Dear Authors,

thanks for posting your authors comments and providing an updated version of the manuscript. You mostly addressed in your comments the issues raised by the 2 reviewers, but I would like to have those answers more included in the manuscript itself. After all, most readers will only read the manuscript.

This certainly applies to the comments raised by both reviewers about the applicability of the method for homogeneity checks of the datasets. Based on the log-files of the different IGS stations, some obvious breaks can be detected in the time series and have already been reported for some stations. Therefore, it would be nice if you could give some examples (not the full statistical linking between your representativeness statistic and the results from a homogenization method you are referring to) for some stations about a possible link between the values of the representativeness statistics and a break point in the GPS time series. Moreover, in this context, it can also be very instructive to show the time series (probably best the monthly mean time series, although most of the representativeness statistics are calculated from daily values) of both the GPS and ERA-Interim for some example outlying sites (section 4). This can make the discussion of the characteristics of the outlying site differences between GPS and ERA-Interim less heavy to read and might give you the opportunity to illustrate why your method might be an interesting tool to identify the stations for which the GPS-ERA-Interim IWV differences cannot be used for breakpoint identification.

Also the point raised by the second reviewer about ”the current knowledge of the spatial representativeness error of the GPS derived PWV and how this study tackles this issue” has not been treated in the manuscript. This is however a very important point. Also the first reviewer has some difficulties with interpreting the definition of representativeness error (P2,L32-34: Are representativeness errors never to be referred to model errors? I interpret the definition of a representativeness error as that the only cause is the limited model resolution? If this is correct it can be stated explicitly, because I can also argue that the limited resolution of a model can be the cause of “model errors”) and your answer has not been implemented in the manuscript itself. In this context, I miss a reference in your manuscript to earlier papers investigating the concept of representativeness error (the second reviewer already provided a reference to the work by Liou et al. 2001). One such a publication that I missed in particular is the ACP paper by Buehler et al. (2012). In their section 3.5, they give a nice description of the concept of representativeness error, in general terms, and distinguish between two effects: (i) the measurements may not be perfectly collocated in space and time, and (ii), they may have different sampling characteristics – one may be an instantaneous point measurement and the other an average over some distance in space and time. To my opinion, you should be more specific about which representativeness you are considering in the manuscript, which not (or are cancelled out by exact temporal matching e.g.) and how this related to earlier concepts of representativeness. As already mentioned, I also missed references to past studies explaining for which area (cone) a GPS IWV measurement is representative. With this more specific definition of representativeness error, you can also be more specific about the motivation and applications of your study, another remark made by the two reviewers (“applications are not very appealing” and “the stated motivation for the work was to identify GPS stations where ERA Interim is not recommended to be used when searching for inhomogeneities in the GPS time series of IWV”).

Other examples where you should include your answer to a reviewer comment in the manuscript are:

“RC1: Related to this it is clearly stated that the method is subjective. Nevertheless, if the method is to be applied by others, it would be informative to also document the reasoning behind the choices. For example, why did you choose non-symmetric limits for the mean differences in Figures 2 and 3?”

“ RC1: P4,L9-11: I think you should mention that the GPS time series used have passed some kind of quality check, because a very large break should have an impact on the overall standard deviation of the differences GPS – ERA Interim”: be more specific on which data screening has been applied to detect e.g. very large breaks.

“RC1: P6,L5: It cannot be taken for granted that the discrepancy is not due to GPS errors just because the formal errors do not increase. For example, a nearby installation of say a metallic structure may introduce significant multipath errors without affecting the formal errors.”

“RC1: P7,L2-5: Can you compare this standard deviation of 0.81 kg/m2 to what is obtained for stations located in the same area of the present study, in order to quantify the improvement obtained for the higher resolution model?”

“RC1: P7,L13-15: Perhaps the GPS sites that do not show an improvement using bi-linear interpolation are located close to one of the four grid points that is more representative compared to the others?”

Some other specific comments

P6,L13-15: to which representativeness “differences” in the reanalysis humidity field are you referring to here? Please be more specific.

P6,L23-24: ”a mistake which has been made in several past studies and led to the erroneous statement that IWV differences increase towards the equator due to the increasing mean IWV”: without a proper referencing to those several past studies in which this statement is literally present, this is a very suggestive statement.

P7,L14-16: ”The slight shift of the ensemble of results below the 1:1 line is reflecting the fact that a majority of sites exhibit small positive bias (0.47 kg m-2 on average) as already noticed in Figs. 2a, c, 15 which is not due to representativeness differences”: give the reason for the small positive bias once again here.

Dear authors,

thank you for taking most of my considerations into account. Although I personally am more in favor of the conceptual description of the error terms between two datasets by Buehler et al. (and their systematics in describing one by one the different contributions to the representativeness error term) and find your distinction between the concept of representativeness error in the context of data assimilation/reanalysis and between two true observational datasets a bit artificial (Buehler et al. also mixed up the differences between GPS and observational dataset with differences between GPS and ERA-Interim), I think that the description of the concept of representativeness error/differences used in the paper should be clear enough for most readers now.

With kind regards,

Roeland Van Malderen