A new method to quantify surface urban heat island intensity
, Xiaoma Li
, Xun Wang
, Sahar Sodoudi
Institute of Meteorology, Freie Universität Berlin, Berlin, Germany
Department of Geological and Atmospheric Sciences, Iowa State University, Ames, IA, USA
Forestry Experiment Center of North China, Chinese Academy of Forestry, Beijing, China
•Quantifyingsurface urbanheat island in-
tensity using the relationship between
LST and Impervious Surface Areas.
•The impervious surface areas was re-
gionalized within the footprint of remote
sensing observation using a Kernel Den-
sity Estimation method.
•Linear functions of LST were well ﬁtted
using the regionalized impervious sur-
•Slope of the linear function of LST was
deﬁned as the surface urban heat island
Received 17 August 2017
Received in revised form 30 November 2017
Accepted 30 November 2017
Available online xxxx
Editor: SCOTT SHERIDAN
Reliable quantiﬁcation of urban heat island (UHI) can contribute to the effective evaluation of potential heat risk.
Traditional methods for the quantiﬁcation of UHI intensity (UHII) using pairs-measurements are sensitive to the
choice of stations or grids. In order to get rid of the limitation of urban/rural divisions, this paper proposes a
new approach to quantify surface UHII (SUHII) using the relationship between MODIS land surface temperature
(LST) and impervious surface areas (ISA). Given the footprint of LST measurement, the ISA was regionalized to in-
clude the information of neighborhood pixels using a Kernel Density Estimation (KDE) method. Considering the
footprint improves the LST-ISA relationship. The LST shows highly positive correlation with the KDE regionalized
). The linear functions of LST are well ﬁtted by the ISA
in both annual and daily scales for the city of
Berlin. The slope of the linear function represents the increaseinLSTfromthenaturalsurfaceinruralregionstothe
impervious surface in urban regions, and is deﬁnedas SUHII in this study.The calculatedSUHII show highvalues in
summer and during the day than in winter and at night. The new method is also veriﬁed using ﬁner resolution
Landset data, and the results further prove its reliability.
© 2017 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://
Surface urban heat island
Land surface temperature
Impervious surface area
Kernel density estimation
Footprint of remote sensing observation
Urban areas show higher temperature than the surrounding rural
areas, which is well known as Urban Heat Island (UHI) effect. Since its
ﬁrst observation by Howard in London (Mills, 2008), UHI phenomenon
has been widely reported in different sized cities (e.g. Arnﬁeld, 2003;
Zhang et al., 2010; Zhou et al., 2017). Warmer air caused by UHI increases
heat load stress of urban residents, potentially raising the threat of mor-
tality (e.g. Tan et al., 2010; Constantinescu et al., 2016). Meanwhile,
higher temperature increases energy consumption and associated
greenhouse gas emissions due to the use of air conditioning (e.g. Zhou
and Gurney, 2010; Zhou et al., 2012). Under the background of fast ur-
banization (e.g. Kuang et al., 2013, 2016b) and global change (e.g.
Science of the Total Environment 624 (2018) 262–272
E-mail address: firstname.lastname@example.org (S. Sodoudi).
0048-9697/© 2017 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
Contents lists available at ScienceDirect
Science of the Total Environment
journal homepage: www.elsevier.com/locate/scitotenv
Grimm et al., 2008), residential living in cities is suffering from higher
risk of heat wave (e.g. Zhou et al., 2014, 2015; Yang et al., 2017).
Concerning the increasing possible hazards cased by UHI, more and
more attention has been paid to the studies of UHI (Inouye, 2015;
McDonnell and MacGregor-Fors, 2016; Lee et al., 2015).
