Directional and Zonal Analysis of Urban Thermal Environmental Change in Fuzhou as an Indicator of Urban Landscape Transformation

Round 1

Reviewer 1 Report

The authors have adequately addressed most of my previous comments, I am only missing a comment/modification in relation to the following:

Is the chosen method sensitive to whether the selected years (2000 and 2016) were relatively cold/warm? For example, if 2000 was a relatively cold year and vice versa for 2016, is the method able to normalize for this, or will this just end up as an increase in LST? Would the results have been very different if e.g. 1999 and 2015 were analyzed. Please comment on this and add a sentence in the paper.

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

General comment:

The authors provide a revision that should be further improved in some parts, as detailed below. Some responses are not fully persuasive, and therefore the authors should modify some parts.

Specific comments:

-The abstract was not modified. In the previous revision round, it was commented that “the methodology of the LST aggregation should be better explained. Some sentences should be changed/improved to provide a greater appealing and understanding from a technical point of view: in fact, some parts appear too generic and vague.”. The authors did not provide any change. For example: line 23.24 is not clear from a methodological point of view. Line 32: “The four levels of hotspot density…”. In the abstract, that is the first part to read, any mention is made on the four levels.

Therefore, I suggest to carefully improve the abstract.

-End of introduction: a brief sentence reporting study area, dataset, time of the study is necessary at the end of the introduction since, otherwise, it is not clear for a reader which sensors are used and, in particular, that only two images were used. This information should be clear from the Introduction, before reading the other main sections.

-In the final discussion the authors seem to explain the choice of only two satellite images as a merit whilst, instead, it should be considered a limitation. Even if they justify with cloud covers and some invariances in subtropical humid climates, if a researcher has the possibility to provide thermal analysis with more images, even in the same season or across more intermediate years, it is an added value and the work is stronger. Also, this allows authors verifying their assumptions. Therefore, the authors must highlight this limit in their work.

-The authors stated: “About the climate background and the yearly or seasonal (summer in this case) average LST, it is beyond the scope of our research (we cannot solve all the problems in one paper)”. I think that this response is not serious and scientific. Currently, a researcher can download and process Landsat images in a very short time. For instance, having two or three images in summer 2000 and consider a mean LST map for the analysis is not a hard work that requires to write another paper.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

This paper is relevant because it suggests a methodology for characterizing the urban thermal environment and a scientific basis for sustainable urban development

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The paper can be considered suitable for publication. The content can be seen as a first stage for future studies (preferably in cities with less rainy and cloudy conditions) by using a wider satellite dataset. The strenght of a work is also linked with the statistical significance of the data/measurements. 

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.

Round 1

Reviewer 1 Report

In general, this research paper presents an interesting and novel methodology for analyzing the relationship between urban development and temperature hotspots in urban areas. However, some issues needs to be addressed before I can recommend the paper for publication.

 

General comments:

The results of the paper (ISA change analysis, LST change and LST hotspot analysis) needs to be compared and discussed in relation to the findings of previous research efforts within this area. Impact of the choice of case study years 2000 and 2016 on the results of especially the LST change and hotspot analysis. Would the results have been very different if e.g. 1999 and 2015 were analyzed? Is the chosen method sensitive to whether the selected years were relatively cold/warm? For example, if 2000 was a relatively cold year and vice versa for 2016, is the method able to normalize this, or will this just end up as an increase in LST? Please elaborate on the impact of the selected years for the results of your study and include in the paper.

 

Influence of spectral difference between L5 (TM) and L8 (OLI) on the results of your change analysis (both ISA, LST and LST hotspot). There is a difference in the location and width of the spectral bands in the radiometric spectrum (wavelengths) between Landsat 5 (TM) and Landsat 8 (OLI), which can affect some types of change analysis considerably, especially when using vegetation indices. Please elaborate on whether this is the case for your change analysis (ISA, LST and hotspot) and include a sentence on this in the paper. If you find that the difference between TM and OLI may be important for your analysis, please adjust either the reflectance in the L5 data or L8 data, so that they are comparable, and re-run the entire analysis again!!!

 

Accuracy of the high resolution reference data. I doubt that the un-supervised classification (ISODATA) of 4m IKONOS and 1m GEOEYE data is accurate enough to estimate the reference (ground truth) data. Either conduct manual digitalization of the end-members if interest or at least conduct a supervised classification. In the latter case or if using the current ISODATA method, a reference is needed to previous research where it is shown that the methods are very accurate and acceptable for use as reference/ground truth data.

Specific comments:  

Page 2, line 75: References to previous urban thermal environment analysis is needed here.

Page 5, line 150: Does NLST values above a certain threshold indicate hotspot? Please include if relevant. Is the NLST sensitive towards very high/low values within the Landsat scene? Please elaborate if relevant.

Page 5, line 172: Bare soil and ISA have similar spectral signatures and are often confused. Is this a major issue in relation to your case study area (does it contain large areas of bare soil). Please include a sentence on this.

Table 2: Please include an estimate of per pixel accuracy for the different end-members. Use for example Mean Absolute Error

 

Table 2:Why are the total areas not exactly the same and does this affect the results? Especially for there is a considerable difference for the year 2000. TM Image 2000 = 3.86km2, IKONOS 2000=3.82km2, OLI 2016 = 3.80, Geoeye-1 2016 = 3.81.

Figure 5: In 2000 (figure 5a) there are more hotspots in the north-east of the map compared to 2016 (figure 5b). Why is this the case? You would expect the opposite. Please include a sentence on this.

