The Impact of Urban Renewal on Land Surface Temperature Changes: A Case Study in the Main City of Guangzhou, China

: To improve land use e ﬃ ciency, urban renewal must also consider urban microclimates and heat islands. Existing research has depended on manual interpretation of high-resolution optical satellite imagery to resolve land surface temperature (LST) changes caused by urban renewal; however, the acquired ground time series data tend to be uneven and unique to speciﬁc frameworks. The objective of this study was to establish a more general framework to study LST changes caused by urban renewal using multi-source remote sensing data. Speciﬁcally, urban renewal areas during 2007–2017 were obtained by integrating Landsat and yearly Phased Array type L-band Synthetic Aperture Radar (PALSAR) images, and LST was retrieved from Landsat thermal infrared data using the generalized single-channel algorithm. Our results showed that urban renewal land (URL) area accounted for 1.88% of urban land area. Relative LST between URL and general urban land (GUL) of Liwan, Yuexiu, Haizhu, and Tianhe districts dropped by 0.88, 0.42, 0.43, and 0.10 K, respectively, whereas those of Baiyun, Huangpu, Panyu, and Luogang districts presented opposite characteristics, with a rise in the LST of 0.98, 1.03, 1.63, and 2.11 K, respectively. These results are attributable to population density, building density, and landscape pattern changes during the urban renewal process.


Introduction
Human-oriented urbanization and related land-use change processes significantly affect the thermal environment of cities and their surrounding areas by transforming the natural landscape into an impervious surface [1][2][3]. The most prominent urban thermal environment problem is the development of urban heat islands (UHIs). As one of the most obvious characteristics of urban climate change caused by construction and human activity, UHIs constitute heat accumulation in urban areas due to their higher surface temperature compared with the surrounding suburbs and rural regions [4,5].

Study Area
As the most important political, economic, cultural, and technological center in southern China, Guangzhou City has undergone rapid economic development and urbanization since its opening and reform [36]. With large-scale human and land development activity, the urban landscape and climate have experienced significant changes, putting Guangzhou City at great environmental risk from thermal changes [37]. The study area includes a core region, consisting of Liwan, Yuexiu, Haizhu, and Tianhe districts, and a peripheral region that includes Baiyun, Huangpu, Panyu, and Luogang districts ( Figure 1). According to the statistics, the land area of the core region is about 279.63 km 2 , and its urban population and gross domestic product (GDP) in 2017 were 5.47 million and 1033.73 billion, respectively. The land area of the peripheral region is about 1809.90 km 2 , and its urban population and GDP in 2017 were 4.61 million and 702.99 billion, respectively. The combined effect of accelerated urbanization and insufficient land resources has led to rapid urban renewal processes, resulting in microclimate and thermal environmental changes [33].

Imagery
Free and open datasets including time series Landsat data and yearly ALOS PALSAR data were collected from the United States Geological Survey and the Japan Aerospace Exploration Agency. Landsat imagery included available Landsat Thematic Mapper (TM), Enhanced Thematic Mapper Plus (ETM+), and Operational Land Imager (OLI)/Thermal Infrared Sensor (TIRS) images (path/row 122/044) from 2007 to 2009 and from 2015 to 2017 for identifying urban land change from surface reflectance data; 69 images were obtained over the first three-year period (2007)(2008)(2009)) and 77 images were obtained during the latter three-year period (2015)(2016)(2017). In addition, we collected Landsat 5 image (path/row 122/044) on 26 July, 2008, and Landsat 8 image (path/row 122/044) on 22 July, 2018, for LST measurements, which included raw data, top-of-atmosphere reflectance data, and surface reflectance data. Before using the optical imagery provided by Landsat TM, Landsat ETM+, and Landsat OLI, images with poor quality were identified and removed: (1) the Fmask algorithm was first applied to develop cloud and cloud shadow layers for time series analysis of Landsat TM, Landsat ETM+, and Landsat OLI images [38]; and (2) to improve the continuity and usability of the

