In their manuscript, ‘Soil organic matter and labile fractions depend on specific local parameter combinations’, Ortner et al. present their work on the analysis of factors controlling the soil organic carbon (SOC) concentration in topsoils of the region around Trier, Germany. The authors collected topsoil samples in arable land and grassland in 4 regions with different parent material, and determined the organic carbon (OC) concentration, hot water extractable carbon (HWEC), microbial biomass carbon (MBC) and multiple soil properties on these samples. They used PCA to cluster the soil samples based on parent material and soil texture into different clusters. The aim of their study was to assess the main factors controlling topsoil organic carbon concentration, HWEC and MBC using two modelling approaches: a bivariate model and mixed effects models. The main findings are that (i) mixed effect models outperformed bivariate (linear) models in predicting OC%, HWEC and MBC, (ii) at the local scale, site-specific parameters explained OC variability better than landscape-related variables and (iii) using the ‘local’ model resulted in better results when predicting the OC% of a specific cluster compared to the ‘global’ model.

The results of the present study help to improve our understanding of the factors controlling topsoil organic carbon concentrations at the landscape scale, which is needed e.g. in order to improve soil organic carbon models. The authors have constructed a valuable dataset which may benefit other researchers. I would therefore encourage the authors to make this data available through an online repository, instead of making it only available upon request.

Overall, the manuscript is well-written. However, at multiple locations very long sentences are used, which does not benefit a smooth reading. Splitting those sentences and using more commas would improve the readability of the manuscript considerably. In addition, I would encourage the authors to use subsections in the Results and Discussion sections, which will provide a better overview to the reader of what is being presented and discussed.

One of my main concerns about the present manuscript is related to the quantification of the goodness-of-fit of the different models, which is now done using R-square. This is a measure to quantify the proportion of variation in a dependent variable that is explained by an independent variable, but is not a measure for the goodness-of-fit of a model. For example, a very poor model can have a high R-square value, while a good model can have a relatively low R-square value. Therefore, the authors should use a different measure to quantify the goodness-of-fit of their model when comparing measured with modelled data, such as the (root) mean square error or similar.

In addition, I missed a discussion about the broader implications of the results and the implications for future research. For example, do the authors suggest that researchers should use ‘local’ model whenever possible? And how about regions where local information is not present? It would also be very informative if the authors would quantify the difference in predicted SOC% when using a global versus local model. To how much of an over or underestimation would this lead? Is that difference significant enough to invest more resources in the collection of local data?

Another concern is related to the title, which I find not very informative. For example, it will not be clear to someone who has not read the manuscript what ‘specific local parameter combinations’ are. Also, it would be good to be more specific about what they mean with ‘soil organic matter and labile fractions’. From the title it is not clear if the authors mean SOC concentrations, stock, spatial distribution etc. In addition, the manuscript discusses soil organic carbon, and not all soil organic matter, so would be good to change this in the title.

Lastly, it would good if the authors specify in the beginning of the manuscript that they discuss SOC concentrations, and not stocks. Throughout the manuscript, the authors talk about ‘SOC’ without specifying that it concerns concentrations, not stocks. This is an important difference, which should be clear to the reader from the abstract onwards, and repeated throughout the manuscript. For example, the authors could change ‘SOC’ to ‘SOC%’ to make this clear.

Specific comments

L18: Would be good to explain here what you mean with ‘global’ and ‘local’ clusters (and models).

L19: define that you assess SOC concentrations, and thus not stocks

L20: would be good to explain here which ‘labile fractions’ you study

L21-22: would be good to make it clear that bivariate models were used to determine these correlations

L24-25: you cannot use R-square values to compare modelled versus measured data (see above)

L27: here you use the term ‘organic matter’, while until here you used ‘SOC’. Please be consistent with these terms, and only use one

L29-30: ‘showing that labile fractions depend less on soil properties than on organic matter input and turnover in soil’. The latter were not studied, so you cannot say this with certainty. Would be better to end the abstract with a statement about the broader implications of your results.

L41: another important labile fraction of SOC is particulate organic carbon. Would be good to justify why you did not study this fraction

L45: ‘MBC is expedient to explain SOC dynamics’: this is rather vague, please be more specific

L45-46: ‘much less research and attempts for quantitative modelling of these labile fractions […]’: recently, multiple mechanistic models have been used to simulate labile carbon fraction such as MBC and POC, e.g. Ahrens et al. (2015), Wieder et al. (2015) and Zhang et al. (2021)

L58: ‘In addition or even instead of’: choose one

L63: please clarify what you mean with parameters versus factors, as you use these terms throughout the manuscript

L87: please define what you mean with ‘global models’

L99: the term ‘sufficient quantification’ is rather vague, please clarify this

L107: ‘similar numbers of samples’: how many per region?

