Characterizing spatial and temporal trends of surface urban heat island effect in an urban main built-up area: A 12-year case study in Beijing, China

Remote Sensing of Environment xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

Remote Sensing of Environment journal homepage: www.elsevier.com/locate/rse

Characterizing spatial and temporal trends of surface urban heat island effect in an urban main built-up area: A 12-year case study in Beijing, China Qingyan Menga,c, Linlin Zhanga,b,c,⁎, Zhenhui Suna,b,c, Fei Mengd, Liang Wanga,b,c, Yunxiao Suna,b,c a

Institute of Remote Sensing and Digital Earth, Chinese Academy of Sciences, Beijing 100101, China University of Chinese Academy of Sciences, Beijing 100101, China c Sanya Institute of Remote Sensing, Sanya 572029, China d College of Surveying and Geo-Informatics, Shandong Jianzhu University, Jinan 250101, China b

A R T I C L E I N F O

A B S T R A C T

Keywords: Impervious surface distribution density Remote sensing Urban heat island intensity level Urban main built-up area

Accurately characterizing spatiotemporal changes in surface urban heat islands (SUHIs) is a prerequisite for sustainable urban development. Although urban administrative boundaries have typically been used for SUHI modeling, they are inaccurate, because urban main built-up areas (UMBAs) are not well characterized. In this study, we developed an UMBA extraction method based on impervious surface distribution density (ISDD), to better differentiate suburban boundaries and ensure the integrity of land cover types. Additionally, we propose a new intensity classification method to analyze SUHI spatial distribution and variation. The UMBA was extracted using LANDSAT-8 data, and the temporal dynamics of SUHI intensity (i.e., daily, monthly, seasonal, and yearly changes) were extracted from MODIS data. A case study for Beijing showed that the mean daytime and nocturnal SUHI intensities vary at multiple time scales. In the daytime, SUHI intensities in Beijing UMBA were mainly level-2 and level-3, with central-south Beijing, a high incidence area, at level-3 in spring and summer. At night, with the rise of SUHI intensity levels, the frequency of SUHI intensity levels increased from the periphery to the center within the same season. ISDD had a marked influence on the frequency of SUHI intensity levels during the daytime, and the frequencies of level-1 to level-4 intensities increased with ISDD. This influence tended to weaken when ISDD exceeded 50%.

1. Introduction Temperature distribution depends mainly on latitude, height above sea level, topography, city size, and atmospheric stability (Stewart and Oke, 2009). With rapid urbanization, natural landscapes are replaced by impervious surfaces, which can alter surface radiation, thermal properties, and humidity over urban areas. Among these effects, the urban heat island (UHI) is a phenomenon in which urban areas tend to have higher atmospheric or surface temperatures than their surroundings. In the early 19th century, the UHI effect was discovered (Howard, 1833). Later, many researchers studied it by observing the air temperatures of urban and suburban areas in cities of different latitudes and types (Carlson et al., 1977; Carnahan and Larson, 1990; Matson et al., 1978; Oke and East, 1971; Rao, 1972; Weng et al., 2004; Zhao et al., 2014). With the development of remote sensing technology, UHI effects are typically estimated from thermal infrared remote sensing techniques (Hung et al., 2006; Imhoff et al., 2010; Melaas et al., 2016; Ogashawara and Bastos, 2012; Stathopoulou and Cartalis, 2009; Yusuf ⁎

et al., 2014). Due to easy access and wall-to-wall continuous coverage of urban areas, the surface urban heat island (SUHI) has gained increasing attention in recent decades (Deng and Wu, 2013; Du et al., 2016; Haashemi et al., 2016; Schwarz et al., 2011; Takebayashi and Moriyama, 2007; Xu, 2009). Therefore, land surface temperatures (LSTs) derived from thermal infrared remote sensors are among the most commonly used indicators for SUHI analysis. SUHI intensity is commonly measured by comparing the difference between urban and suburban temperatures (Heaviside et al., 2016; Mohan et al., 2013; Tomlinson et al., 2012). The difference in mean LST between urban areas and water surfaces (Chen et al., 2006), the Gaussian volume model (Quan et al., 2014; Streutker, 2002; Flores R et al., 2016), areas with LST higher than a mean-plus-one standard deviation (Zhang and Wang, 2008), and LST magnitude (Rajasekar and Weng, 2009) are also used in SUHI analysis. Although different types of indicators for heat island intensity have been proposed, the mean temperature difference between urban and suburban area is the most commonly used indicator (Zhou et al., 2013). However, differentiation

Corresponding author at: Institute of Remote Sensing and Digital Earth, Chinese Academy of Sciences, Beijing 100101, China. E-mail addresses: [email protected] (Q. Meng), [email protected] (L. Zhang).

http://dx.doi.org/10.1016/j.rse.2017.09.019 Received 1 November 2016; Received in revised form 9 September 2017; Accepted 16 September 2017 0034-4257/ © 2017 Elsevier Inc. All rights reserved.

Please cite this article as: Meng, Q., Remote Sensing of Environment (2017), http://dx.doi.org/10.1016/j.rse.2017.09.019