Imagery
Free and open datasets including time series Landsat data and yearly ALOS PALSAR data were collected from the United States Geological Survey and the Japan Aerospace Exploration Agency. Landsat imagery included available Landsat Thematic Mapper (TM), Enhanced Thematic Mapper Plus (ETM+), and Operational Land Imager (OLI)/Thermal Infrared Sensor (TIRS) images (path/row 122/044) from 2007 to 2009 and from 2015 to 2017 for identifying urban land change from surface reflectance data; 69 images were obtained over the first three-year period (2007-2009) and 77 images were obtained during the latter three-year period (2015-2017). In addition, we collected Landsat 5 image (path/row 122/044) on 26 July, 2008, and Landsat 8 image (path/row 122/044) on 22 July, 2018, for LST measurements, which included raw data, top-of-atmosphere reflectance data, and surface reflectance data. Before using the optical imagery provided by Landsat TM, Landsat ETM+, and Landsat OLI, images with poor quality were identified and removed: (1) the Fmask algorithm was first applied to develop cloud and cloud shadow layers for time series analysis of Landsat TM, Landsat Remote Sens. 2020, 12, 794 4 of 15 ETM+, and Landsat OLI images [38]; and (2) to improve the continuity and usability of the images, we built a scan line corrector (SLC) off layer to mask no-observation pixels in Landsat ETM+ images due to the failure of the on-board SLC from 31 May 2003.
Yearly global 25 m resolution ALOS PALSAR imagery provided L-band dual-polarization data, including horizontal transmit/horizontal receive (HH) and horizontal transmit/vertical receive (HV) polarizations, which are capable of full-polarization and multi-view Earth observation. Here, ALOS PALSAR images from 2007-2009 and from 2015-2017 were collected and subjected to radiation and geometric correction in google earth engine (GEE), which is a free and open platform for batch processing of satellite image data. The PALSAR HH and HV digital number (DN) values were then converted into gamma-naught to reduce the influence caused by the change of the backscattering coefficient in the distance direction by the following calibration coefficient [39]: To reduce noise, a 3 × 3 pixel median filter was applied to PALSAR HH and HV images, which were resampled into 30 m images to match Landsat images by the nearest neighborhood interpolation [40].

Ground Reference Data for Approach Training and Validation
We collected 50 completed urban renewal project files from Guangzhou Municipal People's Government during the study period as ground reference data. The specific information in the project files includes the location, scope, duration, transformation types, change in land use type, and so on. According to the urban renewal project files and very high-resolution Google Earth images of the two time periods (approximately 2007 and 2017), we selected 50 urban renewal land (URL) points and 50 general urban land (GUL) points as training data (Figure 2a), which were generated in 60 × 60 m 2 regions of interest (ROIs). images, we built a scan line corrector (SLC) off layer to mask no-observation pixels in Landsat ETM+ images due to the failure of the on-board SLC from 31 May 2003. Yearly global 25 m resolution ALOS PALSAR imagery provided L-band dual-polarization data, including horizontal transmit/horizontal receive (HH) and horizontal transmit/vertical receive (HV) polarizations, which are capable of full-polarization and multi-view Earth observation. Here, ALOS PALSAR images from 2007-2009 and from 2015-2017 were collected and subjected to radiation and geometric correction in google earth engine (GEE), which is a free and open platform for batch processing of satellite image data. The PALSAR HH and HV digital number (DN) values were then converted into gamma-naught to reduce the influence caused by the change of the backscattering coefficient in the distance direction by the following calibration coefficient [39]: To reduce noise, a 3 × 3 pixel median filter was applied to PALSAR HH and HV images, which were resampled into 30 m images to match Landsat images by the nearest neighborhood interpolation [40].

