Understanding the Spatial Distribution of Urban Forests in China Using Sentinel-2 Images with Google Earth Engine

Round 1

Reviewer 1 Report

 The paper covers an important topic, however the study itself does not very innovative.

The used methodology is quite simple, even if performed for the large area. Still, the methodology is in many places do not describe properly (e.g. why only 8 bands/why using indices/how they contribute the urban forests classification).  The second objective of the study - to assess the spatial distribution of urban forests in China, is described very shortly.

In the introduction section there is no references about existing studies on urban forests/green spaces assessment with remote sensing data. You should provide them.

The second challenge that you mentioned in introduction - the requirement of high resolution images is not a challenge anymore since 2015, or at least 2017 when two Sentinel-2 are in orbit. I would rewrite that. 

The same about the fourth challenge - to develop a good enough algorithm for urban forest extraction - currently there are thousands of classification algorithms like machine learning or deep learning.

Please use the term indices, as it is commonly used in remote sensing literature, not indicators (lines 19 and 258)

I suggest that in the area description there should be more focus put on the differences between 3 studied zones. I assume, that the climate of the zone C has a big impact on the lack of vegetation/trees in this region. Furthermore, information on population in zones is important.

Regarding Sentinel-2 data, revisit time is 5 days since 2017, not 10 days (line 106).

There is no information if, and if not - why, the atmospheric correction of S2 was performed.

There is no information in chapter 2.2.1 which S2 bands were used - in abstract there is information that only 8 bands were used. Why? I suggest to use all the bands with 10- and 20-meters resolution.

In lines 171-173 you write about the best band/indices combination. Please provide information how these different combination have been tested.

In the results section I suggest to put the accuracy assessment chapter first before the distribution of China urban forests, as it is also written in this sequence in the objectives.

Please change the method in the map in figure 8 for example to choropleth map

It would be good to provide a table with values for each city e.g. in appendix.

In Accuracy assessment there is Kappa coefficient mentioned in lines 213-214 although it was not mentioned in the methodology section.

I suggest to perform more tests which indices/bands are indeed important for accurate urban forests discrimination. There is only slight improvement in OA when adding indices

Please rewrite sentence in lines 259-260 cause it is unclear, what is important and why?

Lines 289-290 add before "within" which city you mean!

Author Response

Point 1: The paper covers an important topic, however the study itself does not very innovative. The used methodology is quite simple, even if performed for the large area. Still, the methodology is in many places do not describe properly (e.g. why only 8 bands/why using indices/how they contribute the urban forests classification). The second objective of the study - to assess the spatial distribution of urban forests in China, is described very shortly.

Response 1: Many thanks for your recognition of the significance of our study. A dataset of urban forests is very important for relatively studies, but the data gap actually exists. We used the new urban area and established a classifier for urban forest extraction, the accuracy of classification result was well and this work is of great significance for future research.

For the methodology, we added a test table about the selection of bands and indices, to demonstrate that we chose the best combination, according to your detailed suggestions below. We hope this effort can make the selection reasonable.

For the second objective of the study, we added the description about different urban forest cover in three different zones as following:

“For the three zones in China (the spatial division was presented in Fig. 1), the urban forest cover in zone A was 16.0%, zone B had the highest urban forest cover at 23.0%, and zone C had the lowest at 8.5%. Zone A and C are both below the national average level. Nearly 60% of cities in zone C has the urban forest cover below 10%, which is much higher than the proportion of cities in zone A (28.9%) and zone B (4.2%); and nearly 85% of cities in zone B have the urban forest cover between 10% and 30%. Furthermore, the per capita urban forest area in zone C in 2016 was 10.3 m², which was approximately one third of the corresponding value in zone A (34.9 m²), and about one quarter in zone B (39.1 m²).”

Thanks for your kind comments, which bring our attention to some details that we have ignored before. We hope the revised manuscript could explain these problems clearly.

Point 2: In the introduction section there is no references about existing studies on urban forests/green spaces assessment with remote sensing data. You should provide them.