Remote Sensing of Environment xxx (xxxx) xxx–xxx

Q. Meng et al.

2.2. Data set

between urban and suburban areas remains problematic in the literature (Schwarz et al., 2011; Stewart and Oke, 2009). Within the SUHI intensity-computing process, there are various methods to delineate boundaries between urban and suburban areas, such as urban administrative boundaries, pixels around urban and rural weather stations, city ring, roads, high-LST areas versus areas with rural land cover, or city land cover versus other types of land use (Roth et al., 1989; Tomlinson et al., 2012; Jin et al., 2011; Mathew et al., 2016; Shen et al., 2016; Walawender et al., 2014). However, urban boundaries extracted using these methods do not perfectly represent the urban main built-up area (UMBA), which refers to relatively centralized areas of constructed and intensive land use. Different definitions of the boundary between urban areas and suburbs yield different UHI intensity results (Wang et al., 2007). Thus, accurate determination of urban-suburban boundaries directly influences the results of SUHI intensity calculation. The spatial and temporal variation of the UHI is one of the most important themes in UHI studies (Quan et al., 2014). After the SUHI intensity is calculated from the difference between urban and suburban temperatures, the SUHI intensity of different periods must be compared using the same standard. Currently, two methods are used to accomplish this. One is equal interval classification, which classifies LST or its alternatives according to specific rules (Zhang et al., 2006; Xu and Chen, 2003; Wang and Sun, 2004; Zhang et al., 2005; Wu et al., 2006). Although this method can reflect the spatial distribution of LST in some degree, the determination of the best dividing point and number of classes is full of uncertainty. Different dividing points and numbers of classes can result in different SUHI structures, and these choices greatly affect SUHI quantitative studies (Chen and Wang, 2009). The second method involves combing the mean value and standard deviation (Zhang, 2006). Few in-depth analysis of this method have been conducted, and still fewer have compared different classification methods using various angles. These methods cannot be applied for long-term time series data. However, accurate comparisons of UHI intensity data are very important in monitoring SUHI growth (Streutker, 2003). Longterm time series data provide insights into the mechanism behind the evolution of the SUHI effect and its relationship with land use/land cover data and/or climate change (Li et al., 2012), and can aid decisionmakers in developing and executing rational land-use policies (Zhang et al., 2013). Therefore, exploring scientific SUHI classification methods for long-term time series data is crucial. The main objectives of this study were: 1) to extract UMBAs based on impervious surface distribution density (ISDD); 2) to explore the mechanism of SUHI evolution at diurnal, monthly, seasonal, and yearly scales, creating a new SUHI intensity classification method for longterm time series data to analyze SUHI spatial characteristics; and 3) to explore the relationship between UMBA and ISDD to provide references for SUHI mitigation. Based on long-term remote sensing time series data, a SUHI evolution analysis was performed for Beijing.

UMBA were extracted from LANDSAT-8 data. Landsat 8 satellite carries two instruments: the Operational Land Imager (OLI) sensor and the Thermal Infrared Sensor (TIRS). These sensors provide improved signal-to-noise ratio (SNR) radiometric performance quantized over a 12-bit dynamic range. Improved SNR performance allows better characterization of land cover state and condition. This study employed MOD11A2 and MYD11A2 Level-3 8-day 1-km LST products from March 2003 to February 2015, which were downloaded from LAADS Web (https://ladsweb.nascom.nasa.gov). These products adopt a universal split-window algorithm by optimizing the observation angle and range of water vapor column contents. The accuracy of the LST algorithm for terrestrial materials of known emissivity approaches 1 K (Wan et al., 2004). During data processing, the process of joint, tailor, and projection transformation were completed using the MODIS Reprojection Tool (MRT). The results were then multiplied by a scale factor of 0.02 to obtain the real temperature of the land surface (Wan, 2006). Through this method, we acquired 2056 MODIS LST products for Beijing (UTM Zone 50N) from 2003 to 2015. 3. Methods UMBA were extracted using Landsat 8 data, which included impervious surface extraction based on the biophysical composition index (BCI) and distance-weighted ISDD calculations, to obtain the Beijing UMBA boundary. Based on MODIS data, we examined the SUHI evolution process at diurnal, monthly, seasonal, and yearly scales. A new SUHI intensity classification method was then used to study SUHI spatial characteristics. The relationship between SUHI intensity and ISDD was explored to provide references for SUHI mitigation. The flow chart for the current study is as follows (Fig. 2): 3.1. Biophysical composition index The BCI proposed by Deng and Wu (2012) was used to extract urban impervious surfaces. BCI is an urban environment index used to distinguish urban terrestrial materials, and is based on the vegetation–impervious surface–soil (V–I–S) model, a segmented model of the urban land surface (Ridd, 1995). BCI can be calculated as follows:

BCI =

(H + L) 2 − V , (H + L) 2 + V

(1)

where TC refers to the components of Tasseled Cap transformation (Kauth and Thomas, 1976), H is ‘high albedo’ or normalized TC1, L is ‘low albedo’ or normalized TC3, V is ‘vegetation’ or normalized TC2. H, V, and L can be described as follows:

2. Study area and data set

H=

TC1 − TC1min , TC1max − TC1min

(2)

V=

TC 2 − TC 2min , TC 2max − TC 2min

(3)

L=

TC 3 − TC 3min , TC 3max − TC 3min

(4)

2.1. Study area This study was conducted in Beijing (Fig. 1), which is located within 39.4–41.6°N, 115.7–117.4°E, and belongs to the northern part of the northern China plain, which possesses sixteen districts and counties. The city's permanent population totaled 21,689,000 by the end of 2014. The climate in Beijing is a typical sub-humid north-temperate continental monsoon climate characterized by hot rainy summers, cold dry winters, and a short spring and autumn. During the past 20 years, Beijing has experienced great changes in urban development. In 2008, the Olympic Games were held in Beijing, which necessitated the construction of many buildings. The UHI effect has become increasingly serious in Beijing during recent years. It is necessary to study the spatiotemporal changes in UHI intensity based on ISDD to solve ongoing problems related to Beijing's development.

where TCi (i = 1, 2, and 3) represents the first three TC components. TCimin and TCimax are the minimum and maximum values of the ith TC component, respectively. Before the BCI index can be derived from Landsat 8 data (November 2014), three preprocessing steps were conducted: radiometric calibration, water masking, and Tasseled Cap transformation (Liu et al., 2014). Water masking was performed by adopting the normalized difference water index (NDWI) threshold value method. The Beijing BCI index was then calculated based on Eqs. (1)–(4), and the impervious surface was 2

Remote Sensing of Environment xxx (xxxx) xxx–xxx

Q. Meng et al.

Fig. 1. Study area.

closer a pixel is to the center, the larger the weight of the impervious surface. This method can measure the density of distribution of impervious surfaces within the radius and can be described as follows:

(

n

Densitys (r ) =

∑i = 1 Bsi⋅ 1 − n

(

∑i = 1 1 −

Di 2r

Di 2r

)

), (5)

where s is the central pixel, Bsi is the value of ith pixel within radius r (impervious surface pixel = 1; permeable surface = 0), Di is the distance between ith pixel and the central pixel, and the summation refers to all pixels within a circle with radius r. The spatial resolution of MODIS 8 day LST products is 1000 m. We selected radii of 500 m and 1000 m to represent the calculation window of one pixel and one-half pixel, respectively. Fig. 2. Flow chart for the current study. BCI: biophysical composition index; UMBA: urban main built-up area; ISDD: impervious surface distribution density; SUHI: surface urban heat island.

