RID—Roof Information Dataset for Computer Vision-Based Photovoltaic Potential Assessment

Round 1

Reviewer 1 Report

The manuscript “RID – Roof Information Dataset for CV-Based Photovoltaic Potential Assessment” is difficult to digest due to unclear statement on intention.

It appears to be on one hand the description of a scheme for “deep learning for PV potential analysis” and its validation (including a discussion on validation scheme) and on the other hand a making available of a data set, that apparently other researchers may use as “troth” to train their own scheme.

It would be reasonable to present two independent manuscripts, one discussing the own annotation scheme and the proposed validation scheme, and a second on the presentation of the data sets offered to other researchers a training set with a discussion on possible applications and limitations.

 

Suggestions referring to details:

- concerning manuscript preparation  : control section numbering a reference scheme.

 

- the definition of IoU eqn.(1) has to be given much earlier in the manuscript to avoid the readers to speculate on what is discussed when talking about IoU.

 

-  the discussion in the discussion of H2-H3 it is not clear why the Hypothesis test scheme is necessary in discussion the quality of the schemes. In addition H3 is not stated clearly.  Is it that “segmentation of superstructures increases the accuracy of PV “, or that “Neglecting superstructures leads to overestimation, as existing superstructures decrease PV potential by around 20 %”, or that “Furthermore, the effect of superstructures is highly dependent on the roof segment, calling for a building specific potential analysis instead of a statistical analysis.”

The manuscript “RID – Roof Information Dataset for CV-Based Photovoltaic Potential Assessment” is difficult to digest due to unclear statement on intention.

 

It appears to be on one hand the description of a scheme for “deep learning for PV potential analysis” and its validation (including a discussion on validation scheme) and on the other hand a making available of a data set, that apparently other researchers may use as “troth” to train their own scheme.

It would be reasonable to present two independent manuscripts, one discussing the own annotation scheme and the proposed validation scheme, and a second on the presentation of the data sets offered to other researchers a training set with a discussion on possible applications and limitations.

 

Suggestions referring to details:

- concerning manuscript preparation  : control section numbering a reference scheme.

 

- the definition of IoU eqn.(1) has to be given much earlier in the manuscript to avoid the readers to speculate on what is discussed when talking about IoU.

 

-  the discussion in the discussion of H2-H3 it is not clear why the Hypothesis test scheme is necessary in discussion the quality of the schemes. In addition H3 is not stated clearly.  Is it that “segmentation of superstructures increases the accuracy of PV “, or that “Neglecting superstructures leads to overestimation, as existing superstructures decrease PV potential by around 20 %”, or that “Furthermore, the effect of superstructures is highly dependent on the roof segment, calling for a building specific potential analysis instead of a statistical analysis.”

Author Response

Thank you very much for your kind review of our article and your proposals for its improvement. We feel that by including your suggestions, we were able to increase the quality of the paper! We hope the following responses address your review comments to your satisfaction.

You will find all changes described in the responses in the adapted manuscript file. In addition, the revised version includes changes that were made in reaction to all reviewers and may thus include changes that you did not suggest.

Author Response File: Author Response.pdf

Reviewer 2 Report

The article is interesting and represents a valuable work, but the manuscript has some editorial mistakes that should be fixed.

Title:

RID – Roof Information Dataset for CV-Based Photovoltaic Potential Assessment

It is not so good applying abbreviations in the title when they are not common or multi used. consider:

Roof Information Dataset for Computer Vision based PV Potential Assessment

or

Roof Information Dataset for Computer Vision based Photovoltaic Potential Assessment

Abstract:

Consider rewriting by decreasing the “what and how” parts and focusing on the achieved results, then it can be shortened as well.

There are severe reference source error indicated in the text, might be due to the docx -> pdf conversion or missing or improper reference numbers in the original docx file.

L50, 52: The superstructures are mentioned here, consider move there the later introduction what type of elements were maintained in the manuscript. Later it is clear, but as this section introduces superstructures as a subject of the work, it would better be mentioning here as well and how they appear in figure 1. (For example, adding explanation to the figure 1 caption.)

L148: F1 score is known as Sorensen-Dice coeff. consider adding, like:

“. . . F1 Score (Sorensen-Dice coefficient).”

L202: Here appears first the chapter citation error, as the mouse moving shows the page number, but the parentheses for reference are empty:

“Chapter 0 gives an overview on the datasets in Section 0 and its key indicators in Section 0. Chapter 0 describes the annotation experiment in Section 0, the training of the CNNs in Section 0 as well as a PV potential assessment case study in Section 0. “

Check it all over in the manuscript.

L417: the figure number wrong: -> Figure 7.

L419: figure 1 -> figure 7.

L434: figure 1-> figure 7.

L543: kW / a m2 -> kW/m2a

L567: “mean specific yearly energy generation of 108 kWh,”  -> kWh/m2 (?)

L579: figure 2 -> figure 11

L597: figure 2 -> figure 11

Otherwise, good work was done, and results are remarkable.

Author Response

Thank you very much for your kind review of our article and your proposals for its improvement. We feel that by including your suggestions, we were able to increase the quality of the paper! We hope the following responses address your review comments to your satisfaction.

You will find all changes described in the responses in the adapted manuscript file. In addition, the revised version includes changes that were made in reaction to all reviewers and may thus include changes that you did not suggest.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The revision has improved the manuscript considerably.

Points to be stated more clearly:

• for the step from the PV potential as determined by by available unobstructed roof area to the numbers of kWh/m2 more information is needed on how the "superstructures" are taken into account. Is this done on the single pixel level or are "non-local" effects due to shadow casting by chimneys, ladders etc. taken into account (depending time of day).
• relate this to the state of the art of kwh-prediction
• state which assumptions on system losses are set for the PVGIS results
• state to which dataset the statement : "Furthermore, we select data from the year 2014, because this year exhibits average yearly
  solar radiation" refers to.
• The sentence "Flat roofs are assumed to be south oriented." should be reformulated.

 

Author Response

Thank you very much for your kind second review of our article and your proposals for its improvement. We feel that by including your additional suggestions, we were able to increase the quality of the paper even further. We hope the following responses address your review comments to your satisfaction.

You will find all changes described in the responses in the adapted manuscript file. In addition, the revised version may include changes that were made in reaction to all reviewers and may thus include changes that you did not suggest.

Author Response File: Author Response.pdf

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.