General Comments

The manuscript presents a valuable analysis of sea ice emissivity measurements, including novel observations at submillimeter wavelengths, which will be of considerable interest to the remote sensing community, given the forthcoming AWS, ICI and Sterna satellite missions. Importantly it relates the variability on the different scales observed by airborne and satellite sensors. The authors could consider further analysis to include a quantification of the scene variability on other scales - e.g. through the use of variograms/structure functions. This could extend the application of the results to other applications.

It is generally well-written and the conclusions, in particular, are clear. However, there are several cases where key details of the methodology are missing, which would make it very difficult to reproduce the results. Examples are given below. Furthermore, more attention needs to be paid to uncertainties - especially those introduced by various assumptions (see below).

Specific Comments

106: Please provide a reference to a document describing the calibration and bias correction procedures.

116: Does this uncertainty combine systematic and random effects? It is important that they can be treated separately in evaluating the uncertainty of the emissivity estimates on different spatial scales.

153: What uncertainty is added by this assumption? Could it introduce significant biases? E.g. during strong surface inversions - in this case, would it be better to assume a linear change from the flight level to the surface?

158: What infrared emissivity is assumed to open water? Is this a function of sea state?

163: Is this bias compensated for? How is it accounted for in the uncertainty analysis?

165: It would be helpful to include a short summary of how the NE23 analysis is derived. It should also be clarified whether the NE23 analysis is used instead of the KT-19 measurements for emissivity calculations airborne as well as satellite measurements, and, if so, why, and how this contributes to the overall uncertainty - especially in the context of surface temperature gradients through sea ice and any overlying snow layers.

169: Is there any potential here for confirmation bias? I.e. if the multiyear ice concentration maps are derived based on satellite observations, and assumed emissivity spectra?

177: What variability is typically observed within the ±2h window?

205: It should be clarified whether this transmissivity refers to the layer between the aircraft and the surface. And what about atmospheric emission?

221: It would still be interesting to compare the results for all 183GHz channels - albeit with increased uncertainties. 

225: This is an underestimate in the case of strong surface inversions, which are common over sea ice.

232: How could the uncertainties be estimated for satellite observations?

246: It is interesting to note that the results from the specular assumption appear more Gaussian. Why could that be?

Figure 3: I found this figure confusing - it took several readings before I understood it. It could also be expanded to a full page width.

Table 3: How exactly is the mean relative uncertainty calculated?

Figure 5: I would also be interested to see the emissivity plotted as spectra for each cluster.

377: What are the implications of this assumption? Noting the results of Wang et al., 2017b differ from Harlow (2007): and Haggerty and Curry (2001), who found an increase in emissivity with frequency for sea ice between 150 and 220 GHz.

Figure 6b: This is a very useful result. But would it be better to divide by reflectivity (1-emissivity)? This might normalise the distribution.

396: How exactly are the observations averaged to ensure equal spatial sampling?

Figure 7: How exactly is the MiRAC emissivity resampled to satellite footprints? 

Figure 7: It would also be useful to plot the IQR for the emissivity derived from MiRAC and satellites independently. This figure could also be expanded to full page width.

401: It is not clear how the Lambertian assumption introduces a bias less than 2% for MiRAC.

Table 4: Is the relative bias here calculated from the mean difference? It does not seem to match the difference of the median values.

Figure 8: This is potentially a very useful figure, but is confusing to interpret - especially the labelling on the x-axis should be improved. It may also help to more clearly distinguish V & H polarisation.It would be better supplemented by a table of values. 

• Why are multiple values shown for 243GHz (2) 340GHz (3)?
• Any idea why the 89V results for AMSR2 are out-of-family?
• AMSR2 and SSMIS are conical scanners, with V and H polarisations, not QV as shown.
• Why are no ATMS results are given for ACLOUD?

465: How much difference is expected from nadir to 25°?

I would not expect much, following Hewison and English (1999): Airborne Retrievals of Snow and Ice Surface Emissivity at Millimetre Wavelengths. IEEE Trans. Geosci.Remote Sensing, Vol.37, No.4, 1999, pp.1871-1879, doi:10.1109/36.774700

Technical Corrections

263: Typo: Word -> World

473: I suggest “resolve” instead of “capture”, noting it will still affect the mean emissivity.