3.3. City clustering algorithm The city clustering algorithm (CCA) extracts the urban area boundary using land cover instead of demographic data or administrative definitions (Rozenfeld et al., 2008). CCA establishes the impervious surface distribution area based mainly on impervious surface distribution, in three steps: (1) select impervious surface pixels randomly as central pixels and calculate connected components covered by impervious surfaces using the eight-neighborhood method; (2) set a distance threshold value of L and merge connected components when the distance is less than L; and (3) set the distribution threshold value of S and remove connected components when the number of pixels is less than S to obtain the impervious surface distribution area.

extracted by setting a threshold value of 0.7 based on visual interpretation.

3.2. Distance-weighted impervious surface distribution density computing The extraction of the urban–suburban boundary directly influences SUHI intensity calculations during remote sensing SUHI monitoring. An urban area is a relatively concentrated land use area where construction has been completed, including urban impervious surfaces and functional areas such as green belts and parks. If only impervious surface areas are regarded as urban areas for the purpose of SUHI analysis, then the cooling effect produced by vegetation and water can be ignored. Therefore, it is crucial to extract urban areas accurately. ISDD represents the density of impervious surfaces within a certain distance. Because the impervious surface of urban areas is greater than that of suburban areas, we have developed a new method to extract urban boundaries by measuring ISDD and verifying the integrity of the underlying surface type in the study area. A pixel's ISDD value, Densitys(r), is described as the degree and density of impervious surface distribution of the pixel's surrounding area within a specific radius. The mean ISDD value can reflect the density of the area within the radius, using distance as a weight. The

3.4. SUHI intensity calculation SUHI intensity is calculated as the difference in mean temperature between UMBA and the suburb boundary area. Peng et al. (2011) found that the minimum influence area of the SUHI effect is 150% of the urban area. Thus, this study takes the UMBA extracted by ISDD as Beijing's urban area, and 150% of this UMBA as the boundary. Because the study area is flat, the terrain's effect on the SUHI intensity calculation can be neglected. SUHI intensity (ΔT) is given by the following (Schwarz et al., 2011):

ΔT = TUrban − TBoundary, 3

(6)

Remote Sensing of Environment xxx (xxxx) xxx–xxx

Q. Meng et al.

4.1.2. Impervious surface distribution density The impervious surface distribution (Fig. 3(B)) was used as input to calculate ISDD, as shown in Fig. 4. At r = 500 m and r = 1000 m, the distributions were similar, each having a ring-shaped border zone with eastern and western districts at the center. The ISDD of the eastern and western districts was 10–25%. The “outer ring” of these districts (including: Haidian District, Shijingshan District, Fengtai District, and Chaoyang District) is a more concentrated developing area of Beijing; thus, the ISDD exceeded 25%. Outside of this region (the “outer ring” of Haidian District, Shijingshan District, Fengtai District, and Chaoyang District) lies a planned developed area; we anticipated a lower ISDD in this area of 10–25%. When r = 500 m, distribution density reflected accurately the distribution characteristics of impervious surfaces on a fine scale; however, due to the high degree of fragmentation, it is difficult to demonstrate the overall situation. When r = 1000 m, the ISDD patch was large and showed the distribution characteristics of Beijing large-scale impervious surfaces. Thus, we chose Fig. 4(B) as the input to extract UMBA.

Table 1 Definition of urban heat island levels. Level

Definition

Level-1 Level-2 Level-3 Level-4 Level-5

TB_ij ≤ Tij(k, x, y) < TB_ij + ΔTij TB_ij + ΔTij ≤ Tij(k, x, y) < TB_ij + 2ΔTij TB_ij + 2ΔTij ≤ Tij(k, x, y) < TB_ij + 3ΔTij TB_ij + 3ΔTij ≤ Tij(k, x, y) < TB_ij + 4ΔTij Tij(k, x, y) ≥ TB_ij + 4ΔTij

where TUrban is the mean LST of the UMBA and TBoundary is that of the boundary area.

3.5. SUHI intensity level A direct comparison of SUHI intensity data for different periods is inappropriate. To measure SUHI intensity in different seasons using the same standard, we defined SUHI intensity levels for a pixel (x, y) for the ith year, jth month, and kth day (Table 1), where Tij(k, x, y), TBij and ΔTij refer to the LST, mean boundary area temperature, and mean SUHI intensity in the jth month and ith year, respectively.

4.1.3. Urban main built-up area We input ISDD with 10–25% and > 25% pixel distribution (Fig. 4(B)), based on the CCA and empirical tests. We obtained the UMBA for medium and high distribution densities in Beijing using the CCA parameters L = 2000 m and a patch threshold of S = 10,000. The low distribution density area comprised medium and high distribution density areas with patch numbers less than the threshold S. The medium- and high-density distribution areas in Beijing were then defined as the UMBA (Fig. 5(A)). The minimum influence area of the UHI effect is 150% of the urban area (Peng et al., 2011). Thus, this study took the UMBA as Beijing's urban area and 150% of that as the boundary area. Fig. 5(B) shows the distribution of the central district of Beijing from 1973 to 2013 (Zhang, 2014), which is similar to the UMBA boundary contour; this indicates that the proposed UMBA extraction technique is feasible and accurate.

4. Results and discussion 4.1. Extraction of UMBA based on ISDD 4.1.1. Impervious surface information extraction Fig. 3(A) is the BCI distribution for Beijing. The BCI value in urban areas was high and brown; soil and mixed land cover types were close to 0 and yellow-green; and vegetation was low, generally < 0, showing a blue-green hue (Fig. 3(A)). According to the BCI value distribution of different land cover types, the threshold value was 0.7; therefore, we extracted the impervious surface of Beijing as shown in Fig. 3(B) below. As shown in Fig. 3, impervious surface distribution was characterized by gradual outward diffusion, with the eastern and western districts at the center; the impervious surface distribution became sparser moving away from the center. To validate the impervious surface extraction result, 300 sample test points were generated randomly using ArcGIS software. By comparing the automatic extraction points with high-spatial resolution remote sensing data from Google Earth software and a visual interpretation method, 268 points were accurately extracted, providing an extraction accuracy of 268/300 = 89.34%.

