The paper describes the assimilation of along-track CryoSat-2 (CS2) sea ice freeboard observations into a sea ice-ocean coupled model. The along-track CS2 data are not commonly assimilated in the previous studies due to limited available tracks per day, but however are important for real-time forecasts that pursues smaller latency. The paper shows the conversion from freeboard to thickness and validates the 5-day forecasts against both the assimilated data and in-situ observations such as BGEP, IceBridge and EM sea ice thickness. I found that the paper is generally well organized and illustrative. The conclusions are convincing. Some paragraphs, I think, could be more compact. I would recommend a minor revision on the paper. However, the authors may check my comments below at first.

1. Snow thickness is difficult to simulate in sea ice models. The unreliable precipitation and the lack of robust observations could be attributed to as possible reasons. However, since the conversion from the freeboard to thickness relies majorly on how well the snow thickness is simulated, I wonder if the authors could provide some sentences to discuss the sensitivity of the forecast results to the simulated snow thickness.
2. L142: make -> makes
3. Eq (3): The threshold for thickness is 0.7m when having an uncertainty of 8. I realize that the authors have explained how the shape of these functions are obtained, but I feel curious of why 0.7m is used. If it was arbitrarily selected, then this information is necessarily to be present in the context.
4. Paragraph L370: About the unexpected improvement in the mean differences. The authors state that it is caused by the spatial noise introduced during data assimilation. That could be one of the reasons, but from my side, I tend to believe it is caused by the systematic errors of the model. In Figure 5b,c, the model shows negative thickness bias, which indicates a slower growth during the freezing period. However, as suggested in Figure 9a, slightly thicker ice than observations are generally found. That is to say, the assimilation introduced thickness increment is faster than the ice growth by the model physics, i.e., ice grows faster in observations than in models. Could that be the case? I have no evidence about that.
5. The discussion and conclusions could be made compact. I currently read it feeling too much redundant information. I would suggest one or two future plans are already enough for wrap-up.

Overview/summary

The authors present results of a study using the coupled ice-ocean FOAM modeling system (1/4 deg resolution used for this study) in which CryoSat-2 along-track ice freeboard data is assimilated into FOAM for a 3-year period from January 2015 - December 2017. The model assimilates CryoSat-2 alongtrack freeboard and uses snow depth from the model to convert freeboard to ice thickness. They show a reduction in RMSE and bias compared to a control run which did not assimilate ice freeboard data. The paper compares the control and assimilative experiments against NASA IceBridge (2015-2017), BGEP ULS moored ice draft (2015-2017), and Air-EM observations from 2015. A reduction in RMSE and standard deviation is shown against the NASA IceBridge data when comparing 5-day forecasts. Little improvement is shown when examining the BGEP data; the authors attribute this to the control experiment having similar ice thickness to what is found from CryoSat-2, thus showing similar results. The comparison between the Air-EM combined measurements of sea ice thickness plus snow depth showed poorer performance with the assimilative experiment versus the control run in 2015. They authors attribute this in part to uncertainties in the model snow depth, sampling error and possibly the in sity observations themselves.  No data is assimilated in the May-Sept timeframe, but the assimilative model still demonstrated skill during the summer months.

This paper is well written with sufficient details on the FOAM modeling system and NEMOVAR data assimilation system used in the study. Tables and figures are clearly laid out. This paper presents the first known use of along-track CryoSat-2 data in a operational ice-ocean modeling system.

I recommend minor revisions to the manuscript based on my comments below.

Specific Comments:

Line 105: What is horizontal resolution of the atmospheric forcing? Please include this in the text.

Page 14 Fig 4: add 87.5 N gridline on Fig. 4a and 4b to make the area more discernable.

Line 357-358: “Improvement” in Atlantic sector is (5%) is small and difficult to see in plot. Can you make a difference plot between 10c and 10d to show this?