 

Figure 6: Very large areas of hotspots are seen. In the context of adaptation to heat waves, it could be interesting just to see the most extreme hot spots and how these have changed between 2000 and 2016, for example by choosing a significance level of 99% or 99.5% (compared to 95% used now). Please elaborate on this and include if relevant.

Figure 7: Hot spot axis is missing in some of the graphs.

Author Response

Dear Editor-in-Chief, We want to thank the reviewers for their comments and have modified our manuscript accordingly, as indicated below. We believe that our modifications have strengthened the manuscript and hope that it now meets the requirements of the journal. Thank you very much for your assistance. Sincerely, Youshui Zhang Reviewer #1: General comments: The results of the paper (ISA change analysis, LST change and LST hotspot analysis) needs to be compared and discussed in relation to the findings of previous research efforts within this area. Impact of the choice of case study years 2000 and 2016 on the results of especially the LST change and hotspot analysis. Would the results have been very different if e.g. 1999 and 2015 were analyzed? Is the chosen method sensitive to whether the selected years were relatively cold/warm? For example, if 2000 was a relatively cold year and vice versa for 2016, is the method able to normalize this, or will this just end up as an increase in LST? Please elaborate on the impact of the selected years for the results of your study and include in the paper. RE: Thanks for this comment. We did not do a hotspot analysis in this study area before. We have no special reason for the temporal (years and seasons) selection of Landsat data. The study area is on a subtropical plain sandwiched between the Gu and Qi mountains. The vegetation cover in the region is predominantly evergreen and the fractional vegetation cover in different seasons is almost invariable. The increase in populations, coupled with the high summer temperatures (average of 37.5 days per year with a temperature >35°C), makes the city an ideal study area. Compared with the hot summer climate, the weather of Fuzhou in spring, autumn and winter is relatively cool and the weather variations between spring, autumn and winter are not as large. Therefore, the season summer is better for LST analysis than other seasons. In order to reduce the impact of seasonal variations, bitemporal images acquired in summer season were selected to retrieve LST and further quantify the impact of urban expansion on thermal environment. About the choice of years 2000 and 2016, we just selected two same seasonal images acquired in cloud-free weather to do our research. Because the weather of our study area is usually cloudy or rainy, it is difficult to acquire a Landsat image in clear and cloud-free weather. Usually, it takes many years to acquire a Landsat image in clear and cloud-free weather that can be used. Therefore, bi-temporal images in the same season (summer) are more difficult to acquire. Though we can select other seasons to do this research, the summer season in our study is more suitable for thermal environment analysis. The last reason we want to explainis that the focus of our study is to analyze the variation of LST aggregation and area proportion of hotspot area to gain new insights into the thermal impacts of urbanization. We emphasize our method and we think our analytical method may be helpful for readers. Compared with our research method, LST acquired in which years is not important. In the near future, we plan to select other cities to do further analysis on seasonal variation. We have supplemented some sentences in section 2.1 to explain and strengthen our manuscript as below (we think these sentences supplemented in the location of Section 2.1 are better): The city has a subtropical humid climate, and the vegetation cover in the region is predominantly evergreen with almost no seasonal variations. The weather of Fuzhou in spring, autumn and winter is relatively cool and the weather variations between spring, autumn and winter are not as large. The summers, however, are very hot with an average of 37.5 days of temperatures exceeding 35°C. Therefore, in order to reduce the impact of seasonal variations, …(lines 135-140) We hope the reviewer can be satisfied with our reply and modifications. Influence of spectral difference between L5 (TM) and L8 (OLI) on the results of your change analysis (both ISA, LST and LST hotspot). There is a difference in the location and width of the spectral bands in the radiometric spectrum (wavelengths) between Landsat 5 (TM) and Landsat 8 (OLI), which can affect some types of change analysis considerably, especially when using vegetation indices. Please elaborate on whether this is the case for your change analysis (ISA, LST and hotspot) and include a sentence on this in the paper. If you find that the difference between TM and OLI may be important for your analysis, please adjust either the reflectance in the L5 data or L8 data, so that they are comparable, and re-run the entire analysis again!!! RE: This is a good comment, thanks for this comment. We derived reflectance from Landsat images by atmospheric correction, calculated NDVI based on reflectance of L5 and L8, and further calculated emissivity from NDVI. We reply to this comment as below: Because spectrum of L5 (red band 0.63-0.69μm, Near IR band 0.76- 0.90μm) and L7 (red band 0.63-0.690μm, Near IR band 0.75-0.90μm) are similar, previous studies have evaluated the agreement in surface reflectance and vegetation indices derived from multiple sensors, mainly focusing on Landsat TM/ETM + and MODIS instruments (Feng et al., 2012, Ju et al., 2012, Maiersperger et al., 2013). Ke et al. (2015) have ever examined the characteristics of the NDVI derived from the Landsat 8 OLI sensor (red band 0.630–0.680μm, Near IR band 0.845–0.885μm) and compared it with other multiple satellite sensors including L7. They got the conclusion ‘The simulation results showed that Landsat 8 OLI, Landsat 7 ETM+, and MODIS NDVIs agreed well given the same atmospheric condition, while aerosol optical depth was a major factor influencing NDVI, especially for water. The results from the four study sites were consistent with the simulation results. Given the same atmospheric correction methods applied to derive surface reflectance, Landsat 8 OLI and Landsat 7 ETM+ NDVIs had overall good agreement, with mean bias errors within ±0.05, the standard deviations of errors <0.1, and R2 from 0.84 to 0.98.’ ‘Landsat8 OLI produced considerably greater spatial variability of NDVIs than Landsat 7 ETM+ within heterogeneous land cover types such as crop, urban, and turbid water areas. NDVIs in clear water area, in contrast, have smaller spatial variability. Such characteristics are mainly associated with higher SNR performance and radiometric sensitivity of Landsat8 OLI instrument’. All the above sentences and the references are included in the paper: Ke, Y.; Im, J.; Lee, J.; Gong, H.; Ryu, Y. Characteristics of Landsat 8 OLI-derived NDVI by comparison withmultiple satellite sensors and in-situ observations. Remote Sensing of Environment, 2015, 164: 298-313. Based on conclusions from the previous studies that different Landsat sensors agreed well given the same atmospheric condition (this is one reason why we selected the images acquired in same season), we did not analyze the influence of spectral difference between the two sensors and directly compared the results derived from L5 and L8 (previous studies has gotten this conclusion. In addition, it's beyond the scope of this research to some extent. We can’t solve all the problems through this paper). In order to strengthen our manuscript based on this comment, we supplemented some sentences at the end of the Section 4 Conclusions as below. We think this modification is instructive to the readers. The supplemented sentences is ‘In addition, though Ke et al. [70] proved that, given that the same atmospheric correction methods, the NDVI derived from multiple sensors such as Landsat TM/ETM+ and OLI instruments had good consistency, further evaluation of the agreement in surface reflectance and vegetation indices derived from multiple sensors is required because of the spectral differences between these sensors.’