Response 2: We are very sorry for this. In the revised manuscript, we provided the references after the general introduction of methods of measuring urban forest cover, in Line 53-60. This part was as followed,

“Among them, remote sensing is efficient and useful to map forested urban areas for large scale city: Canetti et al.[21] used RapidEye and Satellite for observation of Earth (SPOT 5) images, to quantify multi-temporal urban forest cover in Araucaria (a city of Brazil), based on the support vector machine algorithms; Chen et al.[24] drew the urban green space in the neighborhoods of five Chinese megacities using Google Earth images with the spatial resolution of 0.26 m; and Fan et al.[22] quantified tree canopy of Cook County in the United States, using multispectral images with the spatial resolution of 1 m from National Agriculture Imagery Program and Light Detecting and Ranging data.”

Point 3: The second challenge that you mentioned in introduction - the requirement of high resolution images is not a challenge anymore since 2015, or at least 2017 when two Sentinel-2 are in orbit. I would rewrite that.

Response 3: Many thanks! The expression may be not accurate here. The purpose we raised these challenges was to present the problems that need to be considered before mapping urban forest cover at the national level. Maybe the word “challenges” we used was not proper. Therefore, we revised this word as “problems which should be considered”.

For the second one, Sentinel-2 images provide a solution to this problem, and we have mentioned it in the next paragraph “How to address these problems in our study”.

Point 4: The same about the fourth challenge - to develop a good enough algorithm for urban forest extraction - currently there are thousands of classification algorithms like machine learning or deep learning.

Response 4: The same about the fourth one, to establish a classifier for urban forest extraction is important for mapping urban forest cover, even though there has many classification algorithms for land cover, it is necessary to find a better band combination to make urban forest extraction as accurate as possible. Maybe the word “develop” is not proper, we revised this sentence as “The fourth one is to establish a good enough classifier for urban forest extraction”.

Point 5: Please use the term indices, as it is commonly used in remote sensing literature, not indicators (lines 19 and 258)

Response 5: Many thanks for your useful comments! We have revised this word in the manuscript, and we will take note of this in the future.

Point 6: I suggest that in the area description there should be more focus put on the differences between 3 studied zones. I assume, that the climate of the zone C has a big impact on the lack of vegetation/trees in this region. Furthermore, information on population in zones is important.

Response 6: We added this description in the revised manuscript, about differences in natural conditions and urban population for three zones, in line 108-114, as following,

“Their average daily precipitation in growing season (from April to September) was respectively 2.6 mm, 7.5 mm and 1.1 mm, and average daily temperature was respectively 10.4°C, 19.3°C and 10.1°C [36]. These three zones have obvious differences not only in natural conditions, but also in urban development. Zone A had 32.4 thousand km² of urban area in 2016 and 204.7 million urban population; zone B had the largest urban area of 35.9 thousand km², and urban population of 243.7 million; and zone C was far lower than zone A and B, with an urban area of 4.9 thousand km² and only 28.7 million lived in urban area [37].”

For influential factors for spatial distribution of urban forest cover, this is a relatively complex problem according to the study of Chen and Wang (2013), which should to be further explored in our future research.

Chen, W.Y., Wang, D.T., 2013. Urban forest development in China: Natural endowment or socioeconomic product? Cities 35, 62-68.

Point 7: Regarding Sentinel-2 data, revisit time is 5 days since 2017, not 10 days (line 106).

Response 7: Thanks for your reviews. We added the description of revisit period according to Sentinel-2 User Handbook, “with a revisit period of 10 days at the equator, and 5 days with 2 satellites since 2017 and 2-3 days at mid-latitudes.”

Point 8: There is no information if, and if not - why, the atmospheric correction of S2 was performed.

Response 8: The atmospheric correction was not performed in this study, which may have uncertain impact for the result. Many studies based on GEE platform mentioned this limitation (Huang et al., 2017; Tian et al., 2019; Sun et al., 2019). We added this description in the chapter 4.2, as following,

“GEE did not ingest images with atmospheric correction when we execute our classification; moreover, it was difficult to perform atmospheric correction in the GEE platform because of difficulties in parameter acquisition [31,40,48]. Therefore, the top-of-atmosphere (TOA) reflectance data from Sentinel-2 was directly used to extract urban forest cover in this study, which may be affect the results.”