To augment the NASA IceBridge, BGEP ULS moorings and Air-EM observations, the paper would be more comprehensive if Dartmouth/CRREL IMB were included. Specifically 2015D and 2015F (http:// http://imb-crrel-dartmouth.org/archived-data/).  These buoys are outside the Beaufort Sea region, where little improvement was found with the assimilation of CryoSat-2 along-track freeboard data. How do these observations compare with the FOAM ice and snow depths? I would like this analysis added to the paper.

The paper does not show any comparison to IABP ice drift data. Although not a fully coupled air-ocean-ice modeling system, have you examined ice drift and if you have, did you see any improvement in ice drift prediction between the Control and CryoSat-2 assimilative hindcast? If you have not , what is the level of effort to incorporate an ice drift analysis to the existing manuscript?

Review of „Assimilation of sea ice thickness derived from CryoSat-2 along-track freeboard measurements into the Met Office’s Forecast Ocean Assimilation Model (FOAM)“ by Fiedler et al.

Summary

The authors aim to assimilate Cryosat-2 (CS2) winter sea ice thicknesses (SIT) into Met Office’s Forecast Ocean Assimilation Model (FOAM). In contrast to other studies, where mostly gridded monthly or weekly products have been used for assimilation, here, CS2 along track measurements of freeboard are used instead. To convert freeboard into thickness, the snow depth from the model is used instead of a climatology. Validation is carried out with CS2 SIT estimates and a set of independent airborne/moored measurements. The authors conclude that "the assimilation greatly improves the SIT analysis and forecast fields generated by FOAM, particularly in the Canadian Arctic."

General Comments

I find it important to investigate methods for assimilating satellite ice thickness observations into forecast models and therefore I appreciate such work, especially, since it aims to use along track satellite data. However, I have some major concerns, especially with respect to the evaluation of the assimilation results.

It is stated that “the assimilation greatly improves the SIT analysis and forecast fields generated by FOAM, particularly in the Canadian Arctic. Arctic-wide observation-minus-background assimilation statis- tics show improvements of 0.75 m mean difference and 0.41 m RMSD (root-mean-square difference) in the freeze-up period, and 0.46 m mean difference and 0.33 m RMSD in the ice break-up period, for 2015-2017”.

If I understand correctly, the modelled SIT using assimilation of CS2, and the control run (no assimilation) are compared with CS2 SIT observations that are also used for assimilation? But then it seems somehow clear that the modelled SIT using assimilation of CS2 performs better. Perhaps it would be better to evaluate with a more independent product, e.g. CS2SMOS (at least this is a different product).   

I also assume that the modelled snow depth is a very important component here and potentially adds significant uncertainty. A detailed evaluation of the modelled snow depth within FOAM would be important as well for evaluation here, but I understand if this is not within the scope of this paper, and the authors also suggest carrying out such a study in future.

In fact, looking at the validation results using independent SIT measurements, the improvement from the assimilation is not so obvious anymore. Moreover, looking at Fig 7 (at the very beginning of each autumn), it seems that performance significantly decreases through the summer for both the assimilation and control run.

I think this discrepancy in the evaluation needs to be discussed in more detail, so it is clearer to the reader how good this assimilation really works and if it is a benefit using along track data instead of gridded products for assimilation. A comparison using gridded products for assimilation would be beneficial as well. Is it really an advantage to use along-track measurements? This should be discussed in more detail. After all, in view of these points, I recommend major revisions.  

Specific Comments:

L 114: It would be interesting to SIT maps also for the summer period. A figure showing SIT maps with and without assimilation throughout one season would be beneficial.

L 152-153: probably also mention dual altimetry, i.e., Ku/Ka band, e.g., Lawrence et al.

L 170: Why median and not mean?

L206: These are not helicopter measurements. PAM-ARCMIP measurements were carried out with a fixed wing aircraft.

L 339-347 and Figure 7: But there is a quite high mean difference/RMS at the beginning of autumn when CS2 data for assimilation become available again. But that means that the performance decreases significantly within summer months also for the model using assimilation, see also my general comments?

