Quantifying the Effects of Urban Form on Land Surface Temperature in Subtropical High-Density Urban Areas Using Machine Learning
Round 1
Reviewer 1 Report
Reviewer comments
This paper described a study in Ningbo, China using built environment data base and LST dataset obtained at 10:55 on Feb.3 and Jul. 23, 2017. Observations were made on the correlation between urban form attributes and LST intensity. Suggestions were made to urban planning in relation to urban thermal environment.
The study is timely and important to planning policy. The analysis is thorough compared with peer studies. The scientific rigor of this paper are limited by three critical drawbacks:
1. A critical limitation in the methodology is the ignorance of the temporal aspect of LST, or the time dimension of urban thermal environment. Since surface temperature vary considerably by the hour depending on the solar angle, cloud cover and ambient air temperature, etc., it is unclear why the observation at 10:55 should represent the seasonal condition (summer and winter). Usually urban surfaces experience peak temperature in 2 to 3 hours after the peak solar radiation at the middle of the day due to thermal massing effects. The urban heat island effect is the greatest in the evenings from some literature.
2. A second weakness the lack of clarity on two concepts - LST and the urban thermal environment. In the methodology section, the authors used the concept of LST, which is the dependent variable used in regression anslysis; while in the article's title, discussion and conclusion, the urban thermal environmen (or UHI used interchangably) is used. The lack of clarity risks misguiding urban planning practice, for instance, LST dataset is limited in observing building wall temperature. It over samples rooftop surfaces which have limited impact on the thermal environment of pedestrian-level. Meanwhile, LST cannot survey building wall temperature nor road surfaces obstructed by buildings, which arguably have a higher contribution to human scale thermal environment than roof surfaces.
3. The measurement of human activities HTL is not a good measure of anthropogenic heat. It doesn't include industrial activities nor domestic cooking or the use of air conditioning.
The author should address the above drawbacks; A rigorous analysis of these uncertainties is due before drawing urban planning implications on UHI mitigation and urban thermal environment.
.
Other issues:
Line 371 - "Therefore, urban planners and managers should pay more 371 attention on the arrangement of buildings and ecological composition in high-density urban center, 372 rather than restriction of related human activities in order to improving the urban thermal 373 environment and mitigating the UHI effect." This statement is unsupported and potentially misleading - in high density cities such as Hong Kong, Shanghai, or Singapore, the contribution of anthropogenic heat may exeed those of the urban form due to the extreme high density
There are uncertainties associated with the use of LST in China due to the presence of air -- the concentration of aerosols in urban area will distort measurement in the infrared spectrum.
Author Response
The author firstly thanks the reviewer for your valuable comments!
1. A critical limitation in the methodology is the ignorance of the temporal aspect of LST, or the time dimension of urban thermal environment. Since surface temperature vary considerably by the hour depending on the solar angle, cloud cover and ambient air temperature, etc., it is unclear why the observation at 10:55 should represent the seasonal condition (summer and winter). Usually urban surfaces experience peak temperature in 2 to 3 hours after the peak solar radiation at the middle of the day due to thermal massing effects. The urban heat island effect is the greatest in the evenings from some literature.
Response: Thank you for your advice! Yes, it is indeed a fact that urban surface temperature varied with the changes of external environment. The temporal variation of LST should be taken into consideration when we analyze the impact of physical urban from features on urban thermal environment. In situ data were often used in early UHI studies. However, in situ data are usually subject to governmental or costly restrictions. Moreover, the sparse and limited spatial coverage of in situ data make them unsuitable for large-area analyses. Remote sensing (RS) technology provides a practical and promising technique for monitoring large-scale UHIs. Unfortunately, Remote Sensing (RS) data (e.g. Landsat TM/ETM/OLI+, MODIS LST and ASTER imagery) can only capture the instantaneous temperature field at one or several specific time. Thus, most of current UHI studies have to select one or more cloudless imagery to represent typical seasons or annum. Like previous works on UHIs, we obtained two cloud-free Landsat 8 thermal infrared sensor (TIRS) imagery to illustrate the spatial distribution of LST in both summer and winter. The main reason is short of available data for Ningbo city due to the cloudy climate all year round on the eastern coast region of China.