Accurate quantiﬁcation of UHI can help to efﬁciently evaluate the po-
tential heat risk and to guide the city management and development for
government and city planners. Urban heat island intensity (UHII), the
difference in temperature between urban and surrounding rural regions,
is the classical indicator to quantitatively describe UHI effect (Rizwan et
al., 2008; Stewart, 2011). Traditionally, the detection of UHII is conducted
at two ﬁxed in-situ stations, one in urban and the other in rural regions
(e.g. Yang et al., 2013; Earl et al., 2016). Similarly, the study of the surface
UHII (SUHII) using remote sensing data is conducted over selected pixels
that are located in the urban and rural regions, separately (e.g. Stewart,
2011). The estimation of UHII (SUHII) relies on the deﬁnitions of urban
and rural stations or pixels (e.g. Roth et al., 1989; Azevedo et al., 2016;
Du et al., 2016). However, urban regions are strongly affected by
human activities with high heterogeneity over the urban surface, and
even the surrounding rural areas may have different ecosystems
(Buyantuyev and Wu, 2010; Cadenasso et al., 2007). The urban-rural di-
chotomy alone cannot sufﬁciently guide the choice of the stations
(Stewart and Oke, 2012). Schwarz et al. (2011) compared eleven ap-
proaches for quantifying SUHII with MODIS land surface temperatures
for European cities, and found that the calculated SUHIIs using different
rural pixels showed weak correlations. The different deﬁnitions of the
urban/rural regions make the intercomparison study of UHII among dif-
ferent cities challenging. Stewart (2011) argued that previous UHI stud-
ies that used two stations measurements are often not comparable
because of the different deﬁnitions of the measurement stations and
the lack of the crucial description of these stations. On the other hand,
ﬁxed stations or pixels only represent the local micro-climate around
these stations or pixels (Oke, 2006). Limited measurements only reﬂect
parts of the characteristics of UHI (SUHI), and cannot identify the spatial
variation and the structure of UHI within a whole city, especially for the
cities which have multiple UHI centers (e.g. Li and Yin, 2013; Dou et al.,
2015). The shape of cities could signiﬁcantly inﬂuence the amplitude of
UHI (Zhang et al., 2012; Zhou et al., 2017). To overcome the problems
mentioned above, a promising way for the quantiﬁcation of UHII
(SUHII) should try to get rid of the limitation of urban/rural divisions,
and consider the comprehensive conditions of cities by integrating the
urban surface properties.
Land use change caused by urbanization is the primary driving factor
of UHI (e.g. Cheval and Dumitrescu, 2015; Du et al., 2016; Li et al., 2017).
The lower albedo and higher sealing degree of urban areas signiﬁcantly
alter the surface energy budget and lead to higher temperature than
rural areas (Oke, 1982, 1988; Kuang et al., 2015a, 2015b). Near surface
temperatures are closely related to urban indicators, such as Impervious
Surface Area (ISA). Yuan and Marvin (2007) found that there was a
strong linear relationship between LST and ISA for all seasons in Minne-
sota. Rajasekar and Weng (2009) pointed out that the areas with high
heat signatures had a strong correlation with impervious surfaces in cen-
tral Indiana. Imhoff et al. (2010) concluded that ISA was the primary
driver for the increase in temperature, explaining 70% of the total vari-
ance in LST for 38 the most populous cities combined in the continental
United States. Zhang et al. (2010) pointed that N60% of the total LST var-
iance was explained by ISA for urban settlements within forests at mid to
high latitudes globally. Li et al. (2011) reported a strong positive relation-
ship between LST and ISA in Shanghai. Schatz and Kucharik (2014) found
that ISA within the footprint of measurement stations was the dominant
driver of air temperature and accounted for 74% and 80% of the explained
spatial variation of the air temperature at night and during the day, re-
spectively, in Madison, Wisconsin. Kuang et al. (2017) found that the
highly dense impervious surface areas signiﬁcantly increased land sur-
face temperature. Wang et al. (2017) found that ISA was responsible
for 31%–38% and 49%–54% of air temperature variability during the day
and at night, respectively in Beijing. Compared with UHII, SUHII is usually
more dependent on ISA. This is because that the land cover is the single-
most dominant factor of LST, while the air temperature is affected by
land cover, air advection and anthropogenic heat emission, together
(Azevedo et al., 2016). To summer up, as a good indicator of urban
land use, ISA could reﬂect the spatial pattern of UHI (SUHI). The relation-
ship between temperature and ISA can be a potential powerful tool for
the quantiﬁcation of UHII (SUHII).