4.2. Temporal analysis of the SUHI effect in the UMBA 4.2.1. Diurnal variation Statistical analysis of the 2056 MOD11A2 and MYD11A2 8-day LST products from March 2003 to February 2015 was used to determine the monthly mean SUHI intensity at 1:30, 10:30, 13:30, and 22:30 (Beijing time) (Table 2). The evolution of the diurnal mean SUHI intensity in the

Fig. 3. (A). Biophysical composition index (BCI) distribution. (B) Impervious surface distribution.

4

Remote Sensing of Environment xxx (xxxx) xxx–xxx

Q. Meng et al.

Fig. 4. Distance-weighted impervious surface distribution density (ISDD): (A) r = 500 m and (B) r = 1000 m.

4.2.2. Lunar variation Statistical analysis of 1028 MOD11A2 8-day LST products (day and night were each 514) were used to obtain monthly mean SUHI intensity (Table 4). The distribution of monthly SUHI intensity was stable at night, with a minimum of 1.29 k in October and a maximum of 1.79 k in August. However, during the daytime, monthly SUHI intensity fluctuated between 0.5 and 3.83 k. Compared with that during the night, SUHI intensity during the daytime was higher from May to September and displayed slight differences in April and October. To better explore the evolution of monthly SUHI intensity in Beijing, the relationship between SUHI intensity in the Beijing urban area and the temperature of the boundary area is shown in Fig. 7. Pink dots represent the SUHI intensity of all sample data and the temperature distribution of the boundary area, and color triangles and error bars indicate monthly mean values and corresponding standard deviations. The graphical representation of ΔT and TBoundary during the daytime exhibited an anti-clockwise distribution from January to December, which can be described by Fourier functions (Zhou et al., 2013). A clockwise distribution was observed in ΔT and TBoundary at night. The ‘overturning phenomenon’ of day and night is relevant to climate–vegetation interactions (Brazel et al., 2000; Garratt, 1994; Georgescu

Beijing urban area was evident; from January to March and November to December, SUHI in Beijing decreased between 1:30 and 10:30 and increased between 10:30 and 22:30, reaching the lowest and highest SUHI intensities of the day, respectively. From April to October, it increased from 1:30 to 13:30 and decreased from 13:30 to 1:20 the following day. Statistical analysis of all sample data was used to determine the monthly change in mean SUHI intensity in four epochs (Fig. 6(A)). This change varied monthly during epochs 10:30 and 13:30, and was always larger at 13:30 than at 10:30; they both attained a maximum value in August and a minimum in January. However, during epochs 1:30 and 22:30 the change was relatively stable, with values fluctuating between 1 k and 2 k. To better evaluate the evolution of diurnal SUHI intensity, we performed statistical analysis of the mean SUHI intensity in four epochs in four seasons (Fig. 6(B)). It is clear that in spring, summer, and autumn, SUHI intensity increased from 1:30 to 13:30 and decreased from 13:30 to 22:30, with the greatest change in summer at a maximum of 4.2 k. The results differed greatly in winter, when SUHI intensity increased from 10:30 to 22:30 and decreased from 1:30 to 10:30, with a minimum of 0.6 k (Table 3).

Fig. 5. (A) Urban main built-up area (UMBA) and boundary area distribution and (B) urban central area distribution in Beijing from 1973 to 2013.

5

Remote Sensing of Environment xxx (xxxx) xxx–xxx

Q. Meng et al.

Table 2 Monthly mean SUHI intensity in Beijing at 1:30, 10:30, 13:30, and 22:30. Time

1:30 10:30 13:30 22:30

Month 1

2

3

4

5

6

7

8

9

10

11

12

1.46 0.50 0.50 1.65

1.33 0.73 1.21 1.61

1.35 0.93 1.64 1.65

1.38 1.65 2.16 1.65

1.39 2.65 2.99 1.74

1.48 3.00 3.57 1.76

1.51 3.61 4.35 1.69

1.52 3.83 4.49 1.79

1.31 2.67 3.17 1.61

1.15 1.17 1.36 1.29

1.13 0.68 0.84 1.27

1.31 0.67 0.59 1.43

SUHI: surface urban heat island.

et al., 2011). There is a one-to-one correspondence between spring and autumn in the temperature of the boundary area. Taking the night as an example, March and November, April and October, and May and September correspond closely. When TBoundary values were similar, SUHI intensity in spring was higher than autumn at night. The ΔT vs. TBoundary pattern is related to the vegetation coverage in the different seasons, due to phenological characteristics and seasonal human activity (use of air-conditioning and heating) (Zhou et al., 2013).

Table 3 Mean SUHI intensity in four epochs in four seasons. Time

1:30 10:30 13:30 22:30

4.2.3. Seasonal variation During the daytime, Beijing SUHI intensity was characterized by apparent seasonal variation (Fig. 8(A)). SUHI intensity in urban areas attained a maximum in summer (June to August), which was obviously higher than in other seasons, with average SUHI intensity values of 3.47 k and a maximum of 5.88 k (Table 5). These maximum values may be related to the common use of air-conditioning since the mid-1990s. In spring and autumn, the mean SUHI intensity was similar to the values of 1.87 k and 1.63 k, respectively, and the differences in mean values between day and night were both 0.2 k. The mean SUHI intensity reached a minimum in winter (December, January, and February), at < 0.7 k. During the night, seasonal variation in SUHI intensity was relatively stable, with values of 1–2 k during the 12-year study period (Fig. 8(B)). The minimum average value was in autumn (September to November) with a mean value of 1.40 k over the 12-year study period (Table 5), and the maximum average value occurred in spring and summer with a mean value of 1.7 k. The difference in mean SUHI intensity between day and night attained a maximum in the summer, at 1.82 k. The difference in SUHI intensity value between day and night was −1 k in winter.

