Dear Reviewer,

thank you very much for your time and suggestions for the improvement of our manuscript.

Please see your original comments below (in bold) and our responses (in italics):

This study presents intercomparison of four atmosphere reanalysis models to show

the influence of several key parameters including sea ice concentration, surface

temperatures and wind speed on turbulent (sensible and latent) heat flux variability.

It is a well-written and performed study with a clear structure and aim. The quality of

the analysis and illustrations are of high quality. Yet, some minor revisions might be

needed to improve the study. In its current state it compares four reanalyzes but

does not provide much of analysis which of them might be more accurate for

various application. Moreover, the background information might be improved with

regards to physical drivers of turbulent fluxes variability, their parameterization and

validation of the key parameters including surface temperature. The same goes with

sea ice concentration, snow and ice thickness. This follows by a very compact

conclusion paragraph, not providing enough understanding of the advantages of

such intercomparison. I believe that explaining the exact procedure of flux

estimations in various reanalyzes would strengthen future discussion about results

of correlation analysis and expand the potential audience of the study.

General comments:

1) Despite understandable references to reanalyzes names, the less usage of

abbreviations may improve readability of the study. Generally, in comparison to a

few other reanalysis-related papers, this study is partially challenging to read due to

abbreviations and many values without a context.

We have used fewer abbreviations in the revised manuscript (also according to

suggestions in the comment number 14).

2) The discussion of (simplified) representation of snow and ice in the reanalysis

may be improved. For the average Arctic ice and snow thickness of 2.0 m and 0.2 m,

their equivalent total thickness could be 2–3 times higher than assumed 1.5 m.

These parameters are quite important for surface heat balance (for high SIC), and

there are a few studies comparing surface temperatures from in-situ observations

with reanalysis (showing a substantial difference in IST). It would be useful to

mention, how is the difference between different models in comparison to ERA5

bias in comparison to measurements, as the study also covers analysis of areas

with relatively high SIC values. This is especially vital as there are known issues of

strong surface temperature biases of reanalyzes in comparison to observations

(Zampieri, 10.1175/MWR-D-22-0130.1).

We have added a new subsection to the revised manuscript: 3.4 Effects of thin ice on

leads and snow pack on top of sea ice. In this subsection, we have carried out calculations

on the above-mentioned effects on LHF and SHF using mean data from the SHEBA

campaign to study them, and comment on the results.

We have also added a paragraph to the Discussion section of the revised manuscript on

simplified representation of sea ice and its impact on turbulent surface fluxes (subsection

4.2).

3) The effect of cloudiness was not discussed (often reanalysis works better with

clear sky conditions, for example, following Herrmannsdörfer,

10.1525/elementa.2022.00085).

We have added a paragraph to the Discussion section of the revised manuscript on

uncertainties in paramaterization of surface turbulent fluxes including discussion on the

warm surface-temperature bias in clear-sky conditions in cold seasons in the Arctic

(subsection 4.3).

(There is possibly a typo in the reviewers comment – reanalyses (numerical weather

prediction models) usually work worse (simulate too warm surface temperatures) in winter

clear-sky conditions in the Arctic.)

4) Since the largest effect on flux variability comes from the SIC, it would be useful

to discuss a bit more on flux measurements above leads, lead definition, lead

fraction measurements, etc.

We have added a text on lead definition and flux observations over leads to the

Introduction of the revised manuscript. We have also added more discussion on

observations of the lead fraction (SIC) and freezing of leads to the Discussion (subsection

4.2).

Specific comments:

5) Line 8: It would be great to add the time interval for given sensitivity values.

Added ‘using daily means of data’ to the Abstract.

6) Line 11: Is it possible to distinguish decrease from the warming and from the

reanalysis improvement? Can you quantify that difference between 1980-2000 and

2001-2021?

We don’t have evidence that more observations assimilated in reanalyses would improve

the datasets in a way that the sensitivity of turbulent surface fluxes to SIC would decrease

between 1980-2000 and 2001-2021. The assimilation scheme and model are the same for

the entire period covered by each reanalysis.

7) Line 14: Can you specify what is the effect of SIC on wind speed? Is it a physical

effect?

