New methodology to estimate Arctic sea ice concentration from SMOS combining brightness temperature differences in a maximum-likelihood estimator

Gabarro, Carolina; Turiel, Antonio; Elosegui, Pedro; Pla-Resina, Joaquim A.; Portabella, Marcos

doi:https://doi.org/10.5194/tc-11-1987-2017

Articles | Volume 11, issue 4

https://doi.org/10.5194/tc-11-1987-2017

© Author(s) 2017. This work is distributed under
the Creative Commons Attribution 3.0 License.

https://doi.org/10.5194/tc-11-1987-2017

© Author(s) 2017. This work is distributed under
the Creative Commons Attribution 3.0 License.

Articles | Volume 11, issue 4

Research article

|

30 Aug 2017

Research article |

| 30 Aug 2017

New methodology to estimate Arctic sea ice concentration from SMOS combining brightness temperature differences in a maximum-likelihood estimator

Carolina Gabarro, Antonio Turiel, Pedro Elosegui, Joaquim A. Pla-Resina, and Marcos Portabella

Download

Final revised paper (published on 30 Aug 2017)
Preprint (discussion started on 03 Nov 2016)

Interactive discussion

Status: closed

AC: Author comment | RC: Referee comment | SC: Short comment | EC: Editor comment

- Printer-friendly version

- Supplement

RC1: 'review of "Measuring sea ice concentration in the Arctic Ocean using SMOS" by Carolina Gabarro et al.', Anonymous Referee #1, 05 Dec 2016
- AC1: 'Response to Review 1', Carolina Gabarro, 06 Feb 2017
RC2: 'review: Measuring sea ice concentration in the Arctic Ocean using SMOS, C. Gabarro et al.', Anonymous Referee #2, 08 Dec 2016
- AC2: 'Answere to referee 2', Carolina Gabarro, 06 Feb 2017

Peer-review completion

AR: Author's response | RR: Referee report | ED: Editor decision

AR by Carolina Gabarro on behalf of the Authors (07 Feb 2017) Manuscript

ED: Referee Nomination & Report Request started (22 Feb 2017) by Lars Kaleschke

RR by Mohammed Shokr (04 Mar 2017)

Suggestions for revision or reasons for rejection

Review of manuscript “Measuring sea ice concentration in the Arctic Ocean using SMSO” by Gabarro et al.

The manuscript introduces a new algorithm to retrieve sea ice concentration in the Arctic region based on radiometric observations from the L-band SMOS, with its multi-viewing angles. It uses the Maximum Likelihood criterion with input distributions of a couple of radiometric indices, assuming Gaussian distributions with mean and standard deviation obtained from model and observation findings.

The idea is interesting, the study is scientifically warranted and it is an appropriate continuation, with a remarkable progress, of the limited number of similar studies using SMOS in the past 8 years. I would recommend publication and certainly commend the authors on their effort. However, one should keep in mind that the challenge of estimating SIC from remote sensing data remains.

In most studies, sea ice is treated as one entity. Yet, in nature it is manifested in several types with large diversity of physical properties that may lead to consider it as different entities. Thin ice, seasonal ice, perennial ice, summer ice, etc. are different “creatures” within the realm of sea ice; let alone snow-covered ice under different metamorphosed snow conditions that lend itself to different radiative properties, even under the passive L-band radiometry. Given this concept, I would be cautious when considering SIC algorithms trying to approximate the wide range of ice properties into a single set.

With this in mind I am not comfortable with the title of the paper that groups all ice types under one phrase “sea ice in the Arctic Ocean”, but I guess nothing can be done here! But it would be more appropriate to use the word “Estimating” in the title instead of “Measuring” since we don’t really measure SIC.

The method adds to the to the tools of estimating ice concentration from microwave data. One advantage of having different methods is to be able to perform cross-checking. This is important because there is no reliable “truth” data against which we can evaluate each method. All methods suffer from errors and the only way to approach the “truth” is by cross-checking. For example, the current study performs the validation by comparing results against maps from OSI-SAF. But the latter, much like any other operational ice maps, may not be considered truth data either.