. (lines 746-751 ) We hope the reviewer can be satisfied with our reply and modifications. Accuracy of the high resolution reference data. I doubt that the un-supervised classification (ISODATA) of 4m IKONOS and 1m GEOEYE data is accurate enough to estimate the reference (ground truth) data. Either conduct manual digitalization of the end-members if interest or at least conduct a supervised classification. In the latter case or if using the current ISODATA method, a reference is needed to previous research where it is shown that the methods are very accurate and acceptable for use as reference/ground truth data. RE: Thanks for this comment. We have tried different methods to classify the high resolution images and compared the results. We found the accuracy of ISODATA is not lower than other methods. We derived the land cover types through ISODATA and further assessed the accuracy of land cover derived from Landsat imagery in our previous two papers. We also used manual digitization to select test sites. Finally, we calculated the total area and compared with the results derived from TM/OLI imagery by FCLS. We assessed the accuracy of the total areas by comparing the total area of fractional cover for each land cover type with those estimated from the VHR reference data because the resolutions were different between TM/OLI imagery and high-resolution imagery. We have supplemented one sentence and one reference of our previous paper as below: ‘As our previous study in [39, 42], the accuracy of fractional covers derived from TM/OLI imagery by FCLS was assessed…..’(line 262). Considering accuracy assessment is not the focus of this paper and follows fairly standard methods, we do not introduce it in detail. We hope the reviewer can agree with our opinion. Specific comments: Page 2, line 75: References to previous urban thermal environment analysis is needed here. RE: We have supplemented two references [39, 42] of our previous study (line 262). Thank you. Page 5, line 150: Does NLST values above a certain threshold indicate hotspot? Please include if relevant. Is the NLST sensitive towards very high/low values within the Landsat scene? Please elaborate if relevant. RE: We calculated hotspots based on LST, not NLST. We calculated NLST to further analyse it with area proportion of ISA, area proportion of ISA with LSTClass 4 and area proportion of hot spot area in each sector as in Fig. 7 (Section 3.4). About NLST sensitive towards very high/low values, to some extent, it is similar to LST. a pixel with a high LST value may not be a statistically significant hotspot. To be a statistically significant hotspot, a pixel must have a high LST value and must be surrounded by other pixels with high LST values. It is related to the algorithm of hotspot calculation; we introduced it in our manuscript. Page 5, line 172: Bare soil and ISA have similar spectral signatures and are often confused. Is this a major issue in relation to your case study area (does it contain large areas of bare soil). Please include a sentence on this. RE: Bare soil areas are mainly distributed alongside the river of our study area, not urban area; therefore soil does not have a significant effecton the estimation of urban percent ISA. We have supplemented one sentence ‘The high-albedo fraction image also included some bare soil. Because of the spatial distribution of bare soil areas along the river, they do not have a significant effecton the estimation of percent ISA in the city.’ (lines 233-236) Table 2: Please include an estimate of per pixel accuracy for the different end-members. Use for example Mean Absolute Error RE: We derived subpixel land cover types by soft classification method such as spectral unmixing FCLS. We cannot assess the accuracy by per pixel as hard classification method. What’s more, the resolutions between TM/OLI imagery and high-resolution imagery are different. We cannot assess the accuracy by per pixel. Therefore, we assessed the accuracy by comparing the total area of fractional cover for each land cover type with those estimated from the VHR reference data. Table 2: Why are the total areas not exactly the same and does this affect the results? Especially for there is a considerable difference for the year 2000. TM Image 2000 = 3.86km2, IKONOS 2000=3.82km2, OLI 2016 = 3.80, Geoeye-1 2016 = 3.81. RE: This is a good comment. I think it is because of two main reasons. The first reason is the scale effect. As we know, the resolutions between TM/OLI imagery and high-resolution imagery are different, we are absolutely impossible to derive the same area for one test site with different resolutions/scales (it is same as the question ‘Unable to accurately measure the length of Britain's coastline’). The second reason is the classification error; every classification method can result in error. We hope the reviewer can agree with our opinion. Figure 5: In 2000 (figure 5a) there are more hotspots in the north-east of the map compared to 2016 (figure 5b). Why is this the case? You would expect the opposite. Please include a sentence on this. RE: Figure 5 is the LST classes for the study area (classify LST into different classes), not the hotspot areas. As we can see in Figure 1, the north-east of the map is the mountain. Because green spaces attracted more and more attention in urban planning, that area was the vegetation area in 2016 but not in 2000. This is the reason that areas of LST in the north-east of the map in 2000 were more than in 2016. We think it is easy to understand and LST analysis is not our focus. We advise not to supplement a sentence here. We hope the reviewer can agree with our opinion. Figure 6: Very large areas of hotspots are seen. In the context of adaptation to heat waves, it could be interesting just to see the most extreme hot spots and how these have changed between 2000 and 2016, for example by choosing a significance level of 99% or 99.5% (compared to 95% used now). Please elaborate on this and include if relevant. RE: Chinese cities expanded very rapidly in the recent 20-30 years. Especially, it is very common for the provincial capital cities like Fuzhou. Driven by fast economic growth and population increase, Fuzhou has experienced rapid urbanization in the past 20-30 years, along with a drastic transformation of the urban landscape patterns and environment. The thermal environment and urban heat island also changed largely in the past 20-30 years (especially Fuzhou is on a small plain sandwiched between the Gu and Qi mountains, and urban areas expanded into the mountains.). We have compared the results by choosing different significance levels of 95 % and 99%. We found that the results of significance level of 95% were better for analyzing the urban expansion and its impact on urban thermal environment, as setting the type I error to 1% reduces the statistical power of detecting true differences too much. Considering our purpose is to compare the results of two dates, we chosen the significance level of 95% (considering our research purpose and the length limitation of the paper, we think no need to do another comparative analysis of the results of the significance levels of 99% and 95%). We hope the reviewer can agree with our opinion. In addition, other scholars have done hotspot analysis and indicated that at a significance level of 0.05 (95%), a z-score is statistically significant, and they chose the significance level of 95% to do further analysis (Tran, D.; Filiberto, P.; Latorre-Carmona, P.; Myint, S.; Caetano, M.; Kieu, H. Characterizing therelationship between land use land cover change and land surface temperature. ISPRS Journal ofPhotogrammetry and Remote Sensing. 2017, 124, 119–132.). Figure 7: Hot spot axis is missing in some of the graphs. RE: Modified.

Author Response File: Author Response.docx

Reviewer 2 Report

General comment:

The paper deals with the spatio-temporal analysis of hotspot locations reflecting urban land cover changes in different spatial sectors of the Fuzhou city using two Landsat images of 2000 and 2016. The well-known spectral unmixing method was used to derive subpixel ISA. The LST aggregation was analyzed in different directions and concentric zones, and hotspot density was calculated. The paper findings are not new and widely addressed, even if the methodology is quite interesting. Some issues should be better addressed and discussed, following the comments detailed below.

Specific comments:

Abstract:

-The study area is not reported

-The methodology of the LST aggregation should be better explained. Some sentences should be changed/improved to provide a greater appealing and understanding from a technical point of view: in fact, some parts appear too generic and vague.

Introduction:

- Line 47-59: in the state-of-the-art overview, the majority of the references are very old. They do not account for the exponentially growth of the publications in recent years on these topics. Therefore, I suggest adding also very recent papers. Also, a sentence like “LST from Landsat sensors is often used to analyze spatial and temporal relationships between the urban thermal environment and land cover” cannot be explained by a random paper, but by a review paper reporting the several studies employing the Landsat or by more significant references.

-Line 80: “bi-temporal”: please, explain the meaning adding the two years, and pointing out that only two images were processed.

-Lines 81-85: the sentence “hotspots with high LST within the study area are used to identify areas which have experienced urban land cover transformation” is too general: in this form, I doubt that they were never previously studied in other works. The sentence must be better specified pointing out the novelty of the work.

Also, other papers performed analysis of LST and land cover changes with respect to distance and angular directions. The authors should perform a more accurate state-of -the art study in this regard, highlighting the novelty of the presented paper.

-End of the introduction: study area, dataset, time of the study must be reported.

Section 2:

-Figure 1: the two panels are not well formatted. Also, the time of the RGB image must be reported

Table 1: third column name is not correct. Also, the description of the third column is not proper for a table. Please, change and improve.

-The authors used a single image for each year (years very distant, i.e. 2000 and 2016), also of different months. Since one goal of the paper is the analysis of the thermal pattern, a greater number of images for each investigated year is expected. Today, archive/free data dissemination easily allows to download and process a large number of images to provide a more robust and significant study. The habit of considering the thermal pattern of a year/season with only an image is misleading. An LST map of a single day can have values depending on the climate background (soil humidity due to the monthly rainfall, air temperature, wind circulation, solar irradiation) that can differ from the LST measured some weeks before or after. Therefore, a greater number of images should be processed, then considering the yearly or seasonal (summer in this case) average LST. Also, the months considered are very different (June and July), rendering the comparison even more unfair. Probably, a summer mean of LST maps will provide some differences in hotspot locations and areas. It is not clear how the approach described on lines 138-146 can completely overcome this issue.

