The Differences and Influence Factors in Extracting Urban Green Space from Various Resolutions of Data: The Perspective of Blocks

Round 1

Reviewer 1 Report

This article provides an analysis of the differences in the application of data with different resolutions for green space extraction, giving conclusions and corresponding guidelines that may be useful for data selection solutions in urban green space studies. However, this work has some weaknesses, in my opinion, that require a major revision.

Main Criticism 1:

This research aims to compare the effects of different resolution data on the extraction of urban green space. Except for the resolution, the other attributes of the experimental data should be as consistent as possible. However, the bands of the data in the case study are different, corresponding to different extraction methods of GR, which may also affect the extraction results. I suggest using data with the same band instead in this study.

Main Criticism 2:

The experimental data has been processed, including modifications to the DJI resolution and the remote sensing data pre-processing. However, why and how this pre-processing was performed, especially the change of resolution from around 0.1m to 0.5m, is important in determining the correctness of the experimental results. I suggest avoiding processing the data resolution during pre-processing, as this would otherwise make interpreting its effect on the results difficult.

Minor Remarks 1:

Page 11, Line 324 to 358: This is a case study, so the conclusions may only be applicable to this case. More experiments are needed to know whether the findings can be applied to other city-wide studies.

Minor Remarks 2:

Page 1, Line 8 to 24: The Abstract section requires further revision to clarify the innovations in the study. As there are still studies on the selection of suitable data in different application scenarios, the novel aspects of this work, such as the significance of the study from the perspective of blocks, the analysis of the factors affecting the differences, need to be emphasized.

Minor Remarks 3:

Page 6, Line 193 to 212: As written in the article, "the overall difference was minimal", analyzing mean value and standard deviation is probably not a good idea as it could not provide a valid reference for data selection. For comparisons of differences of GR in fishnets with various sizes, it is recommended to use indicators of TT, TF, FT, and FF, similar to remote sensing classification accuracy evaluation.

Minor Remarks 4:

1. Page 8, Figure 4: Could you also provide the spatial distribution of GR grades in a typical region? This would make the analysis more convincing.

Minor Remarks 5:

Some minor issues. Abbreviations only need to be defined once in the main body, please check “urban green space”, “green space ratio”, “degree of difference”. Please check capitalization of table headers. Page 5, Line 162: divided into; Page 9, Figure 5: Using line charts may not be appropriate as there is no consecutiveness among A to J land use types.

Author Response

Thanks so much for your professional comments and suggestions for our manuscript. They are all valuable and instructive to improve the academic quality of this paper. We have studied the comments carefully and made revisions as well as responses accordingly. All revisions have been marked up using the “Track Changes” function in the revised file. Hope we can meet with your approval.

 

Main Criticism 1:

This research aims to compare the effects of different resolution data on the extraction of urban green space. Except for the resolution, the other attributes of the experimental data should be as consistent as possible. However, the bands of the data in the case study are different, corresponding to different extraction methods of GR, which may also affect the extraction results. I suggest using data with the same band instead in this study.

R: It is indeed a problem we didn’t notice earlier. Because DJI image only has three visible light bands, GLI index was finally used to extract UGS from all three kinds of data. The illustration and subsequent analysis results have been changed accordingly in Section 2.4 (Page 4, Line 136-157) and Section 3. 

 

Main Criticism 2:

The experimental data has been processed, including modifications to the DJI resolution and the remote sensing data pre-processing. However, why and how this pre-processing was performed, especially the change of resolution from around 0.1m to 0.5m, is important in determining the correctness of the experimental results. I suggest avoiding processing the data resolution during pre-processing, as this would otherwise make interpreting its effect on the results difficult.

R: The initial consideration of changing resolution was to make sure the difference of three kinds of resolution are not so much, but we missed the possible consequence. To avoid that, the resolution of DJI has been changed back into 0.1m (see Table 1). As for the pre-processing of remote sensing images, we believe those are necessary to exclude the influence of atomsphere and other equipment errors, as DJI drone is more stabilized for its relatively low flight height. And there are a few references to support that the pre-processing of remote sensing data is required.  https://doi.org/10.3390/rs14194859; https://doi.org/10.3390/rs12223845

 

Minor Remarks 1:

Page 11, Line 324 to 358: This is a case study, so the conclusions may only be applicable to this case. More experiments are needed to know whether the findings can be applied to other city-wide studies.

R: It is undeniable that the correlation and regression results may be different because of the sample quantity. But still we believe that the conclusions have certain reference value as the selected area of this study is representative and large enough to cover all types of land use and blocks. And it is revealed that there are factors affecting the difference that should be noticed in the practical application. In the future research, more case studies about the application of the conclusions will be conducted. We have added the explanation about this part in Page 12, line 365-367 and Page 14, Line 457-461.

 

Minor Remarks 2:

Page 1, Line 8 to 24: The Abstract section requires further revision to clarify the innovations in the study. As there are still studies on the selection of suitable data in different application scenarios, the novel aspects of this work, such as the significance of the study from the perspective of blocks, the analysis of the factors affecting the differences, need to be emphasized.

