In this manuscript, Pyles and co-authors present a dataset of frontal ablation at Svalbard's tidewater glaciers. The dataset is comprehensive and highly resolved, and I expect it to be of great value to the science community. The manuscript is well written, detailed, and clearly illustrated with high-quality figures. I commend the authors on an overall enjoyable and insightful read! I have some minor recommendations and questions that I hope will improve the manuscript:

General comment:

The tone of language is at points -- in particular in the abstract and introduction -- a bit informal/florid or leans toward editorializing. Examples of this are: "...the manual labor cost inherent to mapping..." (L14) - manual labor is not strictly speaking inherent to mapping, so I would just simplify this to "...the manual labor cost of mapping..."; "...from the extensive ITS_LIVE velocity database, ..." (L 15), change to "from the ITS_LIVE velocity database, ...";  "unprecedented" (L19, L61) - I would just remove the word or replace with "high"; "is strongly recommended" (L 40) - reword?; I also recommend following the general guideline of avoiding the terms "novel" or "new" (the science should speak for itself, L 12 and 62). I would encourage the authors to go over the full text with this in mind.

Specific comments:

L 44: the statement on "decadal timescales" appears in conflict with the L 48 statement on annual/seasonal resolution by Minowa et al and Fahrner et al - rephrase?

L 64: Could you clarify why you are not using the Li et al (2024) frontal positions for this study?

Fig 2: While this figure is visually striking, I found it a little hard to interpret. I was trying to figure out which elements were schematic illustrations of the quantities used in the dataset, and which were artist's rendering/ aesthestic additions. I would urge the authors to simplify this and only include what is needed to make sense of the dataset/ processing chain.

L 165: Could you clarify in the main text what the OpenStreetMap is used for?

Fig 3: I found the overall layout of this Figure, with the different insets and zoom-ins somewhat confusing and was wondering whether it could be laid out slightly differently, or maybe panel a should be made its own figure?

L 261: Is the smoothing of the polygon boundaries necessary beyond an aesthetic improvement? Would it be worthwhile to also provide the unsmoothed polygons in the dataset?

L 340-375: I struggled a little to follow this argument; would it be worth adding a schematic to illustrate this process, or adjust Fig 7 to help with this?

L 386: It wasn't quite clear to me here whether you are using the Fuerst et al data or method here - could you please clarify?

L 423: How do you handle days without a velocity measurement?

L 460: Could you provide a short justification for using this reduced density of ice?

L 556: Are all the study polygons located in the ablation zone, in which case a negative CMB would be expected? Could you please comment on this?

L 571: The increase in tidewater termini length: might improved data coverage/quality over time contribute to this increase in the termini length? (Part of my question here is because I thought it a surprising finding - I would expect shortening of the termini as they retreat more onto land?)

Wording/Typos/Grammar etc:

L 27: "the ice-water-air interface" - these are actually (3) different interfaces; rephrase?

L 34: "total mass loss" -> "total ice loss" ?

L 80/89: The two sentences on diverse geometries feel repetitive and could be tightened?

L 129/134 and elsewhere: I was a little confused by the use of "inference" vs "prediction"? Are these terms referring to the same thing? If so, I would just use one of them, if not, maybe the difference could be clarified?

L 140: It may be worth spelling out "Randolph Glacier Inventory (RGI)" here, not just in the Fig 1 caption?

L 176: reword: "...the segmentation model, which is pre-trained ..." (?)

L 177: what do you mean by "unseen domain" ?

L 234: "consensus" seems maybe the wrong term?

L 304: "Like the cross-sectional area definition, ..." -> I struggled to make sense of this wording - could be clarified?

L 305: "We assume no uncertainty in the widths ..." -> rephrase: "We assume the uncertainty in the widths is small ..." (?)

Fig 6a: Is the map inset needed here, since you have the same inset in panel b?

L 487: "m^3 W.E. (Fig. 8)" -> the figure legend gives the total mass in Gt rather than the volume in "m^3 W.E." - could this be made consistent or clarified?

L 606: unit of Gt a^-1 (100 km)^-1 - as on L 608

L 655: remove "heavily" ?

L 663: reword: "mass loss due to frontal ablation *has* rapidly *increased* since 2000 ..."

L 696: "both *study* results"

L 701: These values should be negative, to be consistent?

L 734: "...and *its* implications..."

