Reply on RC2

R2.C1 Line 120, could you provide more details in the ocean/lake effect removal? Although the SSMIS observations has been downscaled to 3.125 km resolution, however considering the bigger footprint of 36.5 GHz (4*6 km^2), can the water effect truly be excluded in the pixels near the ocean/lake? As can be seen from Figure 1, at CB for example, there are truly only a few grids that are lake free. How the influence of lake was considered?

For CB, an area with the same spatial coverage but a slightly different location was used since the snowpit area was within 25 km (resampled pixel resolution of SSMIS) from the ocean. The lakes in CB shown in Figure 1 were not considered in the soil emission contribution because most of the water was frozen (4-6) (Mironov et al., 2010), which had a similar permittivity to frozen soil (2-4) (Mavrovic et al., 2021) than liquid water. However, this simplification had importance for 19 GHz given that soil emission has a greater influence on the signal at this frequency, hence the composition of frozen water and soil derived from landcover information should be used instead. Since 37 GHz is more sensitive to snow volume scattering, this step was neglected. The 19 GHz frequency was briefly used in this study in Figure 7 (old version) only for TVC in 2018 to investigate the effect of snow variability which modifies the amount of snow scatterers inside the radiometer's footprint.
C2 Line 120, also, the snowpit measurements were at point scale whereas the Tb data is at 3.1.25 km. Why and how the Tb data was averaged to match the point scale measurements? To which resolution was it averaged?
The following was added in section 2.3.
for both TVC and CB regions. A single value of measured T_B (per frequency) were used by averaging all pixels within snow pits area (CB: 24 pixels, TVC: 14 pixels for 37 GHz). Each pixel with at least one snow pit inside was used. Since all snow pits were aggregated to obtain mean value and distribution of snow properties for SMRT, averaged T_B covering the snow pits area was used. This sentence was removed for clarity.
To my understand, the DHF was determined only by one parameter, i.e., the snow depth. The prior information is the probability distribution of snow depth and the relationship between DHF and snow depth described in Figure 5. Therefore, the generated DHF (posterior DHF field) described in Figure 6 has also some random characteristics. In other words, Figure 6 is only a realization of DHF, one of the possibilities. The scatter points are not fixed, determined values. Therefore, will a different realization influence your TB simulation results?
Your understanding is correct. Different realizations are shown by the ± 2σ region in Figure 8 and 9. It is not explained in the text but the uncertainty (2σ) is estimated by generating the same experiment of simulating T_B for the CV_sd range of 0.1 to 1 (basically Figure 7b) 20 times. The mean and std of those 20 simulations are shown by the middle line and the 2σ range of those realizations. Figure 7, it will be more interesting to provide an estimation of distribution of TB difference between 18.7 and 36.5 GHz. The authors need to explain why the TBthat considers the sub-pixel variability is higher when the standard deviation of snow depth is higher. Is it because when the snow depth is higher, the reduced variability of DHF will result in less samples of strong volume scattering, such that the TB at 36.5 GHz will increase? In addition, will this result be influenced by the soil emission background?
We decided to briefly add 19 GHz in figure 7 (old version) so the small effect (negligeable) from CV on 19 GHz simulation could be shown. See addition from comment R2.C1 about soil contribution and the addition of 19 GHz in the data section.

Simulation of T_B19 showed higher biases at horizontal polarization then vertical polarization.
To address the second questions in the comment, the following paragraph was added in the discussion (section 4) (Takala et al., 2011), using the CV_sd to account for variability of scatterers only affected simulation of 37 GHz with no effect on 19 GHz (Figure 7). If standard deviation of snow increases (more drift) then relatively fewer large scatterers from depth hoar are present within the footprint due to a low DHF in large drifts. The net result is then an increase in T_B at 37 GHz resulting from an increase in CV_sd (Figure 7).

R2.C6 How the effect of vegetation was considered in the simulation?
The effect of vegetation was not considered because it is not accounted in tundra snow retrievals (Saberi et al., 2020). Shrubs and tussock are not considered as trees or tall vegetation with significant interaction. Some studies do account for vegetation interaction with PMW but in sub-arctic areas with trees (Derksen et al., 2012;Larue et al., 2018;Roy et al., 2012). The interaction is based on vegetation product like Leaf Area Index which are not available for small vegetation like shrub. The following sentence was modified by removing Gaussian Processes to avoid confusion.
which suggested the use of a term involving variation in snow depth and microstructure within the footprint instead of a uniform snow depth.
This sentence was also modified in the next introduction paragraph when stating the objectives of the study.
Finally, we perform a Gaussian Process fit to estimate depth hoar fraction (DHF) from snow depth, using probability density functions of snow depth to add variation of snow depth and microstructure within the footprint. This sentence was modified as follows.
Secondly, we presented in situ measurements of snow microstructure and density in both main tundra snow layers (depth hoar and wind slab), mean ratios of layer thickness and the depth hoar fraction (DHF) relative to snow depth.