Revealing a Shift in Solar Photovoltaic Planning Sites in Vietnam from 2019 to 2022
Round 1
Reviewer 1 Report (Previous Reviewer 3)
This is a re-submission of the original work where the authors have reworked considerably the article. This new submission now studies the PV installations in the complete country of Vietnam over a period of 4 years (2019-2022). I find now the study much more robust and the conclussions better supported by the analysis done by the authors.
In general I consider the article is in a good position for publication, but I have a few comments that I think could be addressed on an updated version:
- One limitation as the authors note is the need to use OSM to obtain the Powerline information. If available, it shall be mentioned how old is the OSM information because lines could have been built or decommissioning since then (the authors even mention the case of new lines not available in OSM)
- S2 bands used are 'B2', 'B3', 'B4', 'B6', 'B8', 'B11', 'B12'. If I understand correctly the analysis is done using 10m x 10 m image data, but bands B6, B11 and B12 have a resolution of 20 m. How is this taken into account?
- It is not clear to me from this study what is the minimum PV area detectable by the method presented. Is the minimum detection 10 m x 10 m? or a cluster of pixels is needed due to the post-processing? I think this is relevant in case it could be reducing the efficiency for PV detection in 2022 when the installations are clearly smaller.
- Solar PV detection is performed with a resolution of 10 m based on S2 data, but in section 3.2 the LULC changes are evaluated assuming that each pixel is 30 x 30 m. I think this needs to be better explained
- Figure 5 and 10 have very low resolution.
- In Table 1 it is clear that Transmission capacity of 220 kW is the one that has a larger share every year. However, that could be the result of the number of kilometers of transmission lines with different capacity. What I mean is that if 50% of the network has 220 kW capacity, then it is not surprise that 50% of the plants connect to those lines, right? I am not sure how relevant is the information in Table 1 or what is learnt from that
- In general I would have liked some more details on the UNet performance. For instance, it would be interesting to know the accuracy (user/producer) in order to compare with the accuracy given in the introduction for reference [7]. Also, I am curious about the accuracy of PV detectin over different LULC classes, in order to know if there could be some bias in the detection over certail categories (e.g. the authors mention underperformance in urban areas).
- There is a factor cocerning PV development that I wonder if could also affect, which is the terrain. I assume flat areas are more suitable than mountainous areas due to illumination and construction costs. Have you considered this factor?
Author Response
We have addressed the review comments and modified the manuscript accordingly. Please kindly see the attachment for the detailed point-by-point response.
We appreciate all the comments and suggestions which are really helpful to improve the quality of the manuscript.
Author Response File: 
Author Response.pdf
Reviewer 2 Report (Previous Reviewer 1)
I accept the manuscript in the given form and recommend its publication.
Author Response
Thank you very much for the review.
Reviewer 3 Report (New Reviewer)
The manuscript reveals the implementation of solar photovoltaic (PV) facilities in Vietnam, and identifies the potential sites for building PV in foreseeable future. Some recommendations are also given at the end of the manuscript, and satellite imageries have been adopted for conducting spatial analysis. The idea of this manuscript and topic is very new, and it is very interesting. Nevertheless, the authors should have addressed the following points before the manuscript is considered for publication:
Lines 14-15: "The result showed that the deep-learning model achieved satisfactory performance" - The authors should show some key figures and statistical values obtained in their study, not only in Abstract, but also in different parts of this manuscript.
Lines 35-36: income tax and land lease may not be related to the topic discussed here, please remove relevant texts.
Section 1 (Lines 98-107): The paragraphs here are not clear. Do the study attempt to identify potential sites for future solar PV installment? Or to obtain spatial distribution of facilities and conduct spatial analysis?
Section 1 (Line 104): The authors should add some references related to LULC monitoring via the use of Machine Learning techniques, especially in developing cities, say Vietnam, Pakistan, India etc. Please cite relevant sources.
Line 109: Typo in "2.1"
Line 141: Please explain the "rationale" of using median value here.
Line 143: "The resolution of the images is set to 10 m" How-what type of interpolation has been adopted here?
Line 178: Do you mean 600 and 60?
Section 2.5: Please provide more descriptions of the deep-learning model, describes the input bands and their significance (Line 193), as well as showing the formula of the IOU and F-score metrics (Lines 198-199).
Section 2.6: Please provide the data size / dimension of the eventual vector file (Line 210).
Lines 216-219: What is the PV is built in 2022? How would the authors deal with such case within this study?
The resolutions of some figures could have improved, for example, Figure 6, Figure 7, etc.
Line 382: glid should be "grid"?