Another concern is about a sentence in the Abstract; “We find that sea ice concentration is well determined (correlations of about 0.75) when compared to estimates from other sensors such as the Special Sensor Microwave/Imager (SSM/I and SSMIS).” Retrieval of any parameter from remote sensing data is associated not only to the sensor characteristics but also to the retrieval method. I believe the method used in this study is new, so if there are other method using L-band then the authors can do comparison. To keep the above statement, the authors should qualify it; namely to say correlate well with other passive microwave – but under what condition (when and where). I think it should correlate well with other sensors over mature Arctic sea ice in winter. Other than that I don’t think the correlation would reach 0.75.

A few suggestions for corrections are listed below. It would be nice if the authors consider them while preparing the final submission.

Page 4 Line 21: correct the sentence to be “Hereafter we will use TB to refer to surface brightness temperature, for simplicity”

Page 4 Line 25: correct the sentence to be “is the ratio between reflected and incident radiation”

Page 5 Line 4: no need to mention the refractive index (n); this is for optical remote sensing but here we deal with microwave.

Page 5 line 6: the sentence “The nonlinearity is an advantageous property for remote sensing …” is not explained. How? Also it has no relevance to the text before it. The authors may remove it.

Page 5 Line 7: if Fig. 2 is for the L-band please mention that in this line or in the figure caption. Also, while the authors mentioned the seawater and sea ice parameters that are used in equation 3, they did not mention the snow parameters. Here they have to be careful because it is difficult to characterize the snow by a single temperature value as it is highly responsive to the air temperature. Even for dry snow, it can be lossy because the salinity at the snow base is usually higher than 0; it can be as high as 20 ppt or higher as shown in many studies.

Page 5 line 23: “dry snow still has an effect in (make it “on” not “in”) emissivity that changes with the angle of incidence according to Snell’s law”. Snell’s law is about the angle of refraction, nothing to do with the emissivity.

Page 6: in the set of presented equations I think one equation is missing; that is the one that determines the reflectivity from the ice surface in terms of its dielectric constant (i.e. Fresnel equation). This should be inserted before equation 5.

Page 7 line 7: close the bracket after (resulting from ….unknown physical parameters).

Page 7 line 10: something wrong in the sentence “… among conditions such as deletedwhen a phase change …”

Page 7: just wonder why didn’t you use the polarization ratio instead of the polarization difference? The former is more common and it eliminates the dependence of the brightness temperature of the physical temperature. I am not suggesting to change the present scheme but an inclusion of a sentence to explain why PD and not PR would be useful.

Page 7 Line 25 “we will use tie points as ground truth estimates of sea ice concentration”. What does that mean? Tie points are used to estimate ice concentration based on a set of algebraic equations.

Page 8 Line 7: just a comment on Figure 5, it is good to see the model confirms what we know – that Tb, not PR or PD can be used to estimate ice thickness. The latter are good for estimating ice concentration.

Page 9 Line 3: the statement “AD is the most robust index to retrieve SIC, slightly better than PD, and significantly better than TB, as TB is highly sensitive to ice thickness variations” may need some more thought. The fact that TB in the L-band is sensitive to the thickness has no relevance to its robustness in retrieving SIC of total ice (i.e. concentration regardless of ice thickness), hence the above statement may not be accurate. As mentioned in the text, the L-band has problem in estimating SIC only when the ice is thin (a few centimeter) and becomes partially transparent to the L-band. I think what can be used to comment on the high value of the propagated error in Table 3 when TB is used is the fact that the variability of TB is quite high with the thickness parameter (unlike the temperature and salinity) because of the large penetration of the L-band, and may not conform to the Gaussian assumption.

Page 12 Line 1: change the word “replacedmoenthsepochs”.

Page 12 Line 1: when talking about Fig. 10, the given information is expected, nothing new. I would prefer seeing Fig. 10 generated for data in winter months. Ice in Laptev and Kara seas remains thin during winter. The region remains a marginal ice zone throughout most of the freezing season. So, I believe that we will see the same difference during November and December as we see it during the period 2-5 November shown in Fig. 10. But it is interesting to confirm that. The authors may refer to a publication about thin ice in the Arctic titled “Interannual variability of young ice in the Arctic estimated between 2002 and 2009”.

Page 13 Line 5: Same argument applies here. The text says “… for some days in November, the month of maximum extension of thin young ice”. My argument is that this statement applies to ice extending west through the Beaufort Sea“ but not in the Laptev and Kara Seas area, where thin ice cover continues to exist in winter.

Figure 6: in the caption it should be mentioned that the values for the ice are coming from multi-year ice (as mentioned in the text).

Figure 10: I believe that the difference of SIC is presented in scale of tenth concentration. Please indicate that in the caption. The advantage of this figure is not limited to what is already described in the paragraph. It also marks the area of highly dynamic and ice reproduction, which, again not limited to November.

Finally - the phrase “tie-point regions” is confusing. Better use “regions for generating tie-points”

Hide

RR by Anonymous Referee #4 (08 Mar 2017)

RR by Anonymous Referee #2 (10 Mar 2017)

Suggestions for revision or reasons for rejection

Sea ice concentration observations are a highly relevant topic. In summer, sea
ice concentration products have typically higher uncertainties compared to
winter. Extending the frequency range of passive microwave retrieval
algorithms to lower frequencies such as L-band, has the potential to lower the
overall uncertainties of sea ice concentrations. The Manuscript is well
written and combines theoretical and empirical aspects to derive sea ice
concentration from multi angular observations from SMOS. The presented
algorithm includes an estimation of uncertainties of sea ice concentration
retrievals and is compared a sophisticated operational sea ice concentration
product. The manuscript is suitable for publication in The Cryosphere after
addressing the following specific comments:

P2, L6: remove "AMSR-2 and"

P3, L27: how can it happen that the incidence angle range is not fully covered
within three days? It should be mentioned in the manuscript if also
extrapolation is done or not. "interpolate TB to locations" here is misleading as it sound like
interpolating spatially and not in TB-incidence angle space.

P4, L16-17: all "self-emitted"-->"emitted"

P5, L8-10: remove parenthesis

P5, L12: "models" -> "cases"

P5, L12: remove "larger"

P5, L18: remove "spontaneous"

P5, L11-13: Actually the water under sea ice has a contribution to the
emissivity, as you can easily calculate with your model, but you mean probably
that the emissivity is not getting higher with increasing ice thickness from
about 60cm, i.e., the signal saturates. Please rephrase the sentence. (see
also comment on P6, L25)

P5, L23: Snell's law describes the propagation direction inside a medium with
respect to its permittivity and does not describe transmissivity/emissivity.
Would be good to have either a reference (maybe Schwank et al. (2015)*) or more explanation.

P5, L29-30 (Eq. 3): Burke et al. (1979) actually does not make the step to expand
the multiple reflection to the binomial series. So the source of the Equation
could be Ulaby (1981)** Eq. 4.163, or derived from Ulaby (2014). A follow up on this equation
is also useful as most of the terms disappear or become much simpler once you
assume negligible attenuation in the snow. Also the later used water layer is
not included in this equation.

P6, L5-6: remove "conducting". For sure it is also true for a conducting medium
but you stated the alpha for low-loss-medium, means no- to low- conducting
material. Also the penetration depth is not used further in the document, so
the sentence may be removed.

P6, L7: The dry snow dielectric properties are discussed twice, here and in P5, L22-
It is probably good to put this together for consistency.

P6, L22: Equation for the four layer model is missing.

P6, L25: The surface emissivity of seawater is not a relevant quantity for the
transmissivity of radiation from seawater under the ice into the ice layer.
The boundary between seawater and sea ice has a different reflectivity compared
to the boundary between seawater and air because of the different
permittivities of ice and air. I suggest to just remove the mentioning of the
emissivity of sea water and start with "The net effect ...

P8, L20-23. The variation of a quantity with another within a multi dimensional
space depends on all other variables in case they are not dependent nor
correlated. However, Table 2 shows only a single value for each quantity.
Please give more details on how these values were obtained.

References:
*Schwank M, Mätzler C, Wiesmann A, et al. Snow Density and Ground Permittivity
Retrieved from L-Band Radiometry: A Synthetic Analysis. IEEE J Sel Top Appl
Earth Obs Remote Sens. 2015:1-14. doi:10.1109/JSTARS.2015.2422998.

**Ulaby FT, Moore RK, Fung AK. Microwave Remote Sensing: Active and Passive.
Volume 1 - Microwave Remote Sensing Fundamentals and Radiometry. Artech House Publishers; 1981.