L108-109: the use of the abbreviations throughout the manuscript is not intuitive and confusing for the reader, please use different names to identify the different regions, e.g. the parent material

L119: why were some samples stored at -20 °C and others air-dried?

L134: was the chloroform fumigation extraction performed on samples freshly collected from the field?

L137-140: for how long were the samples incubated? How often was the CO₂ measured?

L143: were all parameters log-transformed? Please clarify this

L146: please provide some examples of the ‘mineral phase parameters’

L146-147: please provide more information about the linear regressions that were performed

L156-157: Please provide information about which parameters were removed from the models

L161: were all parameters checked for collinearity? Please clarify

L163: why a square root transformation? Please justify this

L163: Please clarify how the performance of the models was examined

L170: Please clarify the difference between ‘soil’ and ‘topsoil’ properties

L177: are those differences statistically significant? What are the averages for the different parent materials?

L186: please provide examples for the ‘parameters describing the composition of SOM’

L191-194: this is not clear

L205-206: which 10 parameters?

L213: ‘that largely matched with those found for the complete dataset’: this is not clear

L224: what do you mean with ‘sufficient extent’? Similar wording is used throughout the manuscript, but this is very subjective and should be clarified.

L240: please clarify what ‘equal weight of samples’ means

L237-242: please make clear that you are discussing the results of the bivariate models

L241: what are the ‘complex interactions of several different parameters’?

L243: please clarify what you mean with ‘insufficient’. Which measure do you use to determine if a model performance is sufficient or not?

L249-251: R-square values are no measure for model performance, please provide the root mean square error (or a similar measure). Please show these results in a graph, perhaps in the supplement?

L273: do you mean the bivariate models with ‘linear regressions’ Please be consistent with this terminology

L284: please replace R-square with a measure of model performance

L287: please provide the goodness-of-fit values before concluding that a certain model has an ‘inferior performance’

L309-310: by saying ‘Al- and Fe-oxides were shown to have a relevant influence on sequestration and stabilization of SOC’, it seems like you explicitly studied this, while you only used a statistical model to assess this. Also, since you model SOC concentrations and not stocks, you cannot say anything about C sequestration, as this also depends on bulk density.

L342: please explain what you mean by ‘multidimensional’

L348: ‘to explain SOC’: please clarify which aspect of SOC

L366: how do you conclude that sample size biased the results? Did you test for this?

L369: ‘satisfying extent’: how is it quantified that a model performs satisfying? Please be objective in deciding if a model is good or not

L370: what do you mean with ‘partially practicable’?

L374-375: ‘sufficient quality level’: same remark as L369

L380-381: by saying ‘It became clear that […] with different annual dynamics’, it seems like you tested annual dynamics. Please rephrase

L281-382: You did not take SOC turnover into account, so how can you say that this is a reason for the lower explained variability by the different models?

L403-404: ‘sufficient estimation’: same remark as L369

L405: would be good to end the Conclusions section with a statement about the broader implications of your results

Figures and tables

Fig. 2: the colours in B are difficult to distinguish

Table 3: please make clear in the caption that these are the result for the bivariate regressions

Fig. 3: ‘Predicted vs. measured’: please clarify in the caption which model was used to make these predictions. Please provide a measure for the goodness of fit and remove the R-square values, as this is not measure for model performance

Table 5: please provide more information about the table in the caption, the table should be clear to the reader without having read the entire manuscript. It would be more informative to provide a table with e.g. root mean square errors instead of R-square

Figure 6: it would be informative to show the same graphs for other clusters in the supplement. Please remove the R-square values and replace them by a measure for the goodness-of-fit of the models

Technical comments

L36: driver => drivers

L57: expedient => suitable

L73: space between ‘asCa2+’

L119: it’s not clear what ‘respectively’ refers to

L170: it’s not clear what ‘respectively’ refers to

L172: it’s not clear what ‘thereby’ and ‘essentially’ refer to

L252: it’s not clear what ‘respectively’ refers to

L275: what do you mean with ‘not last’?

L304: ‘the in total very sandy soils’: please rephrase

L320: what is ‘circumneutral’?

L324: please remove ‘respectively’

L343: please remove ‘respectively’

L344: confirmed => is in line with

References

These references were chosen based on their scientific content. I leave it up to the authors to decide if they wish to include them in their manuscript.

