Review Reports

Round 1

Reviewer 1 Report

Overall, an interesting paper to tricky problem.  However, significant edits to the manuscript should be considered before publication.

Major Comments:

1) Clarification of some of the methods are required in parts.  Specifically:

a) The readers need more information about LENS.  Because it is the source of the data and the baseline comparison, understanding the spatial resolution and how visibility is parameterized is important to understand how it is modeling and estimating visibility.

b) Some aspects of how the WMA was fit are a little unclear.  Where the weights calculated once on all available data?  Or were they calculated for each weather station individually?

c) With statistical models, validation schemes are very important to judge how generalizable the results will be.  Are the presented results from evaluating out-of-sample predictions?  Or are the authors evaluating examples of predicting visibility the WMA has been fitted to? 

d) There is a lot of methodological description in the Results section.  Consider creating a subsection in the Methods section just for describing the various analytical methods. This should simplify and focus the presented results. 

2) The Results section should be restructured for clarity.  There are a great many results, and using subsections to organize and structure the information would be very helpful to readers.  Also, not all presented results necessarily show information critical for the readers to appreciate the improvement that the WMA presents over the ensemble predictions.  For example, Figure 3 show reliability diagrams for the ensemble predictions, but no similar plots for WMA are presented.  Focusing the Results on differences between ensemble and WMA predictions should lead to a more impactful presentation.

Minor Comments:

1) Some editing for consistency may be required.  In Table 1, precipitation is listed as a predictor, but this is contradicted in the Conclusions.

2) Although the cited papers that use neural networks to predict visibility don't appear to be doing this, it should be noted that there are parametric methods to adapt neural networks to produce PDFs.

3) It might be nice if the 'Verification Rank' was mapped to values of Visibility in the Rank histograms.  The current numbering scheme does not convey much information.

4) The described weighting scheme is constrained such that all the weights add up to 1.  You might get better bias adjustments if that constraint was loosened.  

 

Author Response

Please find the attached file.

Author Response File: Author Response.docx

Reviewer 2 Report

Please see the attached file.

Comments for author File: Comments.docx

Author Response

Please find attached file.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Thanks for the responses and revisions to the manuscript.  This current draft is significantly improved over the first draft, and in my opinion only a few additional  adjustments should be made before publication.

Specific Comments:

1) The testing regime described in your response to my previous comments sounds great.  But I wasn't able to find a similar description in the paper.  It would be good to clarify the testing regime used in the manuscript, and specify to readers that presented results are from the testing set.

2) Similarly it would be good to mention that WMA was applied to each station individually somewhere in the Methods section.  

3) On line 297, just "Quantitative precipitation..." is fine.  It's not countable, so "A" is not needed.

4) In Figure 3, editing the caption to read some thing like: 'An example WMA...'  could clarify that this is just one PDF generated from this work.

  

Author Response

Please find the attached file.

Author Response File: Author Response.docx