Ground Reference Data for Approach Training and Validation
We collected 50 completed urban renewal project files from Guangzhou Municipal People's Government during the study period as ground reference data. The specific information in the project files includes the location, scope, duration, transformation types, change in land use type, and so on. We also generated 1000 60 m × 60 m rectangles (ROIs) for accuracy assessment through stratified random sampling, with a minimum distance of 1000 m between ROIs (Figure 2b). In total, URL consisted of 100 ROIs, and GUL included 900 ROIs. The ROIs were overlaid with very high spatial resolution images from Google Earth (Figure 2c-g) in the study area for 2007 and 2017, respectively. If more than 50% of the ROIs were URL or GUL, the ROI would be classified as URL or GUL, respectively. Based on these classification results, we constructed a confusion matrix to evaluate the classification accuracy of URL and GUL retrieved from remote sensing images, including producer accuracy, user accuracy, overall accuracy, and the kappa coefficient. We also generated 1000 60 m × 60 m rectangles (ROIs) for accuracy assessment through stratified random sampling, with a minimum distance of 1000 m between ROIs (Figure 2b). In total, URL consisted of 100 ROIs, and GUL included 900 ROIs. The ROIs were overlaid with very high spatial resolution images from Google Earth (Figure 2c-g) in the study area for 2007 and 2017, respectively. If more than 50% of the ROIs were URL or GUL, the ROI would be classified as URL or GUL, respectively. Based on these classification results, we constructed a confusion matrix to evaluate the classification accuracy of URL and GUL retrieved from remote sensing images, including producer accuracy, user accuracy, overall accuracy, and the kappa coefficient.

Methodology
We collected time series Landsat TM, ETM+, OLI images, and ALOS PALSAR images and carried out data preprocessing ( Figure 3). Then, urban land in 2007 and 2017 was identified using annual maximum normalized difference vegetation index (NDVI) calculated by Landsat imagery and HH gamma-naught backscatter data from ALOS PALSAR. We identified urban renewal land using HV gamma-naught backscatter data and 100 ROIs from urban renewal project files and Google Earth images in non-change of urban land. And urban renewal land was validated by 1000 ROIs randomly sampled from Google Earth images. After retrieving the LST from Landsat imagery, we further analyzed the influence of urban renewal on LST.

Methodology
We collected time series Landsat TM, ETM+, OLI images, and ALOS PALSAR images and carried out data preprocessing ( Figure 3). Then, urban land in 2007 and 2017 was identified using annual maximum normalized difference vegetation index (NDVI) calculated by Landsat imagery and HH gamma-naught backscatter data from ALOS PALSAR. We identified urban renewal land using HV gamma-naught backscatter data and 100 ROIs from urban renewal project files and Google Earth images in non-change of urban land. And urban renewal land was validated by 1000 ROIs randomly sampled from Google Earth images. After retrieving the LST from Landsat imagery, we further analyzed the influence of urban renewal on LST.

Identification of Urban Land and Urban Change
Generally, urban land is composed mainly of three-dimensional buildings, mixed with different land cover types (e.g., roads, grasslands, trees, and water bodies). The greenness is relatively low inside urban land areas. To identify urban land from other land use types, e.g., croplands, forests, and wetlands, the NDVI calculated using Landsat imagery and HH gamma-naught backscatter data from ALOS PALSAR were jointly adopted in this study to map urban land, based on earlier successful attempts using this approach [40]. The NDVI is given by

Identification of Urban Land and Urban Change
Generally, urban land is composed mainly of three-dimensional buildings, mixed with different land cover types (e.g., roads, grasslands, trees, and water bodies). The greenness is relatively low inside urban land areas. To identify urban land from other land use types, e.g., croplands, forests, and wetlands, the NDVI calculated using Landsat imagery and HH gamma-naught backscatter data from ALOS PALSAR were jointly adopted in this study to map urban land, based on earlier successful attempts using this approach [40]. The NDVI is given by  ) for each year of the study (2007, 2008, 2009, 2015, 2016, and 2017). PALSAR HH and Landsat NDVI max represent the features of building structure and greenness, respectively [40]. Therefore, we applied both of them to map urban land in the study area from 2007-2009 and from 2015-2017. Urban land over the six years represented in the two time periods was extracted by relatively high HH backscatter values and relatively low NDVI max values according to the mean value and standard deviation of land cover types from Landsat and PALSAR images satisfying the threshold referring to [40]:   [40]. Therefore, we applied both of them to map urban land in the study area from 2007-2009 and from 2015-2017. Urban land over the six years represented in the two time periods was extracted by relatively high HH backscatter values and relatively low NDVImax values according to the mean value and standard deviation of land cover types from Landsat and PALSAR images satisfying the threshold referring to [40]:

