General comments

The article demonstrates that mechanical snow profiles can be statistically evaluated and stratigraphic features reconstructed. Traditional snow profiles are limited in number due to limitation of field work. Using 921 snowmicropen profiles from different expeditions, the authors build on the work of Kaltenborn et al, who demonstrated the use of machine-learning methods to evaluate highly resolved force profiles in snow. Based on their parameterisation and algorithm, significant differences between first-year and multi-year ice floes are quantified.

The statistical quantification of snow properties is  an important step towards synthetic modelling of the snowpack on Antarctic ice floes. The data are also essential to improve detailed snow pack models and develop stochastic modelling tools for these stratigraphically complex snowpacks. Concerning remote sensing, the data help to understand variation in albedo and microwave brightness temperatures. However, the article lacks standard statistical tests between observables.

On these lines, the paper is a major step towards a much more quantitative and statistically testable snowpack.

The paper is well structured. Propositions to improvements are detailed in the "Detailed comments". Figures should be adapted to the final format of TC, and the layout unified.

Detailed comments

Title: could be more specific, eg.

Different spatial scales and snow stratigraphy between first- and multiyear ice floes in the Weddell Sea, Antarctica 

Introduction: compare here the difference to the Arctic sea ice, cite the work of Sturm et al. 

line 22: a short introductory paragraph on the work in the Arctic on sea ice (traditional: Sturm et al, SMP the pioneering work of King, then the MOSAiC ) that forms in many ways the base to improved methods for the Antarctic.

(This is to some degree covered in lines 68 ff, but it much more in detail)

Lines 112-116: are the data of these other measurements available? Then cite data location, references. Otherwise delete

Figure 1: The background sea ice concentration is only for 2018. Likely it was different for 2019 and 2021 - relevant to the interpretation of the data? The geographic separation between PS111/PS124 (southeastern Weddell) and PS118 (northwestern Weddell) confounds the FYI/MYI comparison, since regime and region covary almost completely.

line 155: calculate a histogram of transect length and present. The calculation of autocorrelation are only statistically significant to about half of the total transect length (see Cliff & Ord 1981 on Moran's I confidence intervals).

line 163: why was the unmodified (?) Calonne parameters used? This model is calibrated to alpine snow at rather high snow temperature (average approx -5° C). Explicitly justify this choice and discuss potential biases.

line 214: "weighted Cross-Entropy Loss function": could you add a brief justification for the weighted version?

line 246 -248: repeated statement: "Rare ..." (cf line 209)

line 268: what do you mean by "and the finite vertical support of the SMP signal"? 

line 274: "by a single operator": The authors acknowledge single-operator labelling as a source of subjectivity but do not discuss whether this represents the dominant source of uncertainty relative to the ambiguity of the SMP signal itself. A brief discussion of this hierarchy of uncertainties would strengthen Section 3.3.

line 293-297: paragraph repeated from the methods section (cf lines 120 ff)

Table 3 between FYI and MYI (density, layer thickness, SSA, snow type fractions) lacks any formal significance testing throughout the paper. A Welch's t-test or Wilcoxon rank-sum test would be straightforward additions.

line 320 "is closely related to the snow grain size": this is only the case in theory: see the paper by Painter, Thomas H., Noah P. Molotch, Maureen Cassidy, Mark Flanner, and Konrad Steffen. “Contact Spectroscopy for Determination of Stratigraphy of Snow Optical Grain Size.” Journal of Glaciology 53, no. 180 (2007): 121–27. https://doi.org/10.3189/172756507781833947. 

line 347: replace "in the methods in Section 2.5" by "(see Section 2.5, page xx)"

Figure 7: very interesting data. I guess however that all values and interpretation beyond 30 m are spurious (see comment to line 155).

line 355 ff: Are this differences statistically significant? Perform a test.

5. Discussion

I wonder why you did not apply standard statistical tests too all your comparisons between FYI and MYI. Are the differences significant? Which differences are significant?

The dataset has 19 FYI floes versus only 9 MYI floes, and the MYI floes are almost entirely from a single expedition (PS118). This is a notable limitation, as it could bias the spatial autocorrelation comparisons.

Practical recommendation for sampling in future campaigns?

Code availability: TC data policy request that the code must be made public. Without code results are irreproducible, as the machine-learning part depends on subjective choices. This is strong request for the next round of reviews.