Ahrens, B., Braakhekke, M.C., Guggenberger, G., Schrumpf, M., Reichstein, M., 2015. Contribution of sorption, DOC transport and microbial interactions to the 14C age of a soil organic carbon profile: Insights from a calibrated process model. Soil Biology and Biochemistry 88, 390–402. https://doi.org/10.1016/j.soilbio.2015.06.008

Wieder, W.R., Grandy, A.S., Kallenbach, C.M., Taylor, P.G., Bonan, G.B., 2015. Representing life in the Earth system with soil microbial functional traits in the MIMICS model. Geoscientific Model Development 8, 1789–1808. https://doi.org/10.5194/gmd-8-1789-2015

Zhang, Y., Lavallee, J.M., Robertson, A.D., Even, R., Ogle, S.M., Paustian, K., Cotrufo, M.F., 2021. Simulating measurable ecosystem carbon and nitrogen dynamics with the mechanistically defined MEMS 2.0 model. Biogeosciences 18, 3147–3171. https://doi.org/10.5194/bg-18-3147-2021

Comments on manuscript SOIL-2021-81

This manuscript presents the distribution of SOC and its labile fractions predicted using parent material, land use and soil properties in Southwest Germany. The results indicated that soil properties were clustered by parent materials and soil texture rather than land use. In general, mixed-effect model gave better predictions than bivariate regression. They compared “global model” with “local model” to show that the application of global model on local dataset introduced poorer predictions. Also, the explained variance generally decreased from bulk SOC to its labile fractions.

In general, the objectives were clear and relevant while the scientific value is sufficient. The large sample size contributes to a robust prediction. However, there are several concerns to be addressed.

One concern is the distribution of the sampling points. As mentioned in L47-48, soil formation is also controlled by climate and topography. The clustered locations of the four parent materials are likely to introduce differences in topographical and climate conditions. As climate and topography factors were not included in the models, their effects might be recognized as the effects of parent materials, texture or land use in the predictive models. (Details in comments for Fig. 1) 

Another concern is that the usage of “global/local scale”, “global/local model”, “global/local cluster” and “global/local/entire dataset” may confuse readers because they were used without necessary explanations. In addition, the words “global” vs. “local” give the impression that the study aimed to compare SOC distribution on global vs. local scale, but no investigation on global scale was given in this study.

In addition, in some parts of the manuscript, R^2 was used to estimate whether models are well-fitted, which is not proper. Also, the Results and Discussion can be improved by splitting them into sub-sections and better re-organizing. Finally, the readability of the manuscript can be improved by revising long-complexed sentences and vague expressions.

Title: (1) Although “soil organic matter” is used in the title, the main part of this manuscript is mostly talking about “soil organic carbon”. Please be consistent in using them because soil organic matter contains not only organic carbon but also other elements such as nitrogen.

(2) It is advised to add restrictions on the area/location because the study was performed in western Germany and will not be necessarily applicable in other places.

L14, L18 and L21: It is confusing to mention “local scale”, “global/local cluster” and “global/local dataset” in abstract without further explanation. The usage of “local” vs. “global” gives me a feeling that this study compares SOC distribution on local vs. global scales. Apparently, the distribution of the sampling sites represents a local or sub-regional scale. It is suggested to either give them definitions when they are mentioned for the first time or replace them with more suitable words.

L21: As only regressions were performed in this study, it is recommended not to use both correlation and regression in the text.

L21-23: It is difficult to understand this sentence.  It is not clear between which factors the correlations are significant. What does “partially low” mean? Splitting this sentence into simple ones may help.

L66: In general, organo-mineral associations are considered contributing to the formation of stabilized fractions (not labile fractions) and therefore the accumulation of SOC.

L72: …leading to SOC sequestration…

L85: Please check if surnames and given names are misplaced in this reference.

L90: “Local vs. global models” are confusing. Do they mean models on local vs. global scales?

L102: Is the “entire dataset” equivalent to the “global dataset”?

Materials and Methods

L104: It is recommended to add more information about the study area. In general, most studies show readers climate factors (e.g. annual precipitation and average temperature), soil type/classification and composition of vegetation/crops.

L119: Please explain why soil samples were stored either at -20 °C or air-dried. For different analyses?

L139: More information of the incubation is appreciated. How long the samples were incubated before sampling? What was the temperature? Did you sample for only once or multiple times?

L146: Please give more information of linear regressions. For example, indicate that they only have one predictor. Did you check the normality of residues?

L147: and after: What are the reasons for performing mixed-effect models? Why parent materials, texture group and land use are selected as random effect variables? In general, random effects are used when samples are only a small subset of the group or when limited groups are included. Does it aim to make predictors on a larger scale using the limited dataset?

L162-163: Why was response variable transformed but not predictors?

Result

Overall: The readability can be improved by dividing this section into a few subsections due to a large content in this section.

L170: What are “soils and topsoil properties”? Consider revising.

L177-178: Are they significantly different or different by looking at means/ranges?

L190: “Somewhat different” is vague.

L205: and after: This paragraph is comprised of isolated points, which makes it difficult to follow. A suggestion is to describe Table 3 in a well-organized way to shorten this paragraph. For example, you can follow the order of entire dataset --> land use --> parent materials --> texture, or you can introduce them by the types of predictors. Also, focusing on your key findings helps.

L207-208: The items “global cluster” and “local cluster” are explained here but they appear in previous parts (e.g. L18 and L193). Please give explanations when they appear for the first time.

L208 and L94: Please be consistent for “parent material” or “parent rock material”.

L224: What is “a sufficient extent”? Please specify.

L237 -242: Please indicate that they are from Table 3.

L240: How to know “weight of samples” is equal? Why does it act as global cluster?

L250: It is not clear how to compare R^2 between bivariate regression and mixed-linear model. By the means of each cluster?

L257-258: DCS sites look different from LBS and DLS.

L279-286: Please indicate related Tables and Figures. It is hard to follow.

L282 & L287-288: This gives me a feeling that you are estimating whether the models were well-fitted. If this is true, comparing R^2 does not make sense. Large R^2 means more variation is explained by predictors. Instead, you have to look at the distribution of residue using e.g. root mean square error (RMSE).

Discussion

L304-305: “for the in total very sandy soils …of LBS”. Try to revise this sentence.

L309: “…SOC in soil” --> “in soil”

L314-315: “ECEC, Ca and Mg are suitable predictors for SOC in this study”; L317-318: “The minor ability of ECEC (Ca+Mg) to explain SOC..” They look like contradictory. Also, I missed a point that whether you are talking about entire dataset or specified cluster. Table 3 showed that the predictions using ECEC and (Ca+Mg) are largely dependent on parent materials and texture cluster. A possible explanation is that DCS soils had more sands and lower pH, so that Ca and Mg do not contribute to SOC stabilization, whereas DLS and PSS soils had higher pH, so that Ca and Mg bridging play a role in SOC stabilization (see your cited paper). Please consider re-organizing this part.

L328-333; Grassland had higher SOC contents than arable land, but the PCA showed that they were largely overlapping. This is a good point for discussion. Some explanations will be appreciated.

L334-336: “Several studies with…” has only one citation?

L351-352: Previous explanations are good reasons for using multiple parameter models. However, the reasons for using mixed-effect linear model are not well mentioned. For example, why not multiple fixed-effect model or partial least square regression? My recommendation is to stay in a safe way.

L373-374: To be prudent, I would say models of parent materials explained more variation of SOC because we don’t if the model-fitting was better than others (see comments on L282). The same for L374-375.

L379 and after: A major finding of this study is that the overall explained variance decreased in the order SOC>HWEC>MBC. Some explanations for this would be appreciated.

L395: Please be consistent with “mixed effect model” and “mixed parameter model”.

Figures and Tables:

Fig. 1 The clustered locations of the four parent materials are likely to introduce differences in topographical and climate conditions. For example, DCS and LBS sites are mostly located on the top of the mountain/hill, whereas PSS sites are located in a flatter area. The difference may affect soil formation and SOC accumulation. Also, the different altitudes between DCS and PSS sites may cause differences in climate conditions. Therefore, it is possible that the variation caused by climate and topography factors was explained by parent material or land use in this study. I just wonder whether something has been performed in experimental design, statistics or anything else to deal with this problem.

Table 1: What does the unit for respiration mean? As suggested for L146, more information of the incubation is needed.

A suggestion for Fig. 2: Why not combining Fig. 2 and Fig. S1 if you want to show the readers that parent material and soil texture make good separations while land use make an insufficient separation?

Fig. 2 and 5: The shape of the font might be improved as some of them are narrow but others are wide.  

Fig. 3 It looks like that the residue of MBC is less normally distributed compared to SOC and HWEC. Particularly, MBC in grassland soils is underestimated. Also, HWEC has a similar but less obvious trend. My questions are: (1) Is the model prediction of MBC less reliable than others due to the skewed distribution of residue? (2) Are there any reasons for the underestimation of MBC in grassland soils?

Fig. 4 What do “dataset”, “DCS”, “sand” and “arable” on the left mean?

Table 5: Does the “model” before “global model to local cluster” mean local model?

Fig. 6: Is it a part of Table 5? Is there any reason to make it a new Figure? Maybe try to combine Table 5, Table S3 and Fig. 6 into a good shape, or move unnecessary information to supplementary.