See attached review report

Possibly for the first time, this paper demonstrates good agreement between snow radiative transfer simulations (driven by snow pit measurements) and downlooking microwave observations at higher frequencies, i.e. 89 GHz to 243 GHz. This illustrates a path towards using snow radiative transfer, driven by multi-layer snow models, in the assimilation of microwave observations for both weather forecasting and for the inference of surface snow properties. The coupling of ARTS and SMRT models (and the demonstration of why this is important) is also an important step. The paper also illustrates some of the remaining difficulties to be solved. One of these is the occasionally large mismatches between point snow brightness temperature (TB) measurements and airborne TB measurements with fields of view up to 100m. Another is the drop in TB of up to around 20 K observed from one flight to the next, which was attributed to fresh snow on the surface, and illustrates strong temporal variability.

Overall, this paper is an important step forward, it will be of great interest to many scientists in the field, and it is well presented. However, there are a few areas where the methodology or results could be better explained, there are some possible uncertainties that might deserve more consideration, and it is important that the abstract and conclusions should clearly indicate the scope and limitations of the work. 

Main points

1) As described in the paper’s abstract, coupling ARTS and SMRT is a major developmental step. However, for such an important part of the paper there is very little detail. For example it is not clear how the downwelling atmospheric radiation field is represented by ARTS and then coupled into SMRT (presumably as radiances at the quadrature angles of the discrete ordinates solver used in SMRT, but this is not stated). Assuming ARTS and SMRT are not “fully” coupled, by which I mean a discrete ordinates problem is solved simultaneously in the snow and atmosphere, I imagine that ARTS is called first to simulate the downwelling radiance field, the upwelling radiance at the observation angle, and the surface-to-aircraft transmittance. Then presumably SMRT is called and its output corrected to aircraft level with the paper’s equation 2. These issues should be clearly discussed in the paper in section 2.4. It would also be good to have details of the ARTS radiative transfer solver method, mainly just to exclude the unlikely scenario that atmospheric scattering is being represented too (which could need “full” coupling of the solvers).

2) Some of the descriptions of how Arctic microwave observations are used at NWP centres (with ECMWF as the main example) could be made more precise. Microwave sounding radiances are used over snow and sea-ice surfaces if the surface contribution is small enough. For example the 183+/-3 GHz channels are usually assimilated over sea-ice and snow, whereas 183+/-7 GHz channels are not. Also, one of the main problems with ice and snow surfaces, from an NWP point of view, is that a constant surface emissivity cannot be assumed. Over non-snow land surfaces, the dynamic emissivity retrieval technique typically assumes that an emissivity retrieval can be extrapolated using a constant in frequency approximation (for example an 89 GHz retrieval is used as the surface emissivity for 183 GHz assimilation over non-snow surfaces). A few more detailed points illustrating these issues:

line 28-29: “data over Arctic regions” could more precisely be “surface-sensitive data over Arctic regions” and the reason for the data exclusion is usually the possible presence of snow and ice.

line 30: “potential benefits of .. microwave data over Arctic regions” - but some Arctic microwave data is already being assimilated operationally, particularly in summer, as illustrated in the Lawrence et al. (2019) studies, and as is described clearly on lines 35-38.

line 46: Baordo and Geer (2016) describe the assimilation of only snow-free land surface data, for the SSMIS instrument, and they eliminated surface-sensitive observations at latitudes greater than 60 degrees or for surface temperatures less than 278 K. Hence the point about using atlas in these possible-snow areas is not so relevant. Within Geer et al. (2014) there is a description of subsequent work that extended SSMIS usage over snow and sea ice surfaces following the above-described template. This actively assimilates 183+/-3 GHz and higher peaking channels. The dynamic emissivity retrieval is made at 150 GHz and then assumed to be valid also at 183 GHz, making sure the extrapolation in frequency is as small as possible (but even this small extrapolation induces errors that are too large to permit assimilation of channels that have stronger surface sensitivity, like 183+/-7 GHz). This snow and ice dynamical emissivity retrieval approach started at ECMWF even earlier with clear-sky MHS assimilation following the work of Di Tomaso et al. (2013: Assimilation of ATOVS radiances at ECMWF: third year EUMETSAT fellowship report. EUMETSAT/ECMWF Fellowship Programme Research Report No. 29, available from http://www.ecmwf.int.)

line 48: “microwave emissivity is highly spatially variable” - this could be a place to mention that it is also highly variable in frequency.

Just a discussion point, but these dynamic surface approaches are continuing to be improved for NWP, and in particular we are starting to improve representations of the frequency dependence of surface emissivity. Compared to the more physical approach of the paper under review, the dynamic approach has the advantage of being able to adapt the surface to match what is in the sensor’s the field of view, thus dealing with the time and space heterogeneity issues that are well illustrated in the paper, and hence they may continue to provide strong competition for the fully physical approach for some time to come.

