This is a well-written paper that investigates an important issue in ground-based remote sensing: what improvements arise when additional datastreams are included in the retrieval? For years now, TROPoe and its antecedent AERIoe have included RAP or RUC profiles in the calculations, but the improvement in the retrieval has not been fully quantified. This paper is of the appropriate scope and novelty for inclusion in AMT. I have a few corrections and suggestions that will help improve the readability and utility of this paper, but these should be easy to address. None of these issues rise beyond the level of minor corrections.

Most significant issue:

Minor comments:

1. One small issue I had while reading this paper was understanding how the RASS was integrated into the retrieval. The TROPoe retrieval works in T and q space, and it's relatively easy to understand how the RAP profiles would be able to be included as they exist (or can easily be converted) into those variables. However, RASS is measuring Tv which is a unique variable for the retrieval. How does the retrieval address this? Is it just calculated at the end of each iteration from the interim T/q profile and compared to the RASS observations?

2. When I read line 76, noting that the radiosondes were interpolated to the TROPoe grid, I wondered why the sondes weren't smoothed with the retrieval averaging kernal instead. I see that was addressed later in Line 195.  Still, I wonder if this is the best approach. Would it be better to smooth the sondes, calculate the differences as a function of height, and then interpolate the vertical profile of the differences to a common grid to facilitate the analysis between different combinations of instruments? 

3.  In certain instances, the additional observation vector entries degrade the profiles, either by increased bias or MAE. This is an interesting finding that ought to be discussed more. What is causing this, how consistent is it (are a couple of retrievals dragging everything down, or are most of the retrievals behaving similarly)?

Technical corrections:

Line 42: this sentence is somewhat awkwardly phrased and would be easier to interpret if rewritten.

Line 43:  I don't think it's quite correct to state that in situ sensors only provide point measurements. Surface meteorology stations like ASOS only provide observations at a single point fixed in 3D space, that is true. However, aircraft-based observations like AMDAR are in situ but producing moderately-dense vertical profiles and the sheer density of these observations  next to major airports produces a multidimensional web of observations. 

Line 54 and elsewhere: the name of the radiosonde is the Vaisala RS-41, not Vaisala-41.

Line 96: it is somewhat unclear what the upper range of the IRS is: why does it range from 3000-5000 cm^-1? Did the two IRSes each have a different spectral range?