Review of
An enhancement to sea ice motion and age products - Revision 1
by
Tschudi, M., et al.
I am refering to my first review of this manuscript for the summary of the paper.
I begin with a reply to a few points from the rebuttal letter to my first review. Obviously, you do not need to comment to these but are of course invited to do so.
General Concerns from 1st review:
GC1: No systematic evaluation of the products has been undertaken - neither for version 3 nor for version 4 of the sea-ice motion product. Also the associated newest sea-ice age data set is not evaluated. In your case, it is not sufficient to just compare version 3 and version 4 of the product because a systematic, detailed evaluation of version 3 products is missing in the scientific literature. There is hence no benchmark against which this new version 4 can be quantitatively referenced.
Section 2.2 does not provide new results. There is no indication of a true sea-ice motion retrieval uncertainty provided along with the product, like is done for sea-ice concentration and thickness data sets. You do not present results of an evaluation of the newly derived components of the sea-ice motion entering the gridded product as well as the gridded product.
Reply by the authors:
A) There has been significant validation done on the basic algorithm, particularly the MCC approach (Kwok et al., 1998; Kwok, 2008; Meier et al., 2000), as well as on a previous version of the specific product (Sumata et al., 2014; Sumata et al., 2015).
B) Uncertainty estimates are included in the daily combined motions, based on the optimal interpolation and the relative weights of the source data. We’ve added mention of this to the text.
C) Sea ice age is difficult to directly validate as there is not validation with sufficient accuracy and coverage. We have added a reference (Lee et al., 2017) to an intercomparison study that shows good overall agreement between the ice age data and other ice type/age products.
My comment to the authors' reply:
To A): I can agree to that. However, this is version 4 and you compare version 4 to version 3. Version 3 has not been assessed even though there have been substantial changes between version 2 and version 3, e.g. in the selection of which passive microwave data are used (see Haumann et al., 2016). Particularly when it comes to the long-term stability and the consistency across different satellite sensor products used - which is a mandatory element of product maturity for such a long-term and important data set as the one you are presenting here, it seems sub-optimal to refer to evaluation activities which only target the basic algorithms. One could have expected a considerable move forward here. I therefore appreciate that you now carried out an evaluation of the combined Arctic product with CRREL ice-mass balance buoy data. This is a good start.
To B) I am aware of these uncertainty estimates. These are very much based on the merging process of the data. The mentioned relative weights are only valid / can only be computed where there is a buoy for comparison - plus some correlation length scale around these - leaving the majority of the investigated region void of (detailed / accurate) quality information; this applies in particular to the Antarctic, doesn't it?
What about the uncertainties of the individual products, i.e. the AVHRR ice motion vectors or the AMSR-E ice motion vectors?
To reviewer #4, who stated: "Page 8, line 1: The error for each drift product (and used to calculate C) should be included in a Table.", you replied: "These values were calculated early in the development of the original product. There is not a table of errors for each drift product."
"early in the development of the original product" is exactly what I am referring to. It appears there was very limited effort to re-evalute the newer original data sets input into the merged product. If we'd compare this to the situation in the sea-ice concentration community then this would mean that this community would try to sell a new improved sea-ice concentration data set - referring to the accuracy of the SMMR sea-ice concentration data sets. --> See my general comment GC1 to the revised manuscript.
To C) Certainly I agree that it is challenging to get a reasonable set of evaluation data for this purpose. However, I mean, your group has been THE group to build and maintain this data set for over a decade now and it has received great attention. To my opinion it would have been more than timely to come up with some attempts and advanced ideas to evaluate this data set - at the latest after the publication of Korosov et al. (2018). Therefore I am inclined to rate the efforts carried out into this direction as not sufficient. While the reference added (Lee et al., 2017) mitigates this impression a bit, it cannot replace a decent evaluation and/or inter-comparison study which underlines the improvement of version 4 over version 3 and underpins the credibility of this data set.
GC2: The reader and data set user is informed about user statistics, the importance of the two data sets, some selected bits of the history of the retrievals, and a relatively unspecific description of the changes made to the methods. This is, however, potentially not what a reader of this paper and user of this data set would have expected for the following reason: There is no specific paper in which the various retrieval processes, their uncertainties, the caveats of the different spatio-temporal resolution of the input data sets, a detailed description of the merging (optimal interpolation) approach and its uncertainties have been published so that the full package of detailed, high-quality information is visible at a glance. The retrieval, the input data, the pre-processing steps all these are not transparently described. In other words: A benchmark reference paper containing all bits and pieces is missing so far.
