the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 23 Jun 2025
Improving the relationship between soil texture and large-scale electromagnetic induction surveys using a direct current electrical resistivity calibration
Abstract. Ground-based electromagnetic induction (EMI) surveys can be used to infer soil properties and (by extension) support nutrient loss risk assessments of agricultural fields. The transport of nutrients from an agricultural field to surrounding surface waters depends on the hydrologic connectivity between the two systems, largely controlled by soil texture. Preexisting soil texture maps and associated soil drainage classifications are often used as proxy information to assess the potential for lateral migration of nutrients in the groundwater; however, the resolution of these maps is inadequate at the scale of individual fields. In this study, we evaluated whether the relationship between EMI data and soil texture was improved by calibrating the apparent electrical conductivity measured by an EMI sensor with a 2D electrical resistivity imaging (ERI) survey. The joint geophysical survey was performed across a ~1-ha field in Princess Anne, Maryland, United States. A calibration-inversion-comparison framework is presented that calibrates the EMI measurements using ERI conductivity models and subsequently inverts the EMI data. A robust validation scheme compared the calibrated and not calibrated EMI conductivity models against grain size, core-scale conductivity measurements and an ERI survey performed roughly 80 m from the first. This study shows that the calibration of EMI data with an ERI profile is significantly improves the quantitative relationship between EMI-derived electrical conductivity and representative soil properties, ensuring a finer-resolution proxy soil map for evaluating subsurface nutrient transport from agricultural fields.
- Preprint
(3094 KB) - Metadata XML
-
Supplement
(281 KB) - BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2025-2434', Anonymous Referee #1, 25 Jul 2025
-
RC2: 'Comment on egusphere-2025-2434', Anonymous Referee #2, 01 Aug 2025
The paper addresses a very important issue and tries to make a step forward towards a quantitative use of EMI data for soil characterization. Yet I have found the paper content very weak, for a number of reasons that I try and list below. In particular, there is a clear lack of consideration for the spatial scale of measurements – EMI, ERT, cores, lab samples – that hinders any further serious understanding of the different results that emerge from the study: indeed above and beyond the attempt to correct the EMI inversion on the basis of ERT data – considered as “ground truth”.
Here below are a number of comments:
(1): Table 1: The definition of sigma_a,sim (inv) is unclear: are we talking of the inverted EMI conductivity from calibrated or non-calibrated apparent conductivity data? (the whole thing is a bit obscure)
(2) Too much confidence on ERT – ERI as the authors name it – with no concern for the uncertainty concerning the ERT inversion itself.
(3) The authors attribute the (strong) discrepancy between EMI-derived sigma sections and the ERT-derived sigma sections to (unspecified) strong disturbances in EMI data. I am convinced there is more than this to the difference in electrical conductivity imaging between the two methods. Primarily, I would say, it is a matter of the physical process that excites electrical current in either case: EM induction in one case, galvanic current injection in the second case. I would encourage the authors to use a word of caution in this respect in the revised paper, highlighting the intricacies underlying these physical processes.
(4) Information about homogeneity of soil saturation (lines 152-153: “The soil below 2.5 m depth is likely anaerobic due to periods of persistent saturation. While the water levels were not directly measured during the surveys, a chroma value ≤ 2 likely represents the seasonal-high water table at the site. ") and pore water salinity?
(4) The comparison between ERT lines and samples is only partial, as drillings only go down to 3 meters!
(5) Given the very small range of electrical conductivity in these 3 meters, the calibration towards soil properties is of course very weak.
(6) I would have used much higher resolution for ERT, with shorter ERT lines together with the longer lines, in order to focus on the 3 m depth (mostly unsaturated, as far as it can be deducted from the indirect information reported above).
(7) Fig.2b: the different colors of the soil should have some meaning!
(8) In subsection 3.1 The calibration-inversion-comparison (CIC) framework the references to Figures are probably all relevant to Figure 1 (not to Figure 3), albeit a general confusion exists (Fig.1d – or Fig3.d – does not exist!).
(9) L.116-117: “the study field contains localized regions of elevation change (<0.5 m) that likely influence runoff generation.” This type of microtopography cannot induce runoff, but rather ponding and thus local heterogenous infiltration processes.