Season Spring

Summer

Autumn

Winter

1.39 1.92 2.48 1.71

1.52 3.57 4.18 1.75

1.22 1.63 1.91 1.43

1.39 0.58 0.69 1.60

the mean value was 1.60 k with a standard deviation of 0.33 k (Fig. 8). In Fig. 9(A), pink and blue dots indicate all day and night sample data, respectively. Red and yellow dots represent the annual SUHI intensity in the day and night during the 12-year study period (Fig. 9(B)). The annual mean SUHI intensity in Beijing during the daytime never dropped below 1.8 k, and attained a maximum of 2.32 k in 2008 and a minimum of 1.86 k in 2003. During the night, the annual SUHI intensity was relatively stable, at 1.4–1.7 k, reaching a maximum value in 2011 of 1.7 k with a standard deviation of 0.3 k. 4.3. Spatial analysis of the SUHI effect in UMBA 4.3.1. Frequency of SUHI intensity level The ΔT vs. TBoundary relationship exhibited a seasonal trend (Fig. 7). Because the different levels of SUHI intensity were defined by expanding the temperature range, we provided a measure for comparing SUHI intensity of the same season at the same standard. Specifically, we calculated the SUHI intensity level of each pixel for the MODIS data. The calculated 2056 MODIS data were then classified by season. The number of ith level pixels were counted to calculate their frequency. The statistical analysis results of the frequencies of different SUHI intensity levels are shown in Figs. 10 and 11; the displayed region shows the minimum enclosing rectangle around the boundary area.

4.2.4. Annual variation During the daytime over the 12-year study period, the mean SUHI intensity value was 2.12 k with a standard deviation of 1.29 k; at night,

Fig. 6. Monthly (A) and seasonal (B) distribution of mean surface urban heat island (SUHI) intensity in four epochs.

6

Remote Sensing of Environment xxx (xxxx) xxx–xxx

Q. Meng et al.

Table 4 Monthly mean SUHI intensity in Beijing (k). Time

Day Night

Month 1

2

3

4

5

6

7

8

9

10

11

12

0.50 1.65

0.73 1.61

0.93 1.65

1.65 1.65

2.65 1.74

3.00 1.76

3.61 1.69

3.83 1.79

2.67 1.61

1.17 1.29

0.68 1.27

0.67 1.43

Fig. 7. Distribution of mean surface UHI intensity and temperature of boundary area during day and night.

Fig. 8. Seasonal mean surface UHI intensity values for day and night per year, with range of variation.

level-3 SUHI intensity achieved high frequency only in the south-central area (Fengtai District). Surface solar radiation is larger in summer, and impervious surfaces increase apparent temperatures due to their higher emissivity. Therefore, the Beijing UMBA manifests a regional feature during the daytime in summer. However, different SUHI intensity levels exhibited no obvious spatial pattern in winter. The frequency of SUHI intensity level-5 was higher in the south-western area (Fangshan District and Daxing District). Both the SUHI intensity and the frequency of high intensity levels increased from the periphery to the center with a loop shape in autumn. At night, with no solar radiation affecting the UMBA, the SUHI intensity frequency increased, with a loop shape from the periphery to the center in all seasons. The frequency of level-4 and level-5 SUHI intensity was higher than in the daytime. Taking a winter night as an example, the frequency of SUHI intensity level-1 in the UMBA with medium distribution density was the highest, and the frequency of level-2 in the high distribution density area was the highest. Approaching from the periphery to the center, a level-3 heat island intensity cycle appeared. In the urban central area, the frequency of

Table 5 Seasonal mean surface UHI intensity values for day and night, with range of variation, in Beijing, 2003–2014 (k). Season

Day or night

Mean value

Maximum value

Minimum value

Difference value between day and night

Spring

Day Night Day Night Day Night Day Night

1.87 1.68 3.47 1.75 1.63 1.40 0.65 1.57

4.38 2.61 5.88 2.46 3.96 2.39 2.31 2.25

0.01 0.78 1.92 0.77 0.19 0.78 0.04 0.86

0.19

Summer Autumn Winter

1.82 0.20 − 0.92

During the daytime, SUHI intensity levels were mainly level-2 and level-3. The frequency of level-2 SUHI intensity was the highest, representing over 50% of values in summer in most of the Beijing UMBA during the daytime, with a concentration in the central area. However, 7

Remote Sensing of Environment xxx (xxxx) xxx–xxx

Q. Meng et al.

Fig. 9. Annual mean SUHI intensity values for day and night.

Fig. 10. Frequency of different surface UHI intensity levels during the daytime (from left to right: level-1 to level-5 frequencies).

8

Remote Sensing of Environment xxx (xxxx) xxx–xxx

Q. Meng et al.

Fig. 11. Frequency of different surface UHI intensity levels at night (from left to right: level-1 to level-5 frequencies).

where Pi is the frequency of the ith SUHI intensity level of 2003–2015 and E is the expected value of the ith pixel of the SUHI intensity level. The higher the value of E, the larger the high-level SUHI intensity. The spatial distribution of daytime E values agreed with the impervious distribution density, whereas the agreement for night-time values was comparatively small, as shown in Fig. 12. The expected value of SUHI intensity level in daytime in winter was higher in the boundary area than in the urban central area. This phenomenon occurred mainly due to heat supply in winter. Most thermal power plants are in suburban areas, which change land surface thermal radiance and contribute to heat pollution and thermal anomalies. The value of E is characterized by apparent seasonal variation, and mean SUHI intensity levels in autumn and in the other seasons in the order of winter > spring > summer.

level-4 intensity was the highest, and that of level-5 was approximately 30%. A comparison of Fig. 5(B) and Fig. 11 show that the spatial distribution of SUHI intensity levels at night and of urban expansion display high consistency. Taking winter as an example, level-4 SUHI intensity was observed with higher frequency, which is consistent with the extent of Beijing city center before 1987. With continuous urban expansion from 1987 to 1999, newly built areas exhibited the highest level-3 SUHI intensity frequency. The region that experienced a higher frequency of level-2 SUHI intensity values was more consistent with the extent of built-up areas developed after 1999. The spatial distribution of SUHI intensity levels in spring, summer, and autumn also showed similar consistency. Therefore, the SUHI intensity classification method and spatial pattern analysis presented in this study can better characterize the spatiotemporal variations of the heat island effect in Beijing.