The temperature at each site is also affected by the surrounding envi-
ronment (Rannik et al., 2000). There is a footprint for the temperature
measurement (Oke, 2006). The measured temperature is related to the
overall land use information within the footprint. Schatz and Kucharik
(2014) and Wang et al. (2017) considered the footprint when examining
the relationship between in-situ air temperature and ISA. As for the re-
mote sensing, there is a mismatch between the observation results and
its ground source, especially for the mixed pixels over heterogeneous
areas, due to the variation of the view zenith angles and gridding pro-
cesses. Campagnolo and Montano (2014) found that the width of the
ground-projected instantaneous ﬁelds of view (IFOV) of MODIS products
was larger than the nominal resolutions, and increased with the view ze-
nith angles. The IFOV errors also exist in the Lansat data, especially in the
thermal band (Lee et al., 2004). The satellite observation result in each
pixel also contains information from neighboring pixels. Townshend et
al. (2000) found that parts of the signal in MODIS pixels come from the
surroundings. Peng et al. (2015) found that the size of the signal source
of MODIS pixels is larger than the nominal resolution of the pixel. As
thus, it is necessary to consider the inﬂuence of the footprint of remote
sensing observation when study LST-ISA relationship.
This paper comes up with an approach to calculate SUHII based on
the linear relationship between LST and ISA. Given the footprint of LST
measurement, the ISA was regionalized to include the information of
neighborhood pixels within the footprint using a Kernel Density Estima-
tion (KDE) method. The linear regression function of LST was ﬁtted using
the KDE regionalized ISA (ISA
). The regression slope of the ﬁtted func-
tion was used as SUHII. The temporal variations of the calculated SUHII of
Berlin in 2010 were investigated. In addition, the new developed ISA
was compared with the raw ISA in terms of the ﬁtted functions of LST
and the calculated SUHII. The goal of this paper is to develop a promising
approach for the quantiﬁcation of SUHII.
2. Study area, data, and methodology
2.1. Study area
The study area is Berlin, the capital city of Germany. Berlin (52.34°–
52.68° N, 13.10°–13.77° E) is located in Northeast Germany with a ﬂat to-
pography (34–122 m, altitude), and covers an area of about 900 km
cording to a report in the year of 2015 of the Statistical Ofﬁce of Berlin-
Brandenburg (https://www.statistik-berlin-brandenburg.de), Berlin has
N3.6 million inhabitants, with one third living in the inner city in an
area of about 88 km
. It is the second most populous city in the European
Union. Fig. 1 shows the spatial pattern of CORINE land cover (Feranec et
al., 2007) version 2012 in the study area. Berlin is an urbanized region
with about 35% built-up areas. In addition, transportation and infrastruc-
ture areas cover about 20% of the city. Berlin's built-up areas create a mi-
croclimate with noticeable urban heat island effects, leading to a higher
potential heat stress risk in the central inner-city areas (Dugord et al.,
Berlin has a temperate maritime climate with a mean annual temper-
ature of 9.5 °C and annual precipitation of 591 mm (data based on the
German Meteorological Ofﬁce Dahlem station measurements in the 30-
years period of 1981–2010). Affected by the prevailing westerlies and
abundant water vapor from Atlantic, Berlin has a windy and cloudy cli-
mate (Kottek et al., 2006). Fig. 2 exhibits the seasonal variations of the
cloud fraction, wind speed, and precipitataion in 2010. Most of the win-
ter days were cloudy with more than half of the sky covered by clouds,
263H. Li et al. / Science of the Total Environment 624 (2018) 262–272
while a lot of the summer days had clear skies. The winter days had
stronger winds compared to the summer days. The precipitation was
not very concentrated. More than half of the days had precipitation larg-
er than 0.1 mm/d, and almost one-third of the days had precipitation
larger than 1.0 mm/d. Although the intensity of the precipitation was
not strong, the frequency was high. Most of the typical weather for UHI
with few clouds and calm wind occurred in summer.
