Greenland Ice Sheet late-season melt: investigating multiscale drivers of K-transect events

One consequence of recent Arctic warming is an increased occurrence and longer seasonality of abovefreezing air temperature episodes. There is significant disagreement in the literature concerning potential physical connectivity between high-latitude open water duration proximate to the Greenland Ice Sheet (GrIS) and late-season (i.e., end-of-summer and autumn) GrIS melt events. Here, a new date of sea ice advance (DOA) product is used to determine the occurrence of Baffin Bay sea ice growth along Greenland’s west coast for the 2011–2015 period. Over the 2month period preceding the DOA, northwest Atlantic Ocean and atmospheric conditions are analyzed and linked to lateseason melt events observed at a series of on-ice automatic weather stations (AWSs) along the K-transect in southwestern Greenland. Surrounding ice sheet, tundra, and coastal winds from the Modèle Atmosphérique Régional (MAR) and Regional Atmospheric Climate Model (RACMO) provide high-resolution spatial context to AWS observations and are analyzed along with ERA-Interim reanalysis fields to understand the meso-to-synoptic-scale (thermo)dynamic drivers of the melt events. Results suggest that late-season melt events, which primarily occur in the ablation area, are strongly affected by ridging atmospheric circulation patterns that transport warm, moist air from the subpolar North Atlantic toward west Greenland. Increasing concentrations of North Atlantic water vapor are shown to be necessary to produce melt conditions as autumn progresses. While thermal conduction and advection off south Baffin Bay open waters impact coastal air temperatures, local marine air incursions are obstructed by barrier flows and persistent katabatic winds along the western GrIS margin.


Specific Comments:
L32 "For the unseasonal melt period preceding the DOA": Which unseasonal melt period? This seems to imply a sustained period of unseasonal melt before the DOA, whereas the unseasonal melt usually occurs in transient events. I suggest rephrasing as simply "For the period preceding the DOA".
Thank you for the clarifying suggestion; we have made this change.

L40 "While thermal conduction and advection off south Baffin Bay open waters impact coastal air temperatures, consistent with previous studies": Where is this shown in the manuscript?
Heat transfer off of ice-free ocean waters onto coastal areas is discussed at length in the second paragraph of the Introduction.
L47 "The Greenland Ice Sheet (GrIS) surface mass balance (SMB) decrease has contributed roughly 0.5 mm year-1": It's the overall mass balance decrease (SMB + ice discharge) that has contributed 0.5 mm/year, not the SMB.
We have made this clarification in the manuscript.

L50 "west Greenland waters" -> "ocean waters west of Greenland"?
We have made the suggested change.
L83 "recent, temporally-anomalous GrIS late melt events spanning the end of summer to freezeup of adjacent ocean waters": Which recent events? Please provide a few examples and references.
We have slightly edited the text to reflect satellite era studies of late season GrIS melt in late boreal summer and autumn. Two papers referenced include Stroeve et al., 2017. L112 and L156: Can you provide some further explanation as to why KAN_B was selected to build T+/T-composites, thus limiting the study period to only 5 years? Especially since the objective is to study melt events, why choose a station on the tundra instead of a station on the glacier? KAN_B has a much shorter data record than all the other K-transect stations, so the selected study period [2011][2012][2013][2014][2015] is only a small sub set of the available data. With the short study period and with composites sub-divided into D_60 and D_30 bins, the D_30 composites comprise only 13 days of data. The next closest station, S5, has data going back to 2003, which would allow for a more robust analysis over a 13-year study period. If you repeat the analysis using S5 to build composites over 2003-2015 and S4-S5 change very much? I think it would be preferable to use station S5 and the longer study period for these figures, or at least mention whether the KAN_B-based results are robust when S5 and the longer period are used. The only place I see a need for a shorter study period is in Table 2 and Fig. 2, where all the K-transect stations are compared over an overlapping period, but if KAN_B were excluded then I think the overlapping period could still extend back to 2008 instead of 2011.
Our study provides a temporally homogenous examination of above-freezing temperatures along all current PROMICE and IMAU K-transect stations. The study spans 2011-2015 so that both networks of AWS observations can be included. We use above and below freezing air temperatures from Kan_B, located just off the ice edge, as bins for comparing atmospheric conditions about the transect. The intent is not to build the longest record of unseasonal melt conditions, but the most spatially complete. Table 2 the frequency of T+ events at KAN_B (or S5,if it were used instead of KAN_B to build composites), and mention what percentage of days each represents, i.e., the D_60 bin has 69 events (∼46% of days) and the D_30 bin has 13 events (∼9% of days).

L163: It would be helpful to summarize from
We have briefly summarized the frequency of T+ events from table 2 within the first paragraph of the methods.
L176 "IVT PR is then classified using the SOM technique to produce a matrix of moisture transport patterns, or nodes, that typically occur over the Greenland region": This is very vague. Some additional description should be included for the benefit of readers who are not familiar with SOM -for example, mention that it is an unsupervised machine learning algorithm and that each daily IVT PR field is classified with its closest matching node from the SOM. Are the SOM dimensions and training parameters the same as in Mattingly et al. (2016), or is anything different? It's not clear from this description. It would also be helpful to include in the Supplemental Materials a figure showing the SOM nodes and the wet/dry/neutral groups, so that readers have a better understanding of the SOM classification used in this study.
We have added some clarifying language in the last paragraph of the methods section on the SOM method and application of the Mattingly et al. (2016) SOM classification, from which we assess frequencies of wet, dry, and neutral IVT patterns as identified by the authors. We refer readers back to the aforementioned manuscript to view the SOM matrix referenced here. Mattingly et al. (2016), similar wet (anomalously high), neutral (near climatological median values), and dry (anomalously low) IVT patterns are aggregated, then their frequencies are composited and tested following methods previously summarized": What is meant by "tested" in this context? In the cited paper, the clustering of nodes into wet/dry/neutral groups was subjective, based on visual inspection. I don't recall any testing of this grouping procedure.

L178 "As in
The hypothesis tested is that the frequency of IVT patterns belonging to a given SOM node cluster (i.e. "wet" nodes) differs for T+ events compared to T-events. As you note, the grouping of nodes is based on visual inspection of the IVT patterns for each node, and the grouping procedure itself is not tested in this paper. We have edited the text to clarify this point.
L208-231 I find it difficult to follow the reasoning in these two paragraphs that leads to the conclusion that "late season melt inferred from T+ events may be driven by synoptic patterns as opposed to local marine forcing" (L234). Why exactly do we reject the hypothesis that local marine forcing drives the T+ events? Is it simply because the K-transect wind directions in T+ events are offshore instead of onshore? What about the other aspects discussed here -such as Ktransect wind speeds, sensible and latent heat fluxes over ocean and GrIS, near surface winds in eastern Baffin Bay, etc. -how do they contribute to this argument? These paragraphs should be revised to clarify: if local marine forcing were driving the T+ events, then what would we expect the results here to look like? How do the actual results differ from that?
We have clarified a hypothesis of what physical dynamic and thermodynamic mechanisms would link Baffin Bay open water to GrIS melt in the last paragraph of the Introduction section. Figures S4 and S5 are described here as positive in the upward direction. Is this correct? I would have expected sensible heat fluxes over GrIS to be directed from the atmosphere to the ice sheet surface (i.e. negative SHF if the upward direction is positive). In particular, in the K-transect region during T+ events, if air temperature is greater than 0 C but the melting ice surface is at most 0 C, how can SHF be directed upward?

L214-218 and Figures S4-S5: Sensible and latent heat fluxes in
We have corrected the wording and the positive sensible and latent flux represents downward heat transfer by which the atmosphere heats the ocean and ice sheet.
L240 "In both T+ cases, low-level winds circulate poleward over north Labrador Sea areas of upward, turbulent heat flux (Figs. S4 and S5), aiding the heat and moisture transfer (as shown by heightened IVT values in T+ relative to T-) over western Greenland during D60 and D30 (Fig. 5 Figures  S4 and S5, I am wondering if turbulent fluxes over the north Labrador Sea are actually upward and if this region is actually the source of the heat and moisture transported onto western GrIS? Or do the heat and moisture originate from further south, around 30-35S? Subtropical origin seems more likely and is more consistent with the IVT fields shown in Figure 5.

)": Following from my previous comment about the direction of SHF and LHF in
We have corrected the sign as indicated in the previous comment. While the objective of this work is not to pinpoint origins of subpolar air masses traveling across the ice sheet, we suggest in the Discussions these air masses originate in deeper in the Atlantic than the domain resolved by our synoptic analyses. Upward turbulent heat fluxes (negative sign SHF and LHF in this analysis) over these remote areas may serve enhance the heat and moisture contributions to those air masses.
Of note, we have also changed D30 and D60 to [-30,-1] and [-60,-31], respectively, for consistency between text, figure, and table references to the time periods preceding Baffin Bay DOA.
L254 "higher 500 GPH values and on-ice lower tropospheric mean winds": The phrasing here seems confusing. "Higher" refers to GPH being higher in T+ relative to T-events, but does "lower" refer to wind speeds in T+ relative to T-, or does it mean "lower troposphere"?
"Lower" refers to the troposphere and specifically the winds averaged over the 1000-700 hPa levels. We have clarified the corresponding language in the last paragraph of Results section 4.3. Figure S6: Since each station-year is a separate data point, why not include all years of data for each station in Fig. S6 and related discussion? Only using the 2011-2015 period gives a very limited perspective on whether the temperatures are typically correlated or not, since we see statistically significant correlations in only three of five years.

L267 and
We prefer to include a temporally homogenous correlation analysis of the nearby oceanic SSTs versus AWS air temperatures for the 2011-2015 period. Subsequent analyses may look to capture longer term local oceanic and coastal/ice sheet temperature linkages as well as the effects of the regional ice sheet-ocean pressure gradient in modulating onshore or offshore flows during T+ and T-periods. We have made the suggested change. Table 2: I find the labels here a bit confusing. Although each row is labelled as a relative measure (e.g., "S5 vs. KAN_B"), the columns "T+ n [-60, -31]" and "T+ n[-30, -1]" are not relative to KAN_B, they are the actual counts for each individual station.

Typographic Corrections and Figure/
We have clarified the language of the table to caption to more clearly describe the table contents.   We have made the suggested figure change.
L518: The last four references are not in alphabetic order with the rest of the bibliography.
We have adjusted the placement of references accordingly.

Reviewer 2: Charalampos Charalampidis
General comments. The Discussion Paper by Ballinger et al. is an interesting and novel attempt based on in situ, Automatic Weather Station (AWS) observations and Regional-Climate Model (RCM) output to identify the prevailing atmospheric-circulation pattern that causes unseasonal melt events in the southwestern Greenland ice sheet, and whether sea-ice formation and related turbulent-heat production over Baffin Bay in early autumn is of importance. The study is characterized by an accurate and well-structured methodology, but in my opinion, not also interpretations. As an example, the authors conclude that they "find no evidence to support the hypothesis that local open water and resultant turbulent heating has a demonstrable impact on inland ice melt events", even though the approach is purely statistical; the study does not disprove the hypothesis, but rather proves that it is not the dominant mechanism.
Structurally, the study suffers from obvious overlooks and requires attention. As examples, I note the allegedly examined AWS pressure and humidity according to the Abstract, even though the analysis is based on AWS air temperature and wind proper-ties only, and the mention of a Sisimiut AWS in the Results and of an Upernavik AWS in the Discussion, without being previously introduced or presenting any related figure in the main body.
I recommend a major-revisions status, primarily to provide the authors with ample time to improve the structure and reasoning throughout the text. I also have a comment on the methodology concerning the AWS analysis that, in my opinion, needs to be addressed before publication.
We appreciate the reviewer's comments on the paper and have addressed their comments below and in the paper accordingly.

Specific comments
The linguistic level is good, however sentences tend to be condensed, and would benefit by a more elaborate style. Since the study is not purely meteorological and aims potentially to appeal to a wider Earth sciences audience, some effort should be invested into familiarizing readers with meteorological norms instead of just mentioning parameters, numbers, and units. Additionally, the reader's knowledge over methodological approaches is taken for granted. As a rule of thumb, a second-year Earth-sciences master's student should be able to follow without referring to textbooks for clarifications.
Revisions are aimed at structural improvements to improve readability in line with the journal's cryospheric science audience.
The study has a broadly clear structure, but the argumentation is not straightforward, and complicated -sometimes even speculative -discussion points are widespread in Results and Conclusions. Also, as a rule of thumb, referencing supplementary material in Results or other studies in Conclusions should be avoided, in order for the study to be self-standing.
Effort should be invested into describing the theoretical background and the study area, particularly its wind regime and mechanisms, as well as recent Surface Mass-Budget changes within the Introduction or before Data as a self-standing section, so that readers can easily relate afterwards to the Discussion. Recommended literature: Van Angelen et al. (2011, and the classic GIMEX-90/91 work. The inclusion of a description of the different areas of a glacier, as well as an Equilibrium-Line Altitude definition and how high it is in the study area, is also recommended.
We have touched upon these components through the manuscript where appropriate to complement given research objectives and details of the K-transect study area.
The use of KAN_B observations as basis for the analysis needs to be motivated better than just saying that it records more above-zero Celsius air temperatures than all the other K-Transect AWSs. The way I see it, by the end of August, after the end of the melt season, the increased solar-zenith angle and ice-sheet surface albedo limit the influence of solar radiation on potential ice-sheet surface melt, which is thereafter driven by climatic drivers influencing downward longwave radiation (i.e. cloudiness) and turbulent fluxes (e.g. September 2010; Charalampidis et al., 2015). Also, katabatic-wind flow, driven by the balance of downward sensible heat flux and radiative cooling at the surface, becomes gradually colder and denser, and therefore more intense and laminar, thereby hindering turbulent mixing over large part of the AWS transect that is characterized by decreased surface roughness (Smeets and van den Broeke, 2008; see also specific comments for L208-209 and L211-218). Hence, melt events might occur in cases of weak katabatic-wind flow and concurrent northwesterly atmospheric advection from Baffin Bay, or moderate to strong katabatic-wind flow and concurrent southerly barrier winds along the icesheet margin driven by synoptic circulation.
We have clarified our motivation for using KAN_B above (T+) and below (T-) freezing events for compositing across the K-transect, and the compositing methodology, in the Methods section. Acknowledging the ice-albedo feedback is weak in autumn relative to summer months, we have edited the Discussion section to address local, mesoscale, and synoptic winds supporting T+ events. In concluding that synoptic forcing strongly influences our T+ cases, we have also briefly discussed the implications of enhanced moisture advection and resulting cloud longwave effects on melt.
Located only 1 km away from the ice-sheet margin, KAN_B should be the station most sensitive to both these circulation patterns. In both cases, the local boundary layer will become more humid (the role of humidity on energy balance and as a cause of surface melt should be explicitly described), and KAN_B might record positive air temperatures, in which case it is also worth inspecting all on-ice AWSs for positive air temperatures, which would be indicative of surface melt. (It should be explicitly mentioned in the text that positive near-surface air temperature is used as an indicator of melt at the ice-sheet surface.) A pan-Greenland evaluation of positive on-ice air temperature anomalies (above-freezing or simply above-normal) is beyond the scope of this paper. Future analyses will look to expand the spatial fingerprint of local and remote forcing of marine intrusions on Greenland air temperature events, but here we wish to focus solely on the data-rich K-transect for this proof of concept assessment.
The statistical analysis should reveal which of the aforementioned cases is the most frequent (and not simply differentiate between regional meteorological processes acting during positive and negative KAN_B air-temperature events, as currently stated in L154-155. What goes on during negative KAN_B -and hence K-transect -air temperatures is not the focus per se; it is just used as means of comparison). The above theoretical introduction material along with Figure 2 alone proves, in my opinion, that synoptic circulation in the South of Greenland is the driver of unseasonal melt events mostly in the lower ablation area until the elevation of KAN_L. The persistent katabatic wind regime acts otherwise as a shield against turbulent forcing from Baffin Bay. All the rest should be an elaboration around this key result.
We have adjusted the wording in the methods section to better focus attention on the common timing and physical processes bringing above-freezing air to KAN_B and S5 and KAN_L, thereby generating "unseasonal" melt at latter those on-ice stations.
It might be insightful to make a similar comparison of KAN_B with observations further away from the ice-sheet margin. I do not know what the result might be, but I am guessing weakening southerly wind away from the margin in case of synoptic-driven, barrier-wind occurrences, and otherwise more intense winds than whatever KAN_B is reporting at that time, directed toward the ice sheet.
As previously mentioned, follow-up analyses will take a wider spatial examination of thermodynamics and dynamics of unseasonal melt.
Nevertheless -and this is my major methodological concern -it is not clear in the text how the KAN_B temperature events are defined. Are these definitions based on hourly or daily observations? Is there a time window plus/minus delta_t around a warm/cold KAN_B event t within which the conditions at all other stations are evaluated? Is delta_t selected larger or smaller depending on the duration of the KAN_B event t? Eventually, how sensitive are the statistics, and how conclusive the implications, between hourly and daily analysis? Please, elaborate.
Air temperature composites are based on daily mean air temperature values above and belowfreezing at KAN_B. Therefore, T+ is one bin, and T-represents the other bin. We have taken strides to clarify this partitioning in the manuscript. Duration of T+ and T-events and sub-daily analyses are beyond the scope of the study, but will be examined in follow-up research using climatological seasons.

Technical comments
Please, make use of dashes, or hyphens if appropriate, throughout the text to facilitate readability in case of several nouns in a sequence, e.g. in the Abstract alone We have made recommended changes where appropriate throughout the paper.
We have attempted to use hyphens where appropriate and differentiate dashes and minuses in the text.
Use the more appropriate "area" instead of "zone" throughout the text.
We have made the suggested substitutions.
I recommend the use of "observations" instead of "data", while differentiating between in situ and remotely sensed, and "simulations" for modeling products.
We prefer to use "data" as a more general reference to avoid confusion as the climate science community often tends to characterize reanalysis output, for instance, as observational or modeled based on variable studied.

I recommend the use of "average" instead of "mean."
We prefer to leave the term "mean" as it is used throughout the paper.
We have made the suggested addition of "Baffin Bay" for clarity.
Capitalize the first letters of every word when an abbreviation is introduced, so that the reader's eye can make a quick connection, i.e. L30: Date Of sea-ice Advance (DOA).
We prefer to leave the referencing as it is given in the manuscript preceding the acronym. L27: delete "significant", and reserve it exclusively for describing correlations.
We indicate "significant" as it relates to correlations and t-test differences between T+ and Tevents.