4.4. Relationship between SUHI intensity and ISDD in UMBA Comparing the impervious surface distribution of the Beijing UMBA in Fig. 4(B) with that in Figs. 10 and 11, there is a matching degree of frequency of different SUHI intensity levels, and the ISDD distribution during the daytime was higher than that during the night. The region with higher distribution density remains a high incidence area of strong SUHI effects. To analyze the effect of the impervious surface

4.3.2. Expected value of SUHI intensity level Based on the frequency of different SUHI intensity levels, their expected value E can be described using Eq. (7):

E=

∑ i⋅P,i

(7) 9

Remote Sensing of Environment xxx (xxxx) xxx–xxx

Q. Meng et al.

Fig. 12. Surface UHI intensity level expected values.

Fig. 13. Frequency of surface UHI intensity levels and impervious surface distribution density.

relationship between SUHI intensity and the impervious surface. At night, the frequency of different SUHI intensity levels had no obvious relationship with ISDD. SUHI intensity level-2 to level-5 showed small peaks when ISDD reached 30%, which is related to the increase in SUHI intensity level from the periphery to the center with a loop shape at night; interestingly, the ISDD at the center point was only approximately 30%. The distribution of impervious surfaces had no distinct effect on SUHI in the Beijing UMBA at night, because it was still under the influence of urban population activities and heat emission, such as automobile exhaust and air-conditioning (McCarthy et al., 2010; Taha, 1997).

distribution effect on the SUHI effect, the relationship between frequencies of different UMBA and ISDD SUHI intensity levels from 2003 to 2014 are shown in Fig. 13. During the daytime, ISDD had an influence on level-1, level-2, and level-3 SUHI intensities; the frequency of level-1 SUHI intensity was highest where ISDD was < 10%. With an increase in ISDD, the frequency of level-1 intensity decreased continuously, whereas that of level-2 and level-3 increased. The frequency of level-2 tended to stabilize when ISDD reached 30%. However, the frequency of level-3 increased until ISDD reached 55–60%. The frequency of level-4 intensity rose slightly with an increase in ISDD; however, that of level-5 exhibited no significant change. Thus, when ISDD was below 50%, the distribution of impervious surfaces had an apparent influence on the SUHI effect in the Beijing UMBA in daytime. In contrast, when ISDD was higher than 50%, the feature became ‘saturated’. The curves in Fig. 11 manifest clear oscillations, which indicate the weakening

5. Conclusions The determination of urban and suburban boundaries and SUHI intensity classification are the main limitations in SUHI research. This 10

Remote Sensing of Environment xxx (xxxx) xxx–xxx

Q. Meng et al.

Environ. 33, 65–71. Chen, S., Wang, T., 2009. Comparison analyses of equal interval method and meanstandard deviation method used to delimitate urban heat island. J. Geo-Inf. Sci. 2, 001. Chen, X.-L., Zhao, H.-M., Li, P.-X., Yin, Z.-Y., 2006. Remote sensing image-based analysis of the relationship between urban heat island and land use/cover changes. Remote Sens. Environ. 104, 133–146. Deng, C., Wu, C., 2012. BCI: a biophysical composition index for remote sensing of urban environments. Remote Sens. Environ. 127, 247–259. Deng, C.B., Wu, C.S., 2013. Examining the impacts of urban biophysical compositions on surface urban heat island: a spectral unmixing and thermal mixing approach. Remote Sens. Environ. 131, 262–274. Du, H.Y., Wang, D.D., Wang, Y.Y., Zhao, X.L., Qin, F., Jiang, H., Cai, Y.L., 2016. Influences of land cover types, meteorological conditions, anthropogenic heat and urban area on surface urban heat island in the Yangtze River Delta urban agglomeration. Sci. Total Environ. 571, 461–470. Flores R, J.L., Pereira Filho, A.J., Karam, H.A., 2016. Estimation of long term low resolution surface urban heat island intensities for tropical cities using MODIS remote sensing data. Urban Clim. 17, 32–66. Garratt, J.R., 1994. Review: the atmospheric boundary layer. Earth Sci. Rev. 37, 89–134. Georgescu, M., Moustaoui, M., Mahalov, A., Dudhia, J., 2011. An alternative explanation of the semiarid urban area “oasis effect”. J. Geophys. Res. Atmos. 116, D24113. Haashemi, S., Weng, Q.H., Darvishi, A., Alavipanah, S.K., 2016. Seasonal variations of the surface urban heat island in a semi-arid city. Remote Sens. 8, 17. Heaviside, C., Vardoulakis, S., Cai, X.M., 2016. Attribution of mortality to the urban heat island during heatwaves in the West Midlands, UK. Environ. Health 15, 11. Howard, L., 1833. The Climate of London: Deduced From Meteorological Observations Made in the Metropolis and at Various Places Around it. Harvey and Darton, London. Hung, T., Uchihama, D., Ochi, S., Yasuoka, Y., 2006. Assessment with satellite data of the urban heat island effects in Asian mega cities. Int. J. Appl. Earth Obs. Geoinf. 8, 34–48. Imhoff, M.L., Zhang, P., Wolfe, R.E., Bounoua, L., 2010. Remote sensing of the urban heat island effect across biomes in the continental USA. Remote Sens. Environ. 114, 504–513. Jin, M.S., Kessomkiat, W., Pereira, G., 2011. Satellite-observed urbanization characters in Shanghai, China: aerosols, urban heat island effect, and land-atmosphere interactions. Remote Sens. 3, 83–99. Kauth, R.J., Thomas, G., 1976. The tasselled cap-a graphic description of the spectraltemporal development of agricultural crops as seen by Landsat. LARS Symposia 4B, 41–51. Li, Y.Y., Zhang, H., Kainz, W., 2012. Monitoring patterns of urban heat islands of the fastgrowing Shanghai metropolis, China: using time-serial of Landsat TM/ETM + data. Int. J. Appl. Earth Obs. Geoinf. 19, 127–138. Liu, Q., Liu, G., Huang, C., Liu, S., Zhao, J., 2014. In: A tasseled cap transformation for Landsat 8 OLI TOA reflectance images. Geoscience and Remote Sensing Symposium (IGARSS), 2014 IEEE International. IEEE, New York, pp. 541–544. Mathew, A., Khandelwal, S., Kaul, N., 2016. Spatial and temporal variations of urban heat island effect and the effect of percentage impervious surface area and elevation on land surface temperature: study of Chandigarh city, India. Sustain. Cities Soc. 26, 264–277. Matson, M., Mcclain, E.P., McGinnis Jr., D.F., Pritchard, J.A., 1978. Satellite detection of urban heat islands. Mon. Weather Rev. 106, 1725–1734. McCarthy, M.P., Best, M.J., Betts, R.A., 2010. Climate change in cities due to global warming and urban effects. Geophys. Res. Lett. 37, L09705. Melaas, E.K., Wang, J.A., Miller, D.L., Friedl, M.A., 2016. Interactions between urban vegetation and surface urban heat islands: a case study in the Boston metropolitan region. Environ. Res. Lett. 11, 11. Mohan, M., Kikegawa, Y., Gurjar, B.R., Bhati, S., Kolli, N.R., 2013. Assessment of urban heat island effect for different land use-land cover from micrometeorological measurements and remote sensing data for megacity Delhi. Theor. Appl. Climatol. 112, 647–658. Ogashawara, I., Bastos, V.D.B., 2012. A quantitative approach for analyzing the relationship between urban heat islands and land cover. Remote Sens. 4, 3596–3618. Oke, T.R., East, C., 1971. The urban boundary layer in Montreal. Bound.-Layer Meteorol. 1, 411–437. Peng, S., Piao, S., Ciais, P., Friedlingstein, P., Ottle, C., Bréon, F.o.-M., Nan, H., Zhou, L., Myneni, R.B., 2011. Surface urban heat island across 419 global big cities. Environ. Sci. Technol. 46, 696–703. Quan, J., Chen, Y., Zhan, W., Wang, J., Voogt, J., Wang, M., 2014. Multi-temporal trajectory of the urban heat island centroid in Beijing, China based on a Gaussian volume model. Remote Sens. Environ. 149, 33–46. Rajasekar, U., Weng, Q., 2009. Urban heat island monitoring and analysis using a nonparametric model: a case study of Indianapolis. ISPRS J. Photogramm. Remote Sens. 64, 86–96. Rao, P., 1972. Remote sensing of urban heat islands from an environmental satellite. Bull. Am. Meteorol. Soc. 53, 647–648. Ridd, M.K., 1995. Exploring a VIS (vegetation-impervious surface-soil) model for urban ecosystem analysis through remote sensing: comparative anatomy for cities†. Int. J. Remote Sens. 16, 2165–2185. Roth, M., Oke, T., Emery, W., 1989. Satellite-derived urban heat islands from three coastal cities and the utilization of such data in urban climatology. Int. J. Remote Sens. 10, 1699–1720. Rozenfeld, H.D., Rybski, D., Andrade, J.S., Batty, M., Stanley, H.E., Makse, H.A., 2008. Laws of population growth. Proc. Natl. Acad. Sci. 105, 18702–18707. Schwarz, N., Lautenbach, S., Seppelt, R., 2011. Exploring indicators for quantifying surface urban heat islands of European cities with MODIS land surface temperatures.