Huang, H.; Chen, Y.; Clinton, N.; Wang, J.; Wang, X.; Liu, C.; Gong, P.; Yang, J.; Bai, Y.; Zheng, Y., et al. Mapping major land cover dynamics in Beijing using all Landsat images in Google Earth Engine. Remote Sensing of Environment 2017, 202, 166-176, doi:https://doi.org/10.1016/ j.rse.2017.02.021.

Tian, F.; Wu, B.; Zeng, H.; Zhang, X.; Xu, J. Efficient Identification of Corn Cultivation Area with Multitemporal Synthetic Aperture Radar and Optical Images in the Google Earth Engine Cloud Platform. Remote Sensing 2019, 11, 629.

Sun, Z.; Xu, R.; Du, W.; Wang, L.; Lu, D. High-Resolution Urban Land Mapping in China from Sentinel 1A/2 Imagery Based on Google Earth Engine. Remote Sens-Basel 2019, 11, doi:10.3390/rs11070752.

Point 9: There is no information in chapter 2.2.1 which S2 bands were used - in abstract there is information that only 8 bands were used. Why? I suggest to use all the bands with 10- and 20-meters resolution.

Response 9: Thanks for your comment! The information about the selection of bands was provided in the third point in chapter 2.3, because the band selection is a part of classifier establishment in our opinion. If it is not appropriate, please let us know, thanks!

As for the band selection, we had tried to use all bands with 10- and 20-meters resolution, but the overall accuracy was 92.09%, which is lower than the results we have now. In order to describe clearly why these bands were selected, we added the description in line 190-191, as followed,

“For example, the OA based on these bands (B2-B8, B11) can increase 0.2% compared with all bands of Sentinel-2 with 10 and 20 meters resolution.”

Point 10: In lines 171-173 you write about the best band/indices combination. Please provide information how these different combination have been tested.

Response 10: This comment was exactly useful! Many thanks! We presented the results of different bands and indices which have been tested (Table 2) in revised manuscript, and found that the combination we chose does allow the highest overall accuracy.

Table 2 Overall accuracy for urban forest mapping by RF classifiers using different bands and indices.

Point 11: In the results section I suggest to put the accuracy assessment chapter first before the distribution of China urban forests, as it is also written in this sequence in the objectives.

Response 11: We totally agree with your suggestion and have revised it, thank you!

Point 12: Please change the method in the map in figure 8 for example to choropleth map

Response 12: Thanks for your suggestion! We revised this figure.

Point 13: It would be good to provide a table with values for each city e.g. in appendix.

Response 13: We have provided the table of top 100 prefecture-level cities which have the highest urban forest cover in Appendix A. Supplementary Material; and the downlink of our urban forest cover dataset was also provided.

Point 14: In Accuracy assessment there is Kappa coefficient mentioned in lines 213-214 although it was not mentioned in the methodology section.

Response 14: Thanks for your comments! We added the simple description of Kappa coefficient in the methodology section, Line 155-156, “and the Kappa coefficient. The Kappa value of higher than 0.8 represents strong agreement between classification result and reference distribution [38].”

Point 15: I suggest to perform more tests which indices/bands are indeed important for accurate urban forests discrimination. There is only slight improvement in OA when adding indices

Response 15: Thanks! We added the table about overall accuracies of different bands and indices which have been tested (Table 2) according to your above suggestion. Even though these OA did not have very obvious differences, we still need to choose the combination of the highest OA.

Point 16: Please rewrite sentence in lines 259-260 cause it is unclear, what is important and why?

Response 16: Very sorry, the word “important” may be not appropriate. We meant vegetation, water, and built-up areas are key elements in urban areas. Here we have revised this sentence as “which helped to distinguish between urban forests and non-urban forests because vegetation, water, and built-up areas are the key elements within urban areas.”

Point 17: Lines 289-290 add before "within" which city you mean!

Response 17: Thanks for your comments! The city we referred to was Beijing, and we revised this sentence as “Figure 6 shows that the spatial differences of urban forests distribution within a Beijing are very clear, being concentrated in the north of urban area.”

