Deep Learning-Based Detection of Urban Forest Cover Change along with Overall Urban Changes Using Very-High-Resolution Satellite Images
Round 1
Reviewer 1 Report
In the study, deep learning-based urban forest change analysis is proposed with general changes in the urban environment using VHR bitemporal images. DeepLabv3+ was used to generate binary forest cover masks and a deeply controlled image fusion network (DSIFN) was used to generate a binary exchange mask. Analyzes were made on forest cover changes.
1) (advice) If it is possible, can you compare it to the other studies you did on the open source dataset you used?
2) It will not be sufficient to compare only the results of the NDVI index. It is also very important for scientific soundness by applying and comparing state-of-the-art (SOTA) models on your dataset.
3)In addition to the metrics used in the study, I think that the analyzes to be made with the IoU metric will be very meaningful for this study.
Minor editing of English language required
Author Response
Please find the attached response letter.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The authors proposed a deep learning-based urban forest change detection from VHR bitemporal images. In this study, two networks are used independently; DeepLabv3+ for generating binary 20 forest cover masks, and a deeply supervised image fusion network (DSIFN) for a binary change mask generation. The results are concatenated for semantic CD while focusing on forest cover changes. As for me the manuscript need to address following comments:
1- In the literature review, please include the following reference which is a well-known forest change detection method based on deep learning:
https://doi.org/10.1007/s12145-022-00885-6
https://doi.org/10.3390/rs14071580
https://doi.org/10.1109/TGRS.2018.2863224
https://doi.org/10.3390/rs12060901
2- Additionally, there are other forest change detection methods that utilize NDVI and other vegetation indices, as well as important spectral bands. Please refer to the following paper and include it in the discussion (Lines 48-58). The authors should be cautious not to exclusively rely on NDVI for change detection as there are other effective approaches:
https://doi.org/10.1016/j.jag.2016.06.020
https://doi.org/10.1080/01431161.2021.1995075
https://doi.org/10.3390/rs12020341
3- Regarding Line 59-60 on page 2, where it states "The use of deep learning networks reduces manual steps in monitoring changes by automating feature extraction, avoiding feature selection, and reducing manual steps during CD," please cite the following papers to support this claim:
https://doi.org/10.1109/IGARSS47720.2021.9553859
https://doi.org/10.1007/s12145-022-00885-6
4- The last part of the introduction should provide more clarity on the novelties of the paper. The authors need to explicitly state what the main contributions of their work are.
5- It is essential to conduct an ablation study for the proposed method and compare it with other relevant approaches, not solely relying on the NDVI-based method for verification. An ablation study would demonstrate the effectiveness of using image fusion in the proposed approach.
6- The authors need to provide the validation loss and train loss plots for the trained networks. Additionally, more details should be provided on the image data used, such as whether augmentation techniques were applied, the number of tiles used for training, etc.
7- It is important to consider the generalizability of the method by not only training on a subset of the dataset but also testing on completely separate tiles. Training and testing on the same tiles may reduce the ability to generalize the results to other datasets or regions.
Author Response
Please find the attached response letter.
Author Response File: Author Response.pdf
Reviewer 3 Report
The manuscript handles proposing a new technique in change detection using two different deep learning networks namely; DeepLabv3+ and DSIFN. The study yielded promising scores when compared the NDVI or reference data. The outputs give useful information especially for the researchers and city planners. English quality is good. The similarity index gave as 25% using “Ithenticate” software (20% when excluding references). That value can be considered in accepted levels. Some suggestions/corrections were given as follows to increase the quality of the manuscript:
v Page 1, abstract: Outstanding numerical outputs from this study should be given in the abstract section. The scores or change detection results of the proposed method can be mentioned.
v Page 3, lines 111-116: This part “The three binary…………considerably.” seems more suitable for the conclusion section.
v Page 3, line 122: The 4th contribution given as “transfer learning….” given as 3rd contribution and should be changed to “4)”.
v Page 3, lines 142-143: It stated that “……CD labels were generated through visual inspection and manual digitization of images.” You claim that both pixel based or object based classification requires visual interpretation or manual digitization in the introduction section. However, you use the same procedure for the new suggested approach. It is better to clarify this issue in the introduction section where you mentioned.
v Page 8, lines 263-289: The given information here, related to the validation process (method). Therefore, I suggest you to create a new sub section under the methodology as “3.4. Validation” and give that information here with the last paragraph of the previous section (To assess the overall……….).
v Page 12, line 359: Another term other than “Evaluation” would more fit in the heading of “Evaluation of Forest Cover Changes”. (“Finalizing” might be appropriate).
v Page 14, line 431: Before the “Conclusion” section, it is highly suggested to compare your results (outputs) with the previously conducted similar studies.
v Page 14, 446: It is better to give some numerical outputs from this study such as scores or detected changes in percentages, at the end of the pharagraph.
v Page 16, lines 535-536: Please check the journal or conference name given as “ar Xiv preprint ar Xiv”.
v Page 16, lines 540: Please check the journal or conference name given as “ar Xiv preprint ar Xiv”.
Author Response
Please find the attached response letter.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
In my opinion, it's ready for publication as it stands.
Reviewer 2 Report
The authors well addressed my comments.