(10) Fig.3: the dots indicating EMI measurements are indicated in red in the legend, but are gray in the image.
(11) L. 173-175: “A depth of investigation (DOI) index was computed (Oldenburg & Li, 1999) to define regions of the ERI conductivity model to exclude from the analysis.” The DOI is notoriously an overconservative approach based on a-posteriori sensitivity function: should we adopt it in all ERT investigations, the reliable depth would be of just a couple of electrode separations. In fact an ERT line can provide reliable information down to a much larger depth, namely 1/5 to 1/4 of the ERT line total length. At any rate at this point the authors should specify the average depth at which ERT data have been considered reliable. In comparison with the ERT images of Figure 5, I presume the DOI line (should be shown, it is easily produced by ResIpy – and so it says in the text, i.e. that the DOI line is shown, but I cannot see it in the figure) only goes down a couple of meters.
(12) Note that the electrode spacing is 1 m for both ERT lines: this implies that the maximum attainable resolution is 1 m, and as soon as depth increases such resolution declines sharply. Therefore it is really hard to try correlating the ERT pixel values with the evidence from geoprobe cores. The authors shall spend at least a word of caution in this direction: again, it would have been much better to use shorter, high resolution ERT lines (e.g. with 0.2 m electrode spacing) in roll-along configuration.
(13) EMI inversion goes to 6.3 m instead, beyond the used ERT data, I presume.
(14) L 221-223: “However, a lower α value, away from the elbow of the l-curve, was necessary in this study to ensure minimal smoothing of the calibration effects, which masked the relationship with the vertical soil profiles (discussed later). “ I feel this is forcing the inversion to providing an apparent resolution that is not attainable with the Dualem 421S (that, from the name, has coil separations of 1 m, 2 m and 4 m)!
(15) L.245-247: “To remove the effects of variably saturated soils, measurements were obtained under fully saturated conditions using synthetic groundwater with a constant, low fluid conductivity (0.1 mS/m) to evaluate the soil texture controls on the core-scale electrical conductivity.” This is a bit puzzling, as in situ measurements are indeed run under variably (and unknown!) moisture content conditions.
(16) L. 252: “Therefore, the core-scale measurements were conducted on the last 0.13 of the 3 m depth soil cores to capture conductivity likely from texture variation, given the known saturation and low fluid conductivity.” It is unclear to me how this “known saturation” has been assessed as the water table has not been monitored on the site. Also, it is unclear why the authors only measured electrical conductivity of soil samples from the bottom (3 m below ground) samples. Indeed a more complete characterization of the cores, also in terms of electrical conductivity (both saturated and unsaturated, with a complete characterization of the pressure-saturation curve as well) would have been highly beneficial.
(17) Figure 5: it is visually apparent how there is a strong difference in electrical conductivity values measured on geoprobe samples taken along lines ERT1 and ERT2: yet the values in the corresponding locations in the two ERT lines are practically the same. This alone casts some doubts on any type of correlation attempt. I think the authors need to seriously address this evidence and convince the reader. This is particularly serious as these two datasets (ERT-measured conductivities and lab-based conductivities) are used as “ground-truth” to correct the EMI data, and yet they disagree with each other (in particular, the calibration versus the validation dataset).
(18) The plots in Figures 6 and 7 must obviously show different variables, but while Figure 6 has the axis properly tagged, so is not for Figure 7… and the doubt remains that the scatter plots for uncalibrated sigma_a should plot in the same way in Figures 6 and Figure 7, but this is not so. Please clarify, and make the process clearer to the reader.
(19) L 319: “A striking result of the effect of the EMI calibration is highlighted in Fig. 8a-8d, with marked increases in correlations between all d finer classes and σ EMI [c].” I would be much more cautious in this respect. The correlation remains really poor no matter if the sigma values come from corrected o non corrected EMI inversions. This is obvious, as a visual inspection of the core descriptions in Figure 4 do not find any correlation in the ERT lines in Figure 5 (that are anyway considered as “true” values of conductivity).
(20) I find the comparison in Figure 9 problematic: it is true that the correction seems to indicate a better correlation between lab samples and field EMI results: but the values of either dataset are totally off!
Citation: https://doi.org/10.5194/egusphere-2025-2434-RC2
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 414 | 40 | 14 | 468 | 19 | 11 | 27 |
- HTML: 414
- PDF: 40
- XML: 14
- Total: 468
- Supplement: 19
- BibTeX: 11
- EndNote: 27
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This manuscript is interesting and built upon existing literature on the use of calibration for EMI. The Authors described the process in detail with great pedagogy and highlighted the advantage of calibrated EMI to relate it to soil texture for agricultural application. I believe it will make an interesting manuscript to be published in SOIL, certainly with the greater use of these sensors with the rise of agrogeophysics. Please find below some major and minor comments to be addressed.
The title of the paper focuses on soil texture and EMI but, if I understand well the driver below, it's an agricultural problem that is the reason this study was carried out. I would try to highlight more the agricultural context in the introduction (as you already come back to it in the conclusion). I think highlighting the value of your study in this context can increase its impact. I also propose a rephrased title that highlights this context (see below).
I would more clearly make the distinction between apparent and inverted EC values throughout the manuscript. In literature, often ECa is directly compared to soil texture but here only inverted EC is used for comparison. I would add information about the distinction apparent/inverted and discuss the choice of the use of inverted ECa for comparison.
In general the manuscripts read well but in some places missword endangered the meaning of the sentence. Please carefully re-read the manuscript to ensure its meaning is as you intended.
Detailed comments:
title: the agricultural nutrient loss aspect does not appear in the title while it's still a main driver of the manuscript. In addition, a lot of literature already exists on EMI to map soil texture so it does not feel very new (even though the relationship still remains challenging I acknowledge). Maybe rephrase as: "How resistivity calibrating EMI surveys, improve soil texture prediction and nutrient losses from agricultural fields" or "Agricultural nutrient losses: improvement from resistivity calibrated large-scale EMI surveys" or smth similar.
L23: 'is significantly improves' -> improving?
L32: 'surface drainage tiles' -> also subsurface drainage tiles no?
L46: What about temperature? (2% increase of EC with each degC)
L56: 'electrical resistivity geophysical method' -> 'ERI'?
L56: 'approximate' -> I am not sure about the term here. It actually measures a property linked to sigma, then uses an approximation (LIN) to get an apparent sigma.
L57: 'the electromagnetic inductions methods ...' -> here I would have expected a short explanation of the EMI method (induction of eddy currents proportional to soil sigma, measuring primary and secondary electromagnetic field,...), to put it into perspective with ERI as it's the first part of the sentence. Only then I would go into the difficulty of the environmental influences. Note that ERI too can suffer from 'environmental influence' (e.g. cable coupling) but much less than EMI ofc.
I would specify somewhere here that we are dealing only with frequency domain EMI and not time-domain (TEM).
L70: "the impact of noise" -> from my experience, calibrating EMI with ERI transect will bring two things: (1) it will ensure that the sigma_a from EMI is linked to the soil sigma and (2) it will ensure all different coils configuration have the same relationship to soil sigma. What we often find in commercially available instruments is that each coil is calibrated individually by the manufacturer over a homogeneous ground. But these independent calibrations can drift in time and do not guarantee that the data from all coils together is consistent. When inverting uncalibrated data, you can often see one coil configuration worse fitted than others. Calibrating all coils to the same object (ERI) ensures they can all be fitted together within an EMI inversion routine.
I would also emphasize here the difference between an apparent value and an inverted value. Often in study, apparent values are compared to depth-specific textural property but that's not what is done in this study (and I am happy with that). More explanation on why EMI data are inverted to make the link with soil texture is needed.
I think a difference must be made here if the calibration improves the EMI inversion or if the calibration improves the sigma_a approximation. Von Hebel et al. (2019) and Hanssens et al. (2019) both propose an 'equivalent EC' or 'robust ECa' method to obtain a more robust sigma_a without the limitation of the low induction number (LIN) approximation.
L112: add truncated gps wgs84 coordinates
fig2: add CRS used for the northing/easting (utm?)
L112: "water tables" -> "water table"
fig2: you mention the elevation in the field to change slightly. If you have an elevation map of the field. I would add it here, in addition to subplot b. I would then remove subplot a and just put the us map as inlet on top right of subplot b.
L134: the instrument was "walked in" so it was carried? at which distance from the ground? Was this distance negligible or included in the inversion?
L136: 1 m spacing ERI will lead to a relatively coarse resolution close to the surface where most of the EMI sensitivity is located. This will also have limited the resolution of the ERI to delineate the soil layers. However, it is able to cross the entire field and as such span a large difference in sigma_a from EMI. Von Hebel et al. (2019) uses 0.25 or 0.5 m electrode spacing. Maybe smth to add to the discussion.
It might be interesting to have the original soil map (without EMI interpretation) for comparison or just juxtapose the boundaries of it
fig4: I would add the water table at the time of the survey in addition to the grey zone.
L158: "pre-calibration" -> I would not call it like that, it's not a calibration. The resistance value measured does not need to be calibrated. Just put "Electrical Resistivity Imaging (ERI)"
L165: "between 1200-1500" -> it wasn't constant? or is that because you filter some data out?
L165: "with ~43% containing" -> rephrase like "for each sequence 43% of measurements were added as reciprocal"
L166: 'current and electrodes' -> 'current and voltage electrodes'
L169: 'ResIpy' -> 'ResIPy', please add version used when mentioning code
L176: these eddy currents are more or less strong depending on the soil conductivity (that's the link to soil sigma!)
L195: "average distance" -> you mean all EMI points within 1.5 m from the ERT profile were average for this profile or you mean that you took the 10 closest points and their average distance to the ERI point was 1.5 m?
L197: where do the 816 simulated models come from? Is that because you have a model every 0.5 m along the ERI line?
L203: "8 mS/" -> "8 mS/m"
L209: "forward model (m)" -> "forward modeled data (m)", the 'forward model' is the operator. Usually m is the vector of model parameters (so sigma values here). f(m) is the forward modeled data = simulated data based on m. So if should be ...d_i - f_i(m) is the i-indice to refer to the coil configuration (and you have one forward model for each coil configuration as they have different sensitivity functions).
L252: "below 2.5 m" -> why is that? isn't it just because of lower sensitivity of the EMI? also aren't completely saturated at 2.5 m depth, so soil moisture variation shouldn't matter => or you mean "above 2.5 m depth" that would make more sense actually
fig5: we salute the use of a uniformly perceived color-scale instead of the usual rainbow/jet. Thank you!
fig7: maybe it would also be nice to have a visual comparison of the transect: inverted ERT, uncalibrated inverted EMI on the same transect, vs calibrated EMI on the same transect.
fig8: why don't we have a 0.38 m sample for the nc case? it seems they matters to improve the relationship when you look at the calibrated plots
fig10: I would add the inverted nc and c transect here too for a visual comparison (even if an visual improvement will be difficult to see)
L365: "...reduced RMSE" -> this shows well that the calibration helps the inversion by "binding" all coil to the same calibration (as I commented above)
L383: yes, this should be emphasized from the start I think
fig11: while it's interesting to see if your calibration equations applied outside of your field and that they improve the inversion; the dataset you have is not the best as you don't have any validation data from these other fields (no ERT or cores). So it's not a really quantitative comparison. I would highlight this limitation.
discussion: I would add to the discussion the effect of soil moisture. You mention that in saturated conditions, EC is driven by texture but often you have an unsaturated zone at the surface, where the EMI sensitivity is the highest. What could be the effect of this unsaturated zone on your results and how the calibration interacts with it (as we see it's mainly the shallow coils that have strong calibration equations).
The approach you propose relies on inversion of EMI data and this inversion carries with it its uncertainty. Numerous papers do not invert ECa and directly correlate it with depth-specific soil texture. Why did you choose to invert the data? What could be the uncertainty inherent to the EMI inversion introduced by this process?
ref: there are two McLachlan et al. (2021), one for boxford and one for emagpy, if emagpy is used for the inversion, please mention the software version.
I encourage the Authors to make their code available on an open platform (github, gitlab, zenodo, ...) and cite it in their work. That will ensure their CIC process is reproducible by others.