Algorithms for Identifying Urban Renewal Land through PALSAR Images
In general, because the floors and heights of buildings commonly show considerable change before and after the urban renewal process, this study extracted URL on a large scale according to the height difference of buildings between the former period and the latter period. Previous research demonstrated the sensitivity of synthetic aperture radar (SAR) to building height [41,42]. Therefore, we identified URL from UU based on the building height retrieval from PALSAR imagery in different periods.

Algorithms for Identifying Urban Renewal Land through PALSAR Images
In general, because the floors and heights of buildings commonly show considerable change before and after the urban renewal process, this study extracted URL on a large scale according to the height difference of buildings between the former period and the latter period. Previous research demonstrated the sensitivity of synthetic aperture radar (SAR) to building height [41,42]. Therefore, we identified URL from UU based on the building height retrieval from PALSAR imagery in different periods.  As one of the land use change types of UU, URL was mainly featured by the differences in PALSAR HV backscatter average values before and after urban renewal in our study. To distinguish URL from GUL, we used the following threshold value of the difference between the two periods: where HVformer is the average HV value of each pixel of UU from 2007 to 2009, and HVlatter is the average HV value of the corresponding pixel from 2015 to 2017. Because some pixels of UU were mixed with URL and GUL, the lowest 5% values of ground reference for URL were excluded to reduce false identification and confusion error. Moreover, we removed the pixels that covered less than four independent pixels in the range of 3 × 3 pixels, to reduce URL noise.

Land Surface Temperature Retrieval
Urban renewal aims to replace functionally decaying urban elements with brand-new urban functions by reallocating land use and increasing urban greenness, such as the replacement of heavily polluted factories and thermal power stations with commercial and residential areas. As such, urban renewal was expected to have a significant effect on LST and UHI.
Thermal infrared sensor (TIRS) data from remote sensing is considered to be an effective way to retrieve the LST. We selected two scenes from Landsat images to retrieve the LST; the Landsat Thematic Mapper (TM) on 26 July 2008, and the Landsat Thermal Infrared Sensor (TIRS) on 22 July 2018. The previous image had no cloud coverage in the study area, whereas the latter image had 14% cloud coverage, corresponding to the image with the least amount of cloud coverage of Landsat images in the summer of 2015-2018. Given that the main coverage area of cloud was farmland and woodland, cloud coverage had little impact on this research. Specifically, the LST of the study area As one of the land use change types of UU, URL was mainly featured by the differences in PALSAR HV backscatter average values before and after urban renewal in our study. To distinguish URL from GUL, we used the following threshold value of the difference between the two periods: where HV former is the average HV value of each pixel of UU from 2007 to 2009, and HV latter is the average HV value of the corresponding pixel from 2015 to 2017. Because some pixels of UU were mixed with URL and GUL, the lowest 5% values of ground reference for URL were excluded to reduce false identification and confusion error. Moreover, we removed the pixels that covered less than four independent pixels in the range of 3 × 3 pixels, to reduce URL noise.

