Max Thomas et al. determine if and how a 1D parametrization of brine convection can be expanded to reproduce the laboratory PFA measurements of a previous study (Garnett et al. 2021). The paper consists of roughly four parts. In part one, the authors propose four different methods (A,B,C,D). In part two, they tune the free parameters of methods B, C, and D to minimize the absolute bias. In part three, they compare the model against the observed PFA profiles. Finally, in part four, they discuss if their results would apply to longer sea-ice simulations and other chemicals of interest.

The paper's text is clear and the method is clearly formulated (with one small exception) and applied thoroughly. The main conclusions are clearly stated, relevant to the field, and well supported by the results. The plots are mostly legible and not misleading, and the code has been made fully available. The topic of the submitted manuscript fits The Cryosphere. Based on my experience, the quality of the draft is well above average. 

However, I found the structure of the paper's second half confusing, and the paper is somewhat ambiguous about its scope. Moreover, the figures could be improved upon, and there are a few minor other issues to address. Accordingly, I recommend accepting the submitted paper, but with minor revisions. 

Minor comments in roughly descending importance

• Scope. The introduction clearly states that the paper aims to determine if decoupling can explain the observed properties. However, the methods introduced and the results discussed go beyond that. I feel that one or two paragraphs are missing at the end of the introduction to describe the other main question of the paper, namely if the decoupling is linked to the surface area, brine salinity, or constant. Furthermore, here I feel that the expectations should be clearly stated. From the current draft, I am unsure which methods B, C, and D closest match the known theory.
• Structure. I missed the transitions between results, discussion, and conclusions on my first read. In my view, the tuning of the methods and the analysis of the resulting parameters are the first results. In the current draft, this is a single sentence at the end of 2.2, and is then revisited in Figure 4. Furthermore, I believe the results extend till line 173, and the discussion begins by discussing how general the results are. (I enjoyed the discussion along with the supplementary material.) From lines 129 to 173 I get lost between all the comparisons of B to C to D, and some things are repeated multiple times (e.g. lines 155-159). I recommend breaking down the results into more bite size chunks, answer a question and then move on to the next. One of the questions I would like to see answered is what it means that B and D are so similar. Is there no T dependence in reality? Or, is the data insufficient to distinguish?
• What is the absolute bias |b|? I assume it must be the absolute difference over the vertical sum of the modeled and measured concentrations. But in line 112, it says the difference between the measurements and the co-located model layers, which implies that |b| should only be zero when the model and obs match at all layers. Moreover, how is the absolute bias scaled? What does |b| = 1 mean? Why use the absolute value? Showing b instead could clearly show that the higher alpha is, the higher the total concentration is. It might also make the lines in Subfigure 1b less confusing.
• Figure 1 has many lines that are difficult to distinguish. The readability could be improved by increasing the plots' width to use the paper's full width. The yellow line is also difficult to see; a darker tone would be helpful. There are no subfigure labels (a,b,c),  and shifting the legend outside the area of the subfigure would also help. The current version, in which the legend blocks the lines' view and overlaps with the figure borders, is messy.
• Figure 2 has too many lines and markers in too little space. This figure could be separated into two figures for profiles and scatterplots, but at least make full use of the paper width to make columns 1 and 2 twice as wide. This is now a minor detail, but I was initially confused by the axis choice for the right column. Since they share the same observation data, it makes more sense that the observation data be the x-axis—shared data on the shared axis. For example, one could easily compare where the 2.5 measured C12 is in each plot.
• Lines 170 and 172 reference some tests that can be passed or failed. I have searched the submitted manuscript and find no clue what tests these are.
• "was not a useful method" line 155. "useful" is not a well-defined adjective in this context. I recommend stating that C is worse than B and D and better than A.

I am trying to understand why the authors chose the name method A instead of reference or control. It is not a flaw and does not need to change, but I did find it strange that the first "method of decoupling" is "none". Accordingly, there is alpha_B, alpha_C, and alpha_D, but no alpha_A, and so on.