Impacts of Neighboring Buildings on the Cold Island Effect of Central Parks: A Case Study of Beijing, China

: Urban parks have been considered as an e ﬀ ective measure to mitigate the urban heat island (UHI) e ﬀ ects. Many studies have investigated the impacts of shape, size and interior components on the cold island e ﬀ ect (CIE) of parks, while little attention has been given to the impact of neighboring buildings. Thus, taking twenty-two parks in Beijing as samples, this study investigated the impacts of the neighboring building on the CIE of central parks. The results showed that the average land surface temperature (LST) of parks are 30.98 ◦ C in summer and − 1.10 ◦ C in winter. Parks have a strong CIE in summer, and average cold island footprint (CIF) and LST di ﬀ erence are 0.15 km 2 and 2.01 ◦ C higher than that in winter. The components of the building in the CIF of parks are dominated by middle-rise building (MRB), followed by low-rise building (LRB), and high-rise building (HRB) is the least dominant. The percentage of landscape (PLAND) and landscape shape index (LSI) of MRB, and perimeter area fractal dimension (PAFRAC) of LRB are signiﬁcantly related to CIF in summer and winter. This study could extend scientiﬁc understanding of the impacts of neighboring buildings on the CIE of central parks, and could guide urban planners in mitigating the UHI e ﬀ ects through the rational allocation of buildings.


Introduction
Urbanization transforms the natural surface into impervious surface, while a high population in urban area accelerates this transformation [1][2][3]. This transformation was considered as one of the major causes for the phenomenon of urban heat island (UHI) [4,5]. The UHI refers to urban temperatures being higher than those in rural areas [6]. This negatively affects the quality and health of inhabitants [7,8]. Thus, how to mitigate the UHI effects has attracted wide concern from scholars and government managers.
The urban thermal environment has been extensively investigated by scholars, and the UHI effects is influenced by many factors [9][10][11][12]. It is widely accepted that greater absorption of solar radiation, lower efficiency of longwave radiation emission, energy storage by buildings and paved surface, lower evaporative cooling, and anthropogenic heat release in the urban area together lead to the UHI effects [13]. In addition, many studies have investigated relationships between urban (e.g., water bodies, roads, and built-up areas) and geographic (e.g., Normalized Difference Vegetation Index (NDVI), Normalized Difference Built-up Index (NDBI), and Normalized Difference Water Index (NDWI)) indicators and urban thermal environment [8,[14][15][16][17][18]. Most of the existing research emphasized that built-up areas could increase the UHI effects, while water bodies and vegetation lands decrease it [1,17,19,20].

Study Area
Beijing, the capital of China, is located in the northern of the country (39 • 26 -41 • 03 N, 115 • 25 -117 • 30 E). It covers an area of 164,100 km 2 , with a population of 21.71 million as of 2015 ( Figure 1). This region has a typical warm-temperate, semi-humid continental monsoon climate, with hot and rainy weather in summer and a cold and dry winter. The annual mean temperature and precipitation are 12.3 • C, and 300 to 800 mm, respectively. Beijing has experienced rapid urbanization since the 1978 economic reform and opening up policy. It is also one of the cities with serious UHI effects in China. The study area is located within the 3rd ring road of Beijing metropolitan area, which covers an area of approximately 80,000 km 2 ( Figure 1). There are also many parks inside this region. Twenty-two parks were chosen to investigate the impacts of neighboring buildings on the CIE of central parks. The spatial distribution and details of parks can see in Figure1 and Table 1, respectively. Eleven parks are distributed in the 2nd ring, and the others are distributed in the 3rd ring. The area of parks ranged from 0.04 km 2 (WS) to 1.95 km 2 (TT), with an average area of 0.39 km 2 .   The spatial distribution and details of parks can see in Figure 1 and Table 1, respectively. Eleven parks are distributed in the 2nd ring, and the others are distributed in the 3rd ring. The area of parks ranged from 0.04 km 2 (WS) to 1.95 km 2 (TT), with an average area of 0.39 km 2 .