2. A second weakness the lack of clarity on two concepts - LST and the urban thermal environment. In the methodology section, the authors used the concept of LST, which is the dependent variable used in regression anslysis; while in the article's title, discussion and conclusion, the urban thermal environmen (or UHI used interchangably) is used. The lack of clarity risks misguiding urban planning practice, for instance, LST dataset is limited in observing building wall temperature. It over samples rooftop surfaces which have limited impact on the thermal environment of pedestrian-level. Meanwhile, LST cannot survey building wall temperature nor road surfaces obstructed by buildings, which arguably have a higher contribution to human scale thermal environment than roof surfaces.
Response: Thank you for your advice! As you say, the land surface temperature (LST) can not entirely reflect the 3D features of urban thermal environment (such as the temperature field of building's facades). The two concepts have obviously differences. Based on such consideration, we have changed the term of “urban thermal environment” into “land surface temperature (LST)” in each section of our study. Please see lines 2-3, 27.
3. The measurement of human activities HTL is not a good measure of anthropogenic heat. It doesn't include industrial activities nor domestic cooking or the use of air conditioning.
Response: Thank you for your advice! The nighttime lights products provided by National Oceanic and Atmospheric Administration (NOAA)/National Geophysical Data Center (NGDC) have been widely used in recent studies for modeling the spatial distribution of population density and energy consumption. Jing et al. (2015) confirmed that nighttime lights are closely correlated with the local economic activities. For example, stronger intensities of light are likely to be situated at commercial areas. Furthermore, Chen et al. (2012) found a strong linear relationship between the digital number of nighttime lights and anthropogenic heat flux density over China. The correlation coefficient R2 between the mean anthropogenic heat flux density (AHFD) and the mean digital number was larger than 0.9 as follow:
Based on above studies, we suggested that energy consumption related anthropogenic heat intensity should have a close relation with NTL (night-time light) intensity. Although this may exist a certain error, the NTL intensity could be a spatial proxy for human activities or anthropogenic heat.
References:
X. Jing, X. Shao, C. Cao, X. Fu, L. Yan. Comparison between the Suomi-NPP day-night band and DMSP-OLS for correlating Socio-Economic variables at the provincial level in China. Remote Sens., 2015, 8 (1): 17.
Bing Chen, Guangyu Shi, Biao Wang, Jianqi Zhao, Saichun Tan. Estimation of the anthropogenic heat release distribution in China from 1992 to 2009. Acta Meteorologica Sinica, 2012, 26(4): 507–515.
The author should address the above drawbacks; A rigorous analysis of these uncertainties is due before drawing urban planning implications on UHI mitigation and urban thermal environment.
Other issues:
Line 371 - "Therefore, urban planners and managers should pay more 371 attention on the arrangement of buildings and ecological composition in high-density urban center, 372 rather than restriction of related human activities in order to improving the urban thermal 373 environment and mitigating the UHI effect." This statement is unsupported and potentially misleading - in high density cities such as Hong Kong, Shanghai, or Singapore, the contribution of anthropogenic heat may exeed those of the urban form due to the extreme high density.
Response: Thank you for your advice! Yes, we agree with your suggestion. The results obtained from our study may be a local phenomenon, rather than a universal phenomenon. So we revised the expression of this section. Please see line 385. In the study area, two urban form aspects including ecological infrastructure and building morphology were identified as the dominant factors for the change of LST, while human activities intensity (NTLI) and POI density (POID) have relative weaker impacts on the LST variations.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Dear Authors,
thanks for your interesting subject of your paper.
My main comment is the following: maybe I missed it, but i didn't see the quantification of the variables you included in the regression.
Apart from this major comment/request, I saw that there is space for improvement in the literature review on the topic. You could show more alternative methods/case studies to find out influences on UHI from urban form factors.
Regards
Author Response
The author firstly thanks the reviewer for your valuable comments!
1.My main comment is the following: maybe I missed it, but i didn't see the quantification of the variables you included in the regression.
Response: Thank you for your advice! We have described each explanatory variables in detail. Please see lines 135-165. The seven explanatory variables were calculated using ArcGIS 10.2 software.
2.Apart from this major comment/request, I saw that there is space for improvement in the literature review on the topic. You could show more alternative methods/case studies to find out influences on UHI from urban form factors.
Response: Thank you for your advice! We have added several related previous studies (e.g. conducted by prof. Yang) in the section of literature review. Please see lines 56-59, and 73-75.
Author Response File: Author Response.docx
Reviewer 3 Report
Scale should inserted in the figure 1
Need more discussion in the introduction part regarding Urban heat Island, urban temperature and LST changes with case studies from other parts of the world.
I suggested following paper to considered.
https://rgs-ibg.onlinelibrary.wiley.com/doi/full/10.1111/j.1475-4959.2007.232_3.x
http://www.dept.ku.edu/~biomet/KU_Biometeorology_Lab/Publications_files/oleson_cc13.pdf
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0208949
http://repositori.uji.es/xmlui/bitstream/handle/10234/165600/Tran_2017_Characterizing.pdf?sequence=1
Author Response
The author firstly thanks the reviewer for your valuable comments!
1.Scale should inserted in the figure 1
Response: Thank you for your advice! We have added the scale for the Fig.1 (a).
2.Need more discussion in the introduction part regarding Urban heat Island, urban temperature and LST changes with case studies from other parts of the world.
Response: Thank you for your advice! We have added more discussion in the introduction part based on your recommended papers. Please see lines 55-59.
Author Response File: Author Response.docx
Reviewer 4 Report
General: In this paper, authors investigated the quantities effects of urban form regarding the variations of land surface temperature. Specifically, five aspects, ranging from human activity to ecological infrastructure, have been considered. Experimental results on a case study city-Ningbo were provided, which gives consistent conclusions with start-of-art works. While the impact of urban form problem is of interest, the major drawback of this paper is unclear contribution. The paper work is based on two well-known machine-learning techniques (OLS and RF), while only limited accuracy is achieved (more than half). Although the explanatory power for each urban form is detailed discussed, the conclusion as well as comparison might lack of support, since almost half of LST variation is still unclear and have been explained incorrectly. However, experiment results show similar trends with most of existing works, authors need to clarify the contributions of this work and prove the experiment design to be correct. The second drawback concerns that the manuscript deserve carefully proofreading, with many mistakes (typos etc…). Below, you could find my comments with respect to the rest of paper.
For these reasons, I cannot recommend the paper for publication.
1. With respect to the experiment setup, the authors should give the reason why the model only explain more than 60% of the variations LST in summer and more than 50
% of the variations LST in winter. Is that means the rest of LST variations is independent to these urban form aspects or is still unclear? The conclusions and discussion within this paper would be reliable only if the rest unexplained LST variations is independent, which deserves more evidence and support materials.
2. In this paper, five representative aspects of urban form have been included, while the detail consideration need to be mentioned in introduction section. From line 69-75, authors introduced existing fundamental aspects of urban form, and further comparison is necessary to highlight the comprehensiveness of the employed aspects.
3. With respect to the writing, “urban form” as the key term in this paper, while there are many typos “urban from”, in line 12, line 169, line 345, line 350, line 399 etc. Other typos such as line 98 “64km” should be “64km2”. Equation (1) as well as line 164 is inconsistent. Authors need to improve the paper with more carefully proofreading.
Author Response
The author firstly thanks the reviewer for your valuable comments!
1. With respect to the experiment setup, the authors should give the reason why the model only explain more than 60% of the variations LST in summer and more than 50
% of the variations LST in winter. Is that means the rest of LST variations is independent to these urban form aspects or is still unclear? The conclusions and discussion within this paper would be reliable only if the rest unexplained LST variations is independent, which deserves more evidence and support materials.
Response: Thank you for your advice! It is true that the urban LST is influenced by many factors, such as landscape composition and configuration, climate background, and human activities. As we descripted in the introduction part, currently studies focus on investigating the effects of landscape patterns and urban form on LST at the city-level. For example, Yin et al. (2018) selected three urban form indicators including sky view factor (SVF), building density, and floor area ratio to examine the effects of urban form on LST in Wuhan city based on spatial regression models. The results showed that the R2 can reach up to 0.34~0.62. Another study conducted by Chun et al. (2014) aims to explore the urban determinants of the UHI. The urban variables include building features (building ground floor area, height), sky view and solar radiations, the Normalized Difference Vegetation Index (NDVI), and the area of water bodies. The R2 obtained from OLS regression model were up to 0.647-0.835. The general spatial model (GSM) showed higher performance.
The objectives of our paper are to determine the extent to which urban form metrics influence the LST variations considering the differences of seasonality and observation scales. Firstly, we redefined and disassembled the urban form into five aspects as shown in Fig.3. Seven representative urban form variables were then selected in this paper as follows: (1) Human activity: NTL (night-time light) intensity. (2) Building morphology: building density, and floor area ratio; (3) Transportation system: road density; (4) Public infrastructure: POI (point of interest) density; (5) Ecological infrastructure: water surface ratio, and NDVI. The other influencing factors did not considered in this study.
The statistical parameters of R2, MAE, and RMSE were then calculated using 10-fold cross-validation results were used to compare the performance of the OLS and RF models in different seasons and observation scales. Finally, reliable regression model were received based on model validation. At the three observation scales, the OLS regression models explained more than half (62%~65% in summer and 52%~54% in winter) of the LST variations by selected urban form metrics. However, the RF models showed higher R2 (0.91~0.95) and lower RMSE/MAE values. It can be seen that the RF method showed higher performance than the OLS method, and could capture the most of LST variations. Thus, we suggest that the results obtained from the RF models were reliable and credible.
In order to get more accurately prediction model, we need put more variables into analysis process in future research.
References:
1. Yin, C.H.; Yuan, M.; Lu, Y.P.; Huang, Y.P.; Liu, Y.F. Effects of urban form on the urban heat island effect based on spatial regression model. Sci. Total Environ. 2018, 634, 696–704.
2. B. Chun, J.-M. Guldmann. Spatial statistical analysis and simulation of the urban heat island in high-density central cities. Landscape and Urban Planning, 2014, 125: 76–88.
2. In this paper, five representative aspects of urban form have been included, while the detail consideration need to be mentioned in introduction section. From line 69-75, authors introduced existing fundamental aspects of urban form, and further comparison is necessary to highlight the comprehensiveness of the employed aspects.
Response: Thank you for your advice! We have added several relevant previous studies. Please see lines 56-59, and 73-75. As far as we know, the comprehensiveness of the employed studies in urban form were still limited and unilateral until now. Previous works generally highlighted one specific aspect of urban form. That is why we conducted this comprehensive work to put all five aspects of urban form in the analysis process. It is benefits to distinguish the different effects of urban form metrics on the LST variations.
3. With respect to the writing, “urban form” as the key term in this paper, while there are many typos “urban from”, in line 12, line 169, line 345, line 350, line 399 etc. Other typos such as line 98 “64km” should be “64km2”. Equation (1) as well as line 164 is inconsistent. Authors need to improve the paper with more carefully proofreading.
Response: Thank you for your advice! It is sorry about that several spelling mistakes exist in the manuscript. We have revised them carefully in the whole manuscript.
Author Response File: Author Response.docx
Reviewer 5 Report
The topic of this manuscript is interesting to me, and generally it is well written. Here are my major concerns:
1. About the Figure 1, it may be better to add a subfigure to show the location of the Ningbo city in China.
2. About the Figure 2, where did this figure come from? Is it from a reference paper? If not, how did the authors plot this figure? More descriptions about the temperature data plotted in this figure needs to be added.
3. Section 2.2: Why not use the monthly NPP-VIIRS nighttime light data in July/February, 2017? Please note that the Landsat data is in July/February, 2017, not 2015. The authors need to give more clarification.
Another question is that the data sources of the seven urban form metrics are lacking, please add descriptions about where did you get these data.
4. Figures of the seven urban form metrics are lacking in this manuscript, please add these relevant figures, it is helpful to understand the effects of urban form on LST.
5. Figure 6-8: please keep consistent between the y-axis label and the figure title, predicted? Estimated?
6. Section 3.2 to 3.4 should be reorganized, the authors need to reconsider the logic of this part. For example, in section 3.2, model validation should be emphasized, not the explanatory power for each urban form aspects on LST variations; section 3.3 and 3.4 may be better to be one section, both showing results of variable importance of urban form metrics on the LST using OLS or random forest regression. Furthermore, the focus of this part should be displaying the results, so move the comparisons with existed studies to the discussion part.
7. In the Figures 10a and 10h, LST increased with NDVI, why? Please explain this phenomenon.
Other specific comments:
Line 78 and 117: urban from? I think it’s a mistake, please change it to urban form, and check other places across the whole manuscript.
Line 198-201 and Figure 4: RMSE is not appropriate in this section as RMSE occurred in the next section, please move these sentences and Figure 4 to section 2.5.
About the Figure 5, the title is so long, please shorten the title, like spatial variation of LST in summer (a) and winter (b) in Ningbo city.
Please change “variations LST” to “LST variations” in the whole manuscript.
There are some other grammatical mistakes in the manuscript. Authors need to proofread and make corrections where necessary.
Author Response
The author firstly thanks the reviewer for your valuable comments!
1. About the Figure 1, it may be better to add a subfigure to show the location of the Ningbo city in China.
Response: Thank you for your advice! We have inserted a subfigure in Fig.1 to show the location of the Ningbo city in China.
2. About the Figure 2, where did this figure come from? Is it from a reference paper? If not, how did the authors plot this figure? More descriptions about the temperature data plotted in this figure needs to be added.
Response: Thank you for your advice! We derived daily mean temperature of Yinzhou station in Ningbo city during 1980-2010 from the China Meteorological Data Sharing Service System (http://dataNaNa.cn/). To present the month variation of air temperature in Ningbo, this data set was used to plot the Fig.2. We have revised the figure note. Please see line 119.
3. Section 2.2: Why not use the monthly NPP-VIIRS nighttime light data in July/February, 2017? Please note that the Landsat data is in July/February, 2017, not 2015. The authors need to give more clarification.
Response: Thank you for your advice! Yes, we only derived the annual composite NPP-VIIRS nighttime light data in 2015. Because the NOAA data download Website links (https://ngdc.noaa.gov/eog/viirs/download_dnb_composites.html) cannot be logged in currently. So it is supposed to that slightly changes for night-time population were occurred during the period of 2015-2017.
Another question is that the data sources of the seven urban form metrics are lacking, please add descriptions about where did you get these data.
Response: Thank you for your advice! We have added the detail descriptions of seven urban form metrics including data sources and calculation process. Please see lines 135-165.
4. Figures of the seven urban form metrics are lacking in this manuscript, please add these relevant figures, it is helpful to understand the effects of urban form on LST.
Response: Thank you for your advice! We have added the spatial distribution map of selected urban form metrics in the manuscript. Please see Fig 4.
5. Figure 6-8: please keep consistent between the y-axis label and the figure title, predicted? Estimated?
Response: Thank you for your advice! We have changed the terms of “estimated” into “predicted” in order to keeping consistent between the y-axis label and the figure title. Please see the title of Fig. 7, 8 and 9.