3) It would have been good to discuss the surface characteristics of the Trail Valley Creek site and how they relate to possible uncertainties in the surface radiative transfer. In particular, vegetation, since it appears the surface is being modelled as bare soil. The satellite pictures seem to show trees in the valleys and the possibility of grass or shrubs on the plateaus. Could vegetation have impact on the radiative transfer, particularly if it contains some liquid water, and particularly as the snow cover is not deep, e.g. 20 - 100 cm (lines 109-110)?

4) There could be some more investigation of the way the temperature profile is determined, and whether this has any bearing on the radiative transfer uncertainties. Lines 168-170 describe a linear extrapolation from the air temperature (ultimately from dropsondes?) through the snowpack to a stable lower layer temperature. Is this sufficient to represent the relative complex dependence of the snow temperature profile on the surface air temperature, particularly its insulation properties and speed of heat transfer? For example, when trying to explain the drop in brightness temperature between flights C087 and C090, could this be relevant? Looking at Figure 9, at the time of the C090 flight, could the snow still be cold after a night that dropped below -25 degrees C, and hence has not yet caught up with the rapid rise in the air temperature?

5) It could be worth specifying also the type of seasonal snow in the abstract and conclusions. Currently on line 406 the work is described as relating to “an Arctic tundra snow environment” but that could more precisely be “an Arctic tundra snow environment in late winter”. In order to use satellite observations for weather forecasting globally and in all seasons over snow and sea-ice, we will need to be able to simulate many other snow types such as wet snow and including diurnal cycles of freeze and thaw.

Minor points

line 40 - 19, 37 and 89 GHz channels are extensively used for water vapour, cloud and precipitation assimilation, but the statement that “window frequencies around 19, 37 and 89 GHz are used to obtain information about the surface (e.g. snow)” could be misread to exclude this and to imply that these frequencies are not useful for the atmosphere.

line 44-45 - “forecast and analysis”? rather than “forecast analysis” which is confusing.

line 115 - if the sled measurements are nadir to a snow surface that may be sloping, is any adjustment made when sled measurements are mapped to true nadir aircraft measurements?

line 196 - “representing the layer density and SSA by the largest and smallest observed values” - it’s not clear whether this means within a single pit, or across all pits.

line 197 - the “full range of plateau airborne observations” deserves some explanation, as at this stage it’s really not clear that (presumably) this means across the two flights and incorporating all plateau measurements in the relevant area illustrated in figure 6 and following the comparison strategy described in lines 262-265. It might be worth re-ordering some of this information (e.g. to put it in the section on aircraft data?)

line 221-227 - in the adjustment of the background atmospheric profile to fit aircraft-measured downwelling radiances, can the dropsonde profile below the aircraft be modified to fit observations? It’s not clearly excluded in the text. And how representative is the lowermost dropsonde air temperature of the snow temperature? (See main point 4)

Figure 4 caption (figure 5 similarly) - the significance of the square could be explained in words in the caption (the lines indicating that it is a zoom are faint and easy to miss), The caption should also explain the meaning of the error bars

line 241-242 - linked to Figure 4 and the need to state clearly what the error bars mean, it’s not clear how the “range of simulations” mentioned here is being generated.

Figure 7 and 8 captions - need a careful description of the meaning of the various boxes, whiskers and spots.

Line 383 - this RMSE calculation is a headline result from the paper, quoted in the abstract, so it should be clear how it is obtained. For me the description “RMSE of the base simulation medians by frequency and flight” is not quite clear enough. For example whether this really is the RMS of the SMRT base simulation median minus the observation median and, if I understand correctly, that means the sample over which the RMS is computed is of size ten, e.g. “across 5 frequencies and 2 flights”? Could this be more precisely described in the abstract too, noting specifically the use of medians in the calculation? Because if it is based on medians, then we might expect the RMSE comparing the errors of individual pits to individual surface categories and AOIs to be somewhat higher. That would also be a useful figure to calculate.

Line 396-397 - main point 2 again: “In current numerical weather prediction models, microwave emissivity is assumed to be constant over snow-covered surfaces or derived from a monthly climatology, with errors too large to be able to use satellite observations in the Arctic”: Dynamic emissivity retrievals have been used over snow and sea-ice at ECMWF to allow assimilation of 183+/-3 GHz channels (and higher peaking channels) since the work described in Di Tomaso et al. (2013) and Geer et al. (2014). Hence the snow emissivity does not usually come from atlas and it is not assumed constant in time or space (but it is assumed constant with frequency from 150 to 183 GHz). Nonetheless, the dynamic emissivity retrievals are not yet good enough to permit assimilation of strongly surface sensitive channels (e.g. 183+/-7 GHz) over snow so there still is plenty that can be improved by physical modelling as described in the paper under review.

Line 422-423 - “the addition of fresh, low density precipitation and a later wind event that removed it over the space of a few days caused differences in observed brightness temperatures.” - is the attribution of these changes in TB to the fresh snow event certain enough to be able to say for definite it “caused” it here in the conclusion, rather than to say “likely caused”, for example?