Land Surface Temperature Retrieval
Urban renewal aims to replace functionally decaying urban elements with brand-new urban functions by reallocating land use and increasing urban greenness, such as the replacement of heavily polluted factories and thermal power stations with commercial and residential areas. As such, urban renewal was expected to have a significant effect on LST and UHI.
Thermal infrared sensor (TIRS) data from remote sensing is considered to be an effective way to retrieve the LST. We selected two scenes from Landsat images to retrieve the LST; the Landsat Thematic Mapper (TM) on 26 July 2008, and the Landsat Thermal Infrared Sensor (TIRS) on 22 July 2018. The previous image had no cloud coverage in the study area, whereas the latter image had 14% Remote Sens. 2020, 12, 794 8 of 15 cloud coverage, corresponding to the image with the least amount of cloud coverage of Landsat images in the summer of 2015-2018. Given that the main coverage area of cloud was farmland and woodland, cloud coverage had little impact on this research. Specifically, the LST of the study area was retrieved from Band 6 of the Landsat TM imagery on 26 July 2008, and from Band 10 of Landsat TIRS imagery on 22 July 2018, using the generalized single-channel algorithm [43,44], which has been proven an effective method to retrieve LST from Landsat imagery [45][46][47]. This algorithm mainly calculates LST from a combination of the surface emissivity, at-sensor registered radiance, atmospheric functions, and parameters dependent on Planck's function. Figure 6 showed the LST outcomes retrieved from Landsat TM imagery on 26 July 2008, and Landsat TIRS imagery on 22 July 2018. The overall temperature in the latter period was much higher than that in the former period. Compared with the former period, the relatively high temperature areas showed an obvious outward diffusion trend in the latter period.
Remote Sens. 2020, 12, x FOR PEER REVIEW 8 of 15 was retrieved from Band 6 of the Landsat TM imagery on 26 July 2008, and from Band 10 of Landsat TIRS imagery on 22 July 2018, using the generalized single-channel algorithm [43,44], which has been proven an effective method to retrieve LST from Landsat imagery [45][46][47]. This algorithm mainly calculates LST from a combination of the surface emissivity, at-sensor registered radiance, atmospheric functions, and parameters dependent on Planck's function. Figure 6 showed the LST outcomes retrieved from Landsat TM imagery on 26 July 2008, and Landsat TIRS imagery on 22 July 2018. The overall temperature in the latter period was much higher than that in the former period. Compared with the former period, the relatively high temperature areas showed an obvious outward diffusion trend in the latter period.

Spatial Statistics
To reflect the difference in heat generation between URL and GRL in different periods, we calculated the LST difference between the two types of urban land: where T , is the mean LST of URL in district j during period i; T , is the mean LST of GUL in district j during period i; and D , is the mean LST difference between URL and GUL in district j during period i. If D , < 0, the mean LST of the URL is lower than those of the GUL. Otherwise, the mean LST of the URL is higher than those of GUL.
To further clarify the impact of urban renewal on LST in urban areas, we first calculated the mean LST difference between URL and GUL in different periods, and then calculated the difference between the two differences to show the specific effects of urban renewal on different urban areas: where D , is the mean LST difference between URL and GUL in district j during period i + 1, and D is the difference between the latter period and the former period. If D < 0, urban renewal reduced the LST in specific urban areas. Otherwise, urban renewal led to an increase in LST.

Spatial Statistics
To reflect the difference in heat generation between URL and GRL in different periods, we calculated the LST difference between the two types of urban land: where T URL i,j is the mean LST of URL in district j during period i; T GUL i,j is the mean LST of GUL in district j during period i; and D i,j is the mean LST difference between URL and GUL in district j during period i. If D i,j < 0, the mean LST of the URL is lower than those of the GUL. Otherwise, the mean LST of the URL is higher than those of GUL.
To further clarify the impact of urban renewal on LST in urban areas, we first calculated the mean LST difference between URL and GUL in different periods, and then calculated the difference between the two differences to show the specific effects of urban renewal on different urban areas: where D i+1,j is the mean LST difference between URL and GUL in district j during period i + 1, and D j is the difference between the latter period and the former period. If D j < 0, urban renewal reduced the LST in specific urban areas. Otherwise, urban renewal led to an increase in LST.