Reply by the authors:
Such information is provided is the User Guide for the product, provided by NSIDC. We chose not to include this information in the manuscript in the interest of brevity. We’ve added reference to the NSIDC User Guide. We agree that a peer-reviewed document on the original development of the product would have been useful. However, the original product developer chose not to submit such a paper. As such, our purpose with this manuscript is not to try to recreate a history of which we do not know all of the details, but to document the changes and improvements the current team has made to the newest product as well as giving an overall summary of the processing that is described in the NSIDC User Guide.
My comment to the authors' reply:
I do understand the challenge behind trying to unwrap the bits and pieces of the original (versions 1 and 2) processing of this data set. I would have hoped that you would be able to undertake this tedious work - because nobody else can do it; you are closest to those people and institutions which developed the data set. I see that with my review I wasn't able to convince you that for a journal such as "The Cryosphere" it pays back to double efforts to accept this challenge. I assume that this has been a question of funding and therefore need to leave this issue where it is. It is a pity. As I stated in my original review (actually you never received the long version which made me to reject the previous version of your manuscript) it is very tedious for a data user to climb down into all the web pages that are linked from the NSIDC User Guide and it is in addition relatively frustrating that the information one searches for is partly quite limited or simply not available.
Please also see my GC1 to the revised manuscript in this context.
####################
Now comes the review of the revised version of the manuscript
Following up with the authors' reply to my general concerns formulated in my review of the first version of this manscript, and with the comments of the other reviewers, I am convinced that the manuscript will become acceptable after major revisions.
These revisions should cover the general comments GC1 through GC3 (see below).
While the remaining two general comments (GC4 and GC5) are just ideas to think about, I ask you to consider the specific comments. Some of these - to my opinion - require attention for an adequate understanding of your paper and for having it placed in "The Cryosphere". By following the suggestion I formulate in GC1 you might be able to solve some of the specific comments.
Finally, I found a good number of typos and formulations which require attention which I listed separately in the last part of my review; here you will also find suggestions related to the figures.
General comments:
GC1 refers to the insufficient description of production / evaluation of the "old" basic retrieval of the ice motion. I can understand that it might be too time consuming to dig into all the old documents, web pages, reports, folders, whatsoever left behind by the scientist team responsible for the creation of versions 1 through 3 of the ice motion product. I understand that you don't want to do it and that you don't receive funding for it. Still, this information is mandatory for a reader and/or user to fully understand what this product is about, what its limitations are, and where substantial potential for improvement exists. This would be of high value for other scientists to take over and to develop alternative products.
Since you don't want to write a "benchmark paper" as I suggested in my previous review [even though that one would be good for The Cryosphere, while this manuscript here might be more suitable for Earth System Science Data], I suggest that you provide list of the unknowns, open parameters or assumptions, and original evaluation activities in form of a few tables and/or illustrations put into an appendix to this paper.
Example 1: You cannot find out how C and D in Eq. 1 were computed exactly? What the values are? What the correlation lengths are? Mention this explicitely in one of the tables.
Example 2: You cannot find out whether the 4X over-sampling is carried out for all 37 GHz channel data of all three instruments SMMR/SSMI/SSMIS? Mention this explicitely in one of the tables.
Example 3: Nobody has yet investigated quantitatively the impact of the switching between 37 GHz and near-90 GHz data for SSMI, AMSR-E, SSMIS data on the obtained ice motion? Fine. Mention this in one of the tables (this is something noted by Haumann et al. 2016 for version 2 of the product, creating a big mess in their search for a credible time series of the ice motion for the Antarctic).
Example 4: You cannot find out whether the input PMW ice motion vectors have been evaluated after introducing the 4X oversampling? Mention this in one of the tables.
Example 5: The attempts to evaluate ice motions in the Antarctic are unknown / limited to the following (...) studies? Mention this in one of the tables.
Examples 6: What are the evaluation studies specifically targeting version 3 of the ice motion product? List the references / or NSIDC reports.
I strongly recommend to be as transparent as possible and take over the responsibility to document the status of this unique long-term ice motion product as best as possible.