study demonstrated the UMBA method based on ISDD and a created new SUHI intensity classification method aimed at long-term time series remote sensing data. Landsat 8 data were used to extract the UMBA for the city of Beijing, China. A series of 12-year MODIS data were used to generate LST for characterizing spatial and temporal SUHI trends. The use of UMBA and the new SUHI intensity classification methods improved the accuracy of SUHI intensity calculations, for a better understanding of the SUHI evolution mechanism. Using BCI, we extracted the urban impervious surface. If only impervious surface areas were regarded as urban areas to analyze SUHI, the cooling effect produced by vegetation and water in the urban area would be negligible. The calculated results did not represent the true SUHI phenomenon. Therefore, we created a new method to extract the urban boundary by calculating the ISDD based on a distance weight rule. We found that the ISDDs of urban areas were larger than those of suburban areas. Combining CCA, the UMBAs of Beijing city were successfully extracted, with results similar to the distribution of the central district of Beijing extracted by Zhang (2014). Through a statistical analysis of the 2056 MOD11A2 and MYD11A2 8-day LST products, we explored the evolution of SUHI intensity daily, monthly, seasonal and yearly in the Beijing UMBA. We observed an important characteristic of SUHI intensity. On a temporal scale, ΔT/ TBoundary displayed an anti-clockwise distribution in the daytime from January to December and a clockwise distribution at night. Additionally, we proposed a new SUHI intensity classification method to analyze its spatial distribution and variation. On the spatial scale, the frequency of level-2 SUHI intensity was the highest, with over 50% in summer in most of the Beijing UMBA during the daytime. At night, with the rise in SUHI intensity levels, the frequency of the SUHI intensity level from the periphery to the center increased within the same season. We also provided new insights into the relationship between SUHI intensity and ISDD. Our results indicate that ISDD had a remarkable influence on SUHI intensity level during the daytime in the Beijing UMCA, and the level-1 to level-4 frequency increased with ISDD. This influence tended to weaken when ISDD exceeded 50%. At night, ISDD had no obvious influence on the frequency of different SUHI intensity levels. In summary, this research created a new SUHI monitoring process and can be used to guide further studies in applying the technical framework to other regions. Nevertheless, some aspects are worthy of further investigation, such as the detailed influence of urban boundary for SUHI calculations and the response to critical points between SUHI intensity and ISDD. Finally, Beijing is a typical example of the inevitable trend of increasing urbanization, and the SUHI effect is becoming increasingly serious. Measures to mitigate the SUHI effect will be the primary focus in future research. Acknowledgement This work was funded by the National Natural Science Foundation of China [41471310 and 41271413], the Hainan Province Natural Science Foundation Innovative Research Team: Study on Urban Green Landscape Pattern Remote Sensing Evaluation based on Lidar and Multispectral Data [2017CXTD015]; Guangdong Province Science and Technology Project: Urban Impervious Surface Remote Sensing Extraction System [2016A050502065]; and the Sichuan Province Science and Technology Support Program: Chengdu City Impervious Surface Remote Sensing Extraction System [2016JZ0027]. References Brazel, A., Selover, N., Vose, R., Heisler, G., 2000. The tale of two climates-Baltimore and Phoenix urban LTER sites. Clim. Res. 15, 123–135. Carlson, T., Augustine, J., Boland, F., 1977. Potential application of satellite temperaturemeasurements in analysis of land-use over urban areas. Bull. Am. Meteorol. Soc. 58, 1301–1303. Carnahan, W.H., Larson, R.C., 1990. An analysis of an urban heat sink. Remote Sens.

