the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Technical Note: A validated correction method to quantify organic and inorganic carbon in soils using Rock-Eval® thermal analysis
Abstract. Soils contain large amounts of carbon stored as organic carbon and carbonates. These carbon pools can contribute to climate regulation, and are of primary importance in ensuring proper soil functioning. However, their accurate quantification remains a complex task. Rock-Eval® thermal analysis has emerged as an alternative to classic dry combustion and wet methods, due to its ability to simultaneously provide organic and inorganic carbon measurements on the same subsample. However, it has been observed that Rock-Eval® systematically underestimates the soil organic carbon (SOC), while overestimating the soil inorganic carbon (SIC). In this technical note, we propose a validated correction of both SOC and SIC based on a machine-learning model and using a diverse dataset of 240 soil samples. We show that the proposed correction significantly increases the accuracy of the Rock-Eval® method on the initial dataset, and that it can be successfully applied to data originating from different Rock-Eval® machines, without changing the routine analytical protocol. The transferability of the model allows for its future implementation in the Geoworks software so that Rock-Eval® machines can routinely provide accurate SIC and SOC measurements.
- Preprint
(1347 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
CC1: 'Comment on egusphere-2024-578', Isabel Sonsoles De Soto Garcia, 26 Mar 2024
I find this technical note very interesting. In my opinion, this is an excellent paper because they provide a correction method to determine the concentration of SOC and SIC from Rock-Eval® thermal analysis. Therefore, its future implementation can routinely provide accurate SIC and SOC results without the need to previously manipulate the samples, which greatly facilitates the complex study of C in soils. The proposed methodology can be an alternative to the classic dry and wet standardized methods, especially in carbonate soils, where this methods present a series of drawbacks.
The work is well-written and clear, the reasoning is logical, and the conclusions follow naturally from the results. In addition, they have used an appropriate methodology, using the results of 240 samples of French agricultural soils which cover a wide range of particle size distribution and SIC and SOC content. I consider that the paper should be suitable for publication. Congratulations to the authors on a very nice study
Citation: https://doi.org/10.5194/egusphere-2024-578-CC1 -
RC1: 'Comment on egusphere-2024-578', Anonymous Referee #1, 02 Apr 2024
General comments
This technical note presents a correction method to determine soil organic content and soil inorganic content in single analysis. Using Rock-Eval thermal analysis offers some benefits over dry combustion method with elemental analyzers, where pretreatments or calculation are needed to get the two values of organic and inorganic carbon contents. Rock Eval eliminates the chances of calculation error, and experimental error associated with the elemental analysis method. The main conclusion by the authors is that Rock-Eval analysis can accurately determine SOC and SIC contents thanks to corrections based on a machine learning model. This result is highly promising to facilitate C studies in calcareous soils, however I have some reservations regarding the description, the application and the validation of these corrections. How did you definy the term “validated” ? Please explicit your criteria, the domain of validity and the correction itself? I hope that the authors can amend their manuscript to explicit the corrections, enhance the quality of the methodological work and discuss more their results in order to impede the paper to be taken as an advertisement for Rock-Eval® equipment with the Geoworks software. Please find some specific comments and questions below.
Specific comments
L.8 I did not understand the initial data set? What does it mean?
Please add few lines on SIC, their distribution, the role of SIC in the soil properties, the eventual management of SIC (limestone inputs, Enhanced Rock Weathering…). Why is it important to consider the SIC contents as the SOC contents.
You did mention alternative thermal analysis to quantify SOC and SIC content in a single subsample, but you did not discuss your results with Rock-Eval against this alternative thermal method. Why a CHN elemental analyser (and we do not know which one) was selected for comparison and not an analyser with a thermal analysis ramp (such a LECO)? Does using Rock-Eval thermal analysis offer extra benefits over other thermal analysis that exists? Please add some words on the availability of the Rock-Eval equipment and the cost of the analysis. Is it a method easily available in several soil analysis labs?