GC2 is refering to section 2: When you write about "velocity components" in sections 2.1 and 2.2: do you refer to the meteorological convention, i.e. component u being defined positive from west to east and component v being defind positive from south to north? I am asking because I noted that versions 3, 4 and 4.1 of the ice motion product contain u- and v-velocity components in the direction of the EASE grid used, i.e. a positive u-component at 0degEast is a true west-to-east ice motion while at 180degEast it is an ice motion into the opposite direction. You might want to clarify this at the beginning of your paper.
I note that in section 2.3 you note that the motion is relative to the EASE grid for the gridded product.
I note further that in section 2.4. it is not clear whether you transferred the CRREL buoy ice motion to the EASE grid notation of u and v. By the same token, users being familiar with the meteorological notion of motion might have difficulties to translate your uncertainties into their understanding of how air, water and ice movement is described, i.e. positive west --> east; positive south --> north.
GC3: Information about Antarctic sea-ice motion is VERY limited. You could at least have added similar plots like Fig. 5, Fig. 7 and Fig. 8 to illustrate that you also provide ice motion for the Southern Ocean and how this looks like in comparison to the Arctic. Yes, it will be a challenge to discuss the jumps in the time-series between the different satellite data sources. To my opinion, because your paper is about a bi-polar (or global) data set you have to show these and you have to discuss these and you have to clearly point out the limitations to the users. I rate this a mandatory element which is still missing in the revised version of the manuscript. Alternative: You remove ALL information about the Antarctic.
GC4: What would you say is a typical uncertainty estimate for the ice age (applies to Figs. 9-11 and their interpretation)? Would it be fair to say that ice-age fractions can be determined as accurate as 50 000 sqkm? ... or rather 100 000 sqkm? I would find such an estimate very useful in the context of the interpretation whether recent changes in ice age fraction are significant or within the noise created by the method.
GC5: This comment applies to Page 20, Lines 13-25. In the context of the various data sources with different resolutions, different spatiotemporal coverage, and different relative weights used for the combined ice-motion product which is subsequently used for the derivation of the ice age: How much, to your opinion, do these differences have an impact on the obtained trends in ice age fractions?
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Specific comments:
Page 3, lines 1-6:
- I am wondering whether there isn't knowledge about the relationship between ice age and thickness from earlier studies, i.e. before the NSIDC sea-ice age product era. I am sure there is. What about the book by Untersteiner for instance? Please add a respective reference to the citation of your own work.
Just a comment: It might be true that sea-ice age provides additional information - compared to maps of the multiyear ice cover - about when in a certain region sea ice will melt out during summer. But has this been proven yet in the 10+ years the ice age data set has been out? One could imagine that an accurate first-year ice versus multiyear ice discrimination is sufficient here and that it does not really play a role whether the sea ice is 2, 3, 5 or more years old when it comes to melting. Isn't the main interesting information given by the sea-ice age data set that we have a measure of how long sea ice once formed survives as a function of formation location and ice movement?
Page 3, paragraph starting in Line 22: This is a good start.
- Please add the work of Sumata et al. (2014, 2015) here because it gives a good overview about existing products and how these compare.
- Please add the work of Kwok (2008) about summer-time sea-ice motion estimation using near 19 GHz data. It is important for a reader / data user to know that there are alternatives to using NCEP winds based drift estimates during summer.
- Please add a few sentences about the various attempts to use SAR data for deriving sea-ice motion. This is important for a reader / data user who is interested in small-scale solutions. It is further important for you yourself as this would underpin how valuable your data set is in terms of spatiotemporal coverage.
Page 6, Lines 16-18:
It appears sub-optimal to first state that with 25 km grid resolution one can only estimate the velocity field to the nearest 25 km / day for each motion component (see lines 14-15) and then here state your product obtains useful daily motions by using the described over-sampling.
The point I wish to make here is that the native resolution of the 37 GHz channels of SMMR, SSM/I and SSMIS, as given by the footprint sizes, is coarser than the grid resolution used and, in addition, these footprints are only a good approximation of the reality because of the antenna sidelobes. So what you sell to the user here is: brightness temperature observed at 37 GHz, gridded as a daily mean value into a 25 km grid, show features distinct enough that you can move over these 25 km grid cells with increments of 6.25 km. I am wondering how you assessed the improvement in accuracy stated on Page 6, Line 19. Is there a chance to include the respective results in this paper, e.g. as supplementary material / appendix? Possibly the improvement in accuracy is a function of the frequency because the native resolution is much finer for the near-90 GHz channels of SSM/I and SSMIS.