11

Remote Sensing of Environment xxx (xxxx) xxx–xxx

Q. Meng et al.

abundance relationship for urban heat island studies. Remote Sens. Environ. 89, 467–483. Wu, X.W., Y, B., Zhi, C.G., Gong, W.F., 2006. Study on the heat island effect in Harbin city based on RS and GIS. Geomatics World 05, 63–66 (In Chinese). Xu, S.L., 2009. An approach to analyzing the intensity of the daytime surface urban heat island effect at a local scale. Environ. Monit. Assess. 151, 289–300. Xu, H.Q., Chen, B.Q., 2003. An image processing technique for the study of urban heat island changes using different seasonal remote sensing data. Remote Sens. Technol. Appl. 03, 129–133 (In Chinese). Yusuf, Y.A., Pradhan, B., Idrees, M.O., 2014. Spatio-temporal assessment of urban heat island effects in Kuala Lumpur Metropolitan City using Landsat images. J. Indian Soc. Remote Sens. 42, 829–837. Zhang, J.Q., 2006. Thermal Environment Detection in the Pearl River Delta Area by Remote Sensing and Analysis of Its spatial and Temporal Evolutions. The Chinese Academy of Sciences. Zhang, Z.X., 2014. Urban Expansion Monitoring Atlas by Remote Sensing in China. Star Map Press, Beijing. Zhang, J., Wang, Y., 2008. Study of the relationships between the spatial extent of surface urban heat islands and urban characteristic factors based on Landsat ETM+ data. Sensors 8, 7453–7468. Zhang, Z.M., He, G.J., Xiao, R.B., Wang, W., 2005. A study of the urban heat island changes of Beijing city based on remote sensing. Remote Sens. Inf. 06, 46–48 (In Chinese). Zhang, Y., Yu, T., Gu, X.F., Zhang, Y.X., Chen, L.F., Yu, S.S., Zhang, W.J., Li, X.W., 2006. Land surface temperature retrieval from CBERS-02 IRM SS thermal infrared data and its applications in quantitative analysis of urban heat island effect. J. Remote Sens. 05, 789–797. Zhang, H., Qi, Z.-F., Ye, X.-Y., Cai, Y.-B., Ma, W.-C., Chen, M.-N., 2013. Analysis of land use/land cover change, population shift, and their effects on spatiotemporal patterns of urban heat islands in metropolitan Shanghai, China. Appl. Geogr. 44, 121–133. Zhao, L., Lee, X., Smith, R.B., Oleson, K., 2014. Strong contributions of local background climate to urban heat islands. Nature 511, 216–219. Zhou, B., Rybski, D., Kropp, J., 2013. On the statistics of urban heat island intensity. Geophys. Res. Lett. 40, 5486–5491.

Remote Sens. Environ. 115, 3175–3186. Shen, H., Huang, L., Zhang, L., Wu, P., Zeng, C., 2016. Long-term and fine-scale satellite monitoring of the urban heat island effect by the fusion of multi-temporal and multisensor remote sensed data: a 26-year case study of the city of Wuhan in China. Remote Sens. Environ. 172, 109–125. Stathopoulou, M., Cartalis, C., 2009. Downscaling AVHRR land surface temperatures for improved surface urban heat island intensity estimation. Remote Sens. Environ. 113, 2592–2605. Stewart, I., Oke, T., 2009. A new classification system for urban climate sites. Bull. Am. Meteorol. Soc. 90, 922–923. Streutker, D.R., 2002. A remote sensing study of the urban heat island of Houston, Texas. Int. J. Remote Sens. 23, 2595–2608. Streutker, D.R., 2003. Satellite-measured growth of the urban heat island of Houston, Texas. Remote Sens. Environ. 85, 282–289. Taha, H., 1997. Urban climates and heat islands: albedo, evapotranspiration, and anthropogenic heat. Energ. Buildings 25, 99–103. Takebayashi, H., Moriyama, M., 2007. Surface heat budget on green roof and high reflection roof for mitigation of urban heat island. Build. Environ. 42, 2971–2979. Tomlinson, C.J., Chapman, L., Thornes, J.E., Baker, C.J., 2012. Derivation of Birmingham's summer surface urban heat island from MODIS satellite images. Int. J. Climatol. 32, 214–224. Walawender, J.P., Szymanowski, M., Hajto, M.J., Bokwa, A., 2014. Land surface temperature patterns in the urban agglomeration of Krakow (Poland) derived from Landsat-7/ETM + data. Pure Appl. Geophys. 171, 913–940. Wan, Z., 2006. MODIS Land Surface Temperature Products Users' Guide. ICESS. University of California, Santa Barbara. Wan, Z., Zhang, Y., Zhang, Q., Li, Z.L., 2004. Quality assessment and validation of the MODIS global land surface temperature. Int. J. Remote Sens. 25, 261–274. Wang, J., Sun, S.M., 2004. Research on the method of monitoring the urban heat island by remote sensing technique of satellite. Mod. Sci. Instrum. 06, 27–29 (In Chinese). Wang, K., Wang, J., Wang, P., Sparrow, M., Yang, J., Chen, H., 2007. Influences of urbanization on surface characteristics as derived from the Moderate-Resolution Imaging Spectroradiometer: a case study for the Beijing metropolitan area. J. Geophys. Res.: Atmos. 112 (D22). Weng, Q., Lu, D., Schubring, J., 2004. Estimation of land surface temperature–vegetation

12