L32: "unseasonal melt events. . ."
We are not clear what this comment refers to, but the phrase has been previously defined in the abstract.

L34: "Southwest" or "southwestern"
We have made the suggested change to "southwestern." L34: The influence of synoptic and mesoscale systems on the above-and below-freezing nearsurface air-temperature events. . .
We have altered the referenced sentence as indicated in the following point.
L35: "AWS" -> "the"; "The in situ observations..."; Why are pressure and humidity observations mentioned here? I was unable to spot them later in the study. Why is wind speed and direction not mentioned?
We have removed the reference to pressure and humidity and inserted some commentary on the AWS wind speed and direction.
L35-38: I suppose "MAR, RACMO2, and ERA-Interim are used to provide context to the in situ observations by explaining the air-mass origins and the (thermo) dynamic drivers of melt events." As it is written now, it gives the impression that the in situ observations need to be calibrated against RCMs.
We have altered the wording to clarify the role of MAR, RACMO, and ERA-Interim data. As mentioned, these data provide regional and synoptic context to the local AWS winds. MAR and RACMO have both been extensively calibrated against on-ice AWS observations (e.g. from PROMICE and IMAU station records) as mentioned in the Delhasse et al. (2018) and Noël et al. (2018) references included in the Data section. We have removed this phrase.
L42: Consider removing "pressure-gradient driven" from the Abstract, and describe briefly the mechanism in the Introduction.
We have removed this phrase and paired pressure-gradient discussion with a new composite pressure and 10-m wind figure.
L42: "Katabatic-wind regime", since the word "katabatic" is an adjective referring to a descending object, in this case, wind.
We prefer to leave as "katabatic regime." L41-42: I think this is an overall complicated explanation. I think the primary obstacle of Baffin-Bay marine-air intrusions over the ice sheet should be intuitively the persistent katabatic wind regime that intensifies in the beginning of autumn. Additionally, barrier winds might be present, in which case melt might occur due to air-mass transport from the Atlantic. I propose restructuring as: ". . .are obstructed by the persistent katabatic-wind regime flowing downslope from the ice sheet, and the occasional occurrence of barrier winds along the ice-sheet margin.", or something along these lines.
We have made a similar change as suggested to clarify the role of katabatic winds in "blocking" the marine layer transport onto the K-transect.
L46: "have become sensitive" or "are becoming increasingly sensitive" We have made the change to the suggested latter phrase.
L48: Since all Van or Van de Author have been categorized under V in References, consider capitalizing the first letter of "Van" in the in-text occurrences, as well as that of the first author in the References, in order to allow the reader to navigate through easily.
We elect to leave the author's names as they are given.
L53: Replace "play a key role" by "are of importance" We have made a similar suggested change.

L54: "surface melt at the southwestern GrIS."
We elect to more generally include "western" before "GrIS melt." L55: "Conflicting evidence have been presented in literature over the past decade regarding. . ." We have made the suggested change.
We have made the suggested changes. Ilulissat and Nuuk,approximately 200 [Unit] to the north and south of Kangerlussuaq, respectively. . ."

L63: "in
We have added the appropriate reference to km and "respectively" as suggested.
We have made the suggested change. L67: Please, delete "robust". Statistical quantities are a matter of context, and their interpretation is dependent on the datasets and the research question itself, which in this instance are not present; "(from Markus et al., 2009

L69: "The authors found that significant, positive correlations between Baffin and Labrador SST and coastal SAT often persist. . ."
We have made the suggested change.
L70: "onset of freeze" should suffice; "Applying a similar correlative approach on. . ." We have adopted the first change, but elect to maintain rather than change the second wording recommendation for this line.

L71: "while utilizing melt/freeze product. . ."
We have made a slight change in wording here.
We have made the first change, but elect to keep the second as "upper-level" is commonly used to describe winds in the mid-to-upper troposphere.

L75: Replace "at the limits of" with "toward the end of".
We have made this change.

L81: Replace "research studies" with "literature".
We have made this change.

L86: Delete "automatic weather stations", since AWS has already been introduced.
We have made the change since AWS introduced in the Abstract. We have made this change.
L101: Delete "day of sea-ice advance", since DOA has already been introduced.
We have made this change, but have clarified the DOA metric.

L102: Bliss et al. (in review)
This work is "in press" and we have updated the reference and in-text citation.
L112: "in this study (Fig. 1)." We have added a the reference to Fig. 1.

L113: "KAN_B that is situated approximately 1 km away from the ice-sheet margin. . ."
We have clarified the statement as suggested.

L117: Delete "tundra (KAN_B only) or glacier ice"
We have made the suggested change.
L116-117: The observational distance from the surface decreases during winter season due to the accumulating snowcover around the AWSs by as much as two meters depending on the location along the transect, while it increases again each melt season until the complete ablation of the accumulated snowpack. In the case of KAN_U that is located in the lower accumulation area, this distance tends to become shorter over the course of a few positive massbudget years due to the incomplete ablation of the snow cover, but also due to the sinking of the AWS tripod during ablation in the temperate firn below, as discussed by Charalampidis et al. (2015). After the extreme 2012 melt season, snow accumulated around KAN_U during the 2013 and 2014 positive mass-budget years, while on 3 May 2015 the half-buried tripod was replaced.
We appreciate the author's insights on the K-transect AWS air temperature measurement heights, which are complicated by snow accumulation and ablation through the annual cycle. We have added the suggested citation and an acknowledgement that the sensor heights fluctuate with these mass balance components.

L121: Consider including also Citterio et al. (2015) describing the PROMICE AWS.
There are many papers describing aspects of the IMAU and PROMICE AWS networks, but we elect to mention a concise list here.

L125: Delete "have been shown to"
We have made the suggested change.
L127: "from between 1000-200 hPa height, corresponding to the distance between sea level and lower stratosphere at 67 N (e.g. Zängl and Hoinka, 2001). . ." We appreciate the reviewer's reference, but we elect to keep the original description of integrated vapor transport (IVT), which is elaborated on in the Methods section and follows from work of Mattingly et al. (20016, 2018).
We have made the suggested change.
L132: "10-m above surface and 850 hPa height. . ." We have shortened the sentence, but elect to keep the core "Wind speed and direction at 10-m and 850 hPa" intact as these represent standard reference levels for these dynamic variables.
L143: "ice caps, and improving. . ." We have made minor edits to the sentence to make it clearer.
L145: "to characterize near-surface. . ." We have made the suggested change.

L151: It should be mentioned that above 0 C air temperatures at on-ice AWSs are considered an indicator of ice-sheet melt.
We have integrated the suggested addition.

L152-165: It is not clear how the KAN_B temperature events are defined. I refer to Specific comments
We have made efforts to more clearly explain the compositing approach within the Methods section.
L157: "due to the station's location. . ." As with the previous comment, we have clarified the compositing approach.

L171: "Based on RCM output, . . ."
We clarify that ERA-Interim data is used for IVT calculation and subsequent classification.

L172: Delete "integrated water-vapor transport", since IVT has already been introduced.
We have made this suggested change.

L183: Refrain from mentioning supplementary material as part of the text, and only refer to them in parentheses at the end of sentences. Nevertheless, supplementary material should be referenced outside Results, and if Fig. S1 should be referenced in the first sentence of the Results, perhaps it belongs in the main article.
We elect to keep figure references in their current place with the figures included representative of the main results of the manuscript.
L189-191: Nice, but perhaps also mention that the difference has narrowed primarily due to the prolongation of the melt season over the course of the thirty years.
We have added some remarks with regards to melt season lengthening.

L193-195: It is a bit awkward to see all of a sudden a Sisimiut station being mentioned here in the first subsection in Results even though it has not been introduced earlier, and a case being made based on supplementary material. Please, restructure.
We have introduced the DMI surface air temperature data for Sisimiut and Kangerlussuaq AWS at the conclusion of the Data section. We have also clarified their supplemental usage as they relate to the study.

L195-199: This belongs in the Discussion.
We prefer to leave this section in its existing place.

L201: I am not sure I understand what this first sentence is trying to justify. Also, sounds like it belongs in Methods.
This justification refers to K-transect's observationally-rich environment, and the use of AWS networks to resolve the spatial fingerprint of subgrid-scale melt.
L203: Does the term "composite" essentially refer to the average of all instances? Please, clarify.
We have clarified the term in the Methods section.

L206: Second half of the sentence refers to T above 0 C events? Please, clarify.
This refers to differences in T+ and T-events as indicated in the previous sentence.

L208-209: South-southeasterly winds should have a 157.5 degrees direction, given the orientation of the PROMICE AWSs, so what is now written is incorrect. These winds in autumn are katabatic with a downslope direction (90 degrees) deflected 45 degrees to the right by
Coriolis force (i.e. southeasterly direction; cf. Van den Broeke et al., 2009), plus potential southerly synoptic influence. Nevertheless, this quantification seems to be more prominent above the long-term ELA and KAN_B. Good thing to also mention.
We have added some text clarifying the wind direction and Coriolis influence citing the van den Broeke paper provided. We have also mentioned statistically significant changes to wind direction above the long-term ELA at S9 and KAN_U within the paragraph.

L211-218:
This belongs in the Discussion. Nevertheless, I am not sure I agree with this interpretation, and I explain: Strengthening wind speeds during positive air-temperature events at KAN_B do not reflect strengthening of katabatic flow (i.e. cold, dense wind), but rather that additional wind components inducing turbulent mixing might be present. A strengthened katabatic flow would remain cold, and would be more laminar, and hence less turbulent, than usual. Katabatic wind could enhance turbulent mixing during melt season at the lowermost rough-surface parts of the tran-sect, but we are discussing unseasonal melt events, implying that sunlight is reduced, hence surface melt is not sustained in the way it does during the melt season, while surface roughness below ELA might be substantially decreased due to accumulating snow cover, and above ELA slightly increased for the same reason, i.e. sastrugi formation. The observations from most AWSs within the same averaging periods suggest southerly deviation during positive air-temperature events from the cold-event direction, and the way I perceive it, southerly synoptic influence. I note that Figure 2 and Table 2 suggest that melt might occur primarily at the two lowermost on-ice AWSs, while wind speed as well as wind-speed differences between positive and negative cases within the same averaging periods are more pronounced at higher elevations due to comparatively reduced surface roughness (cf. Van den Broeke et al., 2009).
We have revised our wording here and softened the language. We appreciate the reviewer's feedback and elect to include the present list of factors. As we suggest, these may contribute to the wind speed intensification and therefore link with turbulent flux analysis that is subsequently presented in the paragraph.
L214-215: Positive energy flux that contributes to ice melt is most definitely downward, i.e. directed from the atmosphere toward the ice-sheet surface, and not necessarily the result of increased wind, rather the result of increased turbulence; "Sensible heat fluxes" does not sound nice. Consider removing the plural.
We have made the suggested changes, acknowledging the heat flux sign was incorrectly described.
L219: The first sentence is not a result. Please, delete or relocate to Introduction.
We have added "The K-transect" at the sentence outset to provide context to the result.

L222: "at KAN_M and S9 in the upper ablation area."
We have made the suggested change.
We have made the suggested change.
We have elected to remove part of the sentence that is largely duplicative with text earlier in the section.
L228-229: This sentence, as it is, belongs in the Discussion. You can reformulate and keep in the Results as: "We note that there is a height difference between the RCM 10-m above-surface output and the measuring heights of the AWSs, as mentioned in the Data section.", or something like this.
We have altered the sentence and elect to keep it in its present place.
L233-234: This first sentence is Discussion material. Please, relocate or reformulate, since it sounds like a rather abstract introductory sentence referring to the previously outlined results.
We have integrated this sentence into the Discussion content. L255: The link between moist air masses and how they may facilitate ablation-area melt is not clear. Please, elaborate preferably in the Introduction.
We have added "warm" as is evident when they pass over the GrIS by the increasing height values, tropospheric moisture content, and K-transect surface air temperatures.

L261: Include a short comment on strong negative NAO phases, as seen in monthly observations.
We have added the suggested change.

L266: "turbulent atmospheric heating"
We have made the suggested change.
L267-269: Please, refrain from referring to supplementary material as main part of the text, i.e. delete "As shown in Fig.S6". Instead, include a citation at the end of the sentence. It should be mentioned that it is a moderate to weak link that seems to be year-dependent. The way I perceive it, this differentiation amongst years suggests synoptic-circulation control that may or may not be present each year, and that is a good comment to include before anything else. (For example, Fig. S6 suggests weak correlation in autumn 2013 between local sea-surface temperature and AWSs, when limited melt occurred even at KAN_U (September 2013; Charalampidis et al., 2015).
We adjust the reference to the end of the sentence and restructured the wording. Last in the paragraph we posit that the synoptic control may be stronger in years of higher correlation, and a deeper examination is currently being conducted in related work.
We have changed "robust" to "significant" to be consistent with other correlative findings.
L271-279: This part seems somewhat arbitrary, since no evidence of a different wind regime between 2011-2012 and 2013-2015 was presented. Also, I am not sure I agree with the categorization. Please, revise/clarify.
We are suggesting some physical linkages, supported by literature, that could explain strong versus weak interannual correlations along an "extended transect" at 67ºN between near-shore SSTs and coastal and on-ice air temperatures. In doing, we address some reasons why the coastal response to sea-ice heat flux feedbacks might be strong, yet very localized. We expand on these concepts further in the following paragraphs of the discussion.

L287: Mention also the distance from Kangerlussuaq.
We have added this information.
We have revised this area by highlighting that increased ice sheet radiational cooling over the ice sheet interior versus coast, which helps to propagate katabatic winds that block inland penetration of marine air.
L295: "appears to be of minor importance. . ." We have made a minor wording change.

L304-306: Please, clarify.
We have clarified the language in the sentence.

L306: "Denmark Strait and Irminger Sea at the East coast of Greenland. . ."
We have made a similar change to clarify Denmark Strait and Irminger Sea location with respect to Greenland. We have removed some of the redundant citations, but have kept those in reference to future research avenues that dovetail from our main results. To that point, we have made an effort to further emphasize the main manuscript results in the Conclusions section.

L317-319: Consider simplifying; rephrase "around the limits"
We have changed wording to clarify the limits previously mentioned.

L377: Delete "2018"
We have updated the reference. We have described KAN_B's position in the caption. We have clarified that the counts involve daily mean air temperatures. We have also thickened the line between the columns as suggested.
L540: Replace "30 to 60-day window" with "time window defined by the 60th and 30th day before day of ice advance (DOA)" We have clarified the caption. We have included legends for panels b and c, and added North, East, South, and West accordingly to panel c. We have also clarified the "daily mean" values in panel d.

L551: Define "composites" better.
As the composite methodology has been clarified in the Methodology section, we wish to keep the captions concise.

L553: "selected"
We wish to keep the wording the same.

L569: Spell out IVT and include abbreviation since it is used as such in the figure.
We have made the suggested change. We have included the legends in both panels and have spelled out abbreviations not previously listed in captions.

L754: Define "composites" better.
As the composite methodology has been clarified in the Methodology section, we wish to keep the captions concise.
We have clarified the caption in a similar manner to the Figure 2 caption (L552-553). We have made the suggested change. Figure S2, S3, S4, S5: Spell out DOA initially, and then use abbreviation.
We have spelled out date of sea ice advance (DOA) in all captions after the terms initially appear. Figure S4, S5: Define composites better.
As the composite methodology has been clarified in the Methodology section, we wish to keep the captions concise.

Figure S6: Introduce SST properly in caption.
We have spelled out SST as suggested. Overall, the hypothesis is novel, and I applaud the authors for bringing in multiple datasets to support it. As one of the other reviewers also noted, I am, however, very concerned that the short time series (2011)(2012)(2013)(2014)(2015) limits the conclusions that can be made with any statistical certainty. From the evidence given, I would say the study is suggestive, but not conclusive in its arguments. In my opinion, the authors have a choice: either use a longer time-series of measurements or model data (for instance, keep the 2011-15 analysis of AWS data but make the regional climate data analysis span from 1979-2015, which is possible given that the two datasets aren't even used in direct comparison to each other), *or* soften the conclusions to make it clear they are suggestive but not entirely conclusive given the short time series. With only 5 years at your disposal, random noise can easily be interpreted as interesting new patterns.
As always, I acknowledge that the authors have spent more time considering this paper than reviewers spend reading it. If authors believe my judgements are unfounded or based on a misunderstanding, or if I just missed something in a comment, they may of course make that case upon resubmission to the reviewers or editor. Overall I do think the study, and the hypothesis put forward, is quite suggestive and worth publishing for that reason, even with the concerns cited, but the concerns should be addressed.
We thank the reviewer for constructive comments and have addressed their concerns in an attempt to improve the quality of the manuscript.

Specific comments:
L31-32: "For the unseasonal melt period preceding the DOA" . . . this make it sounds as if it's referring to a *specific* melt event that hasn't yet been mentioned. Generalizing the sentence more to something like "For periods of unseasonal melt preceding the DOA,. . ." may be clearer to the reader.
We have clarified the wording as suggested.
L34-35: "the above and below freezing surface air temperature events. . ." same comment as above.
We have made a similar suggested change.
L83-84: "temporally-anomalous GrIS late melt events" : in much of the text, terms such as "unseasonal" and "temporally-anomalous" are used, but never really defined. Does is refer to melt events after a certain date/season? Recent late-season melt events that are more frequent than previous periods? The descriptor is a bit vague here.
We have clarified the language, relating late-season melt (as in the manuscript title) to "unseasonal" so as show the reader the terms are used interchangeably.
L104: (quick format check) Several headings are not "bolded", while the rest are. Quick fixes.
We have made bolded headings for consistency.
L119: "erroneously low values are filtered out prior to analyses" : Which variables were filtered in this way: temperature, wind speed, others? If you can, be more specific.
We have clarified the which variables were filtered.
L152: "A composite approach is applied to characterize atmospheric conditions. . ." : From what I saw, I couldn't find an explanation in the text of how these composites were put together, or what exactly the reader is looking at when seeing the composite numbers in the tables and figures. I don't expect the authors to provide a textbook lesson on composite statistics, but a brief 1-3 sentence explainer for the reader would help reproducibility and clarity of the text. Right now, if I took the same data and tried to reproduce the results I would have no guidance of how to perform the composite analysis other than the authors' word that they applied a "composite analysis." More specificity is needed here to make the results reproducible. If the method is identical to one used in other papers, a citation may suffice.
We have clarified the composite approach in the Methodology section so that results could be reproduced for additional transects about the GrIS, for instance.