Data Sources
Building dataset, Gaofen2 images, and Landsat 8 image were used in this study, as shown in Table 2. Building dataset was derived from Resources and Environmental Scientific Data Center (RESDC), Chinese Academy of Sciences (CAS). It provides building datasets in Chinese major cities, such as Beijing, Tianjin and Jinan. Some parameters, such as building outline and floors, are included in this dataset. It can reflect the two-and three-dimensional characteristics of the building. Taking Baidu Street View map as criteria, 500 samples were randomly selected to verify the accuracy of this dataset. The result shows that the overall accuracy of this dataset was approximately 97%, which can guarantee the accuracy of building classification and subsequent analysis.
The outline of the park was mapped using GF2 images, which were derived from RESDC. To cover the study area, two GF2 images taken on 2 September 2015, and 12 September 2015, respectively, were used in this study, with a panchromatic (spatial resolution of 1 m) and four multi-spectral (blue, green, red, and near infrared, spatial resolutions of 4 m) bands.
To retrieve the LST, two Landsat 8 OLI images were derived from the United States Geological Survey (USGS) on 22 August 2015, and 14 February 2016, respectively. The thermal infrared band (band 10, spatial resolution of 120 m), was used to retrieve the LST. The images were selected to be as cloud-free as possible. Although the cloud cover of this image was 33.33% in summer, the study area was cloud-free.

Methods
The flowchart of this study is as follows (Figure 2). First of all, the LST was retrieved from two Landast 8 OLI images (summer and winter), and the outline of parks were required by GF2 images. Buildings were classified based on the building dataset. Secondly, the LSTs of parks were calculated, and the LST differences among different parks were analyzed. Thirdly, the CIF of parks were estimated, and CIF of parks were analyzed. Fourthly, average LST of CIFs were calculated, and LST difference between park and CIF were analyzed. Fifthly, the composition of the building neighboring the park was analyzed. Finally, the impacts of neighboring buildings on the CIE of parks was investigated.   The image-based method (IBM) is used for retrieval of LST [15]. The equation is as follows where T s is the LST, T represents the at-satellite brightness temperature, λ and ε are the wavelengths at the center of emitted radiance and spectral emissivity, respectively. ρ is a constant value derived from the headed files of the Landsat 8 OLI image. In this case, ρ equals 1.438 × 10 −2 mK. The estimation of T can divide into two steps. First, the digital number (DN) of thermal band (band 10) is converted to radiation. The equation is as follows where Lλ is the spectral radiance (W·m −2 ·sr −1 ·µm −1 ), ML and AL are rescaled gain (value = 3.342 × 10 −4 ) and rescaled bias (value = 0.1).
After calculating Lλ, the T can calculate as follows where K 1 = 480.89 W·m −2 ·sr −1 ·µm −1 and K 2 = 1201.14 K. The estimation of ε can also divide into two steps. First, the land surface is classified into three groups, water, urban, and natural surface. Then, the ε of water is set as 0.995, while for urban and natural surface, the ε can calculate by fractional vegetation cover (FVC) (Equations (4) and (5)). The FVC is calculated by NDVI (Equation (6)) where ε urban and ε natural sur f ace are the emissivity values for urban and natural surface, respectively. FVC and NDVI are the fractional vegetation cover and Normalized Difference Vegetation Index. NDVI s and NDVI v are the NDVI for the vegetation and soil, respectively.

Extraction of Park and Classification of Buildings
The outline of parks was mapped using GF2 images based on the visual interpretation method. Taking TRT as an example, the outline of TRT was extracted based on GF2 images ( Figure 3). In this study, twenty-two parks were extracted, and the spatial distribution of parks can be seen in Figure 1.

Extraction of Park and Classification of Buildings
The outline of parks was mapped using GF2 images based on the visual interpretation method. Taking TRT as an example, the outline of TRT was extracted based on GF2 images ( Figure 3). In this study, twenty-two parks were extracted, and the spatial distribution of parks can be seen in Figure 1. To evaluate the impact of different building types on the CIE of parks, building dataset was used in this study. This dataset includes the outline of the buildings, and the number of building floors. Furthermore, according to Chinese Residential Building Code, buildings were classified into three types based on building floors, including low-rise building (LRB), middle-rise building (MRB), and high-rise building (HRB), respectively. The classification standard can be seen in Table 3. Table 3. Classification standard of building. To evaluate the impact of different building types on the CIE of parks, building dataset was used in this study. This dataset includes the outline of the buildings, and the number of building floors. Furthermore, according to Chinese Residential Building Code, buildings were classified into three types based on building floors, including low-rise building (LRB), middle-rise building (MRB), and high-rise building (HRB), respectively. The classification standard can be seen in Table 3. Table 3. Classification standard of building.