We have added ‘via surface roughness and atmospheric-boundary-layer stratification’ to

the Abstract.

8) Line 25: Is it always the case, even for relatively thin ice without snow, assumed

in some of the reanalysis?

According to observations that is the case in winter, which we have clarified in the revised

manuscript.

9) Line 46: since there is no direct mention of 20 % difference in the reference, it

would be helpful to explain how it was calculated. In addition, it would be great to

mention the scale of those observations: when/at which conditions SIC can be as

different as 20%? For the current study, SIC concentration is a key parameter, and it

would be helpful to have a more detailed overview of SIC data, algorithms, scales

and uncertainties. For example, one would expect that 20% difference in SIC would

give around 20% difference in turbulent flux differences between various reanalysis

while the actual difference is way larger.

The differences are shown in Figure 7 (right panel) of Valkonen et al. (2008), which we

specify in the revised manuscript.

It was calculated as the difference in sea-ice concentration based on passive microwave

data processed applying Bootstrap and NASA Team processing algorithms (specified in

the Figure description).

The assumption of 20 % difference in SIC resulting in 20 % difference in turbulent surface

flux between reanalyses is not entirely correct – by definition, the same 20 % difference in

SIC (e.g. 0.8 vs. 1) causes already different magnitude of upward SHF and LHF. Also, the

effect of SIC on turbulent surface fluxes is larger between e.g. SIC 1 changing to 0.8 than

SIC 0.2 changing to 0, the former case having larger impact on a change in upward

turbulent flux than the latter. The results of our study (in Figures 4 and 6) show the

modelled change in turbulent surface flux if the SIC change all the way from 0 to 1 within 1

day.

10) Line 72: it would be helpful to add explanation which data in this reanalysis is

based on which measurements or models in addition to Table 1. For example, more

details about snow and ice thickness. Or stating that surface temperature is

calculated from the surface energy budget.

We have added the information about the observations (satellite input for sea-ice

concentration) in each reanalysis to Table 2. Snow and ice thickness representation of the

sea ice is presented in Table 2 in the original version of the manuscript.

Surface temperature over the water and both snow covered and bare sea ice is calculated

from the surface energy budget in each reanalysis, which we have specified in the revised

manuscript.

11) Line 83: It is vital for future analysis to give a better overview of algorithms

behind different models. What is calculated, what is measured among parameters

important for turbulent fluxes.

The turbulent fluxes are prognostic variables in each reanalysis (calculated according to

Eq.1 and Eq.2).

We have added that ‘turbulent exchange coefficients depend on the roughness lengths for

momentum, heat and moisture, and on the stratification of the atmospheric surface layer’.

12) Line 131: It would be helpful to comment on the background of these SIC

algorithms, not only covering their labels.

As addressed in the specific comment number 10, we have added the information about

the observations (satellite input for sea-ice concentration) in each reanalysis to Table 2.

The specific algorithms for obtaining information on SIC (mostly from satellite data) are

rather complex and changing/ evolving in time (during the 42-year study period of our

work) with e.g. reanalyses using different external datasets as described in the end of

section 2 Material and Methods.

We do not think that going very deep into this background information would necessarily

help us interpret the differences in results between various reanalyses (and would expand

the extent of the study considerably).

13) Table 4: The SIC in BS and GS is close to zero, yet the average LHF are typical

for ice thickness of around 0.1 m, LHF would be 2-4 times larger for open water.

Despite this not being your data, can you comment on that effect? It would be

useful to show LHF and SHF as a function of SIC for some average conditions to

quantify the potential range of their values.

Please note that the values presented in Table 4 are both time and area-averaged. The

large average upward fluxes over the Barents and Greenland Seas, in particular in cold

seasons, are due to the high sea-surface temperatures. In the case of 0.1 m ice layer, the

fluxes would be reduced in magnitude.

SHF as a function of SIC in NCEP/CFSR data is shown for example in Figure 7 in the

original manuscript (days in November–December–January 1980–2000): in Laptev Sea

(point A), Beaufort Sea (point B), and Central Arctic (point C).