-The use of the temperature normalization (line 147-151) does not solve this problem, as shown in different published papers. Also, the use of the normalization should be done carefully and deeply analyzed for each map, since min/max outliers could affect the meaning of the values when maps of different years are compared.

-Line 122: it is not clear the algorithm/procedure to obtain atmospherically corrected surface reflectance from TOA radiances in the VIS NIR bands.

-Lines 128-132: the authors refer to Landsat band numbers, without any correspondence with table 1, where the bands are not numbered.

- Figure 2: for the whole image, why did the authors use only two small test areas?

-Lines 209-211 (and figure 1): how are the three zone radii quantitatively established?

-Lines 210-211: the syntax of the sentence must be improved.

Section 3:

-Line 253: “the focus here…”: the authors refer to the overall work or of results of table 2? Since table 2 reports an accuracy study, even if in a very small area (the reason for these small areas should be justified). In fact,  I do not agree with the sentence at line 255 “In general, the accuracy of the sampled plot can represent the accuracy of the whole study area”, since  little samples area can be representative of the study area if they are several and randomly distributed. Therefore, I suggest performing an accuracy test with more samples.

-What is the meaning of the term “plot” in the above sentence of line 255?

-Line 329: delete “of”.

-Lines 365-366: “Therefore, if hotspot locations increase between two dates, the ISA increases regardless of an LST increase or decrease between the two dates.” The authors should better clarify the sentence: is the LST increase/decrease referred to a single pixel than to a cluster (hotspot locations)?

-Figure 7: the size of text of y axes should be slightly increased

-End of section 3: the authors should discuss the limitations of their study.

References

-The numbering is doubled, and not aligned. Please, revise the format of this section

Author Response

Reviewer #2:

General comment:

The paper deals with the spatio-temporal analysis of hotspot locations reflecting urban land cover changes in different spatial sectors of the Fuzhou city using two Landsat images of 2000 and 2016. The well-known spectral unmixing method was used to derive subpixel ISA. The LST aggregation was analyzed in different directions and concentric zones, and hotspot density was calculated. The paper findings are not new and widely addressed, even if the methodology is quite interesting. Some issues should be better addressed and discussed, following the comments detailed below.

Specific comments:

Abstract:

-The study area is not reported

RE: We have supplemented ‘Fuzhou, China’ in Lines 19-20 of abstract.

-The methodology of the LST aggregation should be better explained. Some sentences should be changed/improved to provide a greater appealing and understanding from a technical point of view: in fact, some parts appear too generic and vague.

RE: In Section 2.2.4 Zonal and sectoral analysis on the dynamics of urban landscape pattern and LST aggregation, we have further explained LST aggregation and how it is related to hotspot analysis.

In lines 310-319, we explain ‘The hotspot analysis tool (Getis-OrdGi*) in ArcGIS was used to explore the aggregation of LST in urban area and evaluate how thermal aggregation is related to distance from the city center and change of urban landscape pattern [66]. Unlike directly analyzing LST maps, a pixel with a high LST value may not be a statistically significant hotspot. To be a statistically significant hotspot, a pixel must have a high LST value and must be surrounded by other pixels with high LST values.…….’. The hotspot analysis is calculated based on LST (Getis-Ord Gi* and the Queen's adjacency matrix are calculated based on LST). Getis-Ord Gi* can be calculated in the ArcGIS tool as other scholars did.

[66]Tran, D., Pla,. F., Latorre-Carmona, P., Myint, S., Caetano, M., Kieu, H., Characterizing the relationship between land use land cover change andland surface temperature. ISPRS Journal of Photogrammetry and Remote Sensing. 124, 119–132.

Lines 323-326: ‘In this study, hotspot analysis quantitatively analyzed the spatial autocorrelation of LST by providing a measure of spatial dependence for each LST pixel and indicating the relative magnitudes of the LST values in the neighborhood of the LST pixel [69].’

We hope the reviewer is satisfied with our reply.

Introduction:

- Line 47-59: in the state-of-the-art overview, the majority of the references are very old. They do not account for the exponentially growth of the publications in recent years on these topics. Therefore, I suggest adding also very recent papers. Also, a sentence like “LST from Landsat sensors is often used to analyze spatial and temporal relationships between the urban thermal environment and land cover” cannot be explained by a random paper, but by a review paper reporting the several studies employing the Landsat or by more significant references.

RE: Thanks for this comment. We have modified some sentences in this paragraph (lines 50-64), and some old references are removed. We also read some additional papers published in the recent 3 years and cited these references. The references we cited are:

