See attachment

This manuscript introduces a new near-infrared hyperspectral (NIR HSI) scanning system for fast and continuous laboratory measurements of effective grain radius and infiltration ice content on polar firn cores. The system is based on previous work on seasonal snow and the authors first introduce their modifications to the system for firn core scanning. They then demonstrate that the results are consistent with traditional measurement of grain size and ice layer stratigraphy. They verify that cores should be cut into half rounds before scanning and that the results are largely insensitive to minor changes in objective lens focus. They then discuss the impact of the 2012 extreme melt season on firn structure as manifest in the 14 cores they study.

Overall, this is a well-written paper on an interesting analysis technique that has the potential to greatly improve the availability and resolution of structural data from firn cores, particularly if it can be adapted for use in the field. The paper seems quite technically complete. My only major concerns are with some of the discussion and interpretation of the 2012 melt signatures, but I also have a number of minor comments that I hope can help improve the clarity of the paper.  

Major Comments:

[1] How did you choose a threshold grain size of 1mm for classifying infiltration ice? Given you have a visual stratigraphy record for seven of the cores, I would have liked to see a more quantitative optimization of the grain size cutoff used to choose a threshold that maximizes the agreement between the NIR and visual ice layer stratigraphy.

[2] GreenTRACS cores 10, 11, 12, 14, and 16 were collected at locations that did not have coincident OIB radar collection in Spring 2017. As a result, the lack of 2012 melt layer detections at these sites in the Culberg et al. (2021) dataset is simply because there was no data in those regions to analyze. This means that most of the discussion from lines 374-382 needs to be removed or rewritten since the absence of the ice layer in those regions does not actually say anything about the detection limits of the radar. On the other hand, Core 1 did in fact have 2017 radar coverage, but the 2012 melt layer was not detected. Culberg et al. (2021) speculated that this might be because there was so much infiltration ice throughout the entire firn column that the 2012 melt layer did not form a unique or distinct ice package with strong density contrasts relative to the surrounding firn. The data presented in this paper seems to me to confirm that speculation quite nicely, and that would be an appropriate comparison and discussion that could be included in the paper.

[3] At lines 394-396, you note that Core 11 shows infiltration ice features outside the temporal range expected for melt effects from 2012 and suggest that this indicates deep preferential infiltration and refreezing. However, it seems to me that these features in Core 11 are more likely to have been formed during the relatively high preceding melt years in 2010 or 2011. There is often significant regional variation in exactly which year in the early to mid-2000s had the highest melt volume for any given site, so it’s possible that 2010/2011 is a stronger signal at this location. Core 11 is clearly in a unique region with much higher local accumulation rates than its neighbors, so I would not be surprised if local melt also followed a different pattern. To me, that makes more sense than suggesting that the 2012 melt infiltrated more than 2 m without impacting the firn structure before suddenly leading to large/rapid grain growth at depth (e.g., suggesting somehow no wetting front ever formed, despite high melt volumes).

Minor Comments:

How good is the vertical positioning in the reconstructed core-length images, compared to the spatial resolution in the vertical?

Maybe consider adding a table with all the imaging settings in one place for the reader who is interested in reproducing your setup.

Is the “grain radius” shown in all the figures the effective grain size (r_e)? Perhaps add that label explicitly on the colorbars.

Line 29-31: these two sentences could use a better transition. That is, show the reader how sentence #1 implies or leads to the statements in sentence #2.

Line 32: should be “pore close-off” rather than “pore-close off” as currently written.

Line 34: I am not sure if listing past accumulation rates as something that can easily be obtained from density profiles is quite accurate, since core dating is also required, which is a much more complicated/laborious process. But based on the references, maybe you meant that density profiles enable accumulation estimates from ice-penetrating radar measurements? In that case, I would suggest being precise about that application in the writing.

Line 38: perhaps "not that only important metric to characterize firn structure" would be more correct than “not a perfect proxy for firn structure”? Density certainly is part of the structure, not just a proxy, but it does not tell us all the information we need for sure.

Lines 43-45: grain size is largely irrelevant for radar measurements since the grain scale is orders of magnitude smaller than the wavelength (mm grain sizes vs decimeter to meter wavelengths). It may be relevant for some high-frequency microwave applications (scatterometers, radiometers, SAR, or radar altimeters) operating at center frequencies greater than 10 GHz. I would drop reference to radar and perhaps add a caveat on the frequency range of interest to the statement of on microwave sensitivity.

Line 89: “maps to millimeter – centimeter scale” is a bit ambiguous. Do this mean that pixel sizes in the image are of this order, or that mm-cm scale grains can be resolved?

Lines 100-104: these sentences would be more appropriate for the conclusion than the introduction. Here you just want to present a concise outline of the structure of the paper, not get into discussing the results.

Figure 2: consider adding a label for the Spectralon panels.

Lines 180-181: define “continuum normalized absorption feature” and “continuum reflectance” for the reader.

Line 186: did you run any sensitivity tests with higher impurity concentrations? 0 ppb seems quite unlikely for Greenland cores, and I assume there might be some chemical information available from the GreenTRACS cores to inform a better average value.

Lines 197-198: how do you handle core gaps and gaps from image cropping when reconstructing the full-length core images?

Section 3.1: is there any relationship between ratio of traditional to effective grain size and the degree of past firn wetting? Squinting at Figure 4, it seems like there might be a larger difference in grain size in wetted regions of the cores, but it’s hard to tell. This would be interesting to check in case rapid grain growth in saturated firn produces a different dominant grain geometry than the slower growth in dry firn.

Figure 3: consider adding the 2017 CReSIS OIB flight lines in light grey as an underlay in panel (a) so that it is clearer whether gaps in the ice slab and 2012 melt layer detections are due to lack of detections or lack of radar data.

The legend is a bit confusing here, since you use a black box outline to show the 2012 melt layer region, but the legend labels a white box with black outline as being infiltration ice. I assume the legend is trying to indicate that white colors in the grain size colormap are infiltration ice? Maybe it would be better to add white the top of the colorbar, label it as grain size > X, and then just note in the caption that regions of the core where the grain size exceed X are interpreted as infiltration ice.

Figure 5: there are some interesting spatial offsets here – for example, Core 10 where the NIR stratigraphy seems to be consistently translated downwards by a centimeter or two compared to the visual stratigraphy. What is the uncertainty in vertical positioning for each of the stratigraphic measurement methods and how might that affect your comparison here?

Section 3.3: I would consider moving this section earlier in the results. The best organizational flow would be to first present evidence that your methods are robust (sensitivity tests on core curvature and focusing), then quantitatively verify your results (comparison to traditional grains size and visual stratigraphy), and finish with interpreting those results (2012 melt layer stuff).

Line 336: typesetting issue with the ~

Figure 7: what are the distinct lenses/stripes of low grain radius – for example, in Core 11? Are these physical features, or an effect of core breaks and image splicing?

The histograms in panel b are pretty hard to read with this aspect ratio. It is nice to have them aligned with the cores, so I do not have an immediate easy fix, but consider playing around with some different layouts that might allow for some stretching of the x-axis so that the differences between histogram peaks within each plot are more legible.

Line 432: should be “SMS” not “SMK”?