Using Mid-Infrared Spectroscopy to Optimize Throughput and Costs of Soil Organic Carbon and Nitrogen Estimates: An Assessment in Grassland Soils

Round 1

Reviewer 1 Report

The authors determined how the accuracy of soil organic carbon (SOC) and soil nitrogen (N) predictions using mid-infrared (MIR) spectroscopy is affected by the number of calibration samples, and by different predictive models. They analyzed 1000 samples from grassland soils, where tested the effect of calibration sample size from 100 to 1000 samples, as well as the predictive ability of different prediction tools (partial least square, random forest and support vector machine algorithms) on SOC and N predictions. The samples were obtained from five different farm pairs corresponding to two different grazing types covering a 0–50 cm soil depth.  

Interestingly, the highest predictive performance of the various models occurred with about 400 samples, that is, larger sample sets did not improve the accuracy of the training algorithms tried.  The MIR method was highly accurate for SOC and N estimates and proved cost-effective relative to the more widespread dry combustion. The non-linear models were not able to improve upon the classical partial least square performance.

In my opinion, this report is valuable to the soil and environmental science community especially in the age of grassland restoration, soil loss, and big data as well as predictive modeling. My only request is that the authors try to suggest a way forward, or at least emphasize their preferred sampling suggestion for the issue of overestimation (highest values of SOC and N). Also, same applies to the issue of POM-induced uncertainty. Any suggestions other than this might affect the performance of the models by resulting in less reliable estimates at <10 cm depth?

Author Response

1. (x) English language and style are fine/minor spell check required

Response: The manuscript was carefully checked by a native English speaker to improve grammar and readability. We believe all the minor spelling mistakes were corrected.

2. In my opinion, this report is valuable to the soil and environmental science community especially in the age of grassland restoration, soil loss, and big data as well as predictive modeling. My only request is that the authors try to suggest a way forward, or at least emphasize their preferred sampling suggestion for the issue of overestimation (highest values of SOC and N). Also, same applies to the issue of POM-induced uncertainty. Any suggestions other than this might affect the performance of the models by resulting in less reliable estimates at <10 cm depth?

Response: We thank the reviewer for the useful comments and recommendations. We believe that the observed inability of Mid-Infrared (MIR) to accurately estimate SOC and N at high values, is not an issue that can be addressed by modifying the soil sampling strategy, as indicated by the lack of responsiveness of accuracy to the increased sample size, or exclusion of high SOC samples.   Thus, we refrain from discussing soil sampling suggestions. To address the reviewer’s concern, we revised a paragraph in the discussion (page 26, Line 409-412) and now suggest that for shallow soils with high POM carbon, it may be better to use a conventional combustion approach. Further, we encourage more work to independently quantify POM C via MIR, which can also help address this issue.  

Reviewer 2 Report

The manuscript entitled “Using Mid-infrared spectroscopy to optimize throughput and costs of soil organic carbon and nitrogen estimates: An assessment in grassland soils” is scientifically interesting and original. The manuscript is well written and in scientific language and style.

 However, there are some points in the manuscript that need improvement:

 In my opinion, the words that are in the title of the paper should not be repeated in the keywords. So the words “infrared spectroscopy” and “soil organic carbon” should be deleted or replaced by other words or phrases.

1. Lines 65-77: The proposed paper (among others) referring to the creation of models based on machine learning to predict the values ​​of physicochemical parameters and trace elements may contribute to the theoretical background of the study.: “Soil parameters affecting the levels of potentially harmful metals

in Thessaly area, Greece: a robust quadratic regression approach of soil pollution prediction. Environmental Science and Pollution Research, https://doi.org/10.1007/s11356-021-14673-0.

3. It may be of soil science interest to refer to the soil orders to which the soil samples used for the study belong. The reference to soil orders can give a more complete picture for the analysis of the results. Information on soil orders can be obtained from the articles below:

Vulnerability of Soil Carbon Regulating Ecosystem Services due to Land Cover Change in the State of New Hampshire, USA. Earth 2021, 2, 208–225. https://doi.org/10.3390/ earth2020013

Or Available concentrations of some potentially toxic and emerging contaminants in different soil orders in Egypt and assessment of soil pollution, Journal of Soils and Sediments, DOI: 10.1007/s11368-021-03021-x

Or “Pollution assessment of potentially toxic elements in soils of different taxonomy orders in central Greece. Environ Monit Assess (2019) 191: 106, https://doi.org/10.1007/s10661-019-7201-1”              could perhaps be useful on this subject.

Author Response

1. (x) English language and style are fine/minor spell check required

Response: The manuscript was carefully checked by a native English speaker to improve grammar and readability. We believe all the minor spelling mistakes were corrected.

2. In my opinion, the words that are in the title of the paper should not be repeated in the keywords. So the words “infrared spectroscopy” and “soil organic carbon” should be deleted or replaced by other words or phrases.

Response: We appreciate the comment. We deleted “infrared spectroscopy”, “nitrogen” and “soil organic carbon”. We added “multi-paddock grazing”, “particulate organic matter”, “rangeland monitoring”.

3. Lines 65-77: The proposed paper (among others) referring to the creation of models based on machine learning to predict the values ​​of physicochemical parameters and trace elements may contribute to the theoretical background of the study.: “Soil parameters affecting the levels of potentially harmful metals in Thessaly area, Greece: a robust quadratic regression approach of soil pollution prediction. Environmental Science and Pollution Research, https://doi.org/10.1007/s11356-021-14673-0.

Response: We really appreciate your help with providing references and suggestions. However, we could not include these because we were focused on chemometrics/machine learning tools used for high dimensional and multicollinear variables such as spectra data. Nonetheless, we are grateful for your feedback, and we clarified this point in lines 63-65, page 3.

4. It may be of soil science interest to refer to the soil orders to which the soil samples used for the study belong. The reference to soil orders can give a more complete picture for the analysis of the results. Information on soil orders can be obtained from the articles below:

Vulnerability of Soil Carbon Regulating Ecosystem Services due to Land Cover Change in the State of New Hampshire, USA. Earth 2021, 2, 208–225. https://doi.org/10.3390/ earth2020013

Or Available concentrations of some potentially toxic and emerging contaminants in different soil orders in Egypt and assessment of soil pollution, Journal of Soils and Sediments, DOI: 10.1007/s11368-021-03021-x

Or “Pollution assessment of potentially toxic elements in soils of different taxonomy orders in central Greece. Environ Monit Assess (2019) 191: 106, https://doi.org/10.1007/s10661-019-7201-1”              could perhaps be useful on this subject.

Response: This is a very important point and a valuable comment. Incorporating soil taxonomy is something we missed here. We have added the soil series according to USDA soil taxonomy in lines (90-92, page 4).