He, C.; Gao, B.; Huang, Q.; Ma, Q.; Dou, Y. Environmental degradation in the urban areas of China: Evidence from multi-source remote sensing data. Remote Sensing of Environment. 2017, 193, 65–75. Shen, H.; Huang, L.; Zhang, L.; Wu, P.; Zeng, C. Long-term and fine-scale satellite monitoring of the urban heat islandeffect by the fusion of multi-temporal and multi-sensor remote senseddata: A 26-year case study of the city of Wuhan in China. Remote Sensing of Environment. 2016, 172, 109–125. Yang, J.; Su, J.; Xia, J.;Jin, C.; Li, X.; Ge, Q. The Impact of Spatial Form of Urban Architecture onthe Urban Thermal Environment: A Case Studyof the Zhongshan District, Dalian, China. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing. 2018,11, 2709–2716. Zhou, W.; Wang, J.; Cadenasso, M. Effects of the spatial configuration of trees on urban heat mitigation: Acomparative study. Remote Sensing of Environment. 2017, 195, 1–12. Huang, Q.; Huang,J.; Yang,X.; Fang, C.; Liang, Y. Quantifying the seasonal contribution of coupling urban land use types onUrban Heat Island using Land Contribution Index: A case study in Wuhan,China. Sustainable Cities and Society. 2019, 44, 666-675. Bonafoni, Stefania.; Keeratikasikorn, C. Land Surface Temperature and Urban Density:Multiyear Modeling and Relationship Analysis Using MODIS and Landsat Data. Remote Sensing. 2018, 10, 1471. Chen, Y.; Yu, S. Impacts of urban landscape patterns on urban thermal variations in Guangzhou, China. International Journal of Applied Earth Observation and Geoinformation. 2017, 54, 65–71. Peng, J.; Jia, J.; Liu, Y.; Li, H.; Wu, J. Seasonal contrast of the dominant factors for spatial distribution of land surface temperature in urban areas. Remote Sensing of Environment. 2018, 215, 255–267.

We don’t list other modifications and all the removed sentences and references here.
-Line 80: “bi-temporal”: please, explain the meaning adding the two years, and pointing out that only two images were processed.

RE: We can delete “bi-temporal” here and it won’t impact the meaning of this sentence. There are three main reasons why only two images were selected for our study.

The first reason is the weather of our study area is usually cloudy or raining, it is difficult to acquire a Landsat image in clear and cloud-free weather. Usually, it takes several years to acquire a Landsat image in clear and cloud-free weather that can be used. Therefore, bi-temporal images taken in the same season (summer) are difficult to acquire. Though we can select three or four images to do research, the season of three or four images may not be the same.

The second reason is (as we replied above) that if we select two images in the same season, different Landsat sensors agree well given that they were acquired with very similar atmospheric conditions.

The third reason is that the focus of our study is to analyze the variation of LST aggregation and area proportion of hotspot areas to gain new insights into the thermal impacts of urbanization. We emphasize our method and we think our analytical method may be helpful for readers.

We hope the reviewer will agree with our opinion.

-Lines 81-85: the sentence “hotspots with high LST within the study area are used to identify areas which have experienced urban land cover transformation” is too general: in this form, I doubt that they were never previously studied in other works. The sentence must be better specified pointing out the novelty of the work. Also, other papers performed analysis of LST and land cover changes with respect to distance and angular directions. The authors should perform a more accurate state-of -the art study in this regard, highlighting the novelty of the presented paper.

RE: Thanks for this comment. We have modified the first aspect to ‘First, the spatial patterns of the hotspot densities are used to analyse urban land cover transformation and thermal environment.’ We also modified the second aspect to ‘Second, the direction and magnitude of the urban expansion effect are delineated by area proportion of hotspot area and percent ISA when the whole urban area is classified into different angular sectors.’ (lines 90-94). We need to state here that no other scholars have calculated area proportion of hotspot areas and hotspot density to the best of our knowledge.

-End of the introduction: study area, dataset, time of the study must be reported.

RE: We have introduced the study area, dataset and time of the study in Section 2: Materials and methods (2.1 Study area and data). Some other scholars also introduced these in their papers. Considering that we have introduced them in Section 2, we advise not to introduce these at the end of section 1 (introduction) to avoid repetition. We hope the reviewer will agree with our opinion.

If the reviewer insists on introducing these at the end of the introduction, we will do it at the next stage.

Section 2:

-Figure 1: the two panels are not well formatted. Also, the time of the RGB image must be reported.

RE: The two panels are formatted better now. We have supplemented the time ‘acquired on July 27, 2016’ in the figure title.

Table 1: third column name is not correct. Also, the description of the third column is not proper for a table. Please, change and improve.

RE: Sorry for our typographic error. ‘Spectral and special resolutions’ should be Spectral and spatial resolutions. We have modified the third column name to ‘Description’ as other scholars have done (e.g. Sadegh Jamali et al. Detecting changes in vegetation trends using time series segmentation. Remote Sensing of Environment. 2015, 156: 182-195). We think this modification is better and hope the reviewer is satisfied with our modification.

-The authors used a single image for each year (years very distant, i.e. 2000 and 2016), also of different months. Since one goal of the paper is the analysis of the thermal pattern, a greater number of images for each investigated year is expected. Today, archive/free data dissemination easily allows to download and process a large number of images to provide a more robust and significant study. The habit of considering the thermal pattern of a year/season with only an image is misleading. An LST map of a single day can have values depending on the climate background (soil humidity due to the monthly rainfall, air temperature, wind circulation, solar irradiation) that can differ from the LST measured some weeks before or after. Therefore, a greater number of images should be processed, then considering the yearly or seasonal (summer in this case) average LST. Also, the months considered are very different (June and July), rendering the comparison even more unfair. Probably, a summer mean of LST maps will provide some differences in hotspot locations and areas. It is not clear how the approach described on lines 138-146 can completely overcome this issue.

RE: Thanks for this comment. We have partly relied to this comment as above.