- Another over-arching question to this oversampling: For brightness temperatures of the near 37 GHz channels of SMMR and SSM/I and SSMIS you apply 4X oversampling. How about for the 12.5 km grid resolution near-90 GHz data of SSM/I and SSMIS? And: How about over-sampling and its application to AMSR-E data at 12.5 km resolution? It appears that the description of the over-sampling procedure is not yet complete; other scientists willing to repeat your steps would not be able to do so because of a lack of information.
- How does this over-sampling method compare to the continuous MCC suggested by Lavergne et al. (2010)? It is worth commenting and discussing this issue in this paper because the over-sampling seems to be something introduced relatively recently (?); it deserves to be discussed in the light of alternative choices.
Page 7, Lines 16-23:
- Can you comment on minimum correlation values used by other products? How about the Girard-Ardhuin and Ezraty (2012) products?
- "Various thresholds ... original development of the product ... qualitatively determined ..." --> This is a relatively vague information and leads to many questions: Was this same threshold applied to all satellite data sets described so far? Has the choice of this threshold been revisited in the meantime? Based on which data set(s) this threshold was selected (when?) by the mentioned qualitative determination? Is there a paper or two to which you can simply refer to which illustrate this determination?
- The description of the neighborhood filter should be precised. Neither is it clear how this is done technically (and at which grid resolution). Nor is quantified what is meant by "similar" ... direction? magnitude? How large a difference can be before the vector is considered to be spurious?
Page 8: Lines 1-2: "Another change" --> What kind of a change? Please be more specific. If I understood you correctly then this is a version 4 change - so there is no excuse to not detail what apparently has been done recently by you (and not by the other scientists which developed version 1-2 (3)) of the product. Please note: Without refering to figure 4 the information about the apparent improvement by this post-processing step is completely hanging in the air.
Page 8, Lines 20/21: "combined motion fields" --> How about the retrieval success of satellite derived ice motion during summer? Aren't during summer these "combined motion fields" only based on buoy motion and NCEP/NCAR winds? Please be more specific in your description. Perhaps you adopted the method of Kwok and use the 19 GHz channels now as well?
Page 12, lines 11-24:
I note that the description of how C and D are found is vague. Nobody could re-do this analysis. It is not clear - after your previous statements about the influence spatial resolution appears to have on precision (or accuracy) - why C is set to a constant value when you are operating with individual ice motion resolutions between ~25 km (using SMMR/SSMI/SSMIS 37 GHz), 12.5 km (same but using near-90 GHz or AMSR-E near-37 GHz), 6.25 km (AMSR-E 89 GHz) or 4 km AVHRR. Please try to be more specific in your description.
Also, later on you are refering to the improved interpolation used in version 4, pointing out that the buoy weight drops to zero (e.g. Page 20, Line 2) at a certain distance to the buoy. However, it is not clear what this distance is, whether we speak about 50 km, 500 km, 2000 km of these, presumably, correlation lengths. It is not clear how these were calculated and whether and how much these differ between the single products and input data used as input for the merged product.
Page 13, Lines 5-9:
I strongly recommend to provide more details here. While your Figure 3 illustrates nicely the effect of this amendment from version 3 to version 4 what is missing is information about the distance within which these 15 highest-weighted ice motion vectors are selected. I assume it has something to do with the correlation length scales (please provide examples of these). If not, then one could provocatively say that at any point within the Arctic Ocean the gridded ice motion product is solely determined by buoy motions if at least 15 buoys are reporting - because, as I understand your averaging with C=0.95, these 15 buoys would have the highest weight and would be used no matter how far away they are from the grid cell considered.
Page 14, Lines 18-27:
- Please add information whether you used the daily or the weekly gridded product.
- Please add the fraction of discarded CRREL buoy motion estimates; I assume it is very small.
- Please provide a map of the tracks of these buoys for illustration. In light of the next paragraph you should highlight the track the year 2015 (unless you color code them anyways and unless you decide to use all years instead of just one).