R: The abstract has made revision to highlight the innovation in Line 9-11 and Line 23-25.

 

Minor Remarks 3:

Page 6, Line 193 to 212: As written in the article, "the overall difference was minimal", analyzing mean value and standard deviation is probably not a good idea as it could not provide a valid reference for data selection. For comparisons of differences of GR in fishnets with various sizes, it is recommended to use indicators of TT (Ture-True), TF (Ture-False), FT, and FF, similar to remote sensing classification accuracy evaluation.

R: We agree with your recommendation. As there will be too much data if we present all TT, TF, FT and FF value for each GR grade in all fishnet size, we chose precision index to illustrate the differences, that is Precision = TT / (TT+FT). The revisions have been made in Section 2.5 (Page 5, Line 178-179) and Section 3.1.1 (Page 6-7, Line 203-221).

 

Minor Remarks 4:

1. Page 8, Figure 4: Could you also provide the spatial distribution of GR grades in a typical region? This would make the analysis more convincing.

R: We have provided the spatial distribution of GR grades in a typical region shown in Figure 5 and added the corresponding illustration in Page 7, Line 237-241.

 

Minor Remarks 5:

Some minor issues. Abbreviations only need to be defined once in the main body, please check “urban green space”, “green space ratio”, “degree of difference”. Please check capitalization of table headers. Page 5, Line 162: divided into; Page 9, Figure 5: Using line charts may not be appropriate as there is no consecutiveness among A to J land use types.

R: All abbreviations and table headers as well as other spelling mistakes have been checked. The Figure 5 has been changed into column chart.

Author Response File: Author Response.docx

Reviewer 2 Report

The research is interesting because of the very timely topic it addresses. Census and monitoring of vegetation in the urban environment is a very interesting challenge today, and your research brings an interesting contribution.

In my opinion, however, the research is not exhaustive. Your survey focuses on extracting very general land use classes even though you justify their use. let me try to explain further: evaluating the efficiency in extracting, for example, class "H" - park, Green space etc, without discriminating what kind of vegetation limits the application of your research a bit. The efficiency varies if I detect trees, lawn, bare ground, shrubs, etc.

Perhaps in the results this limitation should be discussed and propose future research developments in this regard.

 

 

Author Response

Thanks so much for your appreciation of our study and professional comments. We agree with you that the identification of different kinds of vegetation is an important part of UGS research. But the present study is more focus on the spatial distribution of UGS in blocks with various land use types. And the different types of vegetation will be the next stage of our future research at fine scale, for which the coarse resolution data is not applicable (such as the 10 m of S2A). We have added this limitation in Page 14, line 463-464.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

This article has been revised according to the comments. I recommend this paper can be published with some revisions.

1. In response to my Main Criticism 1, GLI has been used to extract the UGS. However, more issues emerged. First, there is no scientific basis for using Baidu Map for threshold selection, as the lag in Baidu Map data updates and data errors. Second, why there exist threshold selection? Different thresholds of VC used for DJI, GF1, and S2A may also impact the results. Please elaborate on the threshold selection process and provide further details on how it affects the results. Third, the method of river extraction should be clarified. Moreover, Table 1 shows that the data type of S2A is infrared, while the image taken by S2A included visible light bands. Please check.

2. In response to my Minor Remarks 1, I suggest adding the contents from the response letter to the revised article.

Author Response

We appreciate so much for your recommendation. And we have carefully revised according to the new comments. Sorry we missed some details about the method. Here are the responses for your comment:

1. Just like the threshold for NDVI to identify UGS, there was a threshold for GLI to identify the UGS. It should be noted that VC may not be equal with GR. The reason we calculated the VC was to make the data more easy to understand for its range from 0-1. In order to calculate the GR, the UGS inside the land unit should be extracted first based on a certain threshold, which was selected by repeatedly comparison with the satellite map as well as the high-resolution image of DJI until the accuracy of the extracted UGS was high enough. And the evaluation of accuracy was to select 1000 points randomly and judge whether the extracted UGS was right or not until the Kappa Coefficient was higher than 85%. It was indeed the threshold may not be the same when the images were collected in various conditions, such as sensor types, acquisition time, soil conditions, and so on. Different thresholds may impact the result of UGS extraction along with other correlation result, but the accuracy evaluation has been conducted to make sure the thresholds were suitable in this context so it can still provide some references for other studies. As mentioned in the Section 4.4, line 476-458, the method to define the threshold for UGS extraction will be furtherly explored to increase the accuracy. The revision about this part is in Section 2.4, Line 151-174.

  The method to extract river was NDWI index and the selection of threshold was similar with the UGS extraction as mentioned above. The details have been added in the Section 2.4, Line 160-169. 

  There may be a misunderstanding because of the format in Table 1. In fact, the GF1 and S2A are both “Multispectral, including bands of Red, Green, Blue and Near infrared”. We have changed the format to avoid the problem.

2. We have added the contents from the response letter to the revised manuscript in Section 4.4, Line 469-474.

Author Response File: Author Response.docx