One important reason is the weather of our study area is usually cloudy or rainy, it is difficult to acquire a Landsat image in clear and cloud-free weather. Usually, it takes several years to acquire a Landsat image in clear and cloud-free weather that can be used. What is more, bi-temporal images acquired in the same season (summer) in different years are difficult to acquire. The images need to be acquired in cloud-free weather. We only selected imagery in cloud-free weather because it is helpful for atmospheric correction and LST retrieval. Though we could have selected three or four images to do this research, the seasons of three or four images are not the same and this impacts on the results from different Landsat sensors. If we select two images in the same season, different Landsat sensors agree well given the same atmospheric condition as we stated as above.

In addition, the study area has a subtropical humid climate, the vegetation cover in the region is predominantly evergreen and the fractional vegetation cover is almost invariable in different seasons. Compared with the hot summer climate, the weather of Fuzhou in spring, autumn and winter are relatively cool and the weather variations between spring, autumn and winter are not aslarge (We have supplemented some sentences in lines 135-140 of section 2.1.). Therefore, bitemporal images were selected. We have also indicated this in reference 4 (Zhang,Y.; Balzter,H.; Liu,B.; Chen, Y. Analyzing the Impacts of Urbanization and Seasonal Variation on Land Surface Temperature Based on Subpixel Fractional Covers Using Landsat Images. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing. 2017, 10, 1344–1356.).

We just selected two images acquired in cloud-free weather in the same seasonal conditions to do our research. In addition, the summer is better for LST analysis than other seasons in our study area.

The last reason is that the focus of our study is to analyze the variation of LST aggregation and area proportion of hotspot area to gain new insights into the thermal impacts of urbanization. We emphasize our method and we think our analytical method will be helpful for readers. Therefore, we advise using bi-temporal images in our study. About the climate background and the yearly or seasonal (summer in this case) average LST, it is beyond the scope of our research (we cannot solve all the problems in one paper). In a future study, we can think of this to do further research in another area such as the north of China.

We hope the reviewer is satisfied with our reply and agrees with our opinion.

-The use of the temperature normalization (line 147-151) does not solve this problem, as shown in different published papers. Also, the use of the normalization should be done carefully and deeply analyzed for each map, since min/max outliers could affect the meaning of the values when maps of different years are compared.

RE: In our study, we did not directly analyze normalized LST (NLST), we used NLST to calculate mean NLST of each sector and further analyzed it with area proportion of ISA, area proportion of ISA with LSTClass 4 and proportion of hot spot area in each sector of two zones (Section 3.4 Interpreting distributions of hotspot densities, area proportion of ISA/ISA with high LST and the thermal environment in different sectors). In our study, NLST is an ancillary and unimportant analysis (we used it to ‘adjust the LST of images acquired at different dates [52]’). Considering our research focus and our analytical method, we think our methodology is helpful to the readers in this field though NLST may not be perfect.

We hope the reviewer will agree with our opinion.

-Line 122: it is not clear the algorithm/procedure to obtain atmospherically corrected surface reflectance from TOA radiances in the VIS NIR bands.

RE: Atmospheric correction is very common for image processing, land surface parameters inversion when we used remote sensing imagery. We can explain it here. In the atmospheric correction of VIS/NIR bands we used the atmospheric correction method of the FLAASH package in ENVI, and the visible and near-infrared bands of ETM+/OLI images were converted to land surface reflectance. The algorithm/procedure of FLAASH is MODTRAN (MODTRAN serves as the community standard atmospheric band model). Considering our research focus is not atmospheric correction and there is a limitation of the length of the paper, we suggest not introducing it in detail here (there are many public references describing MODTRAN). We hope the reviewer agrees with our opinion.

In addition, the ‘6s’ model for atmospheric correction (Second Simulation of Satellite Signal in the Solar Spectrum) is usually used to get three parameters xa, xb and xc to get the reflectance of the VIS/NIR bands. We also don’t explain it here because it is a widely adopted method that is described in the literature.

-Lines 128-132: the authors refer to Landsat band numbers, without any correspondence with table 1, where the bands are not numbered.

RE: Thanks for this comment. We have supplemented red band and near infrared band in lines 173-174. We have modified it to ‘…bands 3 and 4 of the Landsat images’ with ‘… bands 3 (red band) and 4 (near-infrared band) of the Landsat images.’

Considering Landsat imagery is well known and to make table 1 more succinct, we do not supplement the band numbers in Table 1. We hope the reviewer agrees with our opinion.

- Figure 2: for the whole image, why did the authors use only two small test areas?

RE: The first reason is the high-resolution IKONOS image for accuracy assessment cannot cover all of the study area, so we selected small test areas to assess the accuracy of subpixel land cover type derivation.

The second reason is that we analyzed it in a previously published paper (Youshui Zhang, Angela Harris & Heiko Balzter (2015) Characterizingfractional vegetation cover and land surface temperature based on sub-pixel fractionalimpervious surfaces from Landsat TM/ETM+, International Journal of Remote Sensing, 36: 16,4213-4232, DOI: 10.1080/01431161.2015.1079344). If we selected four small test areas, the accuracy is nearly the same as the result of two small test areas. Considering the accuracy assessment is not our research focus, we selected two small areas to do the accuracy assessment. The sample areas are representative for the whole study area and provide a good representation of the overall accuracy. In addition, the two small areas are selected in heterogeneous urban area, if we selected these in the fringe of the urban area (homogeneous area), the accuracy will be higher.

We hope the reviewer will agree with our opinion.

-Lines 209-211 (and figure 1): how are the three zone radii quantitatively established?

