WeatherNet: Recognising Weather and Visual Conditions from Street-Level Images Using Deep Residual Learning

Round 1

Reviewer 1 Report

The article is very interesting and well written. It contains all the necessary elements: a broad literature review, well-described research method, interesting conclusions. I’ve got just two remarks:

Drawings should be placed in the part of the text that relates to them, while Figure 1 is at the beginning of the introduction and the reference is in line 73, Figure 2 is on page 5 (it seems that it refers to point 2.3), and the reference is in line 177 – it is misleading In „Remarks and Future Work” the authors did not mention about future work/further directions of studies (while it was mentioned in line 355)

Author Response

The article is very interesting and well written. It contains all the necessary elements: a broad literature review, well-described research method, interesting conclusions. I’ve got just two remarks:

Drawings should be placed in the part of the text that relates to them,

while Figure 1 is at the beginning of the introduction and the reference is in line 73,  Figure 2 is on page 5 (it seems that it refers to point 2.3), and the reference is in line 177 – it is misleading In

Thanks for your comment. The order of the figures is re-arranged to match with the reference in text.

„Remarks and Future Work” the authors did not mention about future work/further directions of studies (while it was mentioned in line 355)

Thanks for your comment. We have further elaborated on future work in section 6.

Author Response File: Author Response.docx

Reviewer 2 Report

I really enjoyed reading the article. It is well written, and it fits very well within the scope of the IJGI. Please find below some comments that would improve the quality of the manuscript:

Lines 44 - 47. I would recommend that this paragraph is shortened, as the discussion on the importance of weather for human behaviour is very well understood, i.e. it brings very little, if any new information to the reader. Line 74 -.... the WeatherNet framework. The comparison of the four approaches is well presented. An overview table which clarifies the advantages and disadvantages of the four distinct approaches would be useful. The quality of Figure 2 is poor. Lines 241 - 242 - The section would benefit from a clarification on the copyright of the images that are used. I suggest that authors elaborate on that aspect to be 100 % sure that no licensing condition is violated. In addition, a discussion on the workload/difficulty associated with labelling and disregarding images that do not belong to any category would be very much appreciated (lines 245 - 248). Figure 5 is missing! The labels of images on figures 7 and 8 are illegible While the results provided in the article are impressive, Section 5 would benefit from some examples where WeatherNet actually failed to correctly classify input imagery. I would appreciate seeing a discussion on the reasoning behind such failures. Another approach that might be investigated (section 5) would be to cross-validate geo-coded and timestamped images with historic meteorological data.

Author Response

I really enjoyed reading the article. It is well written, and it fits very well within the scope of the IJGI. Please find below some comments that would improve the quality of the manuscript:

Lines 44 - 47. I would recommend that this paragraph is shortened, as the discussion on the importance of weather for human behaviour is very well understood, i.e. it brings very little, if any new information to the reader.

Thanks for your comment. We paraphrased this paragraph.

Line 74 -.... the WeatherNet framework.

Thanks for your comment. “the” is added.

The comparison of the four approaches is well presented. An overview table which clarifies the advantages and disadvantages of the four distinct approaches would be useful.

Thanks for your comment. A table is added to show the advantages and disadvantages of the four approaches.

 The quality of Figure 2 is poor.  

Thanks for your comment. The figures is now with a full resolution.

 Lines 241 - 242 - The section would benefit from a clarification on the copyright of the images that are used. I suggest that authors elaborate on that aspect to be 100 % sure that no licensing condition is violated.

Thanks for your comment. We have further explained the copyrights of the image and the purpose we used them for.

 In addition, a discussion on the workload/difficulty associated with labelling and disregarding images that do not belong to any category would be very much appreciated (lines 245 - 248).

Thanks for your comment. We elaborated on the process of manual labelling and inspection of images.

Figure 5 is missing!

Thanks for your comment. There is no figure missing, but the numbering of the figure was incorrect.

The labels of images on figures 7 and 8 are illegible

Thanks for your comment. The images in figure 6 (previously, figure 7) is reduced and we increased the font size, as well as figure 7 (previously, figure 8).

While the results provided in the article are impressive, Section 5 would benefit from some examples where WeatherNet actually failed to correctly classify input imagery. I would appreciate seeing a discussion on the reasoning behind such failures.

Thanks for this comment, we have discussed how the model may misclassify images. We discussed the reasons for this and how it could be also solved for post-prediction phase.

Another approach that might be investigated (section 5) would be to cross-validate geo-coded and timestamped images with historic meteorological data.

Thanks for this comment, we have included this approach for future work in section 6.

 

Author Response File: Author Response.docx