Summary

This paper details a new aerosol retrieval product from the GOMOS instrument. This new retrieval is available only as monthly zonal means because it is performed by averaging the Level 1 transmittances from the stellar occultations and then looking at the residual spectra to solve for aerosol using a spectral smoothing/fitting function. This methodology has its pros (less sensitivity to noise and scintillation) and cons (increased sensitivity to interference from clouds that cannot be filtered later). Comparisons with other instruments seem to suggest an overall improvement in the spectral dependence of the resulting aerosol extinction coefficient profiles from the current production version of GOMOS aerosol data. This paper is well-written and presented and suitable for publication in AMT. My comments below are mostly minor, though I believe there is a source of additional bias being introduced from this new methodology that should be addressed somewhere.

Comments:

I find it strange that one of the wavelengths from the residual spectra that is used for aerosol is one in which the instrument does not even measure (i.e., 750 nm). I understand the motivation is to match wavelengths measured by other instruments for validation purposes, but this wavelength is now essentially an extrapolation of the smoothing that is performed on the residual spectra. Additionally, why is it that the spectra shown in Fig. 2 do not cover the entire range of the IR spectrometer (thus further reducing the reliability of the smoothing out in this spectral region)?

Why not show the comparative Angstrom exponent in Figure 5 as is done for Figure 6?

Pg 10, Ln 210: “The reason for positive bias near the tropical tropopause is GOMOS aerosol retrievals is not fully understood at the moment.”

            How is it not just clouds? If you are averaging all of the transmission profiles without filtering for clouds, then of course clouds are going to bias your aerosol retrievals near the tropopause. Perhaps you cannot easily quantify how much of the bias is from clouds versus any other potential source, but the expected presence of clouds will obviously create a bias like what is shown in Fig. 7. I think the better question is why does the bias appear significantly larger than the averaged SAGE II profiles in Fig. 4?

Pg 10, Ln 213: “We tried to apply various methods for cloud filtering in averaging GOMOS transmittances– according to absolute values of extinction and ratio at different wavelengths.”

            This is no trivial task, and it may not be possible for all but the thickest clouds.

As a curiosity, I wondered what the impact of using averaged transmittances would be for your comparisons. The authors compute an average transmittance profile, then convert it to an average optical depth profile, then perform the retrieval. The comparison profiles from other instruments are averages of individual profiles. As a simple test, the following relationship is true: MEAN(-LOG(T_i)) > -LOG(MEAN(T_i)). In other words, averaging the transmittance profiles first will always result in a slight low bias to your optical depths when compared with the mean of the optical depths derived from individual profiles. I am unsure of how this propagates through the two inversion steps. I would imagine the spectral inversion is less affected by this, allowing the bias to mostly propagate into the residuals. I cannot intuit how this would propagate through the vertical inversion step. If the bias still propagates proportionally (as opposed to inversely in some fashion) into the resulting extinctions, it would mean the positive bias you see in your comparisons is actually smaller than the true comparisons because a small amount of negative bias should be introduced from averaging transmissions first.