14) Line 177: I fully understand the reason to use some abbreviations, but some of

them, lees commonly used (like MB or months) may be removed to increase

readability.

We have replaced the abbreviations of seasons and Mean Biases of Daily Field Means

with words (the latter as Mean Biases).

15) Line 183: Similarly, as BS is almost ice free, one would expect SHF for open

water or 0.1-0.2 m thick ice of negative 100-500 W m^-2 during winter using simple

parameterizations. What could be the reason for much lower values?

In February–April, SHF over the Barents Sea is indeed -100 to -500 W m^-2 during cold-air

outbreaks originating from the Arctic. However, there are also a lot of cases of warm-air

advection from the south. Due to these cases, the seasonal mean loss of sensible heat is

no more than 34 W m^-2 (ERA5) or 44 W m^-2 (NCEP/CFSR).

Please note that in the revised manuscript, we have decided to change the ‘reference’

data set for subsection 3.1. While before, ERA5 was chosen randomly among reanalyses

(as indicated in the original manuscript), we do think that it is more appropriate to put

NCEP/CFSR ‘in the center of the attention’ of this subsection due to the modelled sea-ice

thickness and the snow on top of sea ice in this data set, which makes the data the most

realistic in terms of physical processes (compared to the rest of reanalyses prescribing

constant sea-ice thickness and no snow on top of the sea ice).

16) Figure 4: years and values/difference titles may be also added to the figure to

avoid reading the full caption. It may also be possible to present both data for 1980-

00 and 2001-21, and difference with a separate color bar.

We have added the side titles and data from 2001–2021 to Figures 4 and 6.

17) Line 213: Previously it was reported that LHF for CA is negative 2-11 W/m2, while

for GS it is negative 10-36 W/m2. Assuming these two regions are almost on

different side of SIC range (0.9 and 0.1), the corresponding difference of 10-25

W/m2should roughly represent a slope of monthly average LHF per unit SIC. Here

the slope is one order higher, is it because of different time averaging or why is

that?

The time averaging is the same throughout our study – daily means of data.

However, there is area averaging in the presented ERA5 (NCEP/CFSR) values in Table 3

and Table 4; these are calculated as mean of values from all days from respective season

(~ 2000 days) in all the grid cells of the respective arctic basin (tens to hundreds of

thousands of grid cells in case of GS and CA in ERA5 or NCEP/CFSR). Therefore, the

slope using just these two heavily averaged values is not directly comparable to the slope

coming out of the linear bilateral ODR model using all data from all days of each season in

each grid cell (outputs of which are shown in Figure 4 and Figure 6).

18) Line 241: Despite this being not your assumption, it may be mentioned later that

leads in winter refreeze extremely fast (Petrich, 10.1029/2006JC003466) and such

surface temperature assumption could lead to flux overestimation.

We have addressed the impact of refreezing of leads in the new subsection ‘3.4 Effects of

thin ice on leads and snow pack on top of sea ice’.

19) Figure 7: Maybe the cells could be chosen to be larger and cover wider range of

SIC and SHF.

The ODR model works with SIC/SHF data from original-sized grid cells of each reanalysis

(apart NCEP/CFSR, details explained in the original manuscript) in each season and time

period (~ 2000 days) and returns slope of the regression line (and other information).

Including more grid cells to make the area larger would require averaging of the variable

values and therefore we would be showing different SIC/SHF data than those that the

model works with to compute the slope of the regression line.

As it is mentioned in the text of the original manuscript, the main purpose of Figure 7 is to

show failure cases of the regression algorithm.

20) Line 299: Great you mention the effect from snow and ice thickness. Yet, it

would be very useful to present what would be an expected bias in LHF and SHF

purely from snow/ice thickness inclusion. Zampieri et al. (2023) might be a useful

reference 10.1175/MWR-D-22-0130.1 , especially in the context of MOSAiC

expedition, mentioned in line 415. Additionally, there are products including ice and

snow thickness (for example, KARRA), which may be used to reduce some of

uncertainties or quantify their importance.

We have carried out calculations on the effect of snow pack on LHF and SHF. The results

are presented in a new subsection ‘3.4 Effects of thin ice on leads and snow pack on top

of sea ice’. In this subsection, we used mean data from the SHEBA campaign to study the

above-mentioned effects on LHF and SHF and comment on the results.

