the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Large errors in common soil carbon measurements due to sample processing
Abstract. To build confidence in the efficacy of soil carbon (C) crediting programs, precise quantification of soil organic carbon C (SOC) is critical. Detecting a true change in SOC after a management shift has occurred, specifically in agricultural lands, is difficult as it requires robust soil sampling and soil processing procedures. Informative and meaningful comparisons across spatial and temporal time scales can only be made with reliable soil C measurements and estimates, which begin on the ground and in soil testing facilities. To gauge soil C measurement inter-variability, we conducted a blind external service laboratory comparison across eight laboratories selected based on status and involvement in SOC quantification for C markets. To better understand how soil processing procedures and quantification methods commonly used in soil testing laboratories affect soil C concentration measurements, we designed an internal experiment assessing the individual effect of several alternative procedures (i.e., sieving, fine grinding, and drying) and quantification methods on total (TC), inorganic (SIC), and organic (SOC) soil C concentration estimates. We analyzed 12 different agricultural soils using 11 procedures that varied either in the sieving, fine grinding, drying, or quantification step. We found that a mechanical grinder, the most commonly used method for sieving in service laboratories, did not effectively remove coarse materials (i.e., roots and rocks), thus resulted in higher variability and significantly different C concentration measurements from the other sieving procedures (i.e., 8 + 2 mm, 4 mm, and 2 mm with rolling pin). A finer grind generally resulted in a lower coefficient of variance where the finest grind to < 125 µm had the lowest coefficient of variance, followed by the < 250 µm grind, and lastly the < 2000 µm grind. Not drying soils in an oven (at 105 °C) prior to elemental analysis on average resulted in a relative difference of 3.5 % lower TC, and 5 % lower SOC due to inadequate removal of moisture. Compared to the reference method used in our study where % TC was quantified by dry combustion on an elemental analyzer, % SIC was measured using a pressure transducer, and % SOC was calculated by the difference of % TC and % SIC, predictions of all three soil properties (% TC, % SIC, % SOC) using Fourier Transformed Infrared Spectroscopy were in high agreement (R2 = 0.97, 0.99, 0.90, respectively). For % SOC, quantification by loss on ignition had a low coefficient of variance (5.42 ± 3.06 %) but the least agreement (R2 = 0.83) with the reference method.
- Preprint
(1092 KB) - Metadata XML
-
Supplement
(953 KB) - BibTeX
- EndNote
Status: open (until 21 Jul 2024)
-
RC1: 'Comment on egusphere-2024-1470', Anonymous Referee #1, 24 Jun 2024
reply
In their manuscript the authors present the uncertainty of total carbon, soil inorganic carbon and soil organic carbon measurements depending on sample processing and measurement. The authors show substantial differences that are mainly driven by sieving and measuring methods with LOI being highly variable. It is or great importance to have such comparisons and critical assessments. The need for accurate soil C measurements is getting more important for an evolving C market. A substantial overestimation of C changes would be bad for the actual climate effect and a substantial underestimation of soil C would reduce the economic benefit of the C market. The experimental set up using 11 procedures and the comparison with 8 commercial laboratories is an important approach to reach a better homogenization of analyses approaches and make soil C measurements more consistent.
I have two main concerns:
- The authors need to elaborate their discussion on the application of chemometric approaches by combining MIR and predictive modelling (e.g. Line 505-512, 523-526 and 567-570) It is true that such approaches can work well as reported in the cited literature. However, it needs to be clear that this all depends on the availability of a representative soil spectral library that it large enough to develop models for prediction. The good prediction in this study is expected and bias the generalized conclusion. The model was trained on the KSSL and thus covers the spectral variability of the soils used here. Additionally, the sample pre-treatment was very similar between the P0 method here and the initial data for the model presented in Seybold et al (2019). Seybold et al (2019) measured TC by dry combustion, used the pressure transducer method for SIC and determined SOC by difference. Thus, it is a good model for the soils selected here. However, the transferability of such models is difficult and a major challenge to overcome. For example, sample grinding is important for the transferability. Grinding was also an aspect that motivated the authors to test. It is true, the differences in grinding are not so significant when all samples are similarly prepared and the model trained for the corresponding grinding is applied but transferring a model trained on finely milled sample to samples that are coarser and vice versa brings uncertainty and challenges (Sandermann et al., 2023). Recently, Safanelli et al. (2023) reported that even combining spectra obtained from different devices can be difficult and requires important pre-processing. More importantly, the authors report that sample processing resulted in larger uncertainty of the predictions. Therefore, the authors need to constrain their conclusion here that such approaches are only working when the conditions of a good and regional model are given. Otherwise, the model error (e.g. RMSE) will be too large to detect changes in TC, SOC and SIC.