Estimation of CIF
The area where the CIE of parks occurs is defined as the CIF. The greater the CIF, the stronger the park CIE. According to the urban heat island footprint, the buffer method was used to quantitatively measure the CIF of parks [43][44][45].
Firstly, we constructed buffers of the outline of each park with a distance of 50 m. Then, the mean LST in each buffer were calculated. A scatter plot of the relationship between the buffer number and average LST was drawn. Finally, the CIF of the park was determined by identifying the breakpoint.
Taking TRT as an example, we first constructed buffers of the outline of this park with a distance of 50 m, the buffer diagram is shown in Figure 4. Then, the average LST in each buffer were calculated, and the relationship between the buffer number and average LST was drawn in Figure 5. The result showed that the average LST of TRT is 29.91 • C. As the rings extended outward, the average LST gradually increases, and reached the highest at the fifth rings (35.10 • C). Then, the average LST begins to decrease. Thus, the fifth ring represents the boundary of change in the LST, the area of the five inner rings is determined as the CIF of TRT.
of 50 m, the buffer diagram is shown in Figure 4. Then, the average LST in each buffer were calculated, and the relationship between the buffer number and average LST was drawn in Figure 5. The result showed that the average LST of TRT is 29.91 °C. As the rings extended outward, the average LST gradually increases, and reached the highest at the fifth rings (35.10 °C). Then, the average LST begins to decrease. Thus, the fifth ring represents the boundary of change in the LST, the area of the five inner rings is determined as the CIF of TRT.   and the relationship between the buffer number and average LST was drawn in Figure 5. The result showed that the average LST of TRT is 29.91 °C. As the rings extended outward, the average LST gradually increases, and reached the highest at the fifth rings (35.10 °C). Then, the average LST begins to decrease. Thus, the fifth ring represents the boundary of change in the LST, the area of the five inner rings is determined as the CIF of TRT.

Analytical Methods
Multiple linear regression model was used to explore the effects of neighboring buildings on CIF of central parks. Three building types, including LRB, MRB and HRB, were also considered in this study. To illustrate the composition and configuration characteristics of buildings, some metrics, including percentage of landscape (PLAND), landscape shape index (LSI), perimeter area fractal dimension (PAFRAC), interspersion juxtaposition index (IJI), and aggregation index (AI), which were quantified by Fragstats 4.2, has been used in this study. Particular, PLAND is used to quantify the composition characteristics of buildings, and LSI, PAFRAC, IJI, and AI are used to quantify the configuration characteristics of buildings.

LST Difference among Different Park
Spatial distribution of LST in the study area is shown in Figure 6. The result shows that the LST of the study area ranged from 19.76 to 43.52 • C in summer, and −20.77 to 8.73 • C in winter. The average LSTs were 33.67 • C in summer and −0.33 • C in winter in the study area. It is obvious that LSTs of parks are much lower than the neighboring area, indicating that parks are cold islands. Additionally, compared with winter, LSTs of parks in summer are significantly lower than neighboring areas. This indicates that there are large LST differences between parks and neighboring areas in summer compared to winter.
configuration characteristics of buildings.