We have not added CARRA results in the present study, as CARRA only covers a small

part of the Arctic. The treatment of snow pack is at least equally good in NCEP/CFSR

(included in the study) as in CARRA.

21) Line 329: Please add that NWP stands for Numerical Weather Prediction (or

something else).

Explanation of the acronym added.

22) Line 377: It would be useful to specify for each condition this is the case as SIC

directly changes surface temperature from seawater freezing point to close to air

ambient. What would be the bias if the lead is refrozen (as in winter they typically do

in just several tens of minutes), which gives surface temperature lower than

seawater.

We have carried out calculations on the effect of thin ice on leads on LHF and SHF. The

results are presented in a new subsection 3.4 Effects of thin ice on leads and snow pack

on top of sea ice’.

23) Line 408: The sentence should end with a dot.

Period added.

24) Line 416: I would suggest having a bit longer and clearer conclusion,

underlining what you achieved, what reanalysis capture accurately and what can be

still improved. And why this type of intercomparison work is important.

We have expanded the Conclusion according to the suggestions in the revised manuscript

(Section 5).

25) Line 484: The correct link to the study is https://doi.org/10.1002/qj.3803

Link was changed.

Dear Reviewer,

thank you very much for your time and suggestions for the improvement of our

manuscript. Please see your original comments below (in bold) and our responses (in

italics):

The authors’ study on the impact and sensitivity of Arctic sea ice concentration

on the turbulent surface fluxes of sensible and latent heat in four atmospheric

reanalyses is untouched by others, thus expanding our understanding in the

context of atmospheric reanalyses in Arctic. Therefore, I believe the novelty is

sufficient and of significant importance to this field. The analytical methods used

in this manuscript are clear and convincing, so I believe the conclusions drawn

by the authors are reliable. The structure of the manuscript is also quite

reasonable, and the logic is clear. I believe that this manuscript makes a

significant contribution to the research in the atmospheric reanalyses

intercomparison in Arctic regarding the sea ice concentration and the sensitivity

on turbulent heat fluxes. Therefore, I recommend that the journal accept it for

publication after the authors make minor revisions.

Comments:

26) Lines 51-52: “uniform spatial and temporal resolution” is unclear. Suggest

rewording this.

We have revised this as ‘...spatial and temporal resolutions that are uniform around the

globe…’

27) Lines 113-117: I'm not sure why here switched back to OLSR here. Since ODR

seems more appropriate, I believe it should be continued to make the subsequent

text clearer and the conclusions more coherent. If the only reason for using OLSR

is that it requires fewer computing resources, I think that reason is not

compelling enough.

We do believe that our choice of using OLSR for the multilateral regression analysis is

justified. To be clearer for the reader about our choice, we have revised the respective

paragraph in section 2 Material and Methods as follows:

‘For the bilateral-relationship analysis, we utilised the orthogonal-distance regression

(ODR; Boggs, 1988). Because all variables in reanalyses include uncertainties, we

theoretically considered the ordinary-least-square regression (OLSR), which assumes

no errors in the independent variable, not optimal for this case. Additionally, we carried

out tests on bilateral ODR and OLSR performance using data from several grid cells

from each reanalysis and while we found ‘nearly identical’ (at least five decimal

numbers identical) coefficients of determination (correlation coefficient squared, R2) for

both regression methods, importantly, the slopes of the regression lines varied

considerably. This is attributable to the above-mentioned OLSR's assumption of no

errors in the independent variable (x, in our case SIC) and therefore minimising the

distance only for x data to the regression line, whereas ODR minimises the orthogonal

distance between both x and y data (in our case y is LHF or SHF) and the regression

line. Utilising the same above-described tests comparing ODR and OLSR performance

for multilateral regression analysis, however, we found ‘nearly identical’ values for all

slopes of the regression lines between LHF (SHF) and SIC, Qdiff (Tdiff), and WS10m for

both ODR and OLSR. Values of R2 for all and individual components of the multilateral

regression were ‘nearly identical’ using both ODR and OLSR as well. Based on the

findings that both methods yielded ‘nearly identical’ results for the multilateral

regression analysis (using our reanalyses data), we decided to use OLSR for the

multilateral regression analysis in our work, as it requires much fewer computing

resources to perform.’

28) Lines 118-119: Further explanation on this point is needed.

We have revised the sentence as follows:

‘We used linear model for both ODR and OLSR as we evaluated it as the most

applicable for our purposes, being aware of some non-linearity in the SIC effect on Q2m

(T2m) and LHF (SHF), as shown for near-surface air temperature in e.g. Lupkes at al.

(2008), their Figure 4.’

29) Lines 149-150: You need to specify which figure or table is being described

here.

We have revised the sentence as follows:

‘The mean SIC in NCEP/CFSR ranged from 0.01 in Baffin Bay in August-September-

October in 2001-2021 to 0.96 in the Central Arctic in February-March-April in both

1980-2000 and 2001-2021 (Table 3).’

Please note that in the revised manuscript, we have decided to change the ‘reference’

data set for subsection 3.1 to NCEP/CFSR (more details in response to comment

number 31).

30) Figure 2: “Mean biases of daily field means of sea-ice concentration between

ERA5 and JRA-55 (grey), ERA5 and MERRA-2 (black), and ERA5 and NCEP/CFSR

(light grey).” Which one is subtracted from which? Is it ERA5 - JRA-55 or JRA-55 -

ERA5? This needs to be clarified.

We have revised the part of description of Figure 2 (and analogically the description of

Figure 3 and Figure S2) as suggested, to avoid reader’s confusion:

‘Mean Biases of Daily Field Means of latent heat flux: ERA5 minus NCEP/CFSR (light

grey), JRA-55 minus NCEP/CFSR (grey), and MERRA-2 minus NCEP/CFSR (black).

Horizontal axis refers to Arctic basins as seen in Figure 1...’

Please note that in the revised manuscript, we have decided to change the ‘reference’

data set for subsection 3.1 to NCEP/CFSR (more details in response to comment

number 31).

31) Lines 154 -200: I noticed that both in the Tables/Figures and the texts, you are

comparing other reanalysis data to ERA5, even though you didn't assume ERA5

to be the best in your previous description. I don't think this is appropriate. These

descriptions and graphics overly emphasize ERA5 and neglect the intercomparison

between other reanalyses, for example, JRA55 vs CFSR. I believe this

is neither fair nor accurate. Please modify the text description and figures to

express "inter-comparison" in a more equitable and intuitive manner.

We do believe that the comparison using Mean Bias of Daily Field Means allows us

(and the reader) to compare each reanalysis to the ‘reference’ and, at the same time,

the other reanalyses between each other.

We did, however, reconsider the selection of the ‘reference’ dataset. While before, we

chose ERA5 randomly (as indicated in the original manuscript under ‘We do not

assume that ERA5 is the best reanalysis with respect to turbulent surface fluxes…’), we

do consider NCEP/CFSR the most realistic in terms of physical processes due to its

modelled sea-ice thickness and modelled snow depth on the sea ice.

Still, comparisons between other reanalyses (e.g. JRA-55 and MERRA-2) are clearly

visible in our Figure 2, 3, and S2 – e.g. in cold seasons and most basins in 1980-2000

Mean Bias in sea-ice concentration (Figure 2) JRA-55 minus NCEP/CFSR is positive,

while Mean Bias NCEP/CFSR minus ERA5 or NCEP/CFSR minus MERRA-2 is

negative, therefore we know the sea-ice concentration prescribed in JRA-55 is the

highest of all reanalyses considered.

We have also revised the respective part of the manuscript to be more clear about this

issue as follows:

‘NCEP/CFSR was chosen as the most realistic in terms of physical processes

'reference' due to its modelled sea-ice thickness and the snow on top of sea ice.