Lines 386-390: Could you assess this phenomenon statistically? In particular, how much uncertainty was arisen?
Table 1: The data here is clear and well presented, but what are the insights and political thoughts / conclusions obtained from the Table?
Line 431: "performed well" - there are not much statistical figures / data shown in the entire manuscript, please add relevant statistical values.
Section 4.4 (Lines 554-560): Will the environmental and social impacts be related and connected with pollutions in our atmosphere, especially in less developed area of Vietnam? This would be interesting to mention, as many places of Vietnam were not yet well monitored in terms of air pollution.
Line 571: biggest --> largest
Generally speaking, the manuscript is quite well written, and the topic has huge scientific impacts as well. The aforementioned modifications will have to be addressed in the revision process.
Author Response
We have addressed the review comments and modified the manuscript accordingly. Please kindly see the attachment for the detailed point-by-point response.
We appreciate all the comments and suggestions which are really helpful to improve the quality of the manuscript.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report (New Reviewer)
The modification has greatly improved the quality of the manuscript. However, there are still some minor edits needed.
Point 4: About the reference of LULC - the authors added several application studies of SVM and RF into LULC detection, however they have focused more on the applications rather than the methodology itself. Here are some good references that the authors can focus on for the methodology part, please include how these machine learning approaches could work, and what are the scientific characteristics of these approaches:
https://www.mdpi.com/2072-4292/13/16/3337
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9853425/
https://www.mdpi.com/2072-4292/14/11/2654
Point 6: The authors respond that "Clouds usually have high reflectance, and the cloud shadows show low reflectance so that the “mean” values could be affected by those extreme values" - however, for pixels with clouds or high cloud fraction, have those pixels been already rejected during data manipulation process?
Point 7: The methodology adopted in creating the prescribed dataset is very important, and the authors should search for relevant reading documents / menu
Point 17: Some references have to be added to support some the ideas in the revised manuscript.
Author Response
We have addressed the review comments and modified the manuscript. Please kindly see the attachment for the detailed point-by-point response.
We appreciate to giving us very valuable comments on the manuscript, and it has been updated accordingly.
Author Response File: Author Response.pdf
Round 3
Reviewer 3 Report (New Reviewer)
The authors have addressed my previous comments. Good and interesting work.
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.
Round 1
Reviewer 1 Report
The review of the article is given in the appendix.
Comments for author File: Comments.pdf
Reviewer 2 Report
For a journal article, some new finding should be included and this is not found within this manuscript. Is there any new treatment for image detection in solar plane or parameters adjustment for this particular target? Is this conclusion can only be made vis remote sensing method? Authors should elaborates the scientific portion to improve the content of this manuscript.
Reviewer 3 Report
This article studies the development of PV installation on the Southern Vietnam. The authors use imagery from Sentinel-2 and Landsat to create PV maps over the years 2019, 2020 and 2021 and analyze the land type used, population and availability of powerlines. I do not find any major concerns concerning the methodology or the presentation of results, but I think a larger area of study or more quantitative analysis of PV development in connection with other socioeconomical variables would have increased the interest of this work.
I have only some minor questions/remarks for the authors to consider:
- The interest of the article is limited by the limited area of study. As the authors acknowledge in section 4.5, national scale shall be done in the future. Other studies being published handle larger data volumes. I wonder why it was particular burden for this work. Is it the time series analysis?
- I find that more details on how it is computed the PV potential shown in Figure 7 would be a good addition. I guess the authors do not have data about the actual power installed.
- Results on years 2019 and 2020 seemed to be limited by the training with 2021 data. However, there are a couple of questions that come to my mind. Why was used only 2021 data for training? And what makes the panels in 2021 differ from 2019 and 2020 to explain the difference performance of the classifier? Is the aging of the panels? Different composition?
- It would be interesting to see as well as the area of solar PV installations, what is the total number evolution over the years (I think the number is not given)
- Apart from the accuracy of the PV detection, I think it would be interesting to know something more about eh performance of the detection. For instance, I have the impression looking at Figure 9 that perhaps there is certain overestimation. Did you compare the PV classification with higher resolution aerial imagery, for instance? What is the minimum area that can be classified?
- There is a very big shift in Area installed in 2021 compared to 2020. This could be related to the number of installations being opened. The authors connect this to availability of power lines and land. However, I think there are other factors that could also explain this like national/regional regulations for new installations or economical (prices of land or PV panels, benefits of explotaition).
- I totally agree that the analysis of the PV land versus powerlines is interesting, but it is just qualitative and the data from Open Streetmap might be outdated as the authors acknowledge. Are there no better alternatives?