Hide

ED: Reconsider after major revisions (23 Mar 2017) by Lars Kaleschke

AR by Carolina Gabarro on behalf of the Authors (31 May 2017) Manuscript

ED: Referee Nomination & Report Request started (03 Jun 2017) by Lars Kaleschke

RR by Anonymous Referee #4 (20 Jun 2017)

RR by Anonymous Referee #2 (25 Jun 2017)

Suggestions for revision or reasons for rejection

Sea ice concentration retrieval from satellite based radiometers is a highly
relevant topic. The expansion of currently used algorithms and radiometers to
lower frequencies to investigate in uncertainties and advantages is important
for understanding of microwave emission from sea ice. Satellite based sea ice
concentration data is also used for climate modeling and prediction where it
is highly demanded in todays changing climate. Most retrievals of sea ice
concentration show high uncertainties under melt conditions which makes
investigation and validation in summer important.

The Authors present a algorithm for retrieving sea ice concentration from the
passive microwave MIRAS onboard SMOS working at 1.4 GHz. The retrieval
employs the polarization difference at a fixed incidence angle as well as the
difference between observations of different incidence angles at vertical
polarization. The two variables are brought together using a Maximum
Likelihood Estimator to finally estimate the ice concentration.

Over the course of the review process the Manuscript improved
and is suitable for publication after minor revisions.

General comments:
There are two different emission models used, one for arbitrary number of
layers with 2nd order reflections considered (Burke et al., 1979) and the
other with infinite reflections within one layer surrounded by
infinite half spaces (Ulaby, 1981). Are both models needed to explain the
effects? I suggest to remove the Ulaby model. The difference
between the two will be very small, probably below 0.5 K as long as there is one sno
w and one ice layer.

Specific comments:

L. 147: Eq 4 is referenced before introduction

L. 152-154: just remove ", since the underlying seawater then makes no
contribution to the overall emissivity". The loss of sensitivity to a variation
in ice thickness beyond a certain ice thickness, does not mean that there is no
contribution to the emission from the underlying water. E.g, a sea ice layer
of 60cm thickness in free space would not saturate the signal at L-band
(according the models used in this paper) so that thickness variation would be
visible then in the brightness temperatures.

L. 160: the attenuation coefficient was not defined before this line.
Probably one can resolve this by just saying that the absorption and emission
within the snow is negligibly small without mentioning the attenuation
constant.

L. 163: It is true that the propagation angle inside dry snow is different
compared to that in the air. However at a certain observation angle in the
air, the direction of power propagation in the ice is independent whether
there is a snow layer in between or not, because Snell's law is linear in
refractive index (permittivity^0.5). In contrast the Fresnel equations depend
on the squared difference of refractive indices of the adjacent layers, so
that a double layer transition (with the middle layer having an intermediate
permittivity, like snow between ice and air) will increase the transmitted
radiation through that interface (except at vertical polarization for high
incidence angles). See Figure 2, where this effect is visible also at nadir
observations so Snell's law cannot serve as explanation here since the
propagation angle in ice, snow and air is 0 degree.
In short: I would remove the mentioning of incidence angle and Snell's law
for the effect of a snow layer on the emissivity of sea ice and refer only to
Eq 3.

L. 170-180: The only attenuation there is in Eq. 4 is from the
snow, which is set to zero. Consequently this sets all tau=0 and all
exp(...)=1 which simplifies the equation and makes it more readable. I
suggest to either add the simplified equation or mention this simplification
in the text.

L. 245: permit --> allows

L. 295: It is unclear whether the mentioning of uncertainty distribution of TB
refers to the Table 3 or to something which is not shown here, please clarify.

L. 471: remove "(different scattering response)" or add a reference

References:
Maaß (2013) reference should be:
Maaß, N. (2013). Remote sensing of sea ice thickness using SMOS data. PhD Thesis, Un
iversity of Hamburg, Hamburg. doi:10.17617/2.1737721.

Hide

ED: Publish subject to minor revisions (Editor review) (08 Jul 2017) by Lars Kaleschke

AR by Carolina Gabarro on behalf of the Authors (17 Jul 2017) Manuscript

ED: Publish subject to technical corrections (18 Jul 2017) by Lars Kaleschke

AR by Carolina Gabarro on behalf of the Authors (19 Jul 2017) Manuscript

Short summary

We present a new method to estimate sea ice concentration in the Arctic Ocean using different brightness temperature observations from the Soil Moisture Ocean Salinity (SMOS) satellite. The method employs a maximum-likelihood estimator. Observations at L-band frequencies such as those from SMOS (i.e. 1.4 GHz) are advantageous to remote sensing of sea ice because the atmosphere is virtually transparent at that frequency and little affected by physical temperature changes.