In this manuscript, authors evaluated the different of stabilized soil organic carbon (SOC) between forest and agricultural field along the profile down to 3 m in a subtropical catchment. Authors found that stabilized SOC content was not affected by land use below 90 cm, indicating a limited effect of deforestation on stabilized SOC in deep soil. Therefore, authors suggested that deeper soil layer is unlikely to serve as SOC sink for climate mitigation. Authors also found that stabilized SOC was predominantly controlled by land use, depth, silt and clay, and aluminium ion, while soil weathering degree was not relevant. The results support authors’ hypothesis that the difference of stabilized SOC between forest and agricultural field below 100 cm depth. While it should be noted that this conclusion only remains valid for the regions with highly weather soils in subtropical regions. This manuscript is generally well written with clear objectives, solid methodology and insightful discussion which meets the requirement for the publication in SOIL. Therefore, several issues still need to be addressed before publication.

Line 12: Please either use SOC or OC throughout the manuscript since they share the same meaning in this manuscript.

Line 16: Please indicate how many soil profiles (and soil samples) were used in this study.

Lines 26-27: at the scale of the soil profile? It is not clear.

Line 51: soil organic carbon can be replaced by SOC.

Lines 68-69: Please address the recent MEMS 2.0 model that uses measured SOC fractions for modelling.

Zhang, Y., Lavallee, J.M., Robertson, A.D., Even, R., Ogle, S.M., Paustian, K. and Cotrufo, M.F., 2021. Simulating measurable ecosystem carbon and nitrogen dynamics with the mechanistically defined MEMS 2.0 model. Biogeosciences, 18(10), pp.3147-3171.

Line 75: vegetation(Cotrufo The space is missing here.

Line 80: Is it necessary to separate SOC from TOC? If they have the some meaning, then the use of SOC would be enough.

Line 85: silt (soil particles in 2- 53 μm) would be better.

Line 95: (Alcántara 95 et al., 2016) the font for this text is different from others. Please correct it.

Lines 143-144: I expect to have more information about the approach for soil sampling design. And there you should indicate how many soil profiles were collected not just the number of soil samples.

Lines 158-159: How you get the information of weathering degree before laboratory analysis for choosing the sites for laboratory analysis? More detailed information is needed.

Lines 227: Grain size is rarely used, please use particle size instead. Then the use of silt (2-53 μm) and sand (53-2000 μm) would be clearer.

Lines 239-240: A big concern here is that MIR technique tends to overestimate the low value while underestimate the high value, even the model performance is high. As a result, the high MAOC under forest soil would be underestimated, which potentially leads a close result to the MAOC under agricultural field. I think authors should carefully address this issue in the discussion.

Figure 2: How you conduct the paired comparison at a given depth interval since all the soil profiles were collected from genetic horizons?

Figure 4: The unit is needed for S&C fractions in the x axis. Please also provide the linear equation here.

Line 378: The effect of land use.

Lines 395-397: What are the potential reasons for the difference between this study and previous studies?

Lines 486-487: why 0-20, 25-50 and 45-90 cm were selected here, instead of 0-20, 20-50, 50-90 cm?

Lines 489-490: Maximum potential of SOC is predicted by silt+clay, therefore there is no doubt that silt and clay are important.

This is an interesting study in which the authors assess the stabilized SOC content of soils under forest and agriculture in the humid subtropics of Southern Brazil. The samples they use were sampled down to 300 cm in some cases. The laboratory methods that they use are state of the art. They find that the amount of stabilized OC was mostly controlled by land use and soil depth, in addition silt + clay content and exchangeable Al played a role. They cannot show any land use change effect on stabilized OC below 90 cm depth; but do show that the subsoil has not reached above 50% of the ‘stabilized OC saturation point’. They conclude that in their study area deforestation does not affect SOC content below 90 cm deep and that it is unlikely that deeper soil layers can serve as an OC sink in timescales relevant for climate change.