Precision of the Urban Renewal Interpretation
Based on the ROIs of URL and GUL judged in Section 2.3, we assessed the precision of assigning URL and GUL using the confusion matrix ( Table 1). The confusion matrix showed that both URL and GUL identifications were assessed with high accuracy. The overall accuracy and kappa coefficient of URL were 97% and 0.82, respectively. The error classification of URL may be caused by the complex land cover types and fragmented landscape patterns of the pixels. The user accuracy and producer accuracy of GUL were up to 98%, which was higher than for URL.

Spatial Distribution Characteristics of Urban Renewal Land
During 2007-2017, the URL area was 13.18 km 2 in total, accounting for 1.88% of the area of urban land (Table 2). URL was attached to GUL and widely distributed in the study area ( Figure 7); additionally, URL showed a relatively concentrated distribution in the center of the study area. For the core region, the URL proportion reached 2.49%, which was 0.61 higher than the overall level. The URL proportions for Tianhe and Haizhu districts were relatively high among the core region, at 2.93% and 2.46% respectively. For the peripheral region, the URL proportion was 1.67%, which was 0.21 lower than the overall level. Although the areas of URL in Panyu and Baiyun districts were relatively high, accounting for 45.19% of the total URL, the proportions of URL in Luogang and Huangpu districts were relatively high among the peripheral region.

Spatiotemporal Change in the Land Surface Temperature
The LST differences between URL and GUL in different regions were used to indicate the thermal difference between the two types of urban land in different periods (i.e., D2007 and D2017 in Table 3). The actual impact of urban renewal processes on LST was obtained by calculating the difference in the LST over different periods (i.e., D in Table 3). After urban renewal, the relative LST of Liwan, Yuexiu, Haizhu, and Tianhe districts dropped by 0.88, 0.42, 0.43, and 0.10 K, respectively. Thus, the urban renewal process in the core region was conducive to the decline in LST of urban areas. However, Baiyun, Huangpu, Panyu, and Luogang districts in the peripheral region presented completely opposite characteristics, with a rise in the LST of 0.98, 1.03, 1.63, and 2.11 K. The urban renewal process had the opposite thermal effect on the districts in the core and peripheral regions of Guangzhou from 2007 to 2017. Specifically, urban renewal in areas with relatively high populations and land urbanization (e.g., Liwan, Yuexiu, Haizhu, and Tianhe districts) was beneficial in reducing the LST. In contrast, urban renewal in areas with a relatively low

Spatiotemporal Change in the Land Surface Temperature
The LST differences between URL and GUL in different regions were used to indicate the thermal difference between the two types of urban land in different periods (i.e., D 2007 and D 2017 in Table 3). The actual impact of urban renewal processes on LST was obtained by calculating the difference in the LST over different periods (i.e., D in Table 3). After urban renewal, the relative LST of Liwan, Yuexiu, Haizhu, and Tianhe districts dropped by 0.88, 0.42, 0.43, and 0.10 K, respectively. Thus, the urban renewal process in the core region was conducive to the decline in LST of urban areas. However, Baiyun, Huangpu, Panyu, and Luogang districts in the peripheral region presented completely opposite characteristics, with a rise in the LST of 0.98, 1.03, 1.63, and 2.11 K. The urban renewal process had the opposite thermal effect on the districts in the core and peripheral regions of Guangzhou from 2007 to 2017. Specifically, urban renewal in areas with relatively high populations and land urbanization (e.g., Liwan, Yuexiu, Haizhu, and Tianhe districts) was beneficial in reducing the LST. In contrast, urban renewal in areas with a relatively low population and minimal land urbanization, such as Baiyun, Huangpu, Panyu, and Luogang districts, showed an increase in LST with urban renewal. This phenomenon can be explained by the following two factors. On the one hand, because the initial building density and population density in the core region were already very high, urban renewal had little impact on the building density and population density in that area. The urban renewal process effectively added green landscape, which was conducive to adjusting the urban microclimate and alleviating urban thermal environmental risk [48]. At the same time, most of the target sites for urban renewal were inefficient sites, such as urban villages and old factories and thermal power stations, which directly increased the environmental and thermal threats, compared with other urban land types. Thus, urban renewal alleviated these problems through the transformation of extensive land inefficiencies. On the other hand, the initial building density of edge areas was relatively small in the peripheral region; however, the building density and population increased significantly after urban renewal, which resulted in corresponding changes in the urban microclimate and thermal environment. Some related studies have also pointed out that the increase in the building and population densities had a significant impact on UHI and LST [49,50]. Therefore, the urban renewal process in peripheral region districts promoted higher LSTs, to varying intensities.