L162-165:
The description of the timespans used in the composite approach is good, I would put this at the top of the paragraph before describing why composite approaches were used and what constraints were placed upon them.
We have re-arranged the composite descriptions to introduce the temporal restraints before discussing the temperature bins. Care has also been taken to clarify the methodology as mentioned in the previous comment and following the recommendations of the other reviewers.
L183: "Perspectives on... are shown" is a vague term, almost meaningless. If the sentence is meant to point out something specific that Figure S1 shows, describe it explicitly.
We have edited the wording to be more direct in what Fig. S1 is presenting.
L183-199: The first several paragraphs of the Results refer to figures that are almost exclusively in the supplement (Figs S1-S3, with exception of the Fig. 1 map), which makes it awkward for the reader to follow along. These figures appear to be central to the results, not just supplementary. Unless you are limited by the number of figures, consider putting some of the these in the central text rather than forcing the reader to flip back and forth to an entirely separate document just to follow along with your argument.
We prefer to include figures corresponding to the main manuscript results (i.e. local and regional drivers of K-transect melt events) as they are currently placed, and leave the introductory results on GrIS and Baffin Bay end of melt season overviews, for example, as Supplementary Material. -193: "Relative to the climatology. . . date in 2012-2015 (Table S1)." It is unclear why 2011 is separated from 2012-2015 here, or why this sentence exists at all. Interannual variability has always been present, and no evidence is provided that this juxta-position of years shows any more significant differences than any other random 5-year period in Greenland's climatological history (perhaps it is, but it isn't demonstrated). If not, just omit it. Table S1, the date of Baffin sea ice advance (DOA) in 2011 was the only year of the five years studied with anomalously early ice formation. We wish to briefly note this anomaly as we describe the late melt season characteristics setting the stage for the core results of the paper.

As shown in
L195-199: "Interannual differences. . . found in the east and north (Curry et al., 2014)." These sentences don't show new results, but discuss the context of other literature. Consider moving to Discussion.
We elect to keep these results in their original place to provide context on the ice melt season preceding the primary results of the paper.

L201-202: "
The spatial coherence of observations across the K-transect along with inhomogeneous GrIS Region 3 spatial melt patterns and satellite pixel contamination issues at the tundra-ice interface, lead us to assess the melt events at the station level." It's very unclear what is meant by "inhomogeneous GrIS Region 3 spatial melt pat-terns", or why that would motivate a K-transect station-level approach. Please clarify. Similarly with "the spatial coherence of observations across the K-transect".
In the vast GrIS region (~539,375 km 2 ) melt is not spatially, nor temporally, consistent. For instance, on the same early autumn day the southern region may experience a strong warming signal, but the northern-most bounded area may not. Due to this inconsistency and issues classify melt for pixels intersecting the land surface, we focus on a station-based approach to identifying on-ice melt events. We have further clarified the language in results section 4.2 to further reflect our justification as mentioned.
L208-209: ". . .and comparatively becomes slightly more southerly" One issue with the way these composite records are presented is that there are no uncertainties or spreads presented with them at all, other than which ones are/aren't statistically significant. Thus, comparisons between them, such as "slightly more southerly" are impossible to make without knowing whether or not the difference are simply within the noise of the two datasets, or are just part of statistical noise. This is the case with all the comparisons, actually, and it makes the conclusions difficult to defend.
We acknowledge that there are some shortcomings of the composite methodology as mentioned in your comment, and our comparisons (by time window and temperature threshold) are the main comparisons we explore through the paper.

L223: "Modeled wind speeds are more intense during T+ versus T-events" How much more intense? 2 %? 150 %? As noted in the comment above, is the difference greater than statistical noise?
We have noted that simulated winds during T+ versus T-events are ~20-50% stronger consistent with the differences in the AWS observations.

L227: "with low root mean squared errors (not shown)." Unless I'm misunderstanding this statement, RSMEs for two values should just be two numbers. They can be stated explicitly instead of just saying (not shown). If they weren't computed, this statement should be omitted.
We have summarized the RMSE for the model and observational comparisons mentioned.

L227: A slight positive bias in both models is evident. . ." I assume a slight positive bias in wind speed? Clarify so that the sentence can stand on its own without ambiguity.
We have clarified the wording here.
L235: The "North Atlantic region" is not explicitly defined here, it is just vaguely referenced even though the authors are clearly looking at particular portions of the map. This ambiguity makes the rest of the entire paragraph extremely difficult to decipher from a reader's perspective, given that different portions of "the North Atlantic" behave in different manners in Figure 5.
We have clarified our reference area by mentioning Greenland as the geographic focus within the Northwest Atlantic region.

L235-238: "
Whereas T-events "left panels" tend to be characterized by northerly winds over the 1000-700 hPa layer, . . ." It is extremely unclear what is being referred to here. Are these northerly winds specifically in Baffin Bay, or over the whole figure, or. . . ?? Also, this sentence is disjointed and somewhat convoluted, its meaning unclear to the reader.
The "panels" refer to Figure 5 and to clarify we direct the reader back to these in the text. L238: "found over ice sheet" -> "found over the ice sheet" We have made the suggested change. We have made all contour lines visible and labeled the 540 line so that it is easier for the reader to compare its position between T+ and T-events.

L246-255: I found this paragraph particularly good and compelling.
Thank you for your comment.

L265: "toward earlier (later) melt (freeze)" I understand this sentence construct, and occasionally it's useful, but it's also extremely awkward to read and should only be used when necessary to save space. In this case, "toward earlier melt and later freeze" says the same thing with only one more word, and is much clearer English for the reader to understand.
We have made the suggested wording change.

L265: "This hotspot of melt" It's unclear if you're referring to a particular hotspot of melt referred to earlier (if so, point it out), or making a more general statement here.
We have generalized the wording as suggested.

L267-269: "Sisimiut SSTs fluctuate. . . ablation zone at S9." It's unclear why this sentence is important. Additionally, it's unclear that a statistically significant difference can be inferred at all if it only happened in 3 of 5 years, but not the other two. Even with just 5 years of data it's hard to make firm conclusions about climate patterns. . . moreso if just picking three years selectively out of those five. This needs to be better justified, or omitted completely.
This sentence references correlations between local SSTs and air temperatures extending from the coastal to lower accumulation zone at KAN_U. S9 marks the location from which the correlations in most years drop off and are generally weaker toward the interior. We have clarified the wording in the paragraph to reflect this statistical finding.

L271-279:
The two points made at the end of this paragraph seem difficult to support conclusively with only 5 years at your disposal, comparing the first two years to the second three. The problem with such a short time series is that signals can be easily interpreted from random noise, making such conclusions problematic at best. This is emblematic of the greatest weakness of the whole paper, making strong conclusions about climate patterns from only five years of data. It is unclear that if you had a 30-year record (long enough to infer at least some of the variability of the patterns you describe), that the same inferences could be made in any significant way.
We acknowledge that five years of K-transect case studies provide a limited temporal perspective and therefore must be considered in light of the evidence presented that Baffin open water and onshore winds do not appear to be a primary factor driving late-season melt events. However, the study provides a framework to be expanded upon in future, related studies presuming the on-ice AWS programs continue for years to come.

L289: "near-surface air penetrates at least to" -> "near-surface air often penetrates at least to" (There isn't any indication that it always does.)
We have made the suggested change.
L332-334: It is good that the authors recognize that a longer time series would help with these analyses. It is still unclear that all of the conclusions in the paper can be made so confidently with the short time series available, and that some of the results are more just "suggestive" than "conclusive." As mentioned above, we acknowledge that five years of analysis precludes overconfidence in the results. However, the study provides a framework to further explore unseasonal GrIS melt characteristics amidst ongoing expansion of the Arctic melt season. Atmosphérique Régional (MAR) and Regional Atmospheric Climate Model (RACMO) provide high-resolution spatial context to AWS findings, and are analyzed along with ERA-Interim reanalysis fields to understand the mesoto-synoptic scale (thermo)dynamic drivers of the melt events. Results suggest that late-season melt events occurring in the ablation area of the K-transect are strongly affected by ridging atmospheric circulation patterns that transport warm, moist air from the sub-polar North Atlantic toward west Greenland. While thermal conduction and advection off south Baffin Bay open waters impact coastal air temperatures, local marine air incursions are obstructed by barrier flows and persistent katabatic winds along the western GrIS margin.