6. Section 3.2 to 3.4 should be reorganized, the authors need to reconsider the logic of this part. For example, in section 3.2, model validation should be emphasized, not the explanatory power for each urban form aspects on LST variations; section 3.3 and 3.4 may be better to be one section, both showing results of variable importance of urban form metrics on the LST using OLS or random forest regression. Furthermore, the focus of this part should be displaying the results, so move the comparisons with existed studies to the discussion part.
Response: Thank you for your advice! As you say, we have merged section 3.3 and 3.4 into one part. Please see line 289.
In addition, we revised the contents of section 3.2 “Model estimation and validation”. This part only aims to estimate and compare the model performances for OLS and RF. Table 2 and its explanations were combined into section 3.3. Please see line 290-304.
7. In the Figures 10a and 10h, LST increased with NDVI, why? Please explain this phenomenon.
Response: Thank you for your advice! As shown in Fig. 11, the LST increased with the increasing of NDVI (<0.8%) in the initial stage. However, when the values of NDVI were larger than 0.8%, the LST declined gradually. This is an interesting phenomenon. We suggested it may related with dominant land use types. Smaller value of NDVI reflects that the landscape units were occupied by non-vegetated area, so the LST may increase in this stage.
Other specific comments:
Line 78 and 117: urban from? I think it’s a mistake, please change it to urban form, and check other places across the whole manuscript.
Response: Thank you for your advice! It's a spelling mistake. We have revised them in the whole manuscript.
Line 198-201 and Figure 4: RMSE is not appropriate in this section as RMSE occurred in the next section, please move these sentences and Figure 4 to section 2.5.
Response: Thank you for your advice! This sentence and Fig. 4 aims to present the results of model tuning parameter. So these should be put in section 2.4 “Model estimation”, not the section 2.5 “Model validation”.
About the Figure 5, the title is so long, please shorten the title, like spatial variation of LST in summer (a) and winter (b) in Ningbo city.
Response: Thank you for your advice! We have tried to shorten the title of Fig.6. Please see lines 258-260.
Please change “variations LST” to “LST variations” in the whole manuscript.
Response: Thank you for your advice! We have replaced “variations LST” with “LST variations” in the whole manuscript (total of 11 places).
There are some other grammatical mistakes in the manuscript. Authors need to proofread and make corrections where necessary.
Response: Thank you for your advice! We have carefully proofread and make corrections through the whole manuscript.
Author Response File: Author Response.docx
Round 2
Reviewer 4 Report
Reviewer Comment:
General: At this time, authors addressed several problems. While with respect to the first comment, I think it is still not completely clear. For the detail concern, please find it in the following.
1. Regarding two regression LST models using in this paper, since authors mentioned RF significantly outperforms OLS, why in line 18-19 as well as other part of paper use the statement of “this study indicate that more than half (62%~65% in summer and 52%~54% in winter) of the LST variations”? Shouldn’t it be more than 90% of the LST variations using RF?
2. Similar to the last point, further explanation about the advantages and disadvantages of OLS and RF deserve further discussion, the authors need to explain why RF yields such superior performance as well as the limitation for OLS methods in more details. For example, a similar table for OLS as Table 2 might be helpful.
Comments for author File: 
Comments.pdf
Author Response
The author firstly thanks the reviewer for your valuable comments!
General: At this time, authors addressed several problems. While with respect to the first comment, I think it is still not completely clear. For the detail concern, please find it in the following.
1. Regarding two regression LST models using in this paper, since authors mentioned RF significantly outperforms OLS, why in line 18-19 as well as other part of paper use the statement of “this study indicate that more than half (62%~65% in summer and 52%~54% in winter) of the LST variations”? Shouldn’t it be more than 90% of the LST variations using RF?
Response: Thank you for your advice! Yes, we have revised the statement of this issue. Please see lines 18-19 in abstract, 419-420 in conclusion.
