27–0.54	1.2	0.10	0.10	0.10
2017	32-0-0 ††	1.0	2.3	0.20	0.20	0.20	4.9 × 6.11 m area	9.1 m diameter area	2.3 × 2.4 m area	2.3 × 2.4 m area	2.3 × 2.4 m area
	13-13-13	2.6	5.7	0.48	0.48	0.48
	Gypsum	0.5	1.2	0.10	0.10	0.10
2018	NH4NO3	3.5	4.7	0.44	0.55	0.30	9.1 m diameter area	10.7 m diameter area	2.3 × 3.7 m area	2.3 × 4.6 m area	2.3 × 2.4 m area
	13-13-13	8.5	11.5	1.1	1.4	0.72
	Gypsum	1.8	2.4	0.23	0.29	0.15
2019	NH4NO3	3.5	4.7	0.44	0.55	0.30	9.1 m diameter area	10.7 m diameter area	2.3 × 3.7 m area	2.3 × 4.6 m area	2.3 × 2.4 m area
	13-13-13	8.5	11.5	1.1	1.4	0.72
	Gypsum	1.8	2.4	0.23	0.29	0.15
2020	NH4NO3	3.5	4.7	0.87	0.87	0.30	9.1 m diameter area	10.7 m diameter area	4.6 m diameter area	4.6 m diameter area	2.3 × 2.4 m area
	13-13-13	8.5	11.5	2.1	2.1	0.72
	Gypsum	1.8	2.4	0.45	0.45	0.15
2021	13-13-13	-	-	-	-	0.50	-	-	-	-	2.0 m diameter area

^† Greenkeeper’s Secret 13-13-13 Premium Fertilizer, T&M, Inc., Foristell, MO, USA. ^†† Oakley’s Carefully Blended 32-0-0 Fertilizer, Bruce Oakley, Inc., Little Rock, AR, USA. * Northern red oak (Oak), pecan (Pec), American sycamore (Syc), cottonwood (Cot), and pitch/loblolly pine (Pine).

. Additional site establishment and management details have been reported [15,17,18,19,20,21,22,23,24,25,26,27,28,29,30].

2.2. Survey Equipment and Procedures

Geospatial, EC_a measurements were obtained using a DUALEM-1S sensor (DUAL-geometry Electro-Magnetic; Dualem Inc., Milton, ON, Canada) and a Trimble R2 global positioning system (GPS) unit (Trimble Inc., Westminster, CO, USA; Figure 2). The DUALEM-1S operates at 9 kHz and has one transmitter, which uses a vertical dipole, and two receivers with different orientations [2,31]. The receiver has a 1 m separation from the transmitter and uses a vertical dipole in the horizontal coplanar geometry (HCP), which has a 1.1 m separation from the transmitter and uses a horizontal dipole in the perpendicular geometry (PRP) [2,31]. The depths of exploration (DOE) for HCP and PRP are ~1.6 and 0.5 m, respectively, where the DOE is defined as the depth to which an array accumulates 70% of its total sensitivity. Thus, HCP measures the bulk EC_a of 0 to 1.6 m and the PRP measures the bulk EC_a of the 0 to 0.5 m depth. The conductivity range for the DUALEM-1S is 3000 mS m⁻¹ [2,31]. In this study, a Can-Am Side-by-Side (Defender, BRP US, Inc., Sturtevant, WI, USA) was used to power and pull the DUALEM-1S on a sled during surveying (Figure 2). Measurements from the DUALEM-1S were collected concurrently with the GPS data via a hand-held geoinformation system program (HGIS; HGIS version 10.90, StarPal Inc., Fort Collins, CO, USA) on a Trimble Yuma 2 rugged tablet computer (Trimble Inc., Westminster, CO, USA) [2,31].

Monthly EC_a surveys were conducted between August 2020 and July 2021 to capture the spatial and temporal EC_a variability. To minimize the effects of temperature drift on the DUALEM-1S signal during surveying, surveys were conducted in the early morning and the DUALEM-1S sensor was powered on 30 min prior to each survey to allow the sensor to reach ambient temperature. For each survey, the DUALEM-1S was securely suspended on a sled, 12.7 cm above the sled bottom, and was pulled at a rate of 4.8 to 8.0 km hr⁻¹ (Figure 2). The front of the DUALEM-1S was located 2.1 m behind the Side-by-Side and the center of the DUALEM-1S was located 4.15 m behind the Trimble R2 GPS unit (Figure 2). Each survey was conducted in a looping pattern over two alleys at a time until four parallel drive paths per alley, 2 to 5 m apart, had been achieved. During the survey period, unnecessary stops were avoided. After each survey, a calibration line was driven over all subsequent survey lines so that any drift that occurred in the DUALEM-1S’s measurement during the survey period could be monitored and accounted for. The calibration line was conducted in a “V” shape, starting in the northwest corner, ending in the southwest corner of the AF site, with the midpoint being around the eastern edge of Row 8 or 9 (Figure 1).

2.3. Weather and Soil Property Collection

Air temperature and precipitation was recorded every 30 min on a micrometeorological weather station, using a tipping bucket gauge, located 1.67 km away (36.10°N, 94.173°W) to the northeast of the northeast corner of the AF site. The recorded air temperature was averaged per day to obtain the daily average air temperature and the recorded daily precipitation was totaled to obtain the total daily precipitation. There were three sentinel sites at the study area and were located near the local summit, depression area, and a mid-slope area (Figure 1). In situ soil VWC and soil temperature were recorded every 4 h using 5TM soil moisture sensors (ECH₂O 5TM Soil Moisture Sensor; METER Group, Pullman, WA, USA) at the 15 and 75 cm soil depths from the soil surface, and were logged on a Decagon EM50 data logger (METER Group, Pullman, WA, USA) at the sentinel sites (Figure 1). Additionally, EC_a soil sensors were not able to be installed until 26 January 2021 at the study site; two EC sensors (Teros 12; METER Group, Pullman, WA, USA) were installed at the 15 and 50 cm depths at each of the three sentinel sites (Figure 1), where in situ soil EC_a was measured every 4 h on a Decagon EM50 data logger. At the conclusion of the measurement period, VWC and soil temperature recorded between 1 August 2020 and 31 July 2021 and EC_a recorded between 26 January 2021 and 31 July 2021 were averaged for each day and expressed as daily means.