Introduction 5
Substantial decline in Arctic sea ice extent and mass loss from the Greenland Ice Sheet (GrIS) have been observed for the last four decades (e.g. Serreze and Stroeve, 2015;Bamber et al., 2018). Under sustained climate warming, sea and land ice are becoming increasingly sensitive to changes in the frequency and duration of anomalous weather patterns (e.g. Overland and Wang, 2016;Hanna et al. 2014Hanna et al. , 2018b. The overall mass balance of the GrIS has contributed roughly 0.5 mm year -1 to global mean sea level rise since the early 1990s (van den Broeke et al., 2016), 10 with about 60% of the mass loss attributed to a decline in surface mass balance (SMB) and 40% associated with increased ice discharge (van den Broeke et al., 2017). Observed warming of near-surface ocean waters west of Greenland since at least the early 1990s is linked to accelerated submarine melt and outlet glacier retreat (Holland et al., 2008;Straneo and Heimbach, 2013) concurrent with more frequent summertime air temperature extremes along the coast (Hanna et al., 2012. These findings raise the question of whether Baffin local ocean 15 conditions are of importance in governing the spatial extent and temporal variations in western GrIS melt. Conflicting evidence has been presented in literature over the past decade regarding the importance of the warming of the nearby ocean on GrIS surface melt. Regional climate model simulations have suggested that local sea surface temperature (SST) impacts on GrIS climate and SMB are negligible due to offshore flow arising from a prevailing katabatic wind regime Noël et al., 2014). Using a statistical approach, Rennermalm et 20 al. (2009) found contemporaneous Baffin open water (<15% sea ice concentration (SIC)) and western GrIS melt to be positively correlated in late summer . The authors noted that the strongest, statistically significant correlation (r=0.71) was observed in August near the K-transect (Fig. 1), and was attributed, in part, to wind-driven onshore transport of warm marine air. Hanna et al. (2009) Stroeve et al., 2017). We posit that if Baffin Bay open water were to influence GrIS late-season melt events, the heat accumulated in the marine layer from ocean-to-atmosphere turbulent fluxes, would be transferred directly to the ice sheet by onshore westerly winds. In addressing this hypothesis, our analyses are geographically focused on the western slope of the GrIS with emphasis on the K-transect 75 ( Fig. 1) as this area -with its two on-ice AWS networks (described in Section 2.1) -is rich with in situ records relative to the remainder of the ice sheet. We analyze meteorological data from these in situ sources and additionally from regional climate models and global meteorological re-analyses to address potential links between GrIS late season melt and the local-scale Baffin Bay marine layer for the period of overlapping data, 2011-2015. For completeness, we subsequently expand the scale of meteorological analyses to consider the influence of greater northwest Atlantic 80 synoptic patterns on the melt episodes. The paper is organized as follows: Section 2 outlines data sources, Section 3 describes methods employed, Section 4 covers the local and synoptic scale atmospheric interactions with GrIS melt, Section 5 discusses key results, and Section 6 offers concluding remarks and makes suggestions for future research.

Passive microwave records 85
Sea ice data are from the National Oceanic and Atmospheric Administration/National Snow and Ice Data Center (NSIDC) Climate Data Record of Passive Microwave SIC v3r1 product distributed by NSIDC (Meier et al., 2017). We use the "Goddard Merged" SICs that are produced using a combination of the NASA Team and Bootstrap algorithms applied to satellite passive microwave brightness temperatures (Peng et al., 2013). Daily observations are available over the 1979-2015 period at a 25 km by 25 km nominal grid spacing. The time series of daily SIC at each 90 grid cell were used to identify the sea ice date of advance (DOA), which is the date when SIC increases to 15% for the first time following the sea ice extent (SIE) minima after Steele et al. (2019). Regional mean DOAs for Baffin Bay were obtained from Bliss et al. (2019). The local DOA was determined from the time series of SICs between 1 October and 31 March at each grid cell, then the local mean DOA was computed from 13 grid cells within the domain 66.5 to 67.5°N and 53 to 55°W. SIE was computed by summing the area of grid cells (in km 2 ) where SIC≥15%. At 95 the same spatial resolution as the SIC product, the passive microwave daily GrIS melt time series of Mote (2007 is also used to classify the ice surface environment in a binary manner (i.e. melt/no melt). Following the Ohmura and Reeh (1991) topographic regions, we assess GrIS melt conditions on the west-central portion of the ice sheet bounding the K-transect.

AWS data 100
Meteorological conditions from two AWS networks, the Programme for Monitoring of the Greenland Ice Sheet (PROMICE; stations prefix "KAN") and Utrecht University Institute for Marine and Atmospheric Research (hereafter IMAU; stations prefix "S"), are used in this study (Fig. 1). In order to include the westernmost PROMICE station KAN_B that is situated approximately 1 km away from the ice-sheet margin, analyses begin in 2011 and conclude with the end of the DOA record in 2015. Data are obtained and analyzed from seven weather stations 125 distributed across the K-transect at ~67°N during this period. The transect spans an area from low-elevation tundra, approximately 1 km inland from the ice sheet glacier terminus (KAN_B), to the lower accumulation area (~1800 m) at >140 km from Russell Glacier terminus (KAN_U; see Fig. 1 & Table 1). AWS data used here are recorded at a height of approximately at 2-3 m above the surface, though observed values fluctuate due to snow accumulation and melt season ablation (Charalampidis et al., 2015). Daily mean air temperature (°C), wind speed (ms -1 ), and wind 130 direction (0-360°) are obtained from IMAU and PROMICE station networks. Most data series are complete for the study period, but missing or erroneously low values in the temperature and wind variables are filtered out prior to analyses. Additional details on the respective PROMICE and IMAU AWS programs can be found in van As et al.
(2011) and Smeets et al. (2018). We supplement K-transect data with daily mean DMI AWS surface air temperatures obtained from Sisimiut (WMO code 4234) and Kangerlussuaq (WMO code 4321). These data are used to further 135 evaluate spatial links between local Baffin Bay open water and air temperatures from the tundra regions below the Ktransect (Cappelen, 2018(Cappelen, , 2019 ; Fig. 1).

Atmospheric reanalysis and regional climate model fields
A number of ERA-Interim reanalysis (Dee et al., 2011) variables are selected for the analysis of atmospheric conditions across Greenland and the northwestern Atlantic Arctic sector. ERA-Interim surface and upper-air 140 temperatures, wind speeds, and moisture conditions exhibit relatively small biases compared to Arctic observations (Bromwich et al., 2016). Regional climate models MAR and RACMO2 are also forced with ERA-Interim fields at their lateral boundaries. Tropospheric winds and specific humidity from 1000-200 hPa are used to calculate a moisture flux variable referred to as integrated water vapor transport (IVT); IVT is then classified using a self-organizing map (SOM) approach (see formal description in the Methods section below). Surface-atmosphere interactions and regional 145 circulation are further evaluated with ERA-Interim latent and sensible heat flux data, 500 hPa geopotential heights (GPH), and 1000-700 hPa mean winds at the native 80 km resolution.
Wind speed and direction at 10-m and 850 hPa from MAR and RACMO2 are evaluated to understand lowlevel atmospheric flow over ocean-land-ice sheet areas surrounding the K-transect. Secondarily, we briefly discuss inter-model differences within the planetary boundary layer and biases against AWS observations. Both regional 150 climate models are specifically developed for simulating polar weather and climate, in particular over the Greenland ice sheet (e.g., Fettweis et al., 2011;Noël et al., 2018). MAR v3.9 fields at 15 km are used here (see Fettweis et al. 2017 for a detailed model description). Relative to MAR v3.8 used in Delhasse et al. (2018), the main changes to MAR v3.9 consist of enhanced computational efficiency, adjustments to some of the snow model parameters to better compare with in situ observations, and improved MAR dynamical stability by increasing the atmospheric filtering. 155 RACMO2.3p2 fields at a horizontal resolution of 5.5 km are also used (Noël et al., 2018). Model physics have not changed relative to the previous 11 km version described in Noël et al. (2018). The refined spatial resolution of the host model improves the depiction of topographically complex terrain at the GrIS margins, such as small peripheral glaciers and ice caps, and the representation of near-surface, local winds.

Methods
Above-freezing air temperatures at on-ice AWS locations represent an indicator of ice-sheet melt. A composite approach is applied to characterize atmospheric conditions underlying late-season GrIS melt events, defined here as occurring at the conclusion of boreal summer (i.e. late August) and during autumn preceding sea ice advance 180 on Baffin Bay (Table S1) Table 2. Temporal overlap between above-freezing temperatures at KAN_B versus S5 (77%) and KAN_L (46%) in columns 2-3 of Table 2 suggest spatial coherence in the physical mechanisms forcing melt at least 195 across part of the lower ablation area.
In a similar fashion to Carr et al. (2017), a Wilcoxon test is used to evaluate differences in atmospheric variables and indices between T+ versus T-events. This nonparametric test is intended for continuous data series that do not follow underlying assumptions of the normal distribution making it appropriate for comparative analyses between extreme and non-extreme conditions. The null hypothesis of no difference in atmospheric conditions between 200 cases is rejected at the 95% confidence level when p£0.05.
Following the methodology used in Mattingly et al. (2016Mattingly et al. ( , 2018, daily atmospheric moisture transport about Greenland is resolved from ERA-Interim data by calculating IVT as in Eq. (1): where g is gravitational acceleration, q is specific humidity, V is the vector wind (ms -1 ), and dp represents the 205 difference between atmospheric pressure levels. The percentile rank of IVT (IVT PR) within a 30-day moving window is calculated to account for the seasonal cycle of IVT, and IVT PR is then classified using the SOM technique to Deleted: unseasonal Deleted: First Deleted: regional 210 Deleted: KAN_B is selected to build composite sizes as this location experiences more T+ events annually than any other PROMICE or IMAU AWS on the K-transect, and due to station's location near the ice margin along the transect ( Table 1).

Deleted:
The second limit involves the length of the analysis period prior to DOA. Through lead/lag correlations of recent data series of ³30 years, Hanna et al. (2009) andBallinger et al. (2018a) noted Baffin ice coverage and sea surface temperatures in the 220 preceding two months to be positively and significantly correlated with west Greenland coastal air temperature variations. Composites here are constructed over a similar two-month (60-day) period with data additionally sub-divided into 60-31 day (i.e. D60) and 30-1 day (i.e. D30) bins preceding each Baffin DOA from 2011 to 2015. A 225 paucity of T+ events over the 30 days post-Baffin DOA limits analyses to the aforementioned periods before the development of extensive, seasonal sea ice coverage.