RE: In figure 1, we introduced the text ‘the study area is divided into different sectors from city center to the urban peripheral areas by the yellow lines, which is further illustrated in section 2.2.4’.

In Section 2.2.4, we have introduced some sentences---‘the urban area was divided into three concentric circular zones (Zone 1, Zone 2 and Zone 3) from the city center through comparing urban area and its expansion in bi-temporal images. Zone 1 was the urban core zone in both years, 2000 and 2016. Zone 3 included mainly peripheral urban areas. Zone 2 was a transition zone between Zone 1 and Zone 3. Zone 2 was not included in the urban area in 2000, only in the urban area extent of 2016 because of urban expansion. (lines 293-298)

(We compared the bi-temporal imagery and divided zone 1 as the urban core and Zone 3 as the peripheral urban areas in both dates. Zone 2 was divided as a transition Zone 2 between Zone 1 and Zone 3, Zone 2 in 2016 is in the urban area because urban expansion.)

-Lines 210-211: the syntax of the sentence must be improved.

RE: We have modified it with ‘Zone 1 was the urban core zone in both years, 2000 and 2016. Zone 3 included mainly peripheral urban areas. Zone 2 was a transition zone between Zone 1 and Zone 3. Zone 2 was not included in the urban area in 2000, only in the urban area extent of 2016 because of urban expansion.’. (lines 295-298)

Section 3:

-Line 253: “the focus here…”: the authors refer to the overall work or of results of table 2? Since table 2 reports an accuracy study, even if in a very small area (the reason for these small areas should be justified). In fact,  I do not agree with the sentence at line 255 “In general, the accuracy of the sampled plot can represent the accuracy of the whole study area”, since  little samples area can be representative of the study area if they are several and randomly distributed. Therefore, I suggest performing an accuracy test with more samples.

RE: We partly replied to this comment above. We would like to reply again here.

Because the high-resolution IKONOS imagefor accuracy assessment cannot cover all of the study area, we selected small test areas to assess the accuracy of subpixel land cover type derivation.

We analyzed this issue in a previously published paper. If we select four small test areas, the accuracy is nearly the same as for two small test areas (references 39 and 42). In addition, the two small areas are selected in heterogeneous urban area. If we selected small areas in the fringes of the urban areas (homogeneous area), the accuracy will be higher.

Considering that the accuracy assessment is not our research focus but a procedure for testing the quality of our analysis, we selected two small areas to do the accuracy assessment. Though our method for accuracy assessment may not be perfect, we think the results showed that endmember fractions derived by FCLS can be used for further analysis.

In addition, we have modified further the sentence with ‘The sample areas are representative for the whole study area and provide a good representation of the overall accuracy.’ (lines 353-354) .We think this modification is better.

We hope the reviewer agrees with our opinion.

-What is the meaning of the term “plot” in the above sentence of line 255?

RE:The meaning is area (site), we have modified ‘plot’ with areas in line 353-354 (we have modified all the sentence). We think this modification is better.

-Line 329: delete “of”.

RE:We have deleted ‘of’ here.

-Lines 365-366: “Therefore, if hotspot locations increase between two dates, the ISA increases regardless of an LST increase or decrease between the two dates.” The authors should better clarify the sentence: is the LST increase/decrease referred to a single pixel than to a cluster (hotspot locations)?

RE: In order to clarify the sentence here, we modified this sentence with ‘Therefore, if ISA increases spatially continuously between the two dates, the hotspot locations increase between two dates because LST aggregation is impacted by continuous spatial changes of ISA.’ (lines 547-549). We think this modification is better. We hope the reviewer is satisfied with our modification.

It is related to the algorithm of the hotspot analysis. The hotspot analysis is used to explore the aggregation of LST. The spatial pattern of LST aggregation was evaluated through Getis-Ord Gi*. By calculating Getis-Ord Gi*, the hotspot analysis is applied to evaluate the spatial LST clustering. By providing a measure of spatial dependence for each LST pixel in the neighborhood of the pixel, the hotspot analysis can characterize the presence of hotspots, the thermal aggregation and thermal variation over the study area at different dates (Ord and Getis, 1995). Unlike directly analyzing LST maps, a pixel witha high LST value may not be a statistically significant hotspot. To be a statistically significant hotspot, a pixel must have a high LST value and must be surrounded by other pixels with high LST values. Therefore, hotspot analysis is a technique for characterization of spatial autocorrelation of LST in an image. The identification of hotspot areas does not depend on whether the value of an LST pixel is high or low. We introduced this in Section 2.2.4.

In addition, there are two sentences ‘The hotspot pattern change was impacted by spatial continuous change of urban land cover from pervious to impervious surface. Therefore, hotspot analysis can assess the impact of urban ISA change on the thermal aggregation, rather than focusing only on the high or low LST values separately’ in lines 518-522 (Section 3.3).

-Figure 7: the size of text of y axes should be slightly increased

RE: Modified.

-End of section 3: the authors should discuss the limitations of their study.

RE: Thanks for this comment. We have supplemented the limitations at the end of section 3 as below:

This paper is a contribution to the improvement in the research on the impact of urban expansion on the thermal environment. However, the study has some limitations. The research is focused only on one city and future research needs to compare the findings with other cities and different seasons. Another limitation is that the ISA change threshold needs to be further explored with a view to selecting an optimum threshold value. (lines 702-707)

We hope the reviewer is satisfied with our modifications.

References

-The numbering is doubled, and not aligned. Please, revise the format of this section

RE: Modified