Author Response File: Author Response.docx

Reviewer 2 Report

This study used eight bands (B2–B8, B11) and three indicators of 19 Sentinel-2 in 2016 to map the urban forests of China using the Random Forest (RF) machine learning 20 algorithms at the pixel scale with the support of GEE. The manuscript is well written and organized. Reading through this manuscript, I think the finding is significant and meaningful for both researchers and politicians.

I think the technical part of this paper needs to be improved. The authors stated that RF is used to train an image segmentation model, to segment the urban forests for images. However how to obtain the groundtruth is not clear presented. Besides, how the ground data is used? For images in Figure 2, what are the role for them in the whole pipeline?

minor comments:

Page 7 "Suitable bands of Sentinel-2 images were selected to train data based on the RF classifier "

"The data were trained to map the urban forest distribution in China and assess the178 classification accuracy. "

Data can not be trained. Only the model can be trained.  

Author Response

This study used eight bands (B2–B8, B11) and three indicators of 19 Sentinel-2 in 2016 to map the urban forests of China using the Random Forest (RF) machine learning 20 algorithms at the pixel scale with the support of GEE. The manuscript is well written and organized. Reading through this manuscript, I think the finding is significant and meaningful for both researchers and politicians.

Response: Thanks for your kind encouragement for our manuscript, and the recognition of the significance of this research! We have revised the manuscript according to your suggestion.

Point 1: I think the technical part of this paper needs to be improved. The authors stated that RF is used to train an image segmentation model, to segment the urban forests for images. However how to obtain the ground truth is not clear presented. Besides, how the ground data is used? For images in Figure 2, what are the role for them in the whole pipeline?

Response 1: We are very sorry that this part of our description is not clear enough.

As presented in Figure 4 in the latest manuscript, the training data which showed the ground truth were obtained in the Google Earth, the distribution of samples was presented in Figure 3, and 70% of them were used to train the model. This was explained in Line 152: “Then, 70% of the VHR sample points were used to gather knowledge and train the classifier in GEE”.

Then, the ground data was used to help testing the classification accuracies, together with 30% of the samples obtained in Google Earth. In addition, in Line 138-139 of the revised manuscript, we added the description:

“The field photos were taken (Figure 2), showing different kinds of ground samples for urban forests, and their locations (the latitude and longitude) were recorded to be used in the validation”.

Point 2: minor comments:

Page 7 "Suitable bands of Sentinel-2 images were selected to train data based on the RF classifier "

"The data were trained to map the urban forest distribution in China and assess the178 classification accuracy. "

Data ca not be trained. Only the model can be trained.

Response 2: Many thanks for pointing out this mistake. We have revised the mistake,

Page 7 Line 187 “Suitable bands of Sentinel-2 images were selected to train the RF classifier.”

Line 198 “The established RF model were trained to map the urban forest distribution…”

Meanwhile, we would keep this in mind in future study.

Author Response File: Author Response.docx

Reviewer 3 Report

The authors used 10m Sentinel-2 imagery to map urban forests in Chinese cities. As described, the research is largely sound and will be interesting to the journal's readership. I recommend major revisions mainly due to issue #1 below. I believe the other revisions will be minor and relatively simple to address.

#1. The authors should comply with the journal’s requirements regarding the availability of data and computer code. In the manuscript I received, there was no mention of data availability (the urban forest classification outputs/maps) or the computer code. These must be provided before the article can be published. From the journal’s instructions to authors:

“Data Availability: In order to maintain the integrity, transparency and reproducibility of research records, authors must make their experimental and research data openly available either by depositing into data repositories or by publishing the data and files as supplementary information in this journal.

Computer Code and Software: For work where novel computer code was developed, authors should release the code either by depositing in a recognized, public repository or uploading as supplementary information to the publication. The name and version of all software used should be clearly indicated.”

#2. I understand that 10m imagery was necessary to use to span the entire country of China. But I am curious about how this resolution limits the study, especially in the ability to detect individual trees (in Fig 9, I see some individual trees that were not classified as urban forest). Many other modern urban forest classifications are using imagery at 1m or finer resolution, so they are likely to detect smaller trees and trees growing individually. At a minimum, this issue of pixel size should be discussed in the limitations section (line 263).

It would be interesting to see the authors report the classification accuracy by ground truth tree type such as in Fig 2. For example, does the classification perform better for ‘forests’ compared to ‘street trees’?

The implications of this issue could be meaningful for the conclusions about smaller Chinese cities having lower canopy cover. If these are newer cities, it is possible that many small trees exist in those cities, but they were recently planted so they are small and were not big enough to detect with 10m imagery. But these trees could grow over time and canopy cover would increase along with the ability to detect these trees with 10m imagery. Again, I think this issue is important to discuss if it cannot be addressed in the methods with the available imagery.

#3. Other issues to address:

Abstract: abbreviations are not necessary here, as they are given again in the text Keywords: I recommend avoiding keywords that also appear in the title. For example, ‘urban forests’ could be replaced with ‘urban greening.’ L33: Urbanization is a process, not a noun. I suggest revising to “..China’s population increased from 19.4% urban to 59.6% urban” L43: the phrase “it has become the consensus of urban construction” is awkward and I’m not sure what it means. Construction in many parts of the world, including developed countries, still involves the destruction of trees. L51: suggest changing ‘urban forest extraction’ to ‘measuring urban forest cover.’ Without the remote sensing context, extraction can be confused with resource extraction such as timber harvest. L74: Please clarify how GEE helps develop a suitable algorithm for urban forest extraction. It is not obvious from the text here. L86: how many urban areas? Please provide the number. Figure 1: In the main map panel showing all of China, is the inset map necessary? It doesn’t appear that the islands south of Guangzhou contain any urban areas, so I think this inset map could be removed from the map. L106: What is the approximate revisit period for Chinese cities? The equatorial revisit period is not relevant for China. L114 and elsewhere: check the spelling of Sentinel. Figure 2: What is the difference between ‘trees in dwelling districts’ and ‘street trees’? Based on the photographs, they appear to be the same thing. L144: The abbreviations for NDVI, NDWI, and NDBI are given at L170, but they should be given here instead at first use. L229: suggest rewording to: “This study demonstrated the feasibility of mapping urban forests in China…” L257: replace ‘except’ with ‘along with’ L280: If only a small number of points were included in zone 3, why did the authors not simply add more points in this region to improve the sample depth?

Author Response

The authors used 10m Sentinel-2 imagery to map urban forests in Chinese cities. As described, the research is largely sound and will be interesting to the journal's readership. I recommend major revisions mainly due to issue #1 below. I believe the other revisions will be minor and relatively simple to address.

Response: Thanks for your recognition for our research. We have revised this manuscript according to your kind comments!

Point 1: #1. The authors should comply with the journal’s requirements regarding the availability of data and computer code. In the manuscript I received, there was no mention of data availability (the urban forest classification outputs/maps) or the computer code. These must be provided before the article can be published. From the journal’s instructions to authors:

“Data Availability: In order to maintain the integrity, transparency and reproducibility of research records, authors must make their experimental and research data openly available either by depositing into data repositories or by publishing the data and files as supplementary information in this journal.

Computer Code and Software: For work where novel computer code was developed, authors should release the code either by depositing in a recognized, public repository or uploading as supplementary information to the publication. The name and version of all software used should be clearly indicated.”

Response 1: Many thanks for your comments! We have provided the download link of data and code in Appendix A. Supplementary Material.

Point 2:#2. I understand that 10m imagery was necessary to use to span the entire country of China. But I am curious about how this resolution limits the study, especially in the ability to detect individual trees (in Fig 9, I see some individual trees that were not classified as urban forest). Many other modern urban forest classifications are using imagery at 1m or finer resolution, so they are likely to detect smaller trees and trees growing individually. At a minimum, this issue of pixel size should be discussed in the limitations section (line 263).

It would be interesting to see the authors report the classification accuracy by ground truth tree type such as in Fig 2. For example, does the classification perform better for ‘forests’ compared to ‘street trees’?