The soil samples used in this study were not collected for the purpose that they were used in the present manuscript. Instead they were collected to study how slope gradient affects soil thickness and chemical weathering. I went back to the study by Brosens et al (2020) and learned a few interesting things about the sites which are also relevant for the present study: the study area was chosen because of it relatively homogeneous lithology so that variation in soil depth and weathering are primarily related to topography. Furthermore, site selection was done in a stratified random way, with the goal to cover a wide distribution of slopes. Land use type or history were not accounted for. Only mid-slope positions were sampled. The saprolite in the study area consisted of loose sandy material.

Because of the sampling design that was conducted in this study (largely random), the samples probably included a substantial amount of spatial variability that would have been less if you had collected the samples specifically to detect land use change effects on SOC. We ran into this problem when we conducted sampling for SOC in a montane tropical landscape (de Blecourt et al, 2017). We concluded from this experience that ‘scale-dependent relationships between SOC and its controlling factors demonstrate that studies that aim to investigate the land-use effects on SOC need an appropriate sampling design reflecting the controlling factors of SOC so that land-use effects will not be masked by the variability between and within sampling plots’. Compared to the study by de Blecourt et al. (2017), your sampling design has the advantage that you have a relatively homogeneous lithology, nevertheless, your sampling design did not reflect the potential controlling factors of SOC content.  I think it is therefore safe to conclude that spatial variability caused a larger variance than would have been the case had you sampled specifically with the aim to detect changes in land use (e.g. by doing paired sampling).

This also brings me to your conclusion that you cannot show land use change effects below 90 cm depth. I was wondering how much this conclusion is also affected by your sampling design? Not being able to show differences means that the variance is too large to detect differences. But as I mentioned earlier your sampling design included spatial variability that would not have been sampled if you had purely focused on differences in land uses. So how large are the chances that this is the case in your study? In a study that we conducted in Indonesia we ran into similar problems (Allen et al 2016). We analyzed this problem by conducting an analysis of variance components and a power analysis. It turned out that in the Allen et al., (2016) study a substantial part of the variance was caused by variance within the replicate plot, which explained why we were not able to show land use related changes in some soil characteristics. Furthermore, we were able to show the optimum sampling size using a power analysis with a power of 80%. My point is you write it correct: you were not able to show any land use change effect on stabilized OC below 90 cm depth, but maybe this was simply caused by you sampling design in combination with the relatively low number of samples in the deeper part of the soils. You can analyze this by conducting a power analysis, which I encourage you to do.

A potential additional problem is that the deeper soil samples that you compared comprised a thicker depth interval. I understand the reasoning to do this: you wanted to have a balanced design when you analyze deeper intervals in the soil.  However, since you did not sample the whole depth interval but only discrete samples at different depths, this may have introduced additional variance: Normally SOC contents will decrease with depth and if you consider a depth interval as thick as 160 cm (l. 256) you automatically include systematic variability simply because of the large depth interval takes for your statistical analysis. This again may affect your variance and with that reduce the probability to show differences at larger depths.

If I understand it correctly you are comparing depth intervals independent from the depth of the soil. So, if you compare a 30-40 cm depth interval from a shallow soil on a steep slope with the same interval of a deep soil on a flat area, you may actually compare something close to saprolite (shallow soil) with soil material that is extremely weathered (deep soil). Did I understand this correct? If so how much may this have contributed to the variance of your samples? Normally we would like to compare similar soils, you may not have done this if you put shallow and deep soils together.

A few minor remarks:

-you mention that you land uses are all >30 years old. According to Table s3 you have also more specific ages, and I would have been very interested to see if your conclusion were the same if you only include sites that are 50 years or older?

-All your samples were taking from mid-slope positions (Brosens et al., 2020). How may this have affected the outcome of your study?

-If you sampling design indeed affects your results as much as I anticipate, I suggest that you include more information of your sampling design in your manuscript. It will help the reader to understand the possibilities and limitations of your analyses.

In summary, I think this is an interesting study but the limitations that are caused by the sampling design  may be substantial and should be more prominently discussed. After that is done I expect that it can be published.

I hope you find my suggestions useful.

Best regards, Edzo Veldkamp