In this manuscript, the authors have presented a phenomenal dataset of Svalbard frontal ablation, covering monthly observations over a ten-year period. The dataset itself represents an impressive combination of different approaches/results into a single coherent dataset, and I have no doubt that it will prove useful for the glaciological community. I have one "major"/general comment about the approach and the work, along with a number of line-specific comments.

general comments

I think that at some point, an explanation of how you (or previous studies) selected the glaciers for this study is needed. You mention at line 142 that you start from the 2019 polygons from Kochtitzky et al. (2022b). I wasn't able to track that file down from the supplemental/data availability given in that paper, but I did find the Arctic_marine_terminating_fluxgates.gpkg file (https://github.com/willkochtitzky/FrontalAblation/blob/main/Arctic_marine_terminating_fluxgates.gpkg). From this, I found (I think?) 147 glaciers, based on filtering that dataset by Svalbard and removing fluxgates with "No longer" as a comment. That fits with the "RGI boxes" and "tidewater glacier basins" (line 12) you report, but I'm not sure I understand where the 151 number comes from. It doesn't come from RGI v7.0 (190 glaciers in region 07-01 with term_type = 1), or from RGI v6.0 (134 glaciers in region 07-01 with TermType = 1). Presumably, this is due in part to the RGI outlines often dividing a single terminus into multiple glaciers, combined with removing glaciers that are not marine-terminating over the full study period, but I think this should be stated/explained more clearly - especially if you are then tying this back to the RGI (which version? I can't find this outside of the Fig. 1 caption) with the "RGI-ID".

Looking at the dataset that you have provided, I do see the 2019 polygons, but I am less sure I understand the distinction between the "RGI box"/"glacier basin" and "glacier" as a result. Here, there are some RGI boxes that contain multiple polygons (for example, Nordre Buchananisen). This makes sense given the size of the glaciers included there, but there are also RGI boxes that nearly completely overlap (for example, Nuddbreen and Strongbreen), and I don't understand why these glaciers are not merged as has been done for a number of other RGI glaciers that converge in a single terminus (for example, the Nathorstbreen/Doktorbreen system). Additionally, there are boxes that don't fully encompass the glacier polygon (e.g., Doktorbreen). I think this does need to be clarified/explained in the text, and possibly revised in the dataset itself.

minor/specific comments

l. 36, "Arctic": I'm not sure that Yahtse (Bartholomaus et al., 2013) or LeConte (Sutherland et al., 2019) Glaciers can really be characterized as "Arctic" tidewater glaciers. I think this should either be modified (removing "Arctic"), or additional references specific to Arctic glaciers should be included here. Given that most of this paragraph pertains to tidewater glaciers in general, I would drop "Arctic".

l. 44, 53: while Table 1 in McNabb et al. (2015) only reports results for the full time period (ca. 1985 - 2013), the calculations are done on much smaller timescales, and plotted as an aggregated annual time series (Fig. 7). The utility of this for modelling is low, especially in comparison to the work presented here, but the results/discussion do at least include the interannual variation/trend for frontal ablation in the region.

Fig. 3: I suggest swapping panels a) and b), so that the context map is on the top, labeled a); the large panel showing all of Kongsfjorden could then be labeled b), and the detailed inset of Kronebreen could be labeled c). I also suggest showing the context box for the Kronebreen detail within the larger panel.

l. 221: I think that this assumption is something that you could test using, for example, the rates of advance/retreat reported from previous studies, or from your own results.

l. 222-226: can you provide additional detail about these metrics, or provide references for them? Metrics (2) and (3) seem straightforward enough (assuming that 'median intensity contrast' is just the difference between the median of ice-classed pixels and the median of ocean-classed pixels), but the others are less clear.

l. 233-242: I really like this approach, and the explanation.

l. 287: how were these glaciers chosen?

Fig. 6: it looks like the star for Lilliehøøkbreen is in the wrong place - it should be in the red box shown in the inset

l. 391: how many other glaciers had partial elevation change data?

l. 505-506: as with the comment at l. 221, I think this is something that can be checked/quantified - how big are the changes (as a fraction of the domain area) for all months, for each glacier?

l. 510: can you also report the number of months (out of the 15,500) are gap-filled in this way, and how many glaciers have gapless records?

Table 1 + in-text values: I think you should specify in the table caption that the reported decadal mean uncertainty values are the standard error of the mean uncertainty (i.e., mean(uncertainty) / sqrt(count)). That said, I think the better/more appropriate way to do this would be to calculate the root mean square of the uncertainties, which would give, e.g., -21.57 ± 3.06 for the frontal ablation, -5.60 ± 2.98 for the area change, 17.75 ± 0.65 for the ice discharge, and -1.78 ± 0.28 for the CMB.

Table 2: As for Table 1 - I think the more appropriate way to calculate the decadal uncertainty is as the RMS of the annual uncertainty, rather than the standard error.

l. 605: how is this intensity index derived? Is it just the frontal ablation normalized by the terminus length (in 100s of km)?

l. 609: similar to the intensity index, would it make more sense to compare the CMB correction normalized by the area between the flux gate and the terminus, rather than the magnitude?

l. 634, elsewhere: "Negribreen" is "Negri Glacier", so you can remove "glacier" after Negribreen (and any glacier name that includes "breen")

Table 3: as for Table 1

Section 5.1: I wonder if a portion of the difference between your results and Kochtitzky et al. (2022b) is due to the difference in ice thickness datasets used - either by spatially distributing observations from GlaThiDa, or from filling in with Millan et al (2022)'s results - compared to the SVIFT results (Fürst et al., 2018a) used here. How large are the differences for the thicknesses at your fluxgates compared to the results in Kochtitzky et al., for example? Could you plot, as a supplemental figure, the ice thickness along the flux gate used for both studies, at least for 1-2 glaciers with large differences?