Response to Referee Comment #3 on Correction of wind bias for the lidar on-board Aeolus using telescope temperatures

1. abstract, line 28: "the approach of using ECMWF model-equivalent winds is justified by the fact that the global bias of models u-component winds w.r.t to radiosondes is smaller than 0.3 m/s", This statement may not be representative here since the majority of globe is not covered by radiosondes. Actually, over the large part of remote oceans and lands, NWP models still have large (on the order of several m/s) uncertainty including biases, e.g., in the upper troposphere and lower stratosphere of the Tropics. This comment also applies to line 236-239. However, the regression of O-B to M1 temperatures globally and from all vertical layers makes the M1 correction less sensitive to the considerable latitudinal and vertical layer varying biases or uncertainty between NWP models. It might be helpful to make this point clearer in the paper.

1. abstract, line 28: "the approach of using ECMWF model-equivalent winds is justified by the fact that the global bias of models u-component winds w.r.t to radiosondes is smaller than 0.3 m/s", This statement may not be representative here since the majority of globe is not covered by radiosondes. Actually, over the large part of remote oceans and lands, NWP models still have large (on the order of several m/s) uncertainty including biases, e.g., in the upper troposphere and lower stratosphere of the Tropics. This comment also applies to line 236-239. However, the regression of O-B to M1 temperatures globally and from all vertical layers makes the M1 correction less sensitive to the considerable latitudinal and vertical layer varying biases or uncertainty between NWP models. It might be helpful to make this point clearer in the paper.
It is correct that the statement about the low model bias is difficult to justify in regions with low radiosonde and pilots density. This point is addressed in lines 237 to 244. It is confirmed, for the M1 bias correction altitude varying model bias should not be an issue, because all O-B values of a profile are averaged before the fitting. Moreover, global model averages obtained from 24 hours of observations are used which should mitigate the effect of localized model errors.
The following information was added to Section 2.4 of the manuscript: 2. line 241: Potential wind background uncertainty may be explored by comparing winds from major NWP models, e.g., ECMWF vs. NOAA/GFS. In remote regions, current NWP models still have large uncertainty.
Yes, it is correct that in remote regions, especially in the tropics, model winds can be largely biased. Figure 1 below, for example, indicates the difference of the wind vector between the ECMWF and Met Office mean analysis over a 7-day period in 2015 at 100 hPa. Such analysis indeed helps to identify problem regions. As mentioned above, we try to mitigate the influence of localized model errors by using 24 h of vertically averaged global model winds. It is the preferred solution to use modelindependent ground return winds to avoid model dependency. But results showed that the performance of this approach is not yet stable enough for the operational processing and analysis will continue. For the comparison of different models, the Aeolus CalVal includes the different Met centers that can use in their analysis their own meteorological input data to further quantify the impact of the differences. This is an ongoing effort in a bigger framework. 3. Figure 14 (bottom) is very interesting. I guess this is the O-B average over the entire vertical layer range (0-24km?). It will be helpful to provide this information in the figure caption. Also, it would be greater if the magnitudes and details of the remaining biases could be better visualized, e.g., some kind scatter plots with density distributions (vs. latitude and/or longitude).
Yes, the figure shows vertically averaged E(O-B) values at the L1B observation granularity. The determination of such is also explained in Section 2.4 of the manuscript. For the sake of clarity, further information was also added to the caption of Figure 14. To better highlight the remaining bias, Figure 2 further below shows the residual bias as a function of the argument of latitude. The plot is based on the same data period as shown in Figure 14 of the manuscript. The plot reveals that the binned average (solid red line) is close to zero over the major part of the orbit. However, for the region with particular strong M1 temperature influence, i.e. at 230° and 330° argument of latitude, remaining bias with a binned average of up to 1 m/s is visible. Despite the M1 bias correction being highly effective, the currently used regression approach still can be improved. However, testing more sophisticated regression models, such as random forests (Svetnik et al., 2003) or generalized additive models (Hastie and Tibshirani, 2014), is beyond the scope of this paper and could be considered for our future work. Thus, the following information was added to the summary of the manuscript: 4. It might be helpful to explicitly mention in the abstract and conclusion that the M1 correction has little impact on Mie winds.
It was decided to add this information to the summary: