Review Reports - Mapping Essential Urban Land Use Categories in Nanjing by Integrating Multi-Source Big Data

Round 1

Reviewer 1 Report

Mapping Essential Urban Land Use Categories in Nanjing by Integrating Multi-Source Big Data

The following comments/suggestions should be addressed to improve the quality of the paper

The first sentence of the introduction should be substantiated. It should highlight that about 55% of humanity currently lives in urban areas and the proportion is expected to reach almost 70% by 2050. Urban areas also generate over 80% of the global GDP (https://www.worldbank.org/en/topic/urbandevelopment/overview). Also, according to UN-Habitat, “cities consume 78 percent of the world’s energy and produce more than 60 percent of greenhouse gas emissions. Yet, they account for less than 2 percent of the Earth’s surface”. (https://www.un.org/en/climatechange/cities-pollution.shtml)
Lines 78-81: In critiquing EULUC-China, what is the reason for claiming that the buffer thresholds were not accurately assigned. Also, how many parcel samples were used in the study the authors critiqued and why is the sample size deemed insufficient (lines 80-81)?
I suggest merging sections 2 and 3 to be named material and methods section, consisting of 2.1: Study Area, 2.2: Data and sources, and 2.3: Methods’
To what extent can the results be applied more broadly to other settings (generalizable)?
What are the implications of the study for urban land-use planning?
What are the limitations of the study and future research directions?

Author Response

Response to Reviewer 1 Comments

Point 1: The first sentence of the introduction should be substantiated. It should highlight that about 55% of humanity currently lives in urban areas and the proportion is expected to reach almost 70% by 2050. Urban areas also generate over 80% of the global GDP (https://www.worldbank.org/en/topic/urbandevelopment/overview). Also, according to UN-Habitat, “cities consume 78 percent of the world’s energy and produce more than 60 percent of greenhouse gas emissions. Yet, they account for less than 2 percent of the Earth’s surface”. (https://www.un.org/en/climatechange/cities-pollution.shtml)

Response 1: Thanks. We have rewritten the first sentence of the introduction according to the reviewer's suggestion.

Changes in the paper: See the changes in the line 31-36 in the revised version. The changes have been highlighted in red.

“Urban areas accommodate about 55% of the global population and the proportion is expected to reach almost 70% by 2050. Urban areas also generate over 80% of the global GDP [1]. Besides, according to UN -Habitat, cities consume 78% of the world’s energy and produce more than 60 % of greenhouse gas emissions. Yet, they account for less than 2% of the Earth’s surface [2]. For better environmental monitoring, urban planning and government management, identifying accurate and detailed urban land use patterns is important [3].”

Point 2: Lines 78-81: In critiquing EULUC-China, what is the reason for claiming that the buffer thresholds were not accurately assigned. Also, how many parcel samples were used in the study the authors critiqued and why is the sample size deemed insufficient (lines 80-81)?

Response 2:

(1) According to the attributes of OSM data, there are at least seven level of roads, and the roads at different level have different widths. However, in EULUC-China, the buffer thresholds of roads were only divided into major and minor categories. Therefore, we think that the buffer thresholds setting is not detailed enough, which will result in under- or over-segmentation of urban parcels, and thus lead to reducing the purity of land uses within parcels.

(2) 1795 training samples and 869 validation samples were used in EULUC-China, while the country was divided into 440,798 parcels. Therefore, we think the sample size is insufficient. In addition, in terms of the specific cities, taking Nanjing as an example, there were about 8,000 parcels in the city, but the training samples were only 68 and validation samples were 32 in EULUC-China. So we increased the number of training samples in our research, and found that when the number of training samples reaches 350, the accuracy tends to be stable.

In the revised version, we have rewritten this part and added the reasons.

Changes in the paper: See the changes in line 80-84 and they have been highlighted in red in the revised version.

Point 3: I suggest merging sections 2 and 3 to be named material and methods section, consisting of 2.1: Study Area, 2.2: Data and sources, and 2.3: Methods’

Response 3: Thanks for reviewer’s suggestion. We have merged sections 2 and 3 in the revised version.

Changes in the paper: See the changes in the title of sections 2.

Point 4: To what extent can the results be applied more broadly to other settings (generalizable)?

Response 4: This study recommended that the POI data should be preprocessed before they were used and the spatial features from POIs cannot be overlooked in the national or global urban land use mapping, especially in cities with a lot of tall buildings and there are multiple land use functions in a single building, such as Hong Kong, Shen Zhen, etc.

Changes in the paper: This part of the content is in the line 421-424 in the revised version. It has been highlighted in red.

Point 5: What are the implications of the study for urban land-use planning?

Response 5: The urban land use information provided in this study can be applied to help urban planners monitor urban land use changes, analyze the urban structures [45], and make scientific and reasonable planning for the existing urban land resources, so as to promote the healthy development of the city.

Changes in the paper: We have added this information in sections 5 (line 424-427) in the revised version. It has been highlighted in red.

Point 6: What are the limitations of the study and future research directions?

Response 6: The limitations: (1) The extraction of urban parcels needed to be further refined. (2) The problem of mixed land use was not considered. (3) Different types of POIs used the same weights when constructing parcel vectors. (4) The land use information of MPL data was not deeply mined.

Future research directions: (1) A multi-scale analysis unit, such as a single building, the objects or the parcels segmented from the high-spatial-resolution images will be considered in the urban land use. (2) Attempts will be made to address the weighting problem of different types of POIs. (3) the MPL data and mobile phone positioning data can be fused to derive dynamic trajectory data with high spatial and temporal resolutions to better uncover land use functions in other cities.

Changes in the paper: We have summarized the limitations and the future research directions of this study in sections 5 (line 427-435) in the revised version. The change has been highlighted in red.

Author Response File: Author Response.docx

Reviewer 2 Report

The authors propose an approach to improve the mapping of essential urban land use categories, presenting the Nanjing city as a case study. This approach describes methods for generating urban parcels and dealing with the unbalanced distribution of POIs. At the same time, the authors justify the relevance of integrating the POI features to improve the classification performance and show how their approach achieves better results (Overall accuracy) than the base study (EULUC-China).

Most of the manuscript is well written, structured, and justified, and the scientific contribution is well established. However, there are some typographical mistakes throughout the manuscript. For example, Figure 2, the phase 1 is named as Segementing parcels; Figure 8, the titles of plots are named as Veriables importance.

It is recommended to make a complete revision of the manuscript to replace the references in the first person (we, our, us) with the corresponding ones in the third person or the proposal itself. Similarly, verify that all references are cited in the manuscript in both text and Figures.

Introduction. Add a paragraph describing the structure of the article at the end of this section.
Study Area and Datasets. Add the corresponding references to Section 2.1 and Figure 1.
Method. 1. Add all references to data sources, for example, FROM-GLC10 data (line 170), OSM data (line 172), EULUC classification (line 188), Google Earth (line 253), Caret package (line 265), among others, as well as Figure 2 (image based on [10]). 2. Indicate the meaning of each abbreviated feature for Table 2. 3. Add an extended table with all the 134 features indicated in Table 2. 4. Include a link to download the generated corpus (line 220). 5. Include an example of sample editing (line 256). 6. Indicate how the number of samples for training and validation were determined (line 258). 7. Include a link to download the 680 samples collected (line 257). 8. Indicate all the parameters used in the RF model in a table.
Results. 1. Add a figure to show the Overall Accuracy when features get closer to 134. 2. Verify the percentage of Overall Accuracy for the 68 features in Level I (Table 4) with respect to the percentage displayed in Figure 7. 3. Change Table S4 as a main table (Table 6). 4. Describe the results of Table 5 and make a comparison with the results in Table S4. 5. Describe Figure 8b in the same detail as Figure 8a.
Discussion. Move future work to conclusions.
Conclusion. Check the percentage increase provided for Level I (line 397).

Author Response

Response to Reviewer 2 Comments

Point 1: Most of the manuscript is well written, structured, and justified, and the scientific contribution is well established. However, there are some typographical mistakes throughout the manuscript. For example, Figure 2, the phase 1 is named as Segementing parcels; Figure 8, the titles of plots are named as Veriables importance.

Response 1: Thanks. We have revised it.

Changes in the paper: See the changes in the Figure 2 and Figure 9. It should be noted that Figure 8 is changed to Figure 9 due to the addition of a new figure in the revised version.

Point 2: It is recommended to make a complete revision of the manuscript to replace the references in the first person (we, our, us) with the corresponding ones in the third person or the proposal itself. Similarly, verify that all references are cited in the manuscript in both text and Figures.

Response 2: Thanks. We have done it.

Changes in the paper: See the changes in the line 17, 21, 180, 182, 183, 198-200, 246-248, 260-262, 266, 286-288, 344-346, 355, 358, 364-364, 394, 397-399, 401 and the changes have been highlighted in red.

Point 3: Introduction. Add a paragraph describing the structure of the article at the end of this section.

Response 3: Thanks. We have added it.

Changes in the paper: See the changes in the last paragraph of the introduction (line 92-96). The changes have been highlighted in red in the revised version.

“The structure of this article is as follows. Section 1 describes the background of this research and reviews related work on urban land use classification. The data sources, methodology and techniques are described in detail in Section 2. The experimental results are illustrated in Section 3. Section 4 makes a meaningful discussion of the results and analyzes the future research directions. Finally, this article is summarized in Section 5.”

Point 4: Study Area and Datasets. Add the corresponding references to Section 2.1 and Figure 1.

Response 4: Thanks. We have added the reference 29 to Section 2.1 and reference 12, 31 to Figure 1.

Changes in the paper: See the changes in the line 104 and 108. The changes have been highlighted in red in the revised version.

Point 5: Method. 1. Add all references to data sources, for example, FROM-GLC10 data (line 170), OSM data (line 172), EULUC classification (line 188), Google Earth (line 253), Caret package (line 265), among others, as well as Figure 2 (image based on [10]).

Response 5: Thanks. We have added the reference 25 to FROM-GLC10 data (line 159,161), reference 32 to OSM data (line 151), reference 12 to EULUC classification (line 172), reference 30 to Google Earth (line 267), reference 38 to Caret package (line 283), reference 12 to Figure 2 (line 170). Also, we have verified that all references are cited in the manuscript in both text and Figures.

Point 6: Indicate the meaning of each abbreviated feature for Table 2.

Response 6: Thanks. We have indicated it.

Changes in the paper: See the changes in Table 2 and line 273-274 in the revised version. They have been highlighted in red.

Point 7: Add an extended table with all the 134 features indicated in Table 2.

Response 7: Thanks. Since the table containing 134 features is relatively large, we have added it to the supplementary material (Table S1).

Changes in the paper: See the changes in Table S1 in supplementary material and line 224 in the revised version. They have been highlighted in red.

Point 8: Include a link to download the generated corpus (line 220).

Response 8: Thanks. We have added a link to download it.

Changes in the paper: See the link in line 243 in the revised version. It has been highlighted in red.

“The results of the Parcel-POI corpus, POI category vectors and the parcel vectors can be downloaded from https://pan.baidu.com/s/1jb9lOpqFcqNlVQZ4UrC8rA.”

Point 9: Include an example of sample editing (line 256).

Response 9: Thanks. We have added a Figure 6 to shown the example of sample editing and analyzed it.

Changes in the paper: See the Figure 6 and the analyse in line 270-272. The changes have been highlighted in red.

“For example, as shown in Figure 6, the parcel contained educational and medical two land use types before editing (Figure 6a), while after editing, it was divided into two parcels with high-purity (Figure 6b).”

Point 10: Indicate how the number of samples for training and validation were determined (line 258).

Response 10: Thanks. According to a ratio about 3:1[35], 500 samples were used for training and the remaining 180 samples were used for validation. We have added this information in section 2.3.5.

Changes in the paper: See the changes in line 273-274 in the revised version. They have been highlighted in red.

Point 11: Include a link to download the 680 samples collected (line 257).

Response 11: Thanks. We have added a link to download samples in the revised version.

Changes in the paper: See the link in line 276. It has been highlighted in red.

“The samples can be downloaded from https://pan.baidu.com/s/1jb9lOpqFcqNlVQZ4UrC8rA.”

Point 12: Indicate all the parameters used in the RF model in a table.

Response 12: Thanks for reviewer’s suggestion. We used the random forest model to classification in R. The two most important parameters "ntree "and" mtry " were set to 500 and 3 respectively, and the remaining parameters were the default values. We have added “ntree” and “mtry” parameters information in the modified version.

Changes in the paper: See the changes in line 285-286. The have been highlighted in red.

Point 13: Results. 1. Add a figure to show the Overall Accuracy when features get closer to 134.

Response 13: Thanks. In the revised version, the optimal features were selected from the total of 134 features by using the recursive feature elimination algorithm with the help of Caret package [38] in R. The new Figure 8 which showed the Overall Accuracy when features get closer to 134 was added and the old figure was removed.

Changes in the paper: See the changes in Figure 8.

Point 14: Verify the percentage of Overall Accuracy for the 68 features in Level I (Table 4) with respect to the percentage displayed in Figure 7.

Response 14: The accuracy in Figure 8 was generated by the OOB error of RF model when using the recursive feature elimination algorithm to select features, while the overall accuracy in Table 4 was calculated based on 180 validation samples, so the accuracy was different. We have highlighted this information in the revised version. (It should be noted that Figure 7 is changed to Figure 8 due to the addition of a new figure.)

Changes in the paper: See the changes in 304 and 307. The changes have been highlighted in red.

Point 15: Change Table S4 as a main table (Table 6).

Response 15: Thanks. We have changed it in the revised version.

Changes in the paper: See the change in page of 13. It has been highlighted in red.

Point 16: Describe the results of Table 5 and make a comparison with the results in Table S4.

Response 16: Thanks. We have done it.

Changes in the paper: See the changes in the line 314-319 in the revised version. They have been highlighted in red.

“The residential, industrial, and public land use in Level I achieved a higher accuracy with both producer’s accuracy (PA) and user’s accuracy (UA) of more than 80%, while the commercial land use had a relatively lower user’s accuracy of 68%. As of the Level II category, the industrial land use had a highest accuracy with both producer’s accuracy (PA) and user’s accuracy (UA) of 92%, while the business land use had a lowest accuracy with UA of 60% and PA of 64%.”

Point 17: Describe Figure 8b in the same detail as Figure 8a.

Response 17: Thanks. We have described Figure 8b in the detail. However, it should be noted that Figure 8 is changed to Figure 9 due to the addition of a new figure in the revised version.

Changes in the paper: See the changes in the line 325-328 in the revised version. They have been highlighted in red.

“In the Level II class, six POI spatial features (POIspa_8, POIspa_1, POIspa_2, POIspa_7, POIspa_6, POIspa_9), four POI frequency features (POIp502, POIp501, POI503, POIp201), three texture features (b2entstd, b2cormean, b3entmean) and two spectral features (ndvimean, b4mean) were selected in the top 15 (Figure 9b).”

Point 18: Discussion. Move future work to conclusions.

Response 6: Thanks for reviewer’s suggestion. The last three paragraphs of the discussion section are all about the analysis of future work, and there are many contents. Combined with the suggestions of another reviewer, we finally summarized the future work and added it in the conclusion. Thanks again for the comments of the reviewers.

Changes in the paper: See the changes in the line 430-435 in the revised version. They have been highlighted in red.

“In the future work, a multi-scale analysis unit, such as a single building, the objects or the parcels segmented from the high-spatial-resolution images will be considered in the urban land use. Attempts will be made to address the weighting problem of different types of POIs. Meanwhile, the MPL data and mobile phone positioning data can be fused to derive dynamic trajectory data with high spatial and temporal resolutions to better uncover land use functions in other cities.”

Point 19: Conclusion. Check the percentage increase provided for Level I (line 397).

Response 19: Thanks. We have checked and revised it.

Changes in the paper: See the change in the line 418. The change has been highlighted in red in the revised version.

Author Response File: Author Response.docx

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.