Deleted: tilized
Deleted: integrated water vapor transport ( 230 Deleted: ) produce a matrix of moisture transport patterns, or nodes, that typically occur over the Greenland region. The SOM is based on an unsupervised machine learning algorithm that classifies each daily IVT PR field into the closest matching SOM node. This study uses the same SOM dimensions, training parameters, and IVT composites as Mattingly et al. (2016). We further classify SOM nodes-based on visual inspection of IVT PR patterns over 235 Greenland-into the same "wet" (anomalously high IVT PR over Greenland), "neutral" (near climatological median values), and dry (anomalously low) SOM node groups as these authors, and test whether the frequency of IVT patterns falling into each node group differs across T+ and T-events.

Characteristics of Baffin Bay and west Greenland late-season melt 240
Dates marking the end of the Northwest Atlantic melt season area are shown in Fig. S1. Slow, upward change signifying later sea ice formation is shown in the DOA series, particularly from around 2000. Meanwhile, the start of the last ³3-day sequence of Region 3 (west-central Greenland; see inset in Fig. 1) 2% or 4% melt area also suggest progressively later melt (and a later onset of the freeze season). A significant break in the 2% series is highlighted by a drastic increase in variability from 1979-1999 (s=21.17) to 2000-2015 (s=44.25) that is also present 245 in the annual discharge records from the nearby Watson River and Tasersiaq ice sheet catchments (Ahlstrøm et al., 2017;van As et al., 2018). Differences between the beginning of Baffin Bay sea ice advance and the end of the ice sheet melt season have clearly narrowed in part due to regional melt season lengthening. Some GrIS melt events since 2000 have notably occurred after seasonal sea ice formation (i.e. 2002, 2004-2005, 2010).
Relative to the climatology defined as 1981-2010, sea ice advanced ~11d earlier in 2011 and ~6d later in 250 2012-2015 ( Table S1). Inspection of DOA for individual grid cells adjacent to the Sisimiut AWS (WMO code 4234 in Fig. 1), located ~150 km west of the K-transect ice sheet margin, reveals a northward-extending notch where ice forms ~30-60d later than the Baffin-wide DOA (Fig S2a-e; Fig S3). Interannual differences in ice cover advance often depend on factors including regional winds, ocean heat transport, and water-mass changes (Myers et al., 2009;Ribergaard, 2014). For instance, strong offshore winds and poleward circulation of warm water from the West 255 Greenland Slope Current (WGSC) often contribute to the local open water persistence, while southward Arctic Water transports support earlier ice formation patterns found in the east and north (Curry et al., 2014).

Local meteorology of melt versus non-melt cases
The spatial coherence of AWS observations across the K-transect along with inhomogenous GrIS Region 3 spatial melt patterns (e.g. southern GrIS pixels may pick up melt while those to the north do not experience the same 260 conditions) and satellite pixel contamination issues at the tundra-ice interface, lead us to assess the melt events at the station level. Composites of air temperature, wind speed and direction by KAN_B T+ and T-events are shown in Fig. 2. Across the transect, composite air temperature differences (T+ versus T-events) are warmer by roughly 7 to 8°C in the [-60,-31] window and 12 to 13°C in the [-30,-1] period (Fig. 2a). These differences tend to be least near the coast and increase to S6 in the mid-ablation area (~1000 m asl) where contemporaneous melt occurs ~14-16% of 265 the time ( Table 2).
Wind direction for T+ events are roughly southeasterly (<150°) in [-60,-31] and become more southerly (180°) in the [-30,-1] period (Fig. 2c) regardless of the wind speed (Fig. 2b,d). Above the long-term equilibrium line altitude, S9 [-60,-31] and KAN_U [-30,-1] winds are significantly more southerly during T+ events. These katabatic winds are deflected to the right (southeasterly) by the Coriolis force as they travel toward Baffin Bay (van den Broeke 290 et al., 2009) and may be aided additionally by synoptically-driven southerly winds. There is also evidence of wind speed intensification during [-60,-31] T+ events (~1-2 ms -1 ) and to a lesser magnitude similar relationships hold during the [-30, -1] period (Fig. 2b). Increased wind speeds in T+ versus T-events are likely affected by the seaward enhancement of the pressure gradient aided by nearshore open water persistence (Fig. S4), while increased synoptic cyclone activity and lower MSLP over Baffin Bay during the summer-autumn season transition may also enhance 295 offshore flows (McLeod and Mote, 2015). Wind speed increases may also initiate positive (downward) sensible heat fluxes associated with low elevation ice melt (Fig. S5). Just offshore, sensible and latent fluxes are generally negative in T-events and near zero to slightly positive in melt occurrences with large, positive differences in T+ vs T-events apparent in oceanic areas from the Labrador Sea extending up to the WGSC (Figs. S5 and S6).
The K-transect offshore flows appear to represent a dynamic barrier to Baffin Bay marine layer intrusions. 300 To further examine lower tropospheric flow across the ocean-land-ice interface, we similarly composite MAR and RACMO2 winds at 10-m and 850 hPa (Figs. 3 and 4). In general, the MAR and RACMO2 winds show directional consistency with overlaid PROMICE and IMAU winds with a slight southerly bias in both products for [-30,-1] at KAN_M and S9 in the upper ablation area. Simulated wind speeds are ~20-50% stronger during T+ versus T-events (not shown) as corroborated by most of the AWS observations (Fig. 2b). T+ events in RACMO2 and MAR generally 305 capture AWS observed wind speeds in the upper ablation area at KAN_M (r 2 ³0.77) and lower accumulation area at KAN_U (r 2 ³0.77) with low root mean squared errors (RMSE<1.50 in all cases). A slight, positive bias in modelderived wind speed is evident at the ice sheet edge near KAN_B (MAR r 2 = 0.33, RACMO2 r 2 = 0.58). We note that height and therefore surface roughness differences between the AWS measurements (2-3 m) and regional model 10m winds may explain a portion of the bias at KAN_B. Except during [-30,-1] T+ events, the 10-m winds extending 310 about the transect from the west coast tundra just offshore and into eastern Baffin Bay are notably calm and alongshore (i.e. northerly in T-and southerly in T+; Figs. 3 and 4).

The role of North Atlantic atmospheric patterns
Overplots of 500 hPa GPH, 1000-700 hPa mean wind, and IVT for Greenland and the surrounding northwest Atlantic region are shown in Fig. 5. Whereas T-events tend to be characterized by northerly winds over the 1000-315 700 hPa layer, the T+ events indicate southerly, on-ice transfer of subpolar air aided by the presence of an upper-level trough over Baffin Island and downwind ridging over Greenland (see left versus right panel plots in Fig. 5, respectively). Comparatively higher GPH values are found over the ice sheet with the 540 dam (i.e. 5400 m) contour, which tends to distinguish solid and liquid precipitation, extending across central Greenland in T+ events, but located south of the island in T-events. In both T+ cases, 1000-700 winds circulate poleward over north Labrador Sea areas 320 of upward, turbulent heat flux (Figs. S5 and S6), aiding the heat and moisture transfer (as shown by heightened IVT values in T+ relative to T-) to the western Greenland ice sheet during [-60,-31] and [-30,-1] (Fig. 5). Local IVT maxima in both events are concentrated over the southwest tip of the island, but remain ~100 kg m -1 s -1 or higher near 3 (4-5) during [-60,-31] ([-30,-1]) T+ versus T-events signaling a concentration of moist, onshore flow and 375 condensational processes over the ice sheet that drive the above-freezing air temperatures.
To further characterize and differentiate weather conditions by T± event, we composite SOM-classified daily IVT wet, dry, and neutral patterns identified by Mattingly et al. (2016). Analyses of the aggregated frequencies suggest that the wet patterns (with anomalously high IVT versus climatology) occur significantly more often in T+ versus T-events, and such nodes are more common by a factor of >4.5 in the [-30,-1] period. (Fig. 6a). While some 380 caution should be exercised as the absolute frequency of these patterns decreases from roughly early ([-60,-31]) to late ([-30,-1]) autumn, humid atmospheric conditions appear to enhance late-season melt. Increased incidence of wet patterns coincides with negative (positive) NAO (GBI) (both >|0.50| in T+ events; Fig. 6b), and a synoptic environment characterized by high surface (upper-level) pressure anomalies. This is confirmed by Fig. 5 whereby the higher 500 GPH values and 1000-700 mean winds during T+ events transport warm and moist air masses from 385 Labrador Sea and southerly maritime latitudes to much of the western slope of the ice sheet to facilitate ablation-area melt.