I did not understand in L. 36-37, if Koorneef et al. has proposed some corrections to estimate SOC and SIC and why you did not discuss your results as you did with Hazera et al. corrections with Koorneef corrections. Did you suggest that Koorneef et al. and Hazera et al. did not present validate results in their paper? Please discuss the domain of validity of each study and how you improve the method.
Additional information on the LAS analysis should be add to be more convincing. We have to trust the authors about the accuracy of the LAS data, no mention on the incertitude or if there is any replicates. The method used to get SOC and SIC content are not mentioned (calculation, direct measures with or without pretreatments for SIC or SOC ?). As it is a methodological work, it should be nice to insert some reference geostandard samples in the data set to check the accuracy of the LAS and Rock-Eval® analysis.
Are they any replicates on the LAS and Rock Eval analysis? What is the incertitude associated to each of the analysis? This is a methodological work, so readers expect high standard on the description of the methods.
L.70 Please can you add some indication of the soil aliquots. You only say ca. 60 mg. I presume it is ground soil (200 µm). Is it OK for any soils, even the soil with very high SIC and very low SOC content? Or the inverse? What are the limit of C detection ? limit of C content of each C pools to be properly measured. Is it possible to have a saturation of the signal for one of the C pool and in a same time to be at the lowest detection limit for the other C pool? As your paper is a technical note, this consideration are worth to be add for the future users of your method.
L.96-97 How many values of SIC inferior to zero were in the data set? SIC could be calculated inferior to zero for what range of reference SIC values?
L.98 « all the RE parameters » and L. 106 “after a correlation analysis….” Please specify the RE6 parameters tested. Furthermore if the reader is not familiar with Rock Eval analysis, he/she cannot understand what you mean without a minimum of explanation on these parameters.
L.105 How have you proceed to choose the calibration and the validation sample sets?
L.114 are you sure that the conditions given in the application of the correction suggested by Hazera are OK for all of the soil samples of your data set (soils enriched in poorly degraded organic compounds)? Please specify which the land uses are represented in your data set and why the soils are probably enriched in poorly degraded organic compounds.
Some clarification and homogenisation should be done between the term and the abbreviation using CHN, LAS, Elemental analyser. For example L. 145 “reference CHN value”, but in the figures LAS is mentioned. The same for TOC-RE6 and SOC-RE6. Please try to be clear with all the abbreviation to help the reader to follow.
L.135 The sign “sum” is missing. Please correct the Equation.
L.151 I did not understand if you realised three different models, with three different corrections (non carbonated, carbonated, all soils). I did not see the discussion on the different corrections. One of my problem with your work is that these corrections were not explicit. They will be included in the Geoworks software OK but we do not see how it is calculated. The machine learning procedure is clear but the statistical corrections were somehow cloudy because not explicit in the text. I am not sure that your work could be reproducible.
Why figure 2, 4 and 5 have different scales? (100 g kg-1 and 50 gkg-1 for SIC and SOC respectively in Fig.2 and 70 and 35 in Figure 4). Please be consistent.
The predicted corrections of the ML model were not explicit. Is it possible to do so? Why not?
L.160 “test data” what do you mean? Data in Hazera et al. were compared with reference data, but the data were not statistically adjusted as you did, the corrections proposed were defined from literature (e.g. Disnar et al. 2003).
Please give a domain of validity of the method. Why do you say is it the first method to be validated? The corrective approach of Hazera did not work as well as yours on your data set, but it seems OK for their own data set. Please discuss and give clue to explain what could explain this discrepancy. Is it a question of SOC and SIC range, of type of soil, or of land use or land cover, type of SOC, type of SIC?
Please discuss more your results, e.g. the incertitude associated to the values to the reference data and to the RockEval data. What are the limits and the perspectives of your work. Could these corrections be applied also on soil fractions? On any king of soil with different organic matter and different calcareous minerals? Are 160 French carbonated soils enough? Does it need further calibration, probably yes on the soil fractions as it is tested in Koorneef et al. What are the possible perspective of this work? You also could further understand why some samples are better predicted than others.
