Classifying Urban Climate Zones (UCZs) Based on Spatial Statistical Analysesâ€
Round 1
Reviewer 1 Report
Overall, I think this is a good paper. However, I have a few concerns that I hope the authors can address before publication.
(1) How do you evaluate the classification results of urban climate zones?
(2) As the authors stated that the derived air temperature maps were at 100 m spatial resolution, I was wondering what is the spatial distance between weather stations? If the distance between weather stations is larger than 100 m, I did not think that the interpolation can really provide air temperature maps at a spatial resolution of 100 m. In the latter case, it is just an "interpolation" and may not reflect the real temperature variations in cities particularly as there is a strong heterogeneity in thermal components within urban regions. Please clarify it and add some discussions related to the accuracy of the derived air temperatures.
Author Response
Response to Reviewer 1 Comments
Point 1: How do you evaluate the classification results of urban climate zones?
Response 1:
(Lines 178-195 and line 234-237) The air temperature distributions of each UCZ delineated by clustering analysis were investigated. In addition, the differences of air temperature distribution were identified applying an ANOVA test. (Line 244) As a result, we found that the classification of UCZs in this study shows similar spatial patterns with the air temperature analysis results.
Point 2: As the authors stated that the derived air temperature maps were at 100 m spatial resolution, I was wondering what is the spatial distance between weather stations? If the distance between weather stations is larger than 100 m, I did not think that the interpolation can really provide air temperature maps at a spatial resolution of 100 m. In the latter case, it is just an "interpolation" and may not reflect the real temperature variations in cities particularly as there is a strong heterogeneity in thermal components within urban regions. Please clarify it and add some discussions related to the accuracy of the derived air temperatures.
Response 2:
It will be ideal to have the average spacing of AWSs equal to the analysis spatial resolution.
(Lines 123-133) The average spacing of AWSs in the case study area is 1,087m, which is much shorter than other metropolises. Until now, there has been no research with the density of AWSs as attempted in this study (246 AWSs / 605 km2) in any city in the world. Of course, as noted by the reviewer, there still exists limitations such as reflecting real air temperature with 100m spatial resolution using interpolation methods. In order to reflect the heterogeneity of urban spaces, this study applied universal kriging interpolation methods which consider effects by elevation and water spaces were applied to prepare the air temperature map.
(Lines 143-150) In addition, to verify the accuracy of the interpolated results, RMSPEs were calculated and compared with data from 26 AWSs which were not used in the interpolation process. As presented in line 151, the air temperature maps which show low RMSPE were used to prepare the final air temperature map.
(Lines 268-274) Finally, the limitations and improvement potentials of the universal kriging methods were explained in Discussion and Conclusions.
Author Response File: 
Author Response.docx
Reviewer 2 Report
This is a fairly good paper that can be improved by addressing the following issues.
Line 71: The author says some statistical approaches have been applied when trying to understand the relationship between UHI and urban spatial relationships. Which approaches have been applied?
Lines 169 - 170: Why was K-means clustering algorithm chosen over others? A little more justification is required.
Section 2.4 is very thin. It needs to detail the classification procedure as well as the sensitivity test. The author simply says K-mean and ANOVA techniques were used without justifying these techniques and explaining how they were applied.
K-means does not automatically detect the number of groups in the data. How did the author determine the appropriate number of groups in this analysis?
Author Response
Response to Reviewer 2 Comments
Point 1: The author says some statistical approaches have been applied when trying to understand the relationship between UHI and urban spatial relationships. Which approaches have been applied?
Response 1:
(Line 72) The manuscript was modified by using multiple regression analysis among statistical methods.
Point 2: Lines 169 - 170: Why was K-means clustering algorithm chosen over others? A little more justification is required.
Response 2:
(Lines 179-181) This study employed a large number of samples (N: 52,961). The K-mean clustering analysis was applied because it is effective in finding clusters in large amounts of data.
Point 3: Section 2.4 is very thin. It needs to detail the classification procedure as well as the sensitivity test. The author simply says K-mean and ANOVA techniques were used without justifying these techniques and explaining how they were applied.
Response 3:
(Lines 178-196) The procedure and sensitivity analysis that applied K-mean clustering analysis and the ANOVA test were elaborated. In addition, sensitive analysis methods were explained to find the appropriate number of UCZs.
Point 4: K-means does not automatically detect the number of groups in the data. How did the author determine the appropriate number of groups in this analysis?
Response 4:
(Lines 185-195) As noted by the reviewer, one of the most important points in applying K-mean cluster analysis is determining K (the appropriate the number of clusters). Considering previous studies (Oke, 2006; Ellefsen, 1991), the preliminary number of classes for K-mean clustering analysis (K) was determined from 6 to 12 (for a total of 7 cases), and the appropriate number of K was chosen by sensitive analysis that identified variations of F values from the ANOVA test. The sensitivity analysis results are presented in Appendix A.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
I do not have further comments on this manuscript. I favor its publications.
Reviewer 2 Report
This is a much improved version.