2. Similar to the last point, further explanation about the advantages and disadvantages of OLS and RF deserve further discussion, the authors need to explain why RF yields such superior performance as well as the limitation for OLS methods in more details. For example, a similar table for OLS as Table 2 might be helpful.
Response: Thank you for your advice! The RF algorithm is a non-parametric statistical technique that is capable of synthesizing regression or classification functions based on discrete or continuous datasets. RF has the advantage in modeling the complex nonlinear relationships between the predictors and the response variable instead of detailed numerical expressions. It has successfully been used to enhance the prediction accuracy in the field of ecology.
However, this machine learning approaches is considered black-box with little inferential value. We can’t obtain the regression coefficient from RF model unlike OLS method. Fortunately, RF models computed two qualitative measures that describe the relative importance of the predictor variables: the Increased Mean Square Error (%IncMSE) and Increased Impurity Index (IncNodePurity). We can use the Variable Importance Analysis method to determine the extent to which urban form metrics influence the LST variations. In addition, Partial dependence plots obtained from RF help visualize the average partial relationship between the predicted response and one or more features. Thus, Fig. 10 and 11 presented the main results from RF model.
In the manuscript, we have introduced the fundamentals of design theory of RF method in lines 204-214. Meanwhile, we stated the superior performance of RF in modeling the urban LST patterns. Please see lines 417-418.
Author Response File: Author Response.docx
Reviewer 5 Report
Most of my concers has been adressed well by the authors. However, the authors seem to misunderstand the following two concers:
Please add more descriptions about where did you get these urban form data. I mean that the authors need to tell the readers where they can get these data, from a website? from a data ceter? or other places?
Line 216-219 and Figure 5: RMSE is not appropriate in this section as RMSE occurred in the next section, please move these sentences and Figure 5 to section 2.5. I mean you can not use RMSE in section 2.4, because you define RMSE in section 2.5.
Author Response
The author firstly thanks the reviewer for your valuable comments!
Most of my concers has been adressed well by the authors. However, the authors seem to misunderstand the following two concers:
1.Please add more descriptions about where did you get these urban form data. I mean that the authors need to tell the readers where they can get these data, from a website? from a data ceter? or other places?
Response: Thank you for your advice! Seven urban form metrics, including BD, FAR, RD, POID, NDVI, WSR, and NTLI, were specified as the explanatory variables in estimation model. The details of data sources and calculating process can be found in lines 134-164. The sources of these data listed as follows:
(1) NTL intensity (NTLI): Night-time light imagery in 2015 were obtained from website of NOAA/NGDC (https://ngdc.noaa.gov/eog/viirs/download_dnb_composites.html).
(2) Building density (BD): Building footprint polygon were obtained from the Ningbo Survey and Geographic Information Bureau in ESRI shapefile format.
(3) Floor area ratio (FAR): The number of floor levels for each building footprint polygon were obtained from the Ningbo Survey and Geographic Information Bureau in ESRI shapefile format.
(4) Road density (RD): The road network polyline were obtained from the Ningbo Survey and Geographic Information Bureau in ESRI shapefile format.
(5) POI density (POID): The POIs in 2017 were retrieved from the Location-based Service on the Baidu Map Open Platform (http://lbsyun.baidu.com/index.php?title=lbscloud).
(6) NDVI: Water body information was extracted from Landsat 8 image in 2017 using the Normalized Difference Water Index (NDWI).
(7) Water surface ratio (WSR): Water body information was extracted from Landsat 8 image in 2017.
2.Line 216-219 and Figure 5: RMSE is not appropriate in this section as RMSE occurred in the next section, please move these sentences and Figure 5 to section 2.5. I mean you can not use RMSE in section 2.4, because you define RMSE in section 2.5.
Response: Thank you for your advice! You are right. We have move the sentences and Fig.5 to section 2.5. Please see lines 232-238.
Author Response File: Author Response.docx