The implications of this issue could be meaningful for the conclusions about smaller Chinese cities having lower canopy cover. If these are newer cities, it is possible that many small trees exist in those cities, but they were recently planted so they are small and were not big enough to detect with 10m imagery. But these trees could grow over time and canopy cover would increase along with the ability to detect these trees with 10m imagery. Again, I think this issue is important to discuss if it cannot be addressed in the methods with the available imagery.

Response 2: We totally agree with your suggestion, and added this limitation in the third paragraph of 4.2, Line 311-320. The revised paragraph was as followed,

“Furthermore, we examined the classification results using the 75 ground data (Figure 2) in order to gain misclassification information, and found that eight of them presented errors, mainly for the ground data of individual tree around house and street, and low canopy cover in commercial areas and schools; forests, trees in parks, and cluster trees in schools or roadside showed low misclassification. It should be noted that even though we used Sentinel-2 images with a 10 m resolution to identify the inner city classes of the entire country of China as clear as possible, for some small canopy covers and newly individual trees, the 10 m resolution is still too coarse to be identified. Therefore, this dataset was created to fill the gap in urban forest cover across China, and more suitable for macro analysis at the national level; for individual cities, other images which has higher resolution can be used to obtain more accurate information”

#3. Other issues to address:

Point 3:Abstract: abbreviations are not necessary here, as they are given again in the text Keywords: I recommend avoiding keywords that also appear in the title. For example, ‘urban forests’ could be replaced with ‘urban greening.’

Response 3: Many thanks for your suggestion, we deleted the abbreviations in Abstract. And the ‘urban forests’ in keywords was revised as ‘urban greening’.

Point 4: L33: Urbanization is a process, not a noun. I suggest revising to “..China’s population increased from 19.4% urban to 59.6% urban”

Response 4: This suggestion was very useful, thank you! We revised this sentence as “the proportion of China’s urban population increased from 19.4% to 59.6%”.

Point 5: L43: the phrase “it has become the consensus of urban construction” is awkward and I’m not sure what it means. Construction in many parts of the world, including developed countries, still involves the destruction of trees.

Response 5: On reflection, I agree with you. Our original intention was to present that many countries began to pay attention to the protection of urban forests, in reports from FAO, many countries are increasing urban forests with encouraging results. Here we revised this sentence as, “It has raised concerns about protecting and increasing the urban forests around the world”.

Point 6: L51: suggest changing ‘urban forest extraction’ to ‘measuring urban forest cover.’ Without the remote sensing context, extraction can be confused with resource extraction such as timber harvest.

Response 6: Thanks for suggestion! We have revised it as you suggested.

Point 7: L74: Please clarify how GEE helps develop a suitable algorithm for urban forest extraction. It is not obvious from the text here.

Response 7: Very nice comments. GEE has large computing power, so it can help us timely adjusted the parameters and band selection of the model, to develop the algorithm which could get higher accuracy. We added this description in the revised manuscript, “GEE is a high-performance cloud computing platform with a convenient, fast image selection process and large computing power [30], and provide different classifiers, therefore, the parameters and band selection of the model can be quickly and timely adjusted according to the calculation results.”

Point 8: L86: how many urban areas? Please provide the number. Figure 1: In the main map panel showing all of China, is the inset map necessary? It doesn’t appear that the islands south of Guangzhou contain any urban areas, so I think this inset map could be removed from the map.

Response 8: We added the total number in Line 100-101, “and the total area in 2016 was 73.2 thousand km². There were 31 provincial-level regions, and 334 prefecture-level divisions in Chinese Mainland in 2016.”

As for the inset map, this is the standard for mapping China, please understand that it is necessary. Many thanks!

Point 9: L106: What is the approximate revisit period for Chinese cities? The equatorial revisit period is not relevant for China.

Response 9: Thanks for your reviews. We added the description of revisit period according to Sentinel-2 User Handbook, Chinese cities are most located at mid-latitudes, “with a revisit period of 10 days at the equator with one satellite, and five days with two satellites since 2017 and 2-3 days at mid-latitudes.”

Point 10: L114 and elsewhere: check the spelling of Sentinel. Figure 2: What is the difference between ‘trees in dwelling districts’ and ‘street trees’? Based on the photographs, they appear to be the same thing.