- Please provide information about the processing steps. The combined gridded ice motion product has u- and v-component of the ice motion aligned with the EASE grid. How about the CRREL buoy data? How did you compute the u- and v-components of the CRREL buoy motion in the EASE grid? This description should also include a notion about how many CRREL buoy position observations of one day form one daily estimate.
- Please describe in detail how you compared the data and hence how you ended up with the numbers shown in Table 3, i.e. did you compare absolute values or the "native" positive and negative values? I assume you did the latter. Did you also compare the direction and the absolute value of the ice motion vectors?
- Please provide at least one plot for each ice-motion component which illustrates how the ice-motion values scatter. You could do this as a scatter-plot, a 2-dimensional histogram, regular histograms, ... whatever you like, but please show the reader / user more than just a table.
Page 18, Lines 19-27:
- Just to re-cap: The 4X oversampling is something introduced from version 2 to version 3? Or was this already introduced in version 1?
- Line 25: I don't understand why the different temporal sampling of SMMR explains why the difference in the weekly average drift speed between version 3 and version 4 is near zero.
Page 19, Lines 1-3: How much of the larger differences between version 3 and 4 during summer can be attributed to the change in weights in combination to a predominant usage of NCEP/NCAR winds based ice motion [assuming that in summer there as substantially fewer valid PMW ice motion vectors]?
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
Editoral comments / typos:
Page 4, Line 11:
I'd say Szanyi et al. (2016) refered to the sea-ice motion product as well; therefore I suggest to add "sea-ice motion" here as well. This would also comply better to the "both products" notion in the next line.
Page 4, Line 21:
"monthly" --> It might make sense to either correct the NSIDC web page or the text. Monthly estimates seem not to be available.
Page 5, Line 19/20: "a region around that grid cell" --> Does this mean that the search window is centered at the center of the grid cell center and hence extends by typically 25 km in x- and y-direction?
"typically 50km" --> this reads as if different search window sizes are used. Is this the case? Please be more specific.
Page 5, Lines 22-24: "The highest correlation value, i.e., the correlation peak, is determined to be the offset in the positionof the grid cell between the earlier and the later image; then, the ice velocity is computed by dividing this offset by the time separation between images."
I don't think that the correlation itself gives the spatial offset, does it? How about "The highest correlation value, i.e., the correlation peak, is assumed to coincide with the most likely offset in the position of the grid cell between the earlier and the later image. This offset in the position yields a displacement vector pointing into the direction of the ice motion while the ice velocity is computed by dividing its magnitude by the time separation between the two images used."
Page 6, Line 11:
"daily": SMMR did not provide daily data. This needs to be corrected in the text.
Page 6, Line 14:
"a gridded a resolution" --> "a gridded resolution"
Page 6, Line 15:
"many similar" --> please be more specific and refer to these, e.g., with a reference. Perhaps "many" could be deleted?
Page 6, Line 20:
What about SMMR? Is the oversampling applied here as well?
Page 6, Line 21: I suggest to add "theoretical" to "motion precision".
Page 6, Paragraph starting at line 25:
- Why were AMSR-E data not used for the ice motion product of the Southern Hemisphere? Please either add the reasoning or at least mention that AMSR-E was only used in the Northern Hemisphere.
- This part: "AMSR-E had more than double the spatial resolution of the previous sensor, 6.25 km gridded resolution for some channels, so its motion resolution was likewise improved. So, during this period (2002-2011), it was also used as a source for ice motions." should be re-written as it does not read well. It might help to refer to Table 1 by the way. "of the previous sensor" --> better "than SSM/I and SSMIS"
- On Page 6, Lines 7 and 9, you provide references for the SSM/I - SSMIS data product and the SMMR data product. An adequate reference for the AMSR-E data is missing.
Page 7, Lines 6-14:
- Why were AVHRR data not used for the ice motion product of the Southern Hemisphere? Please either add the reasoning or at least mention that AVHRR data were only used in the Northern Hemisphere.
- Your reasoning that AVHRR data were not used after the year 2000 because AMSR-E became available reads a bit strange given the fact that AMSR-E became available in May 2002. I assume that AVHRR data usage was simply confined to version 1 (and version 2) of the NSIDC sea-ice motion product and/or that the AVHRR data set used those days (...) simply terminates at the end of 2000?
- AVHRR has visible and infrared channels as you state. But which are used for the "Daily gridded composites"? In other words, are the AVHRR ice motion vectors based on visible or infrared data?
- "higher resolution = more accurate motion estimates" --> Is this the case? Or is it mainly the precision which improves as resolution refines? In any case, it might be superb to add information about an inter-comparison between buoy, AVHRR and, e.g., SSM/I based ice motion estimates illustrating this statement in the supplementary material.
- Line 12: "(6.25 km vs. 4 km)" This reads as if the often strongly weather influenced near-90 GHz channels are the backbone of the ice motion estimates using AMSR-E. Is this the case? If not, then you need to mention the 12.5 km grid resolution for the near-37 GHz data. If yes, then your description about the over-sampling of the 25 km gridded data (near 37 GHz data) is given a bit too much weight because then one would assume that also for SSM/I and SSMIS the backbone data set are the near-90 GHz channels which in fact come at 12.5 km grid resolution. Your writing is hence inconsistent and should become more specific.
- On Page 6, Lines 7 and 9, you provide references for the SSM/I - SSMIS data product and the SMMR data product. An adequate reference for the AVHRR data is missing.
Page 10, line 16: I suggest to add "theoretical" in front of "limit of precision"
Page 10, paragraph starting on Line 16:
- This is a very globally written, non-specific paragraph. Please note clearly which versions of the NSIDC ice-motion products were compared here. It would also be important to know whether these comparisons were done using the single ice-motion vectors or using the daily gridded product. It would further be important to learn about the amount of data compared here, i.e. are we talking about several years worth of data or a few months or even only days? Example 1: In Line 21 it only states "the Lagrangian motion product". Example 2: In Line 23 it is not clear whether these "SSMI-derived daily velocity components" are from a gridded product or single motion vectors. The same applies to Line 24, wherein you write about the AMSR-E ice motion. Here, one asks oneself what "error" is meaning in particular. Example 3: In Lines 25/26 it only says "the ice motion data". Neither the time period, nor the gridding or the version number are given. Note: a high correlation is wonderful but how about bias and RMS error for the Sumata et al. studies?
Page 11, Lines 5/6: Please check this first sentence. It gives not clear meaning.
Page 12: Line 2 "and monthly" --> should be deleted as there appears to be no monthly ice motion field available (anymore). At least they are not accessible via the product's web page.
Page 12, Lines 11-24:
- You mention three times that C is a ssigned a value of 0.95 for buoys. One time might be enough.
Page 13, Line 14: Please provide a reference for the correlation length scale of ice motion. Is it the same in the Arctic?
Page 13, Line 20: The perfect place to refer back to Figure 2 which nicely illustrates how smooth the "Winds" ice motion is compared to the PMW ice motion.
Page 13, Line 21: "corrects this over-filtering" --> How? This is your recent work and should be detailed more. See my previous comment on this issue.
Page 14, Lines 1-15:
- I note that in this paragraph finally you do not mention the monthly product anymore.
- Line 10: "discretization effects" --> Could these also be caused by the fact that your 4X downscaling applied to original 25 km gridded resolution (with even coarser footprint) data often cannot resolve the anticipated smaller-scale variations in ice motion but in contrast enhances noise - particularly in the direction of the motion? I recommend to spend a sentence or two about this issue and also take into account the paper by Lavergne et al. (2010).
Page 15, Lines 1-13:
- In Line 1 please refer to the respective paragraph or subsection.
- Line 3: Is the ice speed increasing? Then we are talking about a positive trend and not about an "increasing trend". Or is an already existing positive (or negative) trend increasing? Please be more clear in your formulation.
- What is your explanation for the 10% difference between versions 3 and 4 in the bias of the v-component compared to the near-1 % bias of the u-component?
- Lines 11-13: Frankly speaking I am very surprised about the small impact of using 2% instead of 1% to derive ice motion from NCEP/NCAR winds, but I am quite confident that this can be explained with your choice of data. Therefore, while I appreciate your future plans with respect to regions and long-term trends I am suggesting at first to carry out the comparison for the entire CRREL buoy motion data set (why 2015?) available (as used in the previous paragraph).
Secondly, as you correctly write, the weights used for NCEP/NCAR wind-speed based ice motion are very low - particularly during winter when you have plenty of useful PMW ice-motion estimates. I am sure this changes during summer melt. Therefore it would certainly be much more informative to show the comparison with CRREL buoy ice-motion data alongside with purely NCEP/NCAR based ice motions (i.e. those exemplified in Figure 2 c). If you do this as a time-series you could also account for the fact that during summer NCEP/NCAR winds-based ice-motion estimates potentially play a substantially larger role for the combined gridded ice motion product.
Page 15, Line 15: "the motion product" --> which one? The combined gridded one? Please write specifically what you use and do.
Page 16, Line 7: "1" --> "one"
Page 16, Line 12: Is an "increment" by definition positive?
Page 18, Line 17: "motion age"?
Page 18, Lines 20-21: "motions are smoothed out"? So they are zero?
Page 19, Line 7: "increasing trend" <--> "positve trend"; see my previous comment.
Page 19, Lines 7/8: Isn't it strange to see that for the version 4 product, which is faster than the version 3 product especially for the SSM/I period (see Figure 7) you get a larger positive trend from version 4 than from version 3 data? Perhaps you should mention in the context of Figure 7, that after the AMSR-E period, the difference between versions 3 and 4 is even larger than before the AMSR-E period.
Page 20, Line 11--: What does, to your opinion, explain the fact that the differences between version 4 and version 3 are near zero for all ice-age classes for winters 2005/06 'til 2012/13? Do we really understand why for these winters (dominated by AMSR-E 89GHz PMW ice motion input at 6.25 km grid resolution) the change in the interpolation method between version 3 and version 4 appears to have no influence?
Page 21, Line 8: Please add the information that the winds originate from the NCEP/NCAR atmospheric reanalysis product.
Page 21, Line 28: I suggest to replace the ESA CPOM 2015 link by the recent paper by Salilla et al. from 2019 in The Cryosphere; this paper gives a comprehensive overview about currently existing CryoSat-2 sea-ice thickness products.
Page 22, Line 1: "ice age motion" ?
Page 22, Line 6: Please add "like suggested by Korosov et al. (2018)" behind "... EASE grid cell" to indicate that this is a feasible idea which has been followed up with already by other scientists.
Figure 1: I note that this figure omits the AVHRR data.
Figure 3:
I note that in panel (a) a lot of the circular features with substantially different ice motion than the surrounding do not contain a red dot for a buoy being present. One example is found in the northern East Siberian Sea and another, more pronounced example is north of the Laptev Sea. How do you explain this?
- Is the grey scale the same between panel (a) and (b)? I am asking because I am surprised to see that the area with high positive values in the northern East Siberian Sea has so much increased (both in extent as well as in magnitude of the values) from panel (a) to panel (b). Please comment on this.
- I recommend to also show the respective uncertainty information for these two maps to illustrate to the user what the effect of this step is on this parameter. This might be an important information for users attempting to assimilate your product.
Figure 7: Please state in the text and the Figure caption on which grid cells this difference is computed. Did you use all grid cells? This could mean that you are superposing the true difference between version 3 and version 4 ice motions with an influence of the change in the ice drift distribution due to the sea-ice retreat which (the influence) might be different for version 3 than for version 4. Therefore I ask: How would this figure look like if you limit this comparison to the NSIDC region "Arctic Ocean"? Would the differences still be of the same magnitude?
Figure 9 and its interpretation:
- What is special in winters 1995/96 and 1996/97 causing version 4 and version 3 4+year old ice extent to be similar even though we are still in the middle of the SSM/I period?
- While the previous issue is difficult to understand the one I am refering to now is logical and should be explained in more detail in your text. Apparently, introducing AMSR-E data in summer 2002 did not have an impact on the 4+ years old ice immediately but it took until winter 2004/05 or 2005/06 to see its effect on the difference between both product versions. Likewise, the termination of AMSR-E usage in fall 2011 and hence switch to coarser resolution SSMIS data manifests as late as 2014/2015. This is logical because it takes 3 years for the benefit of first the finer, later the coarser resolution to have an effect on the old ice.
Figure 10:
- Please provide a similar time axis as you used in Figure 9.
Figure 11:
- Please provide a similar time axis as you used in Figure 9.
- What is the rationale behind including Kara and Barents Sea into this plot?
- Caption: I have a problem with usage of the word "trend" here. To me a trend is something I compute from a time series of data, e.g. via a functional relation, and it has, in its simplest form an intercept and a slope. What you plot here are, to my opinion, time series of the fractions of the different ice-age classes. |
- Printer-friendly version
- Supplement