This manuscript addresses the uncertainty in sea ice microwave emissivity representation for numerical weather prediction applications. Quantification of the sea ice contribution to satellite signals is crucial in order to separate surface and atmospheric contributions to satellite signals. This paper identifies sea ice type from microwave emissivity spectra via K-means clustering, demonstrates appropriateness of Lambertian scattering assumptions and investigates scaling issues by resampling airborne observations to satellite resolution and comparing with satellite data, considering resolution, incidence angle, polarisation as well as frequency. This manuscript is well-written and robust with justified assumptions and demonstrates that representative emissivity based on sea ice type is a reasonable approach and consequently that the spatial variability in sea ice properties must be accounted for. This manuscript is suitable for publication with minor amendments, and the following points considered in discussion:

Line 39-41. Please expand on the Hewison study to discuss what was found and how it relates to these results. This is already included around line 280, but what is needed here is to highlight the new frequencies in this approach, particularly given that the higher frequencies are more sensitive to surface type.

Line 79-80 Just to link with the previous section state that the Polar 5 carried the MiRAC and KT-19 instruments (see comment for line 156).

Line 81. ‘1). Various sea ice characteristics were observed… ’: specify this is from the airborne observations as no in situ measurements were made.

Line 84. Is ACLOUD firstyear, multiyear or a mix or ice types? The description for AFLUX was very helpful – please include a comparable description for ACLOUD.

Line 88. How was the integrated water vapour measured? Add a link to (presumably) section 2.4.

Figure 1. Please use a different colour scale to distinguish between RF23 and RF25 and between RF14 and RF15. Perhaps use different line thicknesses or line type.

Table 1. Add ‘Passive’ into the table caption and consider including the KT19 sensor characteristics.

Line 145. It would be useful to remind the reader here that MiRAC 89GHz is only available at 25 deg.

Line 156. This is the first mention of the KT-19 sensor (apart from line 119) – presumably also on the Polar 5, but please clarify.

Line 178. What is the estimated drift rate and how was this determined?

Line 184. Consider moving ‘during ACLOUD (AFLUX)’ to after ‘overflights’ so the meaning is better conveyed before the brackets are used. Could the information in this section be better displayed as a table?

Line 255. Are the numbers in brackets for ACLOUD or AFLUX? In general it’s better to write this out in full for ease of reading.

Line 262. ‘We observe predominantly snow-covered sea ice over the transect’s initial 7 km’ – is this from right to left as per Westerly flight, or left to right as per numbering in Fig 3?

Line 263. Typo: ‘Word’ -> ‘World’

Line 290. ‘The ±8 K surface temperature uncertainty causes the highest emissivity uncertainty for all channels.’ Where is this demonstrated?

Line 304. ‘and we found no significant changes in the shapes of the histograms (not shown)’. What statistical test was used?

Figure 3 – please put this through a colour blind checker, particularly fig 3j, where it’s hard to distinguish between 183 +/- 2.5 and 3.5 GHz bands.

Figure 4 – please use a different colour scheme to distinguish between the two ACLOUD flights.

Line 351. What test of significance was performed?

Line 368. ‘Hence, the satellite footprint contains mean conditions where significant small-scale variability averages out.’ I am unsure what is meant by this and how it relates to the previous sentences  – please could you clarify?

Line 382. ‘The limited spatial coverage of MiRAC causes slightly higher emissivity variability compared to MHS and ATMS, as MiRAC only captures a narrow strip of the satellite footprint’. Why this rather than simply the higher resolution of MiRAC?

Figure 6. Does the cluster colour scheme relate to the emissivity colour palette?

Line 396. As the satellites have different footprints  ‘equivalent spatial sampling’ may be better than ‘equal spatial sampling’

Figure 7(m). What is in the wider satellite footprint that is causing the higher emissivity in the western tip?

Line 416. ‘Additionally, AMSR2 shows higher variability due to its smaller footprint than SSMIS’. This conflicts with ACLOUD IQR being smaller at Vpol for AMSR2 than SSMIS in Fig 8a.

Line 469. ‘Surface temperature assumption: Using the surface skin temperature instead of the emitting layer temperature imposes a frequency-dependent bias on the emissivity during AFLUX’. How much does this assumption influence the conclusion that the emissivity spectra are relatively flat?