During each survey, soil samples from 45 to 55 cm were collected at each sentinel site. Samples from the 0 to 15 cm soil depth were collected during each survey beginning on the 20 November 2020 EC_a survey. Soil samples were placed in a plastic-lined bag and weighed in the laboratory. Afterward, samples were oven-dried in a forced-air oven at 70 °C for 48 h, re-weighed for moisture content, and then mechanically ground and sieved to <2 mm. Then, soil EC and pH were measured potentiometrically in a 2:1 water-volume-to-soil-mass slurry, where 40 g (±0.05 g) of the oven-dried, sieved soil was mixed with 40 mL of deionized water and reciprocally shook on a mechanical shaker for 1 h. The soil–water slurry was then filtered twice (No.1 filter paper, Cat No 1001 125; Whatman Cytiva, Little Chalfont, Buckinghamshire, UK) until there was 40 mL of filtrate; the filtrate was then mixed by pouring the filtrate back-and-forth between two plastic vails. Afterward, the volume of the filtrate was evenly split (20 mL) between the two plastic vials, and the pH and EC were determined by placing a pH (Orion 9107BNMD; Thermo Scientific, Waltham, MA, USA) and EC probe (Orion 013005MD; Thermo Scientific, Waltham, MA, USA) into each of the vials, whereby both were connected to a pH/Conductivity Benchtop Meter (Orion Star A215; Thermo Scientific, Waltham, MA, USA).

2.4. EC_a Survey Data Processing

After collection, EC_a survey data underwent data processing procedures, which included GPS coordinate adjustment, outlier removal, drift calibration, temperature adjustment, averaging of coincidental points, and universal-kriging. A spatial adjustment between the DUALEM-1S sensor and Trimble R2 GPS unit was performed to account for the 4.15 m difference between the location of the GPS unit and the sensor during data acquisition [32]. The EC_a measurements were collected at an average speed of 6.4 km h⁻¹ and the coordinates associated with each measurement were post-corrected to account for the distance between the center of the DUALEM-1S and the location of the GPS receiver. Furthermore, measurement points that had a PRP or HCP EC_a measurement < 0.1 mS m⁻¹ were removed from both the survey and calibration line, as EC_a measurements < 0.01 mS m⁻¹ were likely caused by the presence of metallic objects in the field. Additionally, measurement points that occurred within 2.0 m of soil sensor equipment at the AF site were removed to reduce the effect of magnetic interference in the EC_a data [33,34]. Afterward, a Hampel filter [35] was applied to the PRP and HCP EC_a data with a 10-point, half-width (i.e., 21-point moving data window) and a threshold of 3 using the pracma package [36] in R Studio (version 4.05, R Core Team, Boston, MA, USA) [37]. A Hampel filter normally replaces a determined outlier with the local median; however, in this case, any measurement point that was determined to be an outlier in the PRP and/or HCP EC_a of the survey or calibration line was removed altogether to be consistent with earlier procedures.

Because internal temperatures of the DUALEM-1S increase during the survey period, EC_a data can be slightly skewed and require calibration. To perform calibration, linear regression techniques were used to quantify the difference between the EC_a of the survey points and the average EC_a of the calibration points that occurred within 1.5 m of the survey points [32,34,37]. The projected differences between the EC_a of the survey and the calibration line were subtracted from the EC_a of the survey points at each time point, resulting in a drift-calibrated EC_a survey dataset [32,34,37]. The calibration procedure was performed in R Studio using the geosphere package [38].

Because the EC_a surveys were conducted throughout the year, the EC_a of each survey occurred at a different soil temperature. As a result, it was necessary to standardize the EC_a measurements to a reference temperature. The HCP and PRP EC_a were both standardized to 25 °C (EC₂₅) using the equation of Corwin and Lesch [5] (1):

$${\mathrm{EC}}_{25}={\mathrm{EC}}_{\mathrm{T}}\times (0.447+{1.4034}^{-\mathrm{T}/26.8615})$$

(1)

where EC_T (mS m⁻¹) is the EC_a at a particular soil temperature and T (°C) is the soil temperature [5]. The soil temperature used in Equation (1) was the average soil temperature of soil sensors at the three sentinel sites at the time of a particular EC_a survey. Soil temperature measurements from a depth of 15 cm were used for the PRP EC_a standardization, whereas measurements from 50 cm were used for HCP EC_a standardization. Any measurement point that had an PRP and HCP EC_a value of <0.1 mS m⁻¹ after the EC temperature standardization Equation (1) was applied. Finally, the PRP and HCP EC_a of any coincidental points were averaged, resulting in a final calibrated dataset.

After cleaning and processing, kriging methods were used to generate continuous raster layers from EC_a point data. Twenty percent of the PRP and HCP EC_a survey data were selected at random and set aside for model validation. The experimental semi-variogram was calculated with the remaining 80% and fitted with a nugget and exponential, spherical, Gaussian, Matern, circular, and linear models. The best model minimized the sum of square of error and was selected for universal kriging to a 5 × 5 m grid. Values from the interpolated raster were then extracted to each of the validation locations. The model residuals were calculated using Equation (2):

$$r_i=\widehat{y_i}-x_i$$

(2)

where $r_i$ is the residual value in location i, $\widehat{y_i}$ is the predicted value in location i, $x_i$ is the observed value in location i, and i belongs to all locations selected at random to form the validation dataset. The model error was calculated from model residuals using Equation (3):

$$e_i=\frac{r_i}{x_i}$$

(3)

where $e_i$ is the model error at location i. All interpolation procedures were conducted in R Studio using sp [39], rgdal [40], gstat [41], geodist [42], raster [43], and terra [44] packages.

2.5. Statistical Analyses

The cleaned, GPS-coordinate-adjusted, outlier-removed, drift-calibrated, temperature-adjusted, coincidental point-averaged, universally kriged PRP and HCP EC_a survey datasets are hereafter referred to as PRP and HCP EC_a. Multiple methods were used to accomplish Objective 1. First, the PRP and HCP EC_a were used to determine the mean, SD, and CV among the 5 × 5 m pixels to assess the spatial variability of the AF site’s EC_a during the 12-month study period. Furthermore, a one-factor analysis of variance (ANOVA) was conducted in R Studio to evaluate the effect of season and growing/non-growing season on PRP and HCP EC_a. When main effect differences were found, the means were separated by least significant difference (LSD) at an alpha level of 0.05. The mean, SD, and CV of the PRP and HCP EC_a per growing/non-growing season were determined by averaging the mean, SD, and CV for each survey that fell within the growing or non-growing season groups. The surveys that took place during the growing and non-growing season were determined using soil sensor and air temperature data recorded at or near the AF site along with guidelines provided by the NRCS’s National Water and Climate Center [45].

Table 2. Summary of the seasonal (i.e., weather season (summer, fall, winter, and spring) and growing season (growing season and non-growing season)) groupings and the survey groupings compared against all 12 monthly conducted electromagnetic-induction (EMI) apparent electrical conductivity (EC_a) surveys for the homogeneity of variance assessment. The EMI EC_a surveys were conducted at an agroforestry site in Fayetteville, AR between August 2020 and July 2021.

		Seasonal Groupings		Test Groupings ††
Survey	Survey (Month-Year)	Weather Season	Growing † Season	H1	H2	MS	MFS	MWS
1	August-2020	Summer	GS	X
2	September-2020	Fall	GS		X
3	October-2020	Fall	GS	X		X	X
4	November-2020	Fall	NGS		X
5	December-2020	Winter	NGS	X
6	January-2021	Winter	NGS		X	X		X
7	February-2021	Winter	GS	X
8	March-2021	Spring	GS		X
9	April-2021	Spring	GS	X		X	X
10	May-2021	Spring	GS		X
11	June-2021	Summer	GS	X
12	July-2021	Summer	GS		X	X		X

^† Growing season (GS), non-growing season (NGS)^{. ††} Odd half of surveys in sequential order of conduction (H1), even half of surveys in sequential order of conduction (H2), middle of climate season (MS), middle of fall and spring (MFS), middle of winter and summer (MWS).

summarizes each survey’s seasonal grouping (i.e., weather season and growing/non-growing season).

To accomplish Objective 2, the k-means clustering algorithm [46] was used by overlaying and grouping PRP and HCP EC_a surveys to generate the precision SMZs within the AF site. The optimal number of clusters used in the k-means clustering was determined using the factoextra package [47] in R Studio. Furthermore, Objective 3 was accomplished by conducting Pearson correlations among PRP and HCP EC_a (independently and combined) and in situ, soil-sensor-collected VWC and EC_a and benchtop-measured soil sample GWC, EC, and pH (at all depths) at an alpha level of 0.05. Additionally, to accomplish Objective 4, a Levene’s test for homogeneity of variance was conducted. The non-universally kriged PRP and HCP EC_a survey datasets were used in the Levene’s test. There were five survey groupings of reduced quantity that were used to compare against the variability all 12 monthly surveys: surveys from the odd half of surveys in sequential order of conduction (H1), even half of surveys in sequential order of conduction (H2), middle of the season (MS), middle of fall and spring (MFS), middle of winter and summer (MWS). Table 2 also summarizes the survey groupings used in the Levene’s test.

3. Results

3.1. Weather, Soil Sensor, and Soil Sample Data

Temporal variations in soil properties followed temporal weather fluctuations throughout the sampling period of the study (i.e., 1 August 2020 to 31 July 2021). The AF site received a total of 818 mm of precipitation and the mean overall daily air temperatures at the AF site ranged from −18.7 to 28.3 °C. Additionally, the soil sensor daily mean soil temperature, VWC, and EC_a ranged from 1.6 to 26.5 °C, from 9.9 to 57.7% (v v⁻¹), and from 5.7 to 43.7 mS m⁻¹, respectively (Figure 3). The soil sample GWC, EC, and pH ranged from 7.9 to 38.4% (m m⁻¹), 6.3 to 27.9 mS m⁻¹, and 4.6 to 6.7, respectively.

3.2. EC_a Survey Data

3.2.1. Monthly EC_a Survey Data

The spatial pattern of the measured EC_a was similar across all survey dates within both EC_a configurations. For the PRP configuration, the September 2020 and June 2021 surveys had the lowest (0.2 mS m⁻¹) and largest (29.1 mS m⁻¹) recorded EC_a values at the AF site during the sampling period, respectively (

Table 3. Summary of the perpendicular geometry (PRP) and horizontal coplanar geometry (HCP) apparent electrical conductivity (EC_a) survey dates, number of measurements per survey, semi-variogram information (i.e., experimental model, nugget, sill, and range), and resulting summary statistics (i.e., mean, minimum (min), maximum (max), standard deviation (SD), and coefficient of variation (CV)) of each PRP and HCP EC_a survey after universal kriging at an agroforestry site in Fayetteville, AR between August 2020 and July 2021.

				Semi-Variogram Information				Kriged Survey Summary Statistics
ECa Type	Survey	Date (Month-Year)	Survey Points	Model	Nugget	Sill	Range	Mean	Min	Max	SD	CV
							m	_________ mS m−1 _________				%
PRP	1	August-2020	5271	Exponential	0.61	4.8	39.4	4.2	1.0	17.0	1.9	44.1
	2	September-2020	5422	Matern	0.18	3.5	34.4	3.4	0.4	14.6	1.7	48.1
	3	October-2020	5769	Exponential	0.17	3.7	28.9	4.6	0.9	16.6	1.9	42.1
	4	November-2020	6083	Exponential	0.38	5.0	41.4	7.4	2.3	18.2	2.4	32.2
	5	December-2020	6085	Exponential	0.30	4.8	29.3	6.7	2.6	21.0	2.3	34.1
	6	January-2021	6340	Exponential	0.10	3.6	27.8	5.5	1.4	17.2	2.1	38.0
	7	February-2021	5777	Exponential	0.00	3.7	26.8	6.2	1.7	20.5	2.3	36.3
	8	March-2021	6173	Exponential	0.10	5.6	24.0	5.9	1.3	21.9	2.5	41.8
	9	April-2021	6329	Exponential	0.03	6.0	25.3	5.5	0.6	17.6	2.6	46.8
	10	May-2021	6854	Exponential	0.00	8.4	24.0	7.5	2.2	21.4	2.9	38.6
	11	June-2021	6671	Matern	0.00	13.1	27.6	8.7	2.7	29.1	3.6	41.2
	12	July-2021	6527	Exponential	0.18	4.2	27.7	5.1	0.2	15.9	2.1	40.3
HCP	1	August-2020	5271	Spherical	0.25	16.5	97.9	8.6	1.6	25.5	4.4	51.2
	2	September-2020	5422	Circular	0.14	13.8	84.0	6.7	0.4	19.7	4.1	61.1
	3	October-2020	5769	Spherical	0.32	14.3	91.6	9.9	3.6	25.6	4.1	41.2
	4	November-2020	6083	Matern	0.08	15.9	60.0	9.8	2.9	26.2	4.2	42.1
	5	December-2020	6085	Matern	0.00	15.7	50.0	12.3	3.6	28.4	4.4	35.9
	6	January-2021	6340	Matern	0.00	14.6	48.9	11.7	4.4	29.2	4.2	36.4
	7	February-2021	5777	Matern	0.00	15.2	58.9	8.8	2.2	23.6	4.2	47.5
	8	March-2021	6173	Matern	0.00	16.0	42.1	11.9	4.8	34.2	4.2	35.5
	9	April-2021	6329	Exponential	0.00	16.5	43.2	11.6	4.1	27.2	4.2	36.3
	10	May-2021	6854	Matern	0.00	18.9	46.1	9.7	2.0	31.4	4.4	45.6
	11	June-2021	6671	Spherical	0.13	19.1	69.3	10.1	1.7	28.6	5.1	50.1
	12	July-2021	6527	Spherical	0.18	11.8	77.7	7.2	1.0	19.9	3.9	53.2

; Figure 4). The June 2021 survey had the greatest mean (8.7 mS m⁻¹), SD (3.6 mS m⁻¹), and range (26.4 mS m⁻¹), and the September 2020 survey had the largest CV (48.1%) for the PRP EC_a (Table 3; Figure 4). Furthermore, the September 2020 survey had the smallest mean (3.4 mS m⁻¹), SD (1.7 mS m⁻¹), and range (14.2 mS m⁻¹), and the November 2020 survey had the smallest CV (32.2 %) for the PRP EC_a (Table 3; Figure 4).

For the HCP configuration, the September 2020 and March 2021 surveys had the numerically smallest (0.4 mS m⁻¹) and largest (34.2 mS m⁻¹) recorded EC_a values at AF site during the sampling period, respectively (Table 3; Figure 4). The December 2020 survey had the greatest mean (12.3 mS m⁻¹), the June 2021 survey had the greatest SD (5.1 mS m⁻¹), the September 2020 survey had the largest CV (61.1%), and the March and May 2021 surveys had the greatest range (29.4 mS m⁻¹) for the HCP EC_a (Table 3; Figure 4). Furthermore, the September 2020 survey had the smallest mean (6.7 mS m⁻¹), the July 2021 survey had the smallest SD (3.9 mS m⁻¹), the March 2021 survey had the smallest CV (35.5%), and the July 2021 survey had the smallest range (18.9 mS m⁻¹) for the HCP EC_a (Table 3; Figure 4).

3.2.2. Overall EC_a Survey Data

The spatial patterns of the overall mean, SD, and CV differed greatly between PRP and HCP EC_a. Averaged across all 12 surveys, the mean PRP and HCP EC_a ranged from 1.8 to 18.0 mS m⁻¹, and from 3.1 to 25.8 mS m⁻¹, respectively (Figure 5), and shared a similar spatial pattern as the monthly surveys (Figure 4). The SD for PRP and HCP EC_a ranged from 0.6 to 4.8 mS m⁻¹ and from 1.3 to 4.9 mS m⁻¹, respectively, across all surveys (Figure 5). Additionally, the CV for the PRP and HCP EC_a ranged from 15.1 to 62.9% and from 7.3 to 54.6%, respectively, across all surveys (Figure 5). The PRP and HCP mean EC_a, SD, and CV were unaffected (p > 0.05) by either the weather season or growing season.

3.2.3. SMZ Delineation

Using the PRP and HCP EC_a data, three precision SMZ were determined to be the optimal number of clusters at the AF site and were delineated using k-means clustering (Figure 6). The location and geospatial pattern of the SMZs were similar between the PRP and HCP (Figure 6) and both were visually similar to their respective overall means map (Figure 5). For both PRP and HCP EC_a, SMZ 1, 3, and 2 (Figure 6) occurred where the recorded EC_a was numerically smallest, in between, and the largest throughout the sampling period, respectively (Figure 5). With the exception of the CV, all summary statistics (i.e., mean, minimum, maximum, SD, and SE) for HCP were numerically greater than for PRP (Figure 6;

Table 4. Summary of the perpendicular (PRP) and horizontal coplanar geometry (HCP) apparent electrical conductivity (EC_a) for each cluster generated by the k-means clustering algorithm using 12 electromagnetic-induction EC_a surveys conducted at an agroforestry site in Fayetteville AR between August 2020 and July 2021.

		Summary Statistics †
ECa	Cluster	Mean	Min	Max	SD	CV	SE
		___________________ mS m−1 ___________________				%
PRP	1	3.6	0.2	9.0	1.4	38.3	0.016
	2	9.0	3.1	29.1	2.8	31.3	0.042
	3	6.2	1.3	13.4	1.9	29.9	0.019
HCP	1	5.7	0.4	15.0	2.2	38.8	0.024
	2	16.6	7.5	34.2	3.2	19.3	0.050
	3	10.5	2.4	19.2	2.4	22.6	0.024

^† Minimum (Min), maximum (Max), standard deviation (SD), coefficient of variation (CV), standard error (SE).

). For both PRP and HCP, SMZ 2 was more variable (i.e., greater range, SD, and SE) than the other two SMZs (Figure 6; Table 4). However, when SMZs’ variability was standardized by the mean (i.e., CV), SMZ 1 had the numerically greatest variability relative to the mean compared to the other SMZs (Figure 6; Table 4). Additionally, SMZ 3 was generally the second numerically greatest occurring and second most variable (i.e., range, SD, CV, and SE; Figure 6; Table 4) among the three SMZs.

3.2.4. Correlations among EC_a and Soil Properties

There were positive correlations between the EC_a survey data (i.e., PRP, HCP, and combined (PRP and HCP)) and at least two measured soil properties (p < 0.05;

Table 5. Summary of the resulting correlation coefficients (r) and p-values from Pearson linear correlations evaluating the relationship between the perpendicular (PRP), horizontal coplanar geometry (HCP), and combined (PRP and HCP) apparent electrical conductivity (EC_a) and the upper, lower, and combined (upper and lower) soil-sensor-measured volumetric water content (VWC) and EC_a measurements and the upper, lower, and combined soil-sample-derived gravimetric water content (GWC), electrical conductivity (EC), and pH, respectively. Electromagnetic-induction EC_a survey, soil sensor, and soil sample data were collected from an agroforestry site in Fayetteville AR between August 2020 and July 2021.

Correlation Dataset	r	p
PRP ECa and Upper
VWC	0.09	<0.01 †
ECa	0.84	0.01
GWC	0.70	<0.01
EC	0.57	0.17
pH	0.61	0.13
HCP ECa and Lower
VWC	0.62	0.79
ECa	0.74	<0.01
GWC	0.71	<0.01
EC	0.37	<0.01
pH	0.42	<0.01
Combined ECa and Combined
VWC	0.25	0.24
ECa	0.65	<0.01
GWC	0.36	0.03
EC	−0.22	0.28
pH	0.15	0.49

^† Bold values are significant at p < 0.05.

). The HCP EC_a and soil-sample GWC from the 45–55 cm depth interval (r = 0.71) and the PRP EC_a and the soil-sample GWC in the 0–15 cm depth interval (r = 0.70) had strong, positive correlations (Table 5). The PRP EC_a and the soil-sensor-measured EC_a from the 15 cm soil depth (r = 0.84), the HCP EC_a and the soil-sensor-measured EC_a from the 50 cm soil depth (r = 0.74), and the combined EC_a and the soil-sensor-measured EC_a across soil depths (r = 0.65) also had strong, positive correlations (Table 5).

The PRP EC_a and the 15 cm, soil-sensor-measurement correlation was the only comparison where the soil-sensor-measured VWC was correlated to EC_a survey data, but had a low correlation coefficient (r = 0.09; Table 5). Additionally, the HCP EC_a and the 45–55 cm soil-sample-measurement correlation was the only comparison where the soil-sample-measured EC and pH were correlated to any EC_a survey measurement. Specifically, HCP EC_a and the 45–55 cm EC and pH had moderate, positive correlations (r = 0.37 and r = 0.42, respectively; Table 5).

3.2.5. Homogeneity of Variance Assessment

The homogeneity of variance was used to assess whether fewer, strategically selected EC_a surveys could be conducted to capture the same amount of overall EC_a variance at the AF site as the 12 monthly EC_a surveys captured. The five reduced survey groupings that were used for the homogeneity of variance assessment are summarized in Table 2. For the PRP EC_a, there were differences between the variance of all 12 surveys and the variance of all 5 of the reduced number of survey groupings (i.e., odd half of surveys in sequential order of conduction (H1), even half of surveys in sequential order of conduction (H2), middle of the season (MS), middle of fall and spring (MFS), middle of winter and summer (MWS); p < 0.05; Table 2 and

Table 6. Summary of the homogeneity of variance between all 12 monthly perpendicular (PRP) and horizontal coplanar geometry (HCP) apparent electrical conductivity (EC_a) surveys and five reduced survey groupings (i.e., H1, H2, MS, MFS, and MWS). Additional details of the survey groupings are presented in Table 2. Homogeneity of variance was evaluated using the Levene’s test.

ECa	Survey Grouping Comparison	p	Survey Grouping with Greater Variance
PRP	H1 †	<0.01 ††	H1
	H2	<0.01	All
	MS	<0.01	All
	MFS	<0.01	All
	MWS	<0.01	All
HCP	H1	0.60	-
	H2	0.93	-
	MS	<0.01	All
	MFS	<0.01	All
	MWS	0.87	-

^† Odd half of surveys in sequential order of conduction (H1), even half of surveys in sequential order of conduction (H2), middle of season (MS), middle of fall and spring (MFS), middle of winter and summer (MWS). ^†† Bold values are significant at p < 0.05.

). Except for when compared to the H1 survey group, the variance of all 12 PRP EC_a surveys was greater than the variance of the other four reduced number of survey groupings (i.e., H2, MS, MFS, and MWS) for the PRP EC_a (Table 6). In contrast to the PRP, for the HCP configuration, there were differences between the variance of all 12 monthly surveys and the variance of MS and MFS reduced groupings (Table 6). Specifically, the variance of all 12 monthly HCP EC_a surveys was greater than the variance of both the MS and MFS groupings (Table 6). Conversely, there was no difference between the variance of all 12 monthly HCP EC_a surveys and the reduced H1, H2, and MWS survey groupings (Table 6).

4. Discussion

4.1. EC_a Survey Data

4.1.1. Semi-Variogram Information

The semi-variogram range can be used to help understand the modeled variable (e.g., EC_a in this study). The ranges of the PRP EC_a surveys were 31 to 68% less than the ranges from the corresponding HCP EC_a surveys (Table 3), which is indicative of the PRP EC_a being more spatially erratic and variable than the HCP EC_a [48,49]. DeCaires et al. [50] conducted repeated DUALEM-1S surveys on a 5.81 ha cacao (Theobroma cacao L.) plantation, classified as an alluvially formed silty-clay Inceptisol in Trinidad, and reported that the PRP EC_a generally had a shorter range than the HCP EC_a. The PRP EC_a being more spatially variable than the HCP EC_a is likely to be a result of the HCP measuring deeper into the soil profile than the PRP. Specifically, the PRP incorporates less of the soil profile into its measurement, and the part that is incorporated into PRP EC_a measurement (i.e., 0–0.5 m) is generally much more variable in pedogenesis factors and soil properties (i.e., SWC, OM, nutrients, structure/porosity, and coarse fragments). The variations in pedogenesis and soil properties are the result of the upper portion of the soil being more exposed to surface factors that are also spatially and temporally variable, such as weather (i.e., precipitation, air temperature, wind, and sunlight), environmental processes (i.e., runoff, surface erosion/accumulation, and evapotranspiration), and land management (i.e., fertilizer/manure and type and quantity of trees/grasses/animals).

4.1.2. Monthly EC_a Survey Data

Because PRP and HCP measure the bulk EC_a of the 0–0.5 and 0–1.6 m soil depths, respectively, the measured PRP EC_a was more likely to be a reflection of soil surface management, and the HCP EC_a was more likely to be a reflection of pedogenic factors (i.e., time, parent material, organisms, topography/relief, and climate). Additionally, because different soils have varying properties, it can be expected that different mapped soils would result in varying EC_a measurements across the AF site. However, based on the generated PRP and HCP EC_a maps (Figure 4), soil boundaries were not detailed enough to show visual correlations with the measured EC_a values for either EC_a configurations. Although the soil boundary data were not detailed enough for visual correlations with the EC_a data_, the western portion of the AF site was still generally characterized by a shallower depth to bedrock and an increase in chert coarse fragment abundance (Figure 1). Coarse fragments have been shown to be negatively corelated with EC_a [51]. Thus, the increased chert coarse fragments could have resulted in the decreased PRP and HCP EC_a values in the western portion of the AF site (Figure 4).

An additional potential residual effect is that PL was applied between 2001 to 2007 and can be observed through increased EC_a values occurring in the eastern half of the AF site in both EC_a configurations (Figure 4). Specifically, PL and NH₄NO₃ were annually applied based on the annual soil total-N concentrations to the eastern and western halves of the AF site between 2001 and 2007, respectively [17]. After the PL and NH₄NO₃ applications, Sauer et al. [17] reported numerically larger percent changes in nutrient concentrations in 9 of the 11 soil nutrients measured (i.e., N, P, K, magnesium (Mg), sulfur (S), sodium (Na), iron (Fe), zinc (Zn), and copper (Cu)) in the soils that received PL, than the soils that received NH₄NO₃. Thus, increased nutrient concentrations likely remain in the soils that received PL and may be contributing to increased PRP and HCP EC_a measurements (Figure 4).

In addition to PRP EC_a most likely reflecting surface management, the PRP EC_a likely varied overtime more than the HCP EC_a due to variations in soil properties from fluctuating weather (i.e., soil moisture). Specifically, the SD for HCP EC_a was generally more consistent than SD for PRP EC_a (Table 3). Additionally, the SD for the PRP EC_a was generally larger in the summer months than in the fall, winter, and spring (Table 3), which was most likely a reflection of increased soil moisture [48,52]. During wetter periods (i.e., fall, winter, and spring), the spatial distribution of soil moisture is heavily influenced by topography, where topography strongly controls the lateral distribution of water which increases soil moisture variability, and thus, the variability of measured EC_a [48,52].

Although no surveys for either the PRP or HCP EC_a were exactly alike, each survey per configuration had a similar geospatial pattern to the other (Figure 4). For both the PRP or HCP EC_a, an elevated trail in EC_a values can been seen starting in the northwest corner, peaking in the center (Rows 5 to 7), and trailing off to the bottom-middle of the AF site (Figure 4). This apparent pattern can be attributed to landscape properties that affect water flow and runoff accumulation and factors that affect measured EC_a. Specifically, the trail of elevated EC_a values occurs in the local drainage way of the AF site, where the northwest corner has the greatest elevation and the local landscape depression is in the center of Rows 5 to 7 (Figure 1 and Figure 4) [21,27]. Additionally, the runon from the surrounding, up-slope landscapes enters through a culvert in the northwest corner of the AF site (Figure 1). Overtime, overland flow results in the accumulation of water and transported nutrients, OM, and sediment in channels or low-lying areas, all of which increase measured EC_a [3]. Thus, the elevated PRP and HCP EC_a values in the northwest corner of the AF site towards the center of Rows 5 to 7 were most likely to be the result of greater SWC and accumulated transported materials from the local landscape, but also from the up-slope runoff entering into the northwest corner from the surrounding landscape (Figure 4).

Although more visible in HCP EC_a surveys, two trails of elevated EC_a values starting at the middle-eastern edge and connect to elevated EC_a in the depression area can be observed in both configurations (Figure 4) [15,27]. Unlike at the local depressional area (Sentinel Site 2), no visual signs of soil wetness were observed in these areas during the duration of the study. However, the lower soil sensor at Sentinel Site 3, which was located at the more southern of the two trails of elevated EC_a values, recorded similar VWC measurements as the upper soil sensor measurement at Sentinel Site 2 throughout the winter during this study (Figure 3). Additionally, Smith et al. [53] conducted ground-penetrating radar surveys at the AF site and produced measurements that potentially show groundwater movement in the same areas with elevated EC_a values. Thus, the trails of elevated EC_a values starting at the middle-eastern edge of the AF site are most likely due to increased SWC from a shallow groundwater table (Figure 4).

4.1.3. Overall EC_a Survey Data

The mean HCP EC_a being numerically greater than the mean PRP EC_a was most likely a reflection of how soil properties vary with depth. Specifically, most of the AF site is classified as a Captina silt loam (Figure 1), which is characterized by a large increase in clay content with depth [14]. Additionally, fine-textured soil generally has a greater water-holding capacity and CEC than coarse-textured soil, and SWC generally increases with depth. Thus, the mean HCP EC_a being numerically greater than the mean PRP EC_a was most likely due to the SWC and clay content increasing with depth [3,48]. Additionally, measured soil EC_a is influenced both by stable and dynamic soil properties. Numerous soil properties (i.e., soil texture, clay content, OM content, and depth to bedrock) are temporally stable within short time spans (i.e., within a year). Thus, within humid regions, including the southern Ozark Highlands, temporal fluctuations in the measured soil EC_a are generally reflective of variations in SWC and water movement [48].

For both the PRP and HCP EC_a, the SD was generally elevated in the NW corner and across the middle-eastern edges of the AF (Figure 5), where both were most likely a result of fluctuating SWC. The NW corner is where the runon from the surrounding, up-slope landscape enters the AF site. Therefore, runoff-producing precipitation events would cause the NW corner to have varying SWC, hence, varying EC_a, throughout the year. Additionally, the elevated EC_a across the middle-eastern edges of the AF site are likely to be where subsurface water movement occurs [53], and thus, fluctuation in the SWC in these areas would cause fluctuation in the measured EC_a throughout the year (Figure 5). However, one major difference in the SD between the PRP and the HCP was that the PRP had elevated SD values along the drainage way at the site and the HCP did not (Figure 5) [15,21]. Elevated SD for the PRP, and not the HCP, was also likely to be from fluctuating SWCs throughout the year (Figure 5). Specifically, the PRP measures the upper 0.5 m of the soil, which is going to be more affected throughout the year by changes in weather and surface conditions (i.e., precipitation, air temperature, wind, sunlight, and evapotranspiration) than the 0–1.6 m of the soil profile which the HCP measures.

The CV of the PRP and HCP EC_a generally had similar spatial patterns, where there were increased CVs in the west/southwest portion and decreased CVs occurring in the eastern portion, which slightly increase when moving towards the northeast of the AF site (Figure 5). However, the CV of the PRP EC_a had a less cohesive pattern and was visually more variable across the AF site than that of the HCP EC_a CV (Figure 5). The CV of the PRP EC_a was visually more variable than the CV of the HCP EC_a, which was potentially due to two reasons (Figure 5). First, the PRP EC_a is a reflection of a smaller, upper portion of the soil profile (i.e., 0–0.5 m). Secondly, the 0–0.5 m portion of soil is generally more spatially and temporally variable in soil properties (i.e., SWC, OM, nutrients, structure/porosity, and coarse fragments) than the 0–1.6 m depth from the 0–0.5 m portion of the soil being exposed to surface factors that are also spatially and temporally variable (i.e., weather, environmental processes, and land management; Figure 5). Furthermore, the smallest CVs in the HCP EC_a were generally located in wetter areas at the AF site [21], likely due to the deeper SWC in the soil profile at these locations being consistently moist-to-saturated throughout the year (Figure 5). Additionally, unlike the SD, there were no elevated CVs in the northwest corner or along the middle-eastern edges of the AF site for both EC_a configurations (Figure 5). Thus, the EC_a variability (i.e., SD) in the northwest corner and along the middle-eastern edges was minimal when standardized with the mean for both EC_a configurations (Figure 5).

Elevated HCP, EC_a, SDs, and CVs occurred under the pine tree canopies (Figure 5). There are several possible explanations for this, including a decreased water-holding capacity from a decreased effective soil depth, and an increased coarse fragment concentration in the Nixa cherty silt portion of the site (Figure 1) [22]. Additionally, it is possible that the coniferous trees may be acidifying soils relative to deciduous species at the site, thus, resulting in greater EC_a, as Burgess-Conforti et al. [54] reported that coniferous was more acidic than deciduous leaf litter in northwest Arkansas.

4.1.4. Seasonal Effects on EC_a

As a result of the PRP and HCP mean EC_a, SD, and CV being unaffected (p > 0.05) by either weather season or growing season, there was no difference between the mean or variation (SD and CV) among surveys that were conducted in different weather seasons or surveys that were recorded in the growing season versus the non-growing season. Thus, the hypothesis that there would be significant differences between the EC_a mean, SD, and CV of surveys that were conducted in different weather seasons and in the growing/non-growing season at a 20-year AF system within the Ozark Highlands was rejected. Although SWC is a major contributor to the measured EC_a, other soil properties greatly contribute to the measured EC_a in non-saline soil and are temporally stable (i.e., soil texture, BD, and CEC) [5], potentially helping to buffer EC_a fluctuations caused by SWC variability. Thus, these results highlight that the measured EC_a at the AF site was more reflective of and influenced by temporally stable soil (i.e., soil texture, clay content, OM content, BD, CEC, and depth to bedrock), vegetative (i.e., coniferous versus deciduous trees), and terrain properties than temporally variable soil properties (i.e., SWC) within short time spans (i.e., <1 year). Similar results have been reported by Harvey and Morgan [55] who conducted EC_a surveys in spring 2005, fall 2005, and spring 2006 across four different fields in north-central Texas that primarily consisted of the Miles (Typic Paleustalfs) and Abilene (Pachic Argiustolls) soil series, and had been intensively cultivated with cotton (Gossypium hirsutum L.) and wheat (Triticum aestivum L.). Although the effect of season on the mean EC_a or SD was not formally evaluated, Harvey and Morgan [55] reported relatively similar mean EC_a and SD values for seasonal surveys. Additionally, Harvey and Morgan [55] also reported visual similarities among the kriged maps for the three surveys.

4.1.5. SMZ Delineation

Although SMZ 2 had greater numeric variability in the SD, SE, and range, SMZ 1 had the numerically greatest variability relative to the mean (i.e., CV) for both the PRP and HCP EC_a (Figure 6; Table 4). However, SMZ 1’s numerically largest CV was more likely a result of the PRP and HCP mean EC_a being 42 to 60% and 46 to 66%, respectively, than the other two SMZs, rather than a soil property or terrain attribute causing variability in the measured EC_a within SMZ 1 (Figure 6; Table 4). Except for the CV, SMZ 2 had the numerically greatest mean and variability (i.e., range, SD, and SE) in the PRP and HCP EC_a, which was most likely due to the accumulation of EC_a-contributing soil constituents (i.e., OM, nutrients/salinity, clay, and water) and the largest SWC fluctuations of the three SMZs, from SMZ 2 encompassing the local drainage way and areas of potential subsurface water movement (Figure 6) [21,27,56]. Furthermore, SMZ 1 had the numerically smallest mean and variability in three out of four variability statistics (i.e., range, SD, and SE) in the PRP and HCP EC_a, and was most likely due to SMZ 1 encompassing the area of a shallower depth to bedrock and increased coarse fragment abundance, and SMZ 1 not being a zone of accumulation (Figure 6) [21,26]. Although no soil sensor or sample data were collected from this area, SMZ 1 likely had the smallest VWC among the three SMZs for several reasons. First, SMZ 1 is not an area of water accumulation, thus, soil constituents (i.e., sediment, OM, and nutrients) have been eroded and less water is able to infiltrate into the soil, resulting in decreased SWC and SWC variability than in SMZ 2. Secondly, increased coarse fragments increases soil BD, which, in turn, decreases porosity, soil water-holding capacity, CEC, and VWC [57], and thus, EC_a.

Zhu et al. [48] and Brevik et al. [58] reported larger soil EC_a responses to SWC fluctuations at wetter positions in the landscape than in drier locations, suggesting that soil EC_a is less dependent on SWC in drier locations than in wetter locations. Zhu et al. [48] also reported that at drier landscape positions or during drier time periods, the influence of water flow paths and SWC were masked by other soil and terrain attributes. Thus, the EC_a values that resulted in the delineation of SMZ 1 and 2 could have been primarily influenced by SWC in SMZ 2, and primarily influenced by soil and terrain properties in SMZ 1 (Figure 6). Additionally, because measured EC_a reflects the complex interaction of soil properties, terrain attributes, and vegetation, and the effects of different soil properties and terrain attributes were potentially able to be observed in each SMZ, the hypothesis that the monthly EC_a surveys at a 20-year-old AF system within the Ozark Highlands could be grouped into similar functional populations and delineated into zones for precision soil management was confirmed.

The EC_a-derived SMZs were produced in order to delineate soils with similar properties for precision/site-specific soil management. Adhikari et al. [21] created four topographical functional units (TFUs) at the AF site from a principal component analysis and k-means clustering of 13 terrain attributes (i.e., flow accumulation, slope length factor, mid-slope position, multi-resolution ridge top flatness index, multi-resolution valley bottom flatness index, normalized height, slope percent, slope height, system for automated geoscientific analysis wetness index, valley depth, and altitude above channel network). However, the four-terrain attribute-derived TFUs of Adhikari et al. [21] did not show any major visual similarities with the three EC_a-derived SMZs created in this study (Figure 6). Additionally, for the actual management of this 4.25 ha AF site, the grid size would likely need to be increased from the current 5 × 5 m grid to perhaps a 10 × 10 m grid in order to improve the manageability of the AF site and its SMZs. As the topsoil is the layer that agricultural production is most concerned with, the PRP EC_a configuration would be most beneficial to use when creating SMZs for precision soil management.

4.1.6. Correlations among EC_a and Soil Properties

As a result of correlations between measured soil sensor/sample properties and measured EC_a (Table 5), the hypothesis that monthly EC_a surveys were correlated with soil-sensor- and/or soil-sample-based soil properties at a 20-year-old AF system within the Ozark Highlands was supported for some, but not all, soil parameters. With the exception of the sensor-measured EC_a, GWC had the strongest correlation with EC_a survey data in the PRP, HCP, and combined EC_a datasets (Table 5). Thus, of the measured soil properties, GWC most likely contributed to the greatest extent of the measured EC_a variability for both the PRP and HCP configurations (Table 5). Additionally, PRP and HCP EC_a correlations were likely due to different soil properties (i.e., pH and EC) contributing varying degrees to EC_a variability for PRP and HCP (Table 5).

The resulting EC_a and soil property correlations were similar to previous results. Abdu et al. [2], Martini et al. [34], and Zhu et al. [48] conducted repeated EC_a surveys with SMC measurements at fixed locations across each of their fields. Abdu et al. [2] conducted their study in north-central Utah over a soil of the Millville soil series (Typic Haploxerolls), Martini et al. [34] conducted their study in central Germany over soils primarily consisting of Cambisols, Luvisols, and Gleyols [59], and Zhu et al. [48] conducted their study in central Pennsylvania on Hagerstown (Typic Hapludalfs), Opequon (Lithic Hapludalfs), Murrill (Typic Hapludults), Nolin (Dystric Fluventic Eutrudepts), and Melvin (Fluvaquentic Endoaquepts) series. Similar to the results of this study, Abdu et al. [2], Martini et al. [34], and Zhu et al. [48] reported positive relationships between EC_a and SMC. Using a Spearman rank correlation, DeCaires et al. [50] also conducted repeated EC_a surveys and random soil sample measurements on a 5.81 ha cacao plantation with silty-clay Inceptisol soil in Trinidad and reported a strong positive relationship between both the PRP and HCP EC_a and soil EC. Additionally, DeCaires et al. [50] reported significant relationships between PRP and HCP EC_a and soil pH, and a weak, but significant, positive correlation between the PRP and HCP EC_a and GWC. Furthermore, Corwin et al. [60] conducted one EC_a survey and collected multi-depth soil sample measurements at a site classified as Panoche silty clay (Thermic Xerorthents) located within the San Joaquin Valley in central California. Corwin et al. [60] collected EC_a and soil sample measurements for mapping and correlating EC_a and soil properties to cotton yield variations and reported positive relationships between EC_a at 0–1.5 m and 0–1.5 m GWC, EC, and pH.

4.1.7. Homogeneity of Variance Assessment

Similar EC_a variances between groupings of EC_a surveys with different quantities of surveys have implications on the quantity of EC_a surveys necessary to capture the total temporal EC_a variance of a study site. The variance of all 12 surveys being greater than the variance of the H2, MS, MFS, and MWS survey groupings for the PRP EC_a is indicative that EC_a surveys conducted every month for one year captured more numerical EC_a variability than conducting six, four, or two evenly temporally spaced EC_a surveys within a year (Table 6). Additionally, it is not clear why the H1 grouping for the PRP EC_a had greater EC_a variance than all 12 surveys combined; however, perhaps the greater EC_a variance in the H1 grouping was potentially due to the H1 grouping dataset having fewer observations than the 12-survey dataset (Table 6). Alternatively for the HCP EC_a, the variance of all 12 monthly surveys was greater than the variance of both the MS and MFS groupings, which is an indication that only conducting four HCP EC_a surveys in the middle of the weather seasons (i.e., summer, fall, winter, and spring) or two HCP EC_a surveys in the middle of fall and spring at this AF site would potentially result in not capturing all of the EC_a variability within one year (Table 6). However, there was no difference between the variance of all 12 monthly HCP EC_a surveys and the reduced H1, H2, and MWS survey groupings, which is an indication that only conducting six evenly temporally spaced HCP EC_a surveys within a one-year period, or one HCP EC_a survey in the middle of winter and summer, would be necessary in order to obtain the same amount of numerical EC_a variability as 12 monthly HCP EC_a surveys (Table 6). As a result, the hypothesis that fewer EC_a surveys could have been conducted in one year to capture the same amount of overall EC_a variance as the 12 monthly EC_a surveys was supported for the HCP EC_a, but not for the PRP EC_a.

5. Implications

These results provide support for the applicability and ability of EC_a surveys to look at in-field variability across agricultural systems and ecosystems. A better understanding of the tools and methods (i.e., EMI equipment and EC_a surveys) that can be used to accurately assess soil property variability in spatially complex systems, such as AF, can improve the management and adoption of these conservation practices in areas prone to conservation issues (i.e., Ozark Highlands). Furthermore, EC_a measurements and/or EC_a-derived SMZs can be used as a blueprint for where to collect future soil samples, referred to as EC_a-directed soil sampling [5,6,56]. Because EC_a-derived SMZs delineate areas of potential differing soil properties, EC_a-directed soil sampling can reduce the degree of soil surface disturbance and the quantity of necessary soil samples, and thus, time, energy, and cost spent on soil sampling compared to traditional grid soil sampling schemes [3,4,6,61]. Additionally, collecting soil samples and soil property data from each SMZ allows for the characterization of the most spatially and temporally influential soil properties affecting the measured EC_a within each SMZ.

Apparent EC measurements and EC_a-derived SMZs can also be used to explain crop yield variability. Specifically, the EC_a-directed soil sampling method may be used to help characterize soil properties that are causing crop [4,60], forage, and/or tree growth variations. Through establishing SMZs and identifying soil properties that are most affecting their crop yield, producers have the potential to spatially and temporally target fertilizer, pesticide, and irrigation. As a result, SMZ-targeted resource applications have the potential to increase resource-use efficiency, yield, and profits, while reducing excess resource applications, loss of applied nutrients and pesticides, unintended environmental damage (i.e., surface and groundwater degradation), labor, and producer costs [62]. Additionally, establishing EC_a-derived SMZs with soil property data before the establishment of an agricultural system can allow the producer to not only decide what type of crops to use and where to plant them, but also what type of agricultural system to establish in order to have the most productive agricultural system based on the present soil and land properties.

The ability to conduct fewer EC_a surveys would be beneficial for researchers, agricultural consultants, and producers. Specifically, multiple surveys are necessary to assess the spatiotemporal variability of EC_a measurements or soil properties, requiring additional soil surface traffic and time and resources spent collecting measurements. Thus, the ability to capture the same amount of EC_a variability with fewer surveys would be beneficial for researchers, agricultural consultants, and producers due to the reduced time and energy spent collecting and processing the data; in addition, it would reduce surface traffic, which could positively contribute to maintaining soil surface health.

6. Conclusions

Identifying in-field variability and suitable methods to delineate soil property variations in AF systems may lead to greater adoption and better management in areas prone to conservation issues (i.e., Ozark Highlands). Previously, EC_a surveys have been minimally applied in AF systems; thus, there is little information regarding the spatiotemporal relationship between repeated EC_a surveys and soil property variations in AF systems, let alone AF systems within regions with similar features or conservation issues to the Ozark Highlands. Results of this study did not support the hypothesis that there would be differences between the EC_a mean, SD, and CV of surveys that were conducted during different weather and growing/non-growing seasons. These results demonstrated that repeated EC_a surveys can be used to assess the spatiotemporal pixel variation of the measured EC_a of a 20-year-old AF system within the Ozark Highlands.

Results of this study supported the hypothesis that monthly EC_a surveys can be grouped into similar functional populations and can be made into zones for precision soil management. Results of this study partially supported the hypothesis that monthly EC_a survey data are correlated with soil-sensor-based VWC and EC_a and soil-sample-based EC, GWC, and pH. Results of this study also partially supported the hypothesis that fewer, strategically timed, and evenly spaced EC_a surveys could be conducted to capture the same amount of overall EC_a variance at the AF site as the 12 monthly EC_a surveys conducted. It is recommended that monthly EC_a surveys are necessary to capture the full EC_a variance at the 0–0.5 m soil depth within a 1-year period. However, to capture the full EC_a variance in the 0–1.6 m soil depth of AF site within a 1-year period, it is recommended to conduct either EC_a surveys every other month or one EC_a survey in the middle of winter and summer to limit soil surface disturbance and to save resources. This study presents novel EC_a surveys for use in precision management in a long-term AF system.

Results of this study provided beneficial and valuable insights on the spatial and temporal variability of EC_a over time. Results demonstrated that a variety of methods can be used to assess spatial and temporal changes in measured EC_a, and that the effects of pedogenic and anthropogenic factors on measured EC_a can be observed in an AF system. Results of this study also demonstrated that EC_a survey data are correlated to multiple soil properties and are used to create precision SMZs in an AF system within the Ozark Highlands. Thus, results of this study provide further evidence of the potential versatility, applicability, and ability of EC_a surveys to quickly and accurately delineate in-field variability in different landscapes with different land uses and management systems.