LST Difference among Different Park
Spatial distribution of LST in the study area is shown in Figure 6. The result shows that the LST of the study area ranged from 19.76 to 43.52 °C in summer, and −20.77 to 8.73 °C in winter. The average LSTs were 33.67 °C in summer and −0.33 °C in winter in the study area. It is obvious that LSTs of parks are much lower than the neighboring area, indicating that parks are cold islands. Additionally, compared with winter, LSTs of parks in summer are significantly lower than neighboring areas. This indicates that there are large LST differences between parks and neighboring areas in summer compared to winter. To further investigate LST differences among different parks, average LSTs of parks in different seasons were calculated (Table 4). In summer, LSTs of parks ranged from 29.25 °C (YYT) to 32.84 °C (WS), with an average LST of 30.98 °C. It is lower than the average LST (33.67 °C) of the study area. Among them, the LST of five parks, including YYT, ZZY, LT, BH, and TRT, is lower than 30 °C. Additionally, the LST of five parks, including WS, XW, WFY, DD, and DGY, is higher than 32 °C. Interestingly, the areas of these five parks are relatively small. In winter, the LST of parks ranged from −2.13 °C (SCH) to 0.29 °C (LHC), with an average LST of −1.10 °C. It should be noted that, except for LHC, the LSTs of other parks are lower than 0 °C. Furthermore, compared with summer, there is little difference in the LSTs of parks. To further investigate LST differences among different parks, average LSTs of parks in different seasons were calculated (Table 4). In summer, LSTs of parks ranged from 29.25 • C (YYT) to 32.84 • C (WS), with an average LST of 30.98 • C. It is lower than the average LST (33.67 • C) of the study area. Among them, the LST of five parks, including YYT, ZZY, LT, BH, and TRT, is lower than 30 • C. Additionally, the LST of five parks, including WS, XW, WFY, DD, and DGY, is higher than 32 • C. Interestingly, the areas of these five parks are relatively small. In winter, the LST of parks ranged from −2.13 • C (SCH) to 0.29 • C (LHC), with an average LST of −1.10 • C. It should be noted that, except for LHC, the LSTs of other parks are lower than 0 • C. Furthermore, compared with summer, there is little difference in the LSTs of parks.

CIF of Parks
The CIF of parks is shown in Figure 7. The results show that the CIF of parks ranged from 0.19 to 3.55 km 2 in summer, and ranged from 0.11 to 1.87 km 2 in winter. The average CIFs were 0.67 in summer and 0.53 km 2 in winter in the study area. In summer, the CIFs of YYT (3.55 km 2 ), BZ (1.10 km 2 ), and TRT (1.05 km 2 ) are large. In contrast, the CIFs of NG, DD, WS, and QNH are small, which are all below 0.30 km 2 . It should be noted that the areas of all four parks are small. In winter, the CIF of four parks, including LT (1.87 km 2 ), YT (1.30 km 2 ), YYT (1.07 km 2 ) and TRT (1.05 km 2 ), is higher than 1 km 2 . The CIF of other parks is all below 1 km 2 . In particular, the CIF of JS is the smallest (0.11 km 2 ).

LST Difference between Park and CIF
After determining the CIF of parks, the average LSTs of CIFs were also calculated, and LST differences between park and CIF are shown in Table 5. The results show that there is a significant LST difference between park and CIF. The average LST of CIF is higher than that of park, indicating that the LST of parks is lower than in the neighboring area.  The CIF of parks is different in different seasons. In general, the CIF of parks in summer is larger than that in winter, with an average area that is 0.14 km 2 larger. This indicates that the park has a better CIE in summer than that in winter. It is also associated with a lower average temperature in winter. In particular, there are also six parks, such as YRT, SCH, and DT, with consistent CIF in summer and winter. Additionally, the CIFs of five parks, such as LT, WS, and TT, are smaller in summer than in winter.

LST Difference between Park and CIF
After determining the CIF of parks, the average LSTs of CIFs were also calculated, and LST differences between park and CIF are shown in Table 5. The results show that there is a significant LST difference between park and CIF. The average LST of CIF is higher than that of park, indicating that the LST of parks is lower than in the neighboring area.  LST difference between park and CIF ranged from 1.55 • C (NG) to 4.45 • C (BH) in summer, with an average of 2.87 • C. For winter, the LST difference is relatively small, it ranged from 0.09 • C (SX) to 1.98 • C (SCH), with an average of 0.86 • C. Compared with different seasons, the LST difference in summer is higher than that in winter. It also indicated that parks have a strong CIE in summer.