Figures: I recommend having all relevant information included in the figure. Sometimes this information must be searched for in the caption, which slows down the flow, e.g., Figure 6: colorbar label, Figure 7, 11: dashed lines. I think it is easy to provide that information in the legend.

The article of “Assimilation of sea ice thickness derived from CryoSat-2 along-track freeboard

measurements into the Met Office’s Forecast Ocean Assimilation Model (FOAM)” presents the

3-year assimilation experiment run in FOAM in which the along-track Arctic freeboard

measurements of sea ice have been assimilated into, and further the corresponding SIT/SIC has

been validated by assimilation statics and other independent observations. Clearly, this topic is

very interesting for other operational ocean-sea ice modelling and forecasting systems, and the

present results will be a good reference to compare with the assimilation of traditional gridded

SIT products. Undoubtedly, it is a nice work to show how to use this challenge observation of

SIT. However, there are still some issues affecting the full understanding of their results.

1) In the assimilation experiment, the converting from the freeboard Fi to sea ice thickness is a

key part to affect the SIT forecast performance in FOAM. Compared with the common

converting tool like Round Robin Data Package (ESA, 2013), the authors use the model snow

depth to replace the climatology. If possible, showing these two types of along-track SIT could

be interesting and meaningful to understand the large contrast between the different SIT

observation products in Fig. 12c and Fig. 13c.

2) As Line 286 of “an estimate of 50 km for theminimum SIT correlation length scale.” It can

infer the spatial scale for assimilation of SIT used 50 km. If that is true, it may be one of reasons

why the increments shown in Fig. 14 (a) and (b) are still noise. So there are two related

comments: 1) How about to compare the used horizontal scale for the SIC assimilation in this

system? 2) Have you tried to increase this scale from 50 to 100 km which is a practical order of

the spatial scale for averaging RA data.

3) To convert the sea ice draft from BGEP by dividing observations by 0.89 is not good for the

validations in the ice brake-up and freeze-up months, due to it omits the snow existing. The big

issue is that it will mix with the system bias of SIT and return to be detrimental of the analysis in

Fig. 15. So I suggest to investigate the best fit lines in the scatterplot are divided into the three

interesting periods by the months like MA, ON, MJJAS. It will be helpful to shed light on the MA

evaluation to contrast with the validation in Fig. 13 in Beaufort Sea.

4) In the section 2.4, the assimilation of SIT uses IAU as well, and the SIT increments how to

feedback on the 5 categories ice in sub-grid, although there are some words about “in

proportion to the initial volume distribution”. It should be paid more words or show one

example for the reader to well understand how they can work together.

Other general comments:

1) Line 97 at P4: “No SIT observations are currently assimilatedoperationally". It is better to

use “No SIT observations are currently assimilatedoperationally in the system.”

2) Line 193 at P7: “..for the SIT assimilation experiment period at the time of assessment.”

Replaced by “… for assessment of the SIT assimilation in the experimental period.”

3) Line 246 at P10: the representation uncertainty is set to 0.05 m. Does it mean the

minimal observation error is about 0.1 m as the minimal value around the 3m SIT shown

by the curve in Fig. 3a? If right, it could be better to be presented on this panel. And this

setting could be too small, compared with previous studies and other observation

platforms.

4) Line 353 at P18: “The largest regional improvements are in March-April, where the

mean difference is above 1.30 m and 1.22 m for the RMSD.” Is it possible to specify

where the regions are involved.

5) Line 396 at P 23, the 30-day periods were chose to cover the observation days. As

shown in Table 1 of Section 2.3.1, the observations are located only in one or two days

in the April so the 30-day window could be too wide and far away from the reality

condition.

6) Fig.13: The observations from Air-EM are clearly located into two regions: Beaufort Sea

and north of Canadian. The related scatterplot separated into these two regions may be

more helpful on physic to find something and compared with the result in Fig. 15.