- As far as I understand P0 is the reference method here but also the method used in the authors research lab. It is not clear why the authors are so certain that this method is the most rigorous. For example, in Line 161-163 the authors just argue with their “expert opinion”. Many labs use ball mills that are more efficient in grinding (e.g. <50um), oven drying at 105°C might cause losses of OC in some high C soils (this is only briefly touched at the end) and the pressure transducer methods requires the direct addition of acid to the soil, which can alter the organic matter (fumigation is less harsh). The authors need a reference method here to compare to but they also need to critically discuss the constrains of P0 here. It is even more important to have a good justification here given the conflict of interest that exists here between the research and the commercial lab the authors are part of at the same time.
Specific comments:
Line 14: Please specify what "involvement in SOC quantification for C markets" means
The abstract contains many details but no conclusion of the study.
Line 53: Please specify if the authors mean "quality assurance and quality control"
Line 54: Please specify NAPT for readers that are not familiar with US organisation. This holds true for all other abbreviations that are not explained.
Line 60: Root and rock fragments are not considered as part of the fine soil that is important for the biogeochemical processes. However, rocks and roots are still components of soils.
Line 63-65: Do the authors have any reference that commercial labs do not remove coarse fragments. To my experience, research labs apply sieving and in general same sample preparation for agricultural and non-agricultural soils. Also, soil inventories prepare the fine soil prior to C measurements.
Line 65-67: It is not clear to me why regenerative agriculture results in more coarse fragments in deeper soil. Also, the authors refer here rather to conservational land management rather that regenerative land management, which is a very broad and not well-defined term.
Line 77: Also here, the authors should be specific since it is considered as "fine soil"
Line 97: The authors should specify if near-infrared of mid-infrared regions.
Also, such approaches require a well-trained model based on large enough soil spectral library. This is a critical step for the quantification of soil C using chemometric approaches. Therefore, it follows a different concept compared to the other more direct methods.
Line 121: I would rather expect that the dual homogenisation by sieving to 8 followed by 2 mm would result in lower variability.
Table 1: is pH, %SOC and %SIC are measured with analytical replicates? The authors should add errors to the values.
Table 1 caption: How was pH measured and what are the texture classes applied? It is not clear what "Colorado State University following procedure P0" is. Please provide details of refer to Table 2 here.
Line 218-219: This is not very precise. It is not clear which model and was used and on which data it is trained.
Line 223-224: This is most likely attributed to the fact that the used model for the prediction based on the KSSL is developed with samples of similar degree of grinding. In the cited paper, Sanderman et al (2023) conclude that the model trained on fine milled samples was not well transferable on the coarser samples.
Therefore, the authors used a model that was trained for a certain milling. This makes this testing of grinding here not very useful. in comparison, Sanderman et al (2023) developed separate models for roughly 2400 samples of the KSSL. They conclude that a model that was trained with coarse samples and predicted coarse samples was performing similar to a model that was trained with fine soils and predicted fine soils. However, the transfer of models was not satisfying.
Line 231: Before it was mentioned that acid fumigation was only performed for P9. Here it reads like every sample was fumigated. Please clarify.
Line 241-243: How were CO2 and H2O interferences corrected?
Line 245-248: The authors should add more details regarding the predictions. Seybold et al (2019) developed PLSR based on the NSSC-KSSL. is this also used here? What do the authors mean with "respective geographical region"? Were the models local? Was there any spectral pre-processing like re-sampling, filtering, normalization or bassline correction?
Line 270: "External service labs provided values for % TC, % SIC, and % SOC." can be removed.
Line 271: Looking at Table S3, it seems not fair to just select the extremes here. Most differences are rather lower. It is hard to tell from the table. Maybe boxplots per soil with different symbols for the labs would be easier to read. Anyway, the authors should also mention the range of differences and not only the extremes.
Line 284: Yes, it is an astonishing range of measured values between labs. It is also surprising that the reference measurement (CSU lab) shows a large variability of soil B and H. These are two soils with high pH. I wonder if this could be an effect of the carbonate removal. What is your explanation for large differences between the five analytical replicates? Additional, the external labs did not measure in replicates?
Line 288: Why are the no coarse materials at all in soil L for the P3?
Line 305-307: This relationship seems to be mainly driven by the on P3 point at 0.8 difference in plant material and 1 STD %SOC (right corner). This is in general a very weak correlation and might not add much when the one point is considered as an outlier.
Line 323-324: Do the author mean a relationship to SOC, similar to the plant material in Fig. 3?
Figure 4 and results section: This paper is mainly about errors that are important for the SOC because this will be of interest for the C markets. Therefore, I wonder if Figure S3 with the SOC differences between soils and methods should be the main figure in the manuscript and the current Figure 4 could more to the SI. This might need a restructuring of the section as well.
Line 339: Significances are shown in Table S6?
Figure 6: X axis label, colour and legend are redundant.
Line 396-397: Here the focus is on SOC for the C market.
Line 398-399: Please see my comment regarding the variability on CSU lab for some soils. This is also concerning. Here it would be good to have replicates from the individual labs.
Line 460: This would be a very interesting aspect of the manuscript. The C market needs stocks of C and not concentrations alone. Therefore, the effect of removed or not removed coarse fragments would be most significant. Even the calculation of SOC stocks includes large uncertainties and this would add up with the method uses (e.g. Poeplau et al. 2017). The authors do back on the envelope calculations later in the implications section. Would it be possible to discuss the stock effects even more by estimating the stock differences here for all methods using the soils bulk densities?
Line 465-468: Yes, plant material would be low in mass but might be important in volume and thus could have an impact of stocks as well.
Line 567: This should be Fig. S9
References:
Safanelli, J.L., Sanderman, J., Bloom, D., Todd-Brown, K., Parente, L.L., Hengl, T., Adam, S., Albinet, F., Ben-Dor, E., Boot, C.M., Bridson, J.H., Chabrillat, S., Deiss, L., Demattê, J.A.M., Scott Demyan, M., Dercon, G., Doetterl, S., Van Egmond, F., Ferguson, R., Garrett, L.G., Haddix, M.L., Haefele, S.M., Heiling, M., Hernandez-Allica, J., Huang, J., Jastrow, J.D., Karyotis, K., Machmuller, M.B., Khesuoe, M., Margenot, A., Matamala, R., Miesel, J.R., Mouazen, A.M., Nagel, P., Patel, S., Qaswar, M., Ramakhanna, S., Resch, C., Robertson, J., Roudier, P., Sabetizade, M., Shabtai, I., Sherif, F., Sinha, N., Six, J., Summerauer, L., Thomas, C.L., Toloza, A., Tomczyk-Wójtowicz, B., Tsakiridis, N.L., Van Wesemael, B., Woodings, F., Zalidis, G.C., Żelazny, W.R., 2023. An interlaboratory comparison of mid-infrared spectra acquisition: Instruments and procedures matter. Geoderma 440, 116724. doi:10.1016/j.geoderma.2023.116724
Sanderman, J., Gholizadeh, A., Pittaki‐Chrysodonta, Z., Huang, J., Safanelli, J.L., Ferguson, R., 2023. Transferability of a large mid‐infrared soil spectral library between two Fourier‐transform infrared spectrometers. Soil Science Society of America Journal 87, 586–599. doi:10.1002/saj2.20513
Sanderman, J., Smith, C., Safanelli, J.L., Morgan, C.L.S., Ackerman, J., Looker, N., Mathers, C., Keating, R., Kumar, A.A., 2023. Diffuse reflectance mid-infrared spectroscopy is viable without fine milling. Soil Security 13, 100104. doi:10.1016/j.soisec.2023.100104
Poeplau, C., Vos, C., Don, A., 2017. Soil organic carbon stocks are systematically overestimated by misuse of the parameters bulk density and rock fragment content. Soil 3, 61–66. doi:10.5194/soil-3-61-2017
Citation: https://doi.org/10.5194/egusphere-2024-1470-RC1
Data sets
Soil carbon measurements R. Even https://zenodo.org/doi/10.5281/zenodo.11223422
Viewed
HTML | XML | Total | Supplement | BibTeX | EndNote | |
---|---|---|---|---|---|---|
529 | 208 | 10 | 747 | 29 | 9 | 10 |
- HTML: 529
- PDF: 208
- XML: 10
- Total: 747
- Supplement: 29
- BibTeX: 9
- EndNote: 10
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1