2.2. LST data
Daily MODIS LST products (MOD11A1&MYD11A1) collection 005
with 1000 m resolution grids in 2010 were used in this study (Coll et
al., 2009). MODIS products have high temporal resolution, four observa-
tions per day. Moreover, MODISLST data have high quality. Wan (2008)
reported that the accuracy of MODIS LST product (collection 5) is better
than 1 K in most cases. Rigo et al. (2006) reported an accuracy of MODIS
LST b5%, even in urban environments. The freely available dataare easy
to access. MODIS LST data have been widely used to study SUHI in differ-
ent regions, such as Europe (e.g. Schwarz et al., 2011; Azevedo et al.,
2016), Asia (e.g. Kuang et al., 2015a, 2015b), and North America (e.g.
Zhang et al., 2014; Hu and Brunsell, 2015; Hu et al., 2016).
Affected by the variation in sensing geometry of the MODIS instru-
ments, the effective signal source is larger than the size of a MODIS
pixel. Wolfe et al. (2002) estimated that the width of the instantaneous
ﬁeld of view of the MODIS observation cells reached approximately 2.0
times and 4.8 times the nadir resolution in the track and scan directions
at the scan edge. Campagnolo and Montano (2014) found that in near
optimal locations, the effective resolution of the 250 m MODIS gridded
products varied between 344 and 835 m along rows and between 292
and 523 m along columns, respectively.
Two MODIS sensors are mounted on the Terra and Aqua satellites,
separately. Everyday, the Aqua satellite passed over Berlin at UTC time
around 11:50 and 01:20, and the Terra satellite passed over Berlin at
UTC time around 20:50 and 10:20. As for the MODIS daily LST images,
there is slight difference in the observation time among the pixels,
around 9 & 13 min for the two MOD images and 9 & 14 min during
the study period. Special care needs to be taken into this problemduring
the interpretation of the results. Clouds absorb the longwave radiation
from the earth surface, and then block the observation of LST
(Williamson et al., 2013). Here, the cloud-covered pixels are ﬁltered
out for each image. The images with too much missing pixels, in partic-
ular within urban regions, cannot show the real feature of SUHII well.
Meanwhile, too many clouds could weaken UHI effect (Morris et al.,
2001). In this study, only the images havingN90% of valid pixels within
both the study areas and the urban areas (ISA larger than 25%) are cho-
sen for analysis. Berlin has a cloudy climate. Most days are cloudy days,
especially in winter (Fig. 2). Eventually, a total of 47, 34, 76 and 61 im-
ages at four observation times 10:20, 11:50, 20:50, and 01:20 are kept.
The mean LST for the whole year, summer (May to September) and
winter (October to April) are calculated using these selected images.
2.3. ISA data
High Resolution Layer imperviousness product version 2012 from
Copernicus Land Monitoring Service Pan-European Component (http://
land.copernicus.eu/pan-european) was applied in this study. HRL prod-
ucts are produced from multi-source high resolution satellite imagery
through a combination of automatic processing and interactive rule-
based classiﬁcation (Lefebvre et al., 2013). The imperviousness shows
the percentage of artiﬁcial impervious cover (0–100%), referring to the
built-up areas that are characterized by the substitution of the original
natural land cover or water surface with an artiﬁcial surface (Langanke,
Fig. 1. Spatial pattern of the CORINE land cover in the study area.
50 100 150 200 250 300 350
Wind speed Cloud fraction Precipitation
Wind speed (m/s)
Cloud fraction (%)
Fig. 2. Seasonal variations of the cloud fraction (blue curve), wind speed (red curve), and
precipitation (black curve) in 2010 for Berlin. (For interpