Introduction
Efficient and accurate methods of measuring spatial variations in soil properties are important for precision agriculture. Mobile sensor systems that collect dense datasets in situ can provide several advantages over soil sampling and laboratory analysis, including lower cost, increased capacity and efficiency, and more timely results. Mobile measurements of apparent soil electrical conductivity (ECa) have become widely used to map soil variability. Soil ECa responds to a number of important soil physical and chemical properties. While in saline soils most of the variation in ECa can be related to salt concentration (Williams and Baker, Reference Williams and Baker1982), variations in non-saline soils are primarily a function of soil texture, moisture content, bulk density, and cation exchange capacity (Corwin and Lesch, Reference Corwin and Lesch2005). An empirical model describing the relationship between ECa and soil properties was developed by Rhoades et al. (Reference Rhoades, Manteghi, Shrouse and Alves1989), whereby ECa was defined as a function of soil water content, the electrical conductivity of the soil water, soil bulk density, and the electrical conductivity of the soil particles.
A basic application of ECa data is to provide qualitative information on soil spatial variability. Such information may help guide soil mapping and assist producers in establishing within-field management zones. In a more quantitative application, ECa may be calibrated to, and used to map, variations in soil parameters such as salinity, texture, or moisture. However, such calibration is complicated because ECa sensors respond to the true bulk soil conductivity (EC) down to a sensor-dependent measurement depth, as weighted by a nonlinear response function (Sudduth et al., Reference Sudduth, Kitchen, Wiebold, Batchelor, Bollero and Bullock2005).
In some applications, such as providing spatial data for crop or environmental modelling, the goal may be to use ECa to estimate physical or chemical properties in specific soil layers. If two or more ECa measurements with different depth response functions are available, then mathematical inversion of the theoretical models of ECa instrument response can be used to reconstruct the EC in different layers of the soil profile. Those layer EC values can then be calibrated to layer soil properties. Some inversion solutions (e.g., Hendrickx et al., Reference Hendrickx, Borchers, Corwin, Lesch, Hilgendorf and Schlue2002; Triantafilis and Monteiro Santos, Reference Triantafilis and Monteiro Santos2010) have used data from an ECa sensor held at multiple heights above the ground at each sampling point. Practical application of such approaches for mobilized mapping of large areas is obviously limited. The alternative approach investigated in this study was to apply the mathematical inversion process described by Monteiro Santos et al. (Reference Monteiro Santos, Triantafilis, Bruzgulis and Roe2010) to data from an instrument that provided the required multiple readings simultaneously while traversing a field.
The goal of this research was to evaluate the use of an inversion approach to estimate soil properties for several fields in Missouri, USA. Specific objectives were to:
-
o Implement an inversion solution to extract layer EC values,
-
o Validate layer EC values using data obtained with an ECa sensing penetrometer, and
-
o Use layer conductivities to estimate soil properties.
Materials and methods
Data were collected on four crop production fields in Missouri, USA. Field 1 (F1, 35 ha) and Field 3 (F3, 20 ha) were near Centralia (39.23° N, 92.12° W), within the Central Claypan Major Land Resource Area. Claypan soils are distinguished by a greater than 50% increase in clay from topsoil to subsoil horizons. Topsoil textures at F1 and F3 ranged from silt loam to silty clay loam. The subsoil claypan horizon(s) were silty clay loam, silty clay or clay, and contained as much as 50 to 60% clay. Within-field differences in topsoil depth above the claypan (<0.1 m to>1.0 m), coupled with the marked difference in clay between topsoil and subsoil often lead to large differences in profile clay content (Sudduth et al., Reference Sudduth, Kitchen, Bollero, Bullock and Wiebold2003).
The Marsh and Lee fields were part of the University of Missouri Fisher Delta Research Center near Portageville (36.42° N, 89.69° W). These fields are within the New Madrid Seismic Zone and the combination of alluvial, eolian, and seismic activity over the years has resulted in highly variable soils. Soils of the Marsh field (10 ha) were primarily classified as sandy loam, loam, and silt loam. Typical of the region, the field also contained areas of high sand content (i.e., “sand blows”) too small to show up in a soil map. Soils of the Lee field (8 ha) were finer textured, with the majority of the field classified as clay except for a small area of silt loam in the southwest corner.
Mobile ECa data were collected on all fields in March and April 2016. A Veris 3100 (Veris Technologies, Inc., Salina, KS, USA) sensing system was used to collect data at F1 and F3. The six rolling coulter electrodes of the Veris 3100 allow it to collect two simultaneous ECa datasets, a shallow reading (Vsh) with 90% of the response from the soil profile to a 0.3 m depth and a deep reading (Vdp) with 90% of the response from the soil profile to a 1.0 m depth (Sudduth et al., Reference Sudduth, Kitchen, Bollero, Bullock and Wiebold2003). Additional details on the configuration and operation of the electrode array can be found in Barker (Reference Barker1989). Soil ECa in milliSiemens per meter (mS m−1) was recorded on a 1-s interval along transects nominally spaced 10 m apart. A Veris MSP3 system was used to collect data at F3, Marsh, and Lee, also on a 1-s interval and a 10-m transect spacing. The MSP3 system includes optical and pH sensing; however the ECa sensing components are functionally similar to the 3100 with some differences in spacing of the coulter electrodes (and therefore in theoretical depth response functions) to accommodate the additional sensing elements.
Within several days of mobile ECa data collection, profile ECa data were obtained with a Veris P4000 probe. This probe measured soil spectral reflectance through a sapphire window, ECa from dipole contacts, and penetration resistance from a load cell force sensor; however only the ECa data were used in this study. Measurements were obtained at a nominal 20 Hz rate as the probe was hydraulically pushed into the soil to at least 90 cm depth. Output data were an average of 25 raw measurements and were obtained on approximately 4 to 7 cm depth increments depending on insertion speed of the probe. These profile ECa data were obtained from 7 to 19 calibration locations within each field that were chosen to cover the range in mobile ECa data values obtained. Soil texture data (sand, silt, and clay fractions) from laboratory analysis were available by soil horizon for the calibration locations at F1, Marsh, and Lee. Additional laboratory data including cation exchange capacity (CEC), bulk density, and organic C were available at F1.
Veris ECa data were inverted using the invVERIS software package (EMTOMO, Odivelas, Portugal). This software uses a spatially-constrained one-dimensional (i.e., depth) inversion algorithm as described by Monteiro Santos (Reference Monteiro Santos2004) and Monteiro Santos et al. (Reference Monteiro Santos, Triantafilis and Bruzgulis2011) to generate a quasi-3D solution that can be mapped as EC values of discrete depth slices. The software requires input of the usual X-Y-ECa data collected in a field survey along with auxiliary information including dimensions of the electrode array, initial condition estimates of layer EC, and a damping factor used in the iterative solution. Multiple software runs were done for each dataset, adjusting initial conditions and damping factor to optimize the model fit. Then, layer conductivities on a 0.1 m depth increment were output from the software and values coinciding with the calibration locations were extracted.
Inversion model performance was validated by comparison of these data with ECa data measured by the Veris P4000 at those same locations and averaged to the same 0.1 m increments. Finally, layer EC values at the calibration locations were related to measured soil properties using linear regression.
Results and discussion
Maps of Vdp ECa data for each field (Figure 1) show the extent of spatial variability present. For all fields, Vsh data (not shown) were strongly correlated (r=0.70 to 0.97) with Vdp data. Higher correlations between Vdp and Vsh were seen for the Marsh and Lee fields where soils were generally less layered.
Maps of deep (0–1.0 m) ECa for each study field.
The invVERIS software provided qualitatively reasonable inversion solutions for all fields. For example, Figure 2 is an invVERIS screen capture showing a vertical slice through the inversion solution for F3 using Veris 3100 data. Near the highest elevations, where the field slope is relatively flat, topsoil is seen as a darker, lower EC layer. At the steepest part of the field, near an easting value of 340 m, the surface layer is more conductive, indicating erosion of the topsoil layer and exposure of the underlying claypan, a fact confirmed by field observation. Lower on the slope, areas of thicker topsoil are present, due to deposition of topsoil after movement from the upslope eroded area. At the bottom of the slope the thickest layer of low-EC soil is found, again consistent with field observation.
Cross-sectional view of the landscape of Field 3 showing the inversion solution of Veris 3100 data.
Inversion solutions were quantitatively validated by comparing them with measured ECa profiles from the Veris P4000 probe. Table 1 shows validation statistics for each field and ECa dataset over the 0.9 m depth of the P4000 data. For two of the four fields (F3, Lee), better results were obtained by removing data from the top 10-cm increment. In those cases the surface soil was loose and relatively dry at the time of P4000 data collection, and P4000 ECa was much lower than inversion-estimated EC in the surface 10 cm (Figure 3). Thus, this difference was attributed to measurement issues, not problems with the inversion process. For the Lee field, data were also truncated to a 0.7 m depth as actual and inversion-estimated conductivity generally diverged below that depth. Perhaps the much higher ECa at this field (Table 1), which would cause a reduced response depth (Barker, Reference Barker1989) led to a poorer inversion solution at greater depths.
Relationship of inversion-estimated to measured layer conductivities at Field 3, showing greater divergence for data in the 0–0.1 m depth layer that was attributed to loose soil due to tillage just prior to sensing.
Validation statistics for ECa inversion results for different fields and sensors
* Truncated range: 0.1–0.9 m for F3 and 0.1–0.7 m for Lee
Soil texture components (i.e., clay and sand fractions) were estimated by 0.1-m layer for F1, Marsh, and Lee, using linear regressions on P4000 measurements and inversion-estimated conductivity values. Good texture estimates were obtained for the Marsh and Lee fields, with clay positively related to layer conductivity and sand negatively related (Figure 4). Clay fraction was estimated more accurately than sand fraction (Table 2). Estimates based on inversion data were less accurate than those based on P4000 data, with a greater difference for the Marsh field, where inversion estimates were less accurate (Table 1), likely due to the more pronounced spatial variability at that site (Figure 1). These results suggest potential for ECa inversion results to be used for layer texture estimation on alluvial soils, subject to verification at additional field sites. In contrast, clay fraction estimation based on layer conductivities was not successful for F1, the one claypan soil field where soil data were available. Deeper soil layers there exhibited a higher than expected EC for their clay content, suggesting that another soil variable was confounding the EC-clay relationship (Figure 4). Inversion-estimated EC was not strongly related (r<0.17) to any other measured soil variable (CEC, bulk density, soil moisture, and organic C) in the existing dataset, suggesting those were not the cause of the EC-clay misfit. Furthermore, salinity is not an issue at any of the measured fields. Thus, a more complete set of soil variables needs to be examined in future research.
Comparison of profile texture variations to measured and inversion-estimated layer conductivities for three calibration locations in different fields.
Statistics for soil texture component estimation using different ECa data sources
Conclusions
In this study, a model inversion approach was used to calculate soil layer conductivities for estimation of soil texture fractions.
-
o The invVERIS commercial software package was generally successful in generating reasonable layer EC values. Iterative adjustment of initial conditions and other input parameters was required at times to obtain best results.
-
o Layer EC data were validated by comparison with field data from an ECa-sensing penetrometer. Strong linear relationships (r2≥0.76) were obtained in all cases, although it was sometimes necessary to truncate the depth range because of site-specific sensing limitations.
-
o Layer EC data successfully estimated soil texture fractions in the alluvial fields examined. This was not the case for a claypan soil field where there appeared to be parameters other than texture strongly affecting the EC response.
Acknowledgements and disclaimer
We acknowledge Fernando Monteiro Santos for his assistance with the invVERIS software. Earl Vories provided access to the Portageville fields. He also assisted with ECa data collection, as did Kristen Veum, Matt Yost, Kurt Holiman and Matt Volkmann.
Mention of trade names or commercial products in this paper is solely for the purpose of providing specific information and does not imply recommendation or endorsement by the United States Department of Agriculture.