However, we do not necessarily assume that it is the best reanalysis with respect to

turbulent surface fluxes and use Mean Biases to present overview and comparison of

the typical values in all reanalyses. Mean values (temporal together with spatial) of

NCEP/CFSR variables in Arctic basins, seasons, and periods are shown in Tables 3, 4,

and S1. These Tables provide approximate look into the absolute values of variables in

the other reanalyses in respective Arctic basins (as temporal + spatial means in these

Tables cannot be directly compared to the values of Mean Biases of Daily Field Means

between NCEP/CFSR and other reanalyses presented in Figures 2, 3, and S2).’

32) Lines 214-215: What caused the higher sensitivity of LHF to SIC in this

region? It is not explained here.

In the revised manuscript, we have clarified that this matter is ‘further addressed and

explained in subsection 3.4’.

This is a new subsection in our revised manuscript, where we used mean data from the

SHEBA campaign to study the effects of thin ice on leads and snow pack on top of sea

ice on LHF and SHF and comment on the results.

33) Line 246: The explanation here is not clear - Why would an increase in SIC

variability lead to an increased statistical relationship between SIC and LHF? And

where is the literature supporting the increase in SIC variability? Please add a

reference.

In our data, we clearly see the increase in SIC variability in some regions of the Arctic

(under variability in this case, we mean ‘more days in the season with SIC other than 1’,

which could have been understandably confusing without an explanation).

While it is natural, that ‘more days in the season with SIC other than 1’ increases the

statistical relationship (significance) between LHF/SHF and SIC (variability in former

possible to explain by variability in latter), and it certainly happened like that in some

regions of the Arctic, this mechanism probably more often related to ODR model not

converging in 1980-2000 but returning a value of the slope between SIC/LHF (or

SIC/SHF) in 2001-2021.

Upon further inspection of the differences of SIC/LHF or SIC/SHF relationships between

1980-2000 and 2001-2021 in single grid cells, we found that in cases where the ODR

model converged in both study periods and returned steeper slope of the regression

line between LHF/SHF and SIC in the latter period, the reason for stronger statistical

relationship wasn’t as much caused by ‘more days in the season with SIC other than 1’

but rather just values of SIC/LHF or SIC/SHF forming a steeper slope (shown in

attached Figure S 3 – NCEP/CFSR data).

We have revised the possible explanation of the larger sensitivity of LHF/SHF to SIC as

follows:

‘Mostly in the Central Arctic, however, we found some areas of increased SIC effect on

LHF between 1980–2000 and 2001–2021… This increased SIC effect on LHF may be

explained as follows. We have mentioned before that the effect of SIC on near-surface

air temperature (and specific humidity) is not linear, and it is usually the strongest with

leads opening in SIC very close to 1. As indicated in Table 3 and shown in our

representative grid cells (Figure S 3, data from NCEP/CFSR in November-December-January), SIC in some areas of the Central Arctic increased

between 1980–2000 and 2001–2021 (possible reasons discussed in section

4.5). Therefore, there has been mostly very high SIC in 2001–2021, where even very

small decrease in SIC has a strong effect on near-surface air temperature and specific

humidity. We cannot be sure, however, whether the SIC increased in larger parts of the

Central Arctic in reality in 2001–2021, and only comment on possible physical and

statistical explanations of the phenomena as represented in reanalyses data.’

34) Line 291: There's an extra space here.

The following line should have been new paragraph, we have fixed this.

35) Discussion and Conclusions: This section is too verbose for me and lacks

clarity in its organization. I believe the authors can add subheadings to make the

structure clearer, such as 4.1, 4.2, etc. Some of the content in this section is

repetitive with the previous section; I suggest simplifying it. At the same time,

separating the discussion and conclusion into two parts would make the

structure clearer and more specific.

In the revised manuscript, we have divided Discussion and Conclusion into two sections

(4 and 5), and used subdivision of the Discussion section as following:

4.1 Differences between reanalyses, their importance, and consequences,

4.2 Simplification of the sea ice in reanalyses and its impact on surface turbulent fluxes,

4.3 Other uncertainties in parameterization of surface turbulent fluxes,

4.4 Role of sea-ice concentration and meteorological variables on surface turbulent

fluxes,

4.5 Decadal changes

Some of the subsections or paragraphs were added based on the comments of the

other Reviewer, however, we have also tried to simplify the Discussion section where

possible.