Report 10: Sorry for these spelling mistakes, we have revised them, and thanks for pointing out them. The difference between trees in dwelling districts and street trees is the position of these trees, trees in dwelling districts have more closely relationship with residents in this neighborhood, which could regulate the community climate, the greening cover of the dwelling district is one of the important indicators in the housing construction; and street trees have significance for all pedestrians, and be managed by the city government.

Point 11: L144: The abbreviations for NDVI, NDWI, and NDBI are given at L170, but they should be given here instead at first use.

Response 11: Sorry! That is our negligence. We gave the abbreviations at here.

Point 12: L229: suggest rewording to: “This study demonstrated the feasibility of mapping urban forests in China…”

Response 12: Many thanks for your kind revision, we revised this sentence according to your suggestion.

Point 13: L257: replace ‘except’ with ‘along with’

Response 13: Thanks! We have revised this word.

Point 14: L280: If only a small number of points were included in zone 3, why did the authors not simply add more points in this region to improve the sample depth?

Response 14: The urban area of zone 3 accounted for 6.6% of the total urban area in China, and the urban forest cover of zone 3 is 8.5%, so the urban forests are relatively less. Considering about the quality requirements of training sample points, the acceptable sample points are less.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Two minor comments:

Line 53: You used the term "large scale city" does it mean large cities or large-scale maps?

You could add that information in the text (discussion) to highlight the importance of this topic:

For influential factors for spatial distribution of urban forest cover, this is a relatively complex problem according to the study of Chen and Wang (2013), which should to be further explored in our future research. Chen, W.Y., Wang, D.T., 2013. Urban forest development in China: Natural endowment or socioeconomic product? Cities 35, 62-68.

Author Response

Point 1: Line 53: You used the term "large scale city" does it mean large cities or large-scale maps?

Response 1: Many thanks for your comment! The “large scale city” we used here means large-scale maps, and we revised this term in the manuscript.

Point 2: You could add that information in the text (discussion) to highlight the importance of this topic:

For influential factors for spatial distribution of urban forest cover, this is a relatively complex problem according to the study of Chen and Wang (2013), which should to be further explored in our future research. Chen, W.Y., Wang, D.T., 2013. Urban forest development in China: Natural endowment or socioeconomic product? Cities 35, 62-68.

Response 2: Your comment is very useful! We added this description in the Discussion, Line 335-337, as the following,

“Detecting the influential factors for the spatial differences of urban forest cover is of vital importance for effectively increasing the urban forests. However, it is a relatively complex problem according to previous studies [49-52], which should to be further explored in future research.”

Author Response File: Author Response.docx

Reviewer 3 Report

Thanks to the authors for incorporating changes to the manuscript based on my comments. A couple remaining issues to note:

The authors provided supplemental materials containing their computer code and data files. However, I am not able to open the computer code links because I do not have permission. I will let the journal's editor decide how to proceed with this issue (i.e., the editor can decide if this level of code availability meets the journal's standards). The revised text addresses shortcomings in the original submission, but the revised portions of the text contain some English writing errors that should be fixed before the manuscript is published.

Author Response

Thanks to the authors for incorporating changes to the manuscript based on my comments. A couple remaining issues to note:

Point 1: The authors provided supplemental materials containing their computer code and data files. However, I am not able to open the computer code links because I do not have permission. I will let the journal's editor decide how to proceed with this issue (i.e., the editor can decide if this level of code availability meets the journal's standards).

Response 1: We are sorry for this problem. We have tested that this link can be opened by other colleagues. As long as you have an account of Goole Earth Engine and log in, this code can be opened and running. Here is the running interface.

However, in order to further address this problem, we added the code in the Appendix, download the file of urban area of China in 2016 and copy the code in GEE platform, this can run. If there has any problems, we can actively cooperate to solve it.

Point 2: The revised text addresses shortcomings in the original submission, but the revised portions of the text contain some English writing errors that should be fixed before the manuscript is published.

Response 2: Many thanks for your comment! We have asked for help from English-speaking colleagues, and tried to revise these errors.

Author Response File: Author Response.docx