Comparison with Previous Studies
A large number of studies have shown the close relationship between urbanization and the change in urban thermal environments, as evidenced by higher LST [15,[51][52][53]. But what is the actual impact of urban renewal on LST? The main objective of our research was to identify the key mechanism of urban renewal on LST and to analyze further whether the impact differed with location and/or degree of urbanization. Our research results revealed a difference in the impact of urban renewal on the LST between core and peripheral districts. Specifically, urban renewal in the core region with a relatively high intensity of urbanization was beneficial to reducing LST, whereas in the peripheral region, urban renewal increased the LST.
Despite there being few studies in this field, some useful exploration has been carried out. Hou et al. showed that the LST in several high-temperature areas of central Fuzhou City declined with the development of urban renewal from 2003 to 2016, which is consistent with the changing tendencies of districts in the core region of our study area [32]. Pan et al. examined the Tianhe District of Guangzhou City as a key research area, confirming that changes in land use and spatial structures associated with urban renewal significantly eliminated or weakened the intensity of UHIs [33]. Peng et al. presented a comprehensive mathematical model describing fluid flow and heat transfer characteristics to provide strategies for urban renewal and improving the wind and thermal environments in the old city district of Wuhan, which has a dense population and experiences a strong UHI effect [34]. Existing research has focused on the impact of urban renewal on LST in areas with high urbanization intensity, high populations, and high building densities. However, no prior research has revealed the spatial difference of urban renewal impact on LST, especially in the context of a comparative analysis of the impact difference of areas with different urbanization degrees. Therefore, for regions with relatively low urbanization, such as urban peripheral regions, the impact of urban renewal on LST deserves further study.

Future Work
With the continuous progress of urbanization, the urban landscape continues to evolve [54,55]. Many sites and buildings that are not suitable for the development of modern cities, such as old towns, factories, and villages, must be updated. Thus, with increasing urban populations, there has been great interest in urban renewal and the preservation of resources for a sustainable and healthy living environment [32,34]. Given the lack of extensive research in this area, we hope that the results from this study will provide a framework for future research on LST changes caused by urban renewal processes. Our research group hopes to improve on the spatiotemporal acquisition accuracy of imagery from large-scale urban renewal areas and land type identification for LST calculations. For example, some houses in urban villages have illegally-built additional floors, resulting in obvious changes in building height. Although illegal construction does not belong to the category of urban renewal, it may be identified as URL. In addition, specific landscape changes in the urban renewal process with LST changes can be used to refine the heat exchange processes of urban microclimates driven by different landscape change types, so as to provide more detailed guidance for future urban renewal projects.

Conclusions
Urban renewal is an effective method for solving shortages in urban land resources with urbanization by redeveloping land that is used inefficiently. At the same time, urban renewal affects the urban microclimate and is steadily receiving more attention. In this study, we proposed a general, flexible framework to determine how urban renewal processes influence LST, using multi-source remote sensing images. We presented a case study of core and peripheral regions in Guangzhou City, China, between 2007 and 2017 to analyze the spatial differences in different urbanized areas on the basis of urban renewal area identification and LST remote sensing inversion. Our results indicated that the urban renewal process is conducive to a decline in LST in core region districts with relatively high urbanization, whereas peripheral regions with relatively low urbanization presented the opposite characteristics. The outcomes can be explained by the joint influence of population density, building density, and landscape pattern changes in the process of urban renewal. It is our hope that the results from this study provide a new perspective on the impact of urban renewal on urban microclimates and thermal environmental changes to promote greater awareness of this issue in future planning and management practices.