Composition of Building Neighboring Parks
The compositions of the buildings neighboring parks were calculated, as shown in Figure 8. The results show that the components of the buildings in the CIF of parks are dominated by MRB, followed by LRB, and HRB is the least dominant ( Figure 8). The proportion of buildings in the CIF of each park ranged from 14.40% to 35.90% in summer and ranged from 14.23% to 37.58% in winter. 1.98 °C (SCH), with an average of 0.86 °C. Compared with different seasons, the LST difference in summer is higher than that in winter. It also indicated that parks have a strong CIE in summer.

Composition of Building Neighboring Parks
The compositions of the buildings neighboring parks were calculated, as shown in Figure 8. The results show that the components of the buildings in the CIF of parks are dominated by MRB, followed by LRB, and HRB is the least dominant ( Figure 8). The proportion of buildings in the CIF of each park ranged from 14.40% to 35.90% in summer and ranged from 14.23% to 37.58% in winter.
In summer, the components of the buildings in the CIF of parks are dominated by MRB, which ranged from 4.62% in NG to 21.00% in RDH (Figure 8a). The proportion of LRB ranged from 4.79% to 23.28%, and the proportion of HRB ranged from 0.53% to 14   In summer, the components of the buildings in the CIF of parks are dominated by MRB, which ranged from 4.62% in NG to 21.00% in RDH (Figure 8a). The proportion of LRB ranged from 4.79% to 23.28%, and the proportion of HRB ranged from 0.53% to 14.44%. For all parks, the neighboring buildings in twelve CIFs are dominated by MRB, such as QNH (18.66%), XW (15.55%), and WS (15.09%), respectively. Additionally, the neighboring buildings in seven CIFs are dominated by LRB, such as QCH (23.28%), JS (19.42%), and TT (11.87%), respectively. Furthermore, the neighboring buildings in three CIFs are dominated by HRB, including DD (14.44%), NG (10.55%), and YT (10.41%).

Impacts of Neighboring Buildings on CIE of Parks
Five metrics, including one composition metric (PLAND), and four configuration metrics (LSI, PAFRAC, IJI and AI), were used to quantitatively investigate the impacts of different types of neighboring buildings on CIE of parks. The impact of composition and configuration variables of buildings on CIE of central parks is shown in Table 6. The results show that there some differences in the key variable between summer and winter. For summer, both composition and configuration variables of MRB, including PLAND and LSI, are significantly related to CIF, suggesting that neighboring MRBs with low density and complex patch shape lead to increased CIE of parks. For winter, only one configuration variable of LRB (PAFRAC) is significantly related to CIF, suggesting that neighboring LRBs with complex patch outline lead to increased CIE of parks.

Scale Effects
IB method was used to retrieve the LST in this study, which has a resolution of 30 m. Inevitably, there are scale effects when carrying out the integration analysis between vector (parks and CIF) and raster data (LST), which can lead to some uncertainties [24]. However, the average area of parks and CIFs are approximately 0.39 and 0.67 km 2 in summer and 0.53 km 2 in winter, respectively. This indicates that park LSTs were extracted based on enough LST pixels, which could ensure the accuracy of the results. Furthermore, Liu et al. (2009) pointed out that 30 m was the optimal resolution for investigating the relationship between land use and LST at the patch level in previous studies [46]. Thus, though this may have scale effects, it does not invalidate our results.

The Relationship between CIF and Park Area
Many studies have pointed out that the CIE of a park is related to its size. In this study, the relationship between the CIF and park area was also evaluated. Figure 9 shows the relationship between CIF and park area. The result shows that the CIF has a positive relationship with park area. This result is consistent with other research [24].

Implications for Designing Buildings Neighboring the Park for Urban Cooling
Urban parks are considered an effective measure to mitigate the UHI effects [24,31,34]. However, the CIE varied in shape, size, interior components, and neighboring environments [24,33]. Due to rapid urbanization, urban surface elements, especially three-dimensional buildings, have changed drastically, resulting in changes in urban regional climate [41]. It inevitably has effects on the CIE of parks.
The composition and configuration of neighboring buildings affect the park's CIE. However, the influence is different in different building types. Thus, we suggest that the spatial composition and configuration of the buildings neighboring parks should be emphasized. For MRB, the lower the density and more complex the patch shape, the higher contribution they make to increased CIF in summer. Particularly, due to the high building density and uniform distribution of MRB, it can be considered as the key heat source [38]. For LRB, the more complex patch outline, the higher the contribution they make to increased CIF in winter. Although high building density of LRB in few CIFs can absorb and emit a lot of energy, the density of LRB in most of CIFs are relatively low. Unlike  Regardless of summer and winter, the CIF has a positive relationship with park area, indicating that the larger the park area, the better the CIE. 48% (summer) and 25% (winter) of CIF variation can be explained by park area, and both models passed the significance test. Compared with different seasons, the result also shows that parks have a stronger CIE in summer than that in winter. This is mainly related to the LST differences between parks and neighboring area in winter.

Implications for Designing Buildings Neighboring the Park for Urban Cooling
Urban parks are considered an effective measure to mitigate the UHI effects [24,31,34]. However, the CIE varied in shape, size, interior components, and neighboring environments [24,33]. Due to rapid urbanization, urban surface elements, especially three-dimensional buildings, have changed drastically, resulting in changes in urban regional climate [41]. It inevitably has effects on the CIE of parks.
The composition and configuration of neighboring buildings affect the park's CIE. However, the influence is different in different building types. Thus, we suggest that the spatial composition and configuration of the buildings neighboring parks should be emphasized. For MRB, the lower the density and more complex the patch shape, the higher contribution they make to increased CIF in summer. Particularly, due to the high building density and uniform distribution of MRB, it can be considered as the key heat source [38]. For LRB, the more complex patch outline, the higher the contribution they make to increased CIF in winter. Although high building density of LRB in few CIFs can absorb and emit a lot of energy, the density of LRB in most of CIFs are relatively low. Unlike LRB and MRB, HRB can create more shade, and large interval of HRB also can be beneficial for ventilation [11,27,38,47]. However, there is no significant contribution to CIF in different seasons. Given that the UHI effect is strongest in summer, the CIF is relatively small in winter. Thus, reasonable planning of the density and shape of MRB neighboring parks should be taken into account in the future, which can be a key factor to increase the CIE of parks.

Limitations
In this study, the impacts of neighboring buildings on CIE of parks were explored during summer and winter daytimes. However, due to a lack of nighttime LST data, the impacts and diurnal variation could not be explored. Additionally, this study was carried out in the Beijing metropolitan area, and results may vary in different cities (e.g., coastal cities and industrial cities) or climate conditions (e.g., tropical monsoon climate and subtropical monsoon climate). Therefore, other cities should be further investigated. Furthermore, this study focuses on characterizing the impacts of neighboring buildings on the CIE of central parks. However, many other landscape factors in CIF, including vegetation, pavements and water, can affect a park's CIE. These factors should be considered in future research to comprehensively understand the impacts.

Conclusions
This study investigated the impacts of neighboring building on the CIE of central parks based on Landsat 8 OLI images, GF2 images, and building dataset. The key characteristics of buildings which influence parks' CIE were identified in summer and winter. Thus, this study could extend scientific understanding of degree to which CIE of central parks are affected by neighboring buildings (especially different types of buildings), and could provide guidance to urban planners on how to mitigate the UHI effects through the rational allocation of buildings.
The average LSTs of parks are 30.98 • C in summer and −1.10 • C in winter, respectively. LSTs of parks in summer are significantly lower than neighboring areas. CIF of parks in summer is higher than that in winter, with an average area of 0.15 km 2 larger than that in winter. Average LST differences between park and CIF are 2.87 • C in summer and 0.86 • C in winter, indicating that parks have a strong CIE in summer. The components of the building in CIF of parks are dominated by MRB, followed by LRB, and HRB is the least dominant. Furthermore, the PLAND and LSI of MRB, and PAFRAC of LRB are significantly related to CIF in summer and winter. This study demonstrated that parks have a stronger CIE in summer than that in winter, which can effectively mitigate the UHI effects. This study also demonstrated that the CIE of parks can vary in different types of neighboring buildings, especially MRB. It is emphasized that reasonable planning of different types of neighboring buildings (spatial composition and configuration) should be designed based on the characteristics of parks in the future. Our findings are a good theoretical contribution to the urban thermal environment research, providing useful guidance on optimizing buildings for urban planners and administrators. The findings can provide new insights into further exploration of how to improve the CIE of parks.