To conclude, I think this work is worth to be published after major revisions. It gives promising result and improvement to measure SOC and SIC in single aliquot for a large range of soils. I regret that the corrections were not explicit (a statistical adjustment but the coefficient were not given and discussed against the coefficients existing in the literature for SOC or SOC and SIC content). It works on 260 French soil samples. The Rock Eval must be corrected by statistics… is it not possible to fix the cycle of analysis to avoid these corrections? I also regret that the discussion on the proposed method is poor and no perspectives of the work are proposed. It looks like that because the statistical adjustment is OK for 260 French soil samples, the problem of measuring SIC and SOC content in soil with Rock Eval is no more a problem for any kind of soil materials. Please discuss more the result to avoid the feeling that the paper is an advertiser for RockEval and Geoworks. This very interesting work is really promising despite my many comments, it deserves further discussion.
Citation: https://doi.org/10.5194/egusphere-2024-578-RC1 -
RC2: 'Comment on egusphere-2024-578', Anonymous Referee #2, 03 Apr 2024
The results presented by Stojanova et al. (2024) constitute a significant contribution to the use of Rock-Eval thermal analysis in soil science, as they provide the first comparative study of correction methods for organic carbon (Corg) and inorganic carbon (Cinorg) contents measured by RE in soil samples.
This question is highly relevant because, from the first RE applications to soil samples, Disnar et al. (2003) already observed a significant discrepancy between organic carbon (Corg) contents measured by RE (TOC parameter) and by elemental analysis (LECO). These authors then proposed an "empirical correction" based on simple linear regressions between RE and LECO measurements using a dataset (n = 100) representative of the main types of horizons (organic, organo-mineral, and mineral) sampled under contrasting pedoclimatic conditions.
The manuscript presented by Stojanova et al. (2024) certainly addresses the shortcomings of this initial approach. Firstly, the authors clearly formalize their objectives to minimize discrepancies between organic carbon (Corg) and inorganic carbon (Cinorg) contents by using elemental measurements as a reference. Secondly, the study is conducted on a large panel of samples (n = 240) covering a wide range of Corg (0-50 g.kg-1) and Cinorg (0-80 g.kg-1). Thirdly, the study involves comparing the performances of several models using objective statistical criteria. Results show that these performances, analyzed for different soil categories, allow the identification of a significantly more efficient model than others, thus providing a simple and effective post-analysis "statistical correction" for the studied soil types.
Reading the manuscript, however, raises some incidental questions that can be shared with the authors to strengthen this technical note.
Among the correction procedures tested by Stojanova et al. (2024), four are proposed by the authors, while the fifth is presented as a "correction model" proposed by Hazera et al. (2023), which constitutes a shortcut and raises an attribution issue. Indeed, the work presented by Hazera et al. (2023) focuses exclusively on adjusting the Rock-Eval analytical protocol to improve the accuracy of the initial measurement. The question of post-analysis corrections is addressed as a technical contingency based on the literature. In the "Materials & Methods" section, Hazera et al. (2023) present the empirical correction protocol proposed by Disnar et al. (2003), and then another protocol of "parametric correction" based on a prior interpretation of the RE data (Sebag et al., 2022). Hazera et al. (2023) explicitly state that they use the empirical parameters proposed by Disnar et al. (2003) to correct RE measurements. Therefore, by adopting the protocol used by Hazera et al. (2023), Stojanova et al. (2024) compare the performances of their statistical models to the empirical procedure proposed by Disnar et al. (2003). It seems important to correct this attribution error to avoid any confusion regarding the origin of the correction method.