Discussion
West Greenland summer and autumn air temperature variability and trends during the last 3-4 decades have shown strong response to increased frequency and intensity of Greenland high-pressure blocking, negative NAO 390 patterns, and positive North Atlantic SST anomalies aided by background anthropogenic forcing (Hanna et al., 2016;McLeod and Mote, 2016;Ballinger et al., 2018a;Graeter et al., 2018). The current North Atlantic "warm period" since the mid-1990s is characterized by a positive Atlantic Multidecadal Oscillation phase and rising SSTs around southwestern Greenland (Myers et al., 2009;Ribergaard, 2014), including just offshore of Sisimiut (WMO code 4234 in Fig. 1) (orthogonal trend = +0.03°C year -1 , p<0.05, for 1995-2015 period using SST product described in Ballinger 395 et al., 2018b). Warming waters around the island are influencing Baffin Bay sea ice and west GrIS melt processes (Hanna et al., 2013;McLeod and Mote, 2015;Ballinger et al., 2018a) and seasonality toward earlier melt and later freeze . This melt area about K-transect suggests local low SIC and open water, inferred upward turbulent atmospheric heating, and onshore winds could influence nearby terrestrial melt events. Moreover, Sisimiut SSTs fluctuate with air temperatures (over the 60 days preceding DOA) in a statistically significant fashion for 2013-400 2015 at most K-transect stations with some distance decay noted upslope from the edge of the ablation area at S9 (Fig.   S6). Summer and autumn west Greenland near-coastal air temperatures are modulated by the thermal properties of bordering SSTs (Hanna et al., 2009;Ballinger et al., 2018a). Interannual differences in the strength of SST-air temperature relationships (i.e. 2013-2015 versus 2011-2012) suggest: 1) processes driving warming ocean waters and air temperatures over the GrIS are independent when disparate wind directions occur at or near the ocean-tundra-ice 405 sheet boundaries in years of weak-to-zero correlation (e.g. katabatic flows contrasting near-coastal barrier flows (van den Broeke and Gallée, 1996)), or alternatively 2) large-scale atmospheric circulation forcing (i.e. near-surface and upper-level meridional winds), during years of positive, statistically significant correlations, modulates the near-shore surface open water and ice sheet air temperatures . We recognize that synoptic patterns may not necessarily be mutually exclusive in these examples, but the manuscript objectives do not include comparison of high 410 and low pressure features around Greenland for specific melt and non-melt events.
A number of studies have suggested that Baffin Bay marine layer interaction with the ice sheet boundary 440 layer is obstructed by zonal and meridional flows such as the west coast plateau jet feature and katabatic winds (Hanna et al., 2009;Moore et al., 2013;Noël et al., 2014). Moore et al. (2013) noted a directionally consistent southerly 10m wind field extending over the western half of Greenland in summer and winter, while observational studies similarly indicate a high frequency of southerly-to-southeasterly winds over the K-transect (van den Broeke et al., 2009).
Southerly (easterly) 10-m winds are strongly linked to melt across two thirds (the southern third) of the ice sheet 445 (Cullather and Nowicki, 2018). For an expanded spatial perspective, we briefly examine air temperatures at the next PROMICE station installment approximately 700 km north at Upernavik (UPE; 72.89°N). We find UPE_L (220 m asl on the ice sheet) melt occurs the day of KAN_B T+ events on >50% of occasions in both [-60,-31] and [-30,-1] windows (not shown). This suggests that above-freezing near-surface air often penetrates at least to PROMICE station UPE_L with a relatively warm air mass engulfing much of the west coast. Our observational and regional model 450 analyses further show that homogenous low-level winds extend coastward at least to the tundra-ice sheet interface near KAN_B (see Figs. 2c, 3, and 4) and produce a "blocking effect" that inhibits the inland penetration of nearsurface air from Baffin Bay (Noël et al., 2014). Of note, the katabatic mechanism becomes stronger as Baffin Bay DOA approaches in late autumn, with more pronounced radiational cooling over the ice sheet further supporting winds that prevent incursions of local marine air . 455 If late season K-transect melt is minimally influenced by local Baffin Bay open water, then what physical mechanisms drive the melt events? Composites of the AWS K-transect observations and complementary regional model output indicate that recent late-season melt events tend to be driven by southerly synoptic patterns as opposed to local marine forcing. We provide evidence of this physical forcing by [-60,-31] and [-30,-1] T+ composites that show southerly flows of more warm, moist maritime air of lower latitude origins relative to T-cases. As shown in 460 Fig. 5, during the former event period, air is transferred off northern Labrador Sea to the west coast, while a path of more southerly flow directs moist North Atlantic air masses onto Baffin Bay and southwestern Greenland in the period immediately preceding DOA. These "wet" synoptic patterns occur frequently under anomalous (>|1s|) positive GBI and negative NAO values (Fig. 6a, b). We surmise from Mattingly et al. (2018) that such patterns are particularly moisture-rich (³85 th percentile IVT climatological values) and often accompanied by atmospheric rivers impacting 465 Greenland, and their occurrence causes ablation area melt in non-summer , low insolation months through cloud radiative effects (i.e. increased downward longwave radiation transfer into the ice surface), condensational latent heat release, and liquid precipitation Binder et al., 2017;Oltmanns et al., 2019). The southerly winds that propagate moisture northward off the northwestern Atlantic Ocean are a product of amplified upper-level geopotential height patterns and meridional winds in T+ versus T-events extending from Denmark Strait and Irminger 470 Sea on the east coast of Greenland onto the ice sheet (Fig. 5). Mid-tropospheric ridging, which is more pronounced in [-30,-1] than [-60,-31] events, supports southerly winds that funnel heat and moisture from likely deeper in the Atlantic basin to Baffin Bay and southwest Greenland to stimulate sea ice and GrIS ablation area melt conditions (Ahlstrøm et al., 2017;Ballinger et al., 2018a,b;Hanna et al., 2018). Cullather and Nowicki (2018) similarly find collocated, positive surface and 500 hPa GPH anomalies over Denmark Strait and Irminger Sea tend to be associated 475 with melt events in the basin encompassing the K-transect. Our analyses support North Atlantic-air-ice sheet coupling, rather than localized Baffin Bay ocean-atmosphere processes, as a strong driver of transition season melt before sea ice advances south of the K-transect. Synoptic patterns associated with negative summer NAO and positive GBI incidence strongly influence these melt events (Fig. 6), prompting decreases in SMB, and increase in the K-transect 505 equilibrium line altitude over the last 10-15 years (Hanna et al., 2013;Smeets et al., 2018).

Conclusions
Temporal co-variability between GrIS and Arctic sea ice mass loss suggests a possible feedback whereby adjacent open water conditions, ocean-to-atmosphere heat flux, and on-ice winds affect inland melt during the end of the melt season. In our 2011-2015 analyses bridging the end-of-summer/early autumn melt to the date of first-year 510 Baffin Bay sea ice advance, we find no evidence to support the hypothesis that local open water, resultant turbulent heating, and onshore winds have a pronounced impact on inland ice melt events. These thermodynamic processes, in particular, directly influence coastal air temperatures and have a fingerprint on marine outlet glacier behaviors (Carr et al., 2017), but are shown here to be inhibited by topographically-influenced flows and synoptic patterns whose interactions are not mutually exclusive. Furthermore, Baffin Bay warming coupled with a longer autumn open water 515 period has been hypothesized to stimulate and invigorate upper-level, high-pressure blocking that promotes southerly air advection over the west Greenland coast . This is consistent with the main conclusions of Noël et al. (2014), which suggest that while warming waters around Greenland minimally affect SMB beyond enhancing tidewater glacier retreat rates SST forcing may indirectly influence GrIS SMB changes through impacts on atmospheric circulation. However, in terms of direct forcing by the local marine layer, beyond a near-coastal 520 influence, our AWS, regional, and synoptic wind analyses suggest that Baffin Bay does not represent a substantial advective heat and moisture source to the ice sheet during our 5-year analyses.
Future late season analyses, perhaps reconstructing K-transect meteorological conditions back to the origins of the modern sea ice record, might be insightful in comparing local ocean-ice sheet interactions spanning the 1990s shift from colder to warmer Baffin Bay summer SSTs . Assessing the temperature and pressure 525 gradients and their vertical profiles derived from retrospective analyses would also be useful to categorize the structure, magnitude, and direction of regional winds, including the katabatic regime, in attempting to provide a longer-term perspective of analyses presented in this paper. Noël et al. (2014) hypothesized that future sea-surface warming may exacerbate the division between local ocean and ice sheet by intensifying the temperature and pressure gradient and hence resulting katabatic winds. Baffin Bay climate and cryospheric changes in the last two decades 530 suggest such an increased "blocking" mechanism may already be underway. Moreover, stronger katabatic winds might be enhanced further by an increasing intensity of autumn mid-troposphere high-pressure over Greenland (Hanna et al., 2018a, see their Fig. 1e). Synergistic future research should continue to monitor the spatial extent, drivers, and physical effects of late-season melt through observational products and regional modeling tools, including quantification of late season K-transect mass loss and runoff through the Watson River, contributions to 535 subsurface/firn processes, and preconditioning effects on the following year's melt season.
Author Contributions. TJB and TLM conceived the study. TJB analyzed the observational data, with assistance from MP and SG, and led the writing of the manuscript. TLM developed and processed the satellite-derived GrIS melt data. KSM conducted the IVT classification and assisted with the creation of several figures. ACB developed the Deleted: summer (Rennermalm et al., 2009;Liu et al., 2016)  provided AWS or oceanographic data and support. BN and XF developed and processed regional model wind fields for RACMO2.3p2 and MAR v3.9, respectively. All authors provided valuable insights, feedback, and editing on 555 manuscript drafts.    (Kangerlussuaq (WMO code 4231) and Sisimiut (WMO code 4234)). The inset displays the northwest Atlantic Arctic region with superimposed GrIS topographically-defined boundaries, adopted from Ohmura and Reeh (1991).