Congratulations with the commendable updates to the manuscript, bringing it closer to publication. Both reviewers ask for some additional material documenting Aeolus impact. Please consider and respond to these further suggestions.

Minor revisions to the manuscript have been performed, increasing its scientific value . A few points for further consideration remain in my view.
1) I'd be careful with statements about analysis impact and DFS, as these are related to error covariance settings and not one to one related to the additional information content brought by Aeolus. In particular, it would be useful to more explicitly stress that a large analysis impact is not necessarily a good thing, as it may lead to overfitting and detrimental effects in the subsequent forecast. Please check.
2) In terms of data assimilation, global NWP is followed to a large extent, while particular considerations for mesoscale wind data assimilation in terms of spatial characteristics and sampling, which are very different for the Mie and Rayleigh channel, are not highlighted. For example, a recent publication from the HARMONIE community could be useful to discuss in this respect, i.e., https://doi.org/10.1002/qj.3979 and references therein.
3) In the introduction, wind speed is often highlighted, while the WMO OSCAR wind requirement is for the 2D horizontal wind component of the wind vector. This is much in line with the initialization of 3D turbulence characteristics in the atmosphere on the mesoscale, where the wind vector is relevant and its variability well expressed by its vector components, rather than wind speed and direction. It would be useful to rephrase the introduction with this requirement in mind in my view and avoid the abundant use of "wind speed". Furthermore, Aeolus measures one of the vector components and it would be useful to evaluate its impact in terms of vector components. I agree with reviewer 2 that speed and direction verification are less meaningful. If the capability is not present to track and verify vector component changes in the forecasts, it may be worthwhile to recommend this as a useful capability for the future in the manuscript.

Many thanks for these further improvements.

However, I note a new misquote on Mile et al., which also affects Mie wind assimilation. You write "In the case of observations representing courser spatial scales than the model, this scale difference has been demonstrated to be successfully handled by application of a so-called supermodding approach (Mile et al., 2021)."

Note that ASCAT winds have 25-km resolution, but Mile et al. find 60-km supermodding to work best. This is model aggregation beyond the observation resolution. Mile et al. show furthermore that supermodding changes the spatial scales that are extracted from the observations. The idea of supermodding is that those spatial scales are initialized that best improve the short-range forecasts. Initializing the small-scale noise of the model over the oceans and in the upper air is not thought to be productive. See for a more complete motivation: https://nwp-saf.eumetsat.int/site/download/documentation/scatterometer/reports/High_Resolution_Wind_Data_Assimilation_Guide_1.3.pdf , as referred to by Mile et al.

So, I'd suggest something like: "To deal with model noise and spatial representation of observations, a careful evaluation of data assimilation in terms of initializing targetted spatial scales needs further evaluation for the quiite different spatial characteristics of the Mie and Rayleigh winds. For example, model noise over the ocean has been demonstrated to be successfully handled by application of a so-called supermodding approach (Mile et al., 2021)."

Congratulations!