By using the formulas presented by Hazera et al. (2023), Stojanova et al. (2024) implicitly employ the empirical correction procedure proposed by Disnar et al. (2003) without explicitly stating it. However, this procedure explicitly comprises two distinct and successive steps: the first applies unconditionally to all samples, while the second applies only under certain conditions to specific samples after a prior examination of qualitative RE parameters. In the present form of the manuscript, it appears that Stojanova et al. (2024) systematically applied the second step to all samples without prior verification of the conditions for its application. It is crucial that the procedure proposed by Disnar et al. (2003) is implemented in accordance with its technical recommendations. It is highly likely that the results will not be radically different from those currently presented, but it will minimize any uncertainty when comparing the models’ performances.
To apply the Disnar’s correction extended to Cinorg, Stojanova et al. (2024) propose using a threshold value of 2 gC/kg of SIC to determine which samples are calcareous or non-calcareous. However, the MinC parameter includes a portion of released Corg, particularly during the pyrolysis phase (Hazera et al., 2023; Koorneef et al., 2023). So, does the use of MinC to distinguish calcareous from non-calcareous soils introduce uncertainty in samples with high TOC? Since the comparison of models is conducted for three soil populations (non-calcareous, calcareous, all), have the authors verified the accuracy of these categories through mineralogical analyses (such as XRD) or through a detailed examination of thermograms (as in Pilot et al., 2014)? Another question concerns the minerals. Do the proposed models integrate soils containing minerals other than calcite? Thermograms of dolomite or siderite are quite distinct (Pilot et al., 2014). Could this impact the performance of the machine learning models?
One of the strengths of the work by Stojanova et al. (2024) lies in their explicit and unambiguous statement of the objectives of their approach: to identify the correction method that minimizes the discrepancies between the corrected RE measurements and the standard measurements used as reference. By correcting the RE measurements in this manner, the authors achieve a very satisfactory approximation to standardized measurements (ISO). This significant advancement will facilitate the practical use of RE data while awaiting their possible standardization. Concerning the comparison of models, the analysis is based on their respective performance in minimizing discrepancies with the reference method. However, the results are not analyzed relative to each other. Would it be possible to verify that the differences between the methods are statistically significant considering the analytical precision of the methods used?
However, minimizing the discrepancies between the corrected RE measurements and the standard measurements raises a more fundamental question: it is well-known that standard protocols for measuring SOC and SIC entail several inevitable errors related, on one hand, to sample pretreatment (removal of Corg or Cinorg), and on the other hand, to measurements on two different aliquots. Therefore, in seeking to minimize the discrepancies between RE measurements and the standard method, the authors import errors associated with the latter for calcareous soils. From a methodological perspective, it would be judicious to indicate this limitation in the technical note. Indeed, this inherent limitation to the stated objectives highlights the need for further studies to improve the initial measurement and correct the systematic misattribution of Corg as Cinorg.
This is why the end of the abstract raises perplexity when the authors write: "that the proposed correction significantly increases the accuracy of the Rock-Eval method on the initial dataset, and that it can be successfully applied to data originating from different Rock-Eval machines, without changing the routine analytical protocol." This statement seems to be in contradiction with the objectives presented. The correction did not increase the accuracy of the RE method; it reduced the discrepancies of the measurements with a reference method that has its own errors. The RE method would increase its accuracy if the analytical protocol or calculation methods avoided confusion regarding the forms of carbon.
In conclusion, one may question the format chosen by the authors to publish their results. Does the Technical Note format allow for the development of all the necessary discussions to truly highlight the results? This format, which minimizes the scientific issue in favor of the results, accentuates at the same time the "advertisement for the Rock-Eval device and Geoworks software" commercialized by the company that funded the study, both of which are used for commercial purposes by another party of the study's co-authors (which is not indicated in the section dedicated to potential conflicts of interest).
Citation: https://doi.org/10.5194/egusphere-2024-578-RC2
Model code and software
Technical note: A validated correction method to quantify organic and inorganic carbon in soils using Rock-Eval® thermal analysis Marija Stojanova https://zenodo.org/doi/10.5281/zenodo.10706302
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
239 | 64 | 14 | 317 | 7 | 5 |
- HTML: 239
- PDF: 64
- XML: 14
- Total: 317
- BibTeX: 7
- EndNote: 5
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1