Integrated Interpretation of Electrical Resistivity Tomography for Evaporite Rock Exploration: A Case Study of the Messinian Gypsum in the Sorbas Basin (Almería, Spain)
Round 1
Reviewer 1 Report
First of all, I would like to comment on the English grammar. The authors abused the use of passive voice in their text, which is almost totally written in the passive voice. Yes, it is not a grammar error, but most readers prefer it because it is clearer and more straightforward. I suggest authors utilize any English grammar editing tool or service before sending the revised text. That would enhance your paper a lot.
My recommendation is a major revision. Some critical points need to be addressed before the paper's approval.
I suggest changing the title to a more meaningful one related to the work done: interpreting ERT inversions with a priori information.
Electrical Resistivity Tomography Integrated Interpretation for Evaporite Rocks Exploration. Case Study at Sorbas Basin Messinian Gypsum (Almería, Spain).
I have made several annotations in the pdf. This document will address only the major problems and associated suggestions to correct them.
Introduction:
As the papers deal with integrated geophysical interpretation, authors should cite other methods used for gypsum exploration and then better characterize the problem.
For the same reason authors should introduce two sections:
Data acquisition and data inversion. The way these themes were treated in the text needs to be revised for an article on applied geophysics. These sections are mandatory!
I suggest using a similar paper published recently in the Journal of Applied Geophysics as a guide. Bellmunt, Fabian, et al. "Sediment characterization in deltas using electrical resistivity tomography: The Ebro delta case." Journal of Applied Geophysics 196 (2022): 104520.
Experts have reviewed this paper and helped the authors appropriately treat these two essential sections when dealing with geophysical surveys/interpretation.
Data Acquisition
It is necessary to describe both resistivity and seismic surveys.
Data Inversion
Also important is to describe the inversions procedures. This is critical to correctly understanding the final models and detecting the strengths and weaknesses of your final models (see the corresponding section in the recommended paper). I will come back later when discussing profile 6.
You also have to describe and explain all parameters of the inversion presented in model figures (e.g., Fig 13), including the L2 parameter.
You should also include a description of how you obtained the Vs velocities.
Curiosity: Why do you use Vs only? Vp is always easier to acquire and has several studies correlating velocity and resistivity. Moreover, you could estimate the Vp/Vs ratio, bringing much more geological information.
By the way, why use tables when you could correlate Vs and res by crossploting them? It bothers me to make a semiquantitative interpretation when estimating physical properties' numbers.
Discussion
The major problem is Profile 6 - your interpretation of the inversion results contradicts the basics and good practices of the inversion problem. I will not go through an extensive explanation. Still, when dealing with the inverse problems, the central assumption is that the model MUST fit data within an acceptable small percentage error, and with a 107% error, your model does not explain the data!
Therefore, removing this profile and associated discussions from the paper is mandatory.
Comments for author File: 
Comments.pdf
Author Response
Please see the attached PDF file with the response to the reviewer.
Author Response File: Author Response.pdf
Reviewer 2 Report
Few remarks:
Line 46 : replace beds by uniform layers.
Line 70: this sentence is confusing. At the start of the paragraph, you state that you are talking about ERT, then you state that it is important to know tthe geophysical technique. Possibly, what you mean is that the various geological parameters you mentionned before have an impact on the geophysical technique you selected.
Line 88: replace geological cartograph by geological map
Line 90: replace thin films by thin sections
Author Response
Please see the attached PDF file with the response to the reviewer.
Author Response File: Author Response.pdf
Reviewer 3 Report
The manuscript has been written with great care and presents the contextualization of the study area and objectives in a very clear and illustrative way. The methods are presented satisfactorily and with the necessary rigor. The illustrations are numerous and of high quality, fundamental to the understanding of the study. The cross-referencing of stratigraphic and lithofaciological data with geophysical results proportional to the formulation of feasible and robust discussions. The incorporation of information about the deposit genesis (diagenesis) is quite feasible for the analysis of the geophysical data. The study presents conclusions that reflect the data presented and the local geological and metallogenetic context. Great article, congratulations to the authors. The one issue worthy of addition is the presentation of ERT as a technique and DC Resistivity as a method (first method and then technique).
Author Response
Please see the attached PDF file with the response to the reviewer.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
I congratulate the authors' efforts in improving the paper. They have implemented almost all the suggestions. But unfortunately, they insist on keeping results against the theory of the inversion problem and its good practices.
In their response, they state, "...Thus, we consider that this profile 6 is an interesting topic for discussion, and we do not see that because it has a high RSM value it should "theoretically" be eliminated."
Again, a high RMS error indicates your model does not fit your geophysical dataset. So from the point of view of your geophysical dataset, it's a wrong model. Integrations with other geological-geophysical data for interpretation is a step used only for geophysical models that fits the data!
To improve your knowledge of maths and good practices of ERT data inversion, I suggest reading the seminal paper of Oldenburg and Li - Inversion of induced polarization data - almost all commercial inversions are derived from their work. OLDENBURG, Douglas W.; LI, Yaoguo. Inversion of induced polarization data. Geophysics, v. 59, n. 9, p. 1327-1341, 1994.
It would help if you also examine:
Auken et al. paper - A survey of current trends in near-surface electrical and electromagnetic methods. AUKEN, Esben et al. A survey of current trends in near-surface electrical and electromagnetic methods. Geophysics, v. 71, n. 5, p. G249-G260, 2006.
WISÉN, Roger et al. Experience from two resistivity inversion techniques applied in three cases of geotechnical site investigation. Journal of geotechnical and geoenvironmental engineering, v. 134, n. 12, p. 1730, 2008.
Now you have included the explanation extracted from the EarthImager manual (too similar, you should try to modify the text more), but you need to understand it correctly.
L2 is another measure of data misfit, but it is related to the convergence of the inversion process. See item 4.3.5 L2 norm (page 62) "...When the normalized L2-norm reduces to the unity (1.0) or smaller, the inversion is converged..."
Indeed, L2 is also a stop criterion that users apply to inversion. This is clearly stated in item 4.3.10. (page 63) Percentage of noise "....With the L2-norm stop criterion enabled, a too large noisy estimate produces a too smooth model. A too small noisy estimate forces the inversion to fit data noise and produce artifacts.... ¨" Observe that the parameter L2 influences the model's roughness, depending on the errors.
If L2 is not applicable, the inversion could run indefinitely, as explained in item 4.3.12. Suppress noisy data (page 63) "...If you have noisy data or a slow convergence, you should check this box. This scheme gives less weight to noisy and hard-to-fit data. This will not improve RMS convergence because RMS is not a weighted quantity. Instead, the L2-norm convergence should improve dramatically..."
Finally, L2 is a stop parameter that controls the models' roughness. See item 8.5 Robust Inversion (page 133) - "...Both least squares inversion and smooth model inversion are based on a L2-norm criterion.Robust inversion is based on the assumption of exponential distribution of data errors and minimizes an L1-norm of combined data misfit and model stabilizing functional...¨"
To understand the differences between L2 and L1, their uses, and the comparison of results, please refer to Auken et al. (2006). Note that reaching a low RMS misfit is the only way to accept a model that indicates that you achieved an acceptable one.
Looking more carefully at your data pseudosections in Appendix A, I realized the whole problem is associated with a poor dataset. Too much noise and weird features in the data.
From my experience, this is related to the acquisition method you have chosen in a very resistive ground environment. Traditional ERT systems with metal stakes do not perform well in these situations. Have you measured the contact resistance of the electrodes? It looks like it was very high. I have worked with deep IP surveys for mineral exploration and huge transmitters with 6-10 KVa, that constantly failed at some terrains (similar to yours). To minimize the problem, we had to carry several gallons of saltwater solution to reduce the contact resistance to have an interpretable dataset.
Why did you have not used a pulled array system with capacitive electrodes? They are a much more appropriate system for your geological terrain. See
HÖRDT, Andreas; WEIDELT, Peter; PRZYKLENK, Anita. Contact impedance of grounded and capacitive electrodes. Geophysical Journal International, v. 193, n. 1, p. 187-196, 2013.
My suggestion to work with this dataset: write a paper comparing it with a new one collected at the same sites with a pulled-array system. Such a paper would be influential to the geological community, as many use geophysics as a black box and are unaware that the maxima garbage in, garbage out, is the truth in math-related problems.
Best regards.
Author Response
Thank you very much for all your comments and suggestions that will help us to continue to deepen this type of studies. As the Reviewer 1 suggest, perhaps the best thing to do is to carry out some more tests (more data) in that area for a possible article in the future, about the profile with RMS 106%. Therefore, we are going to remove that profile (ERT 5a) from figure 13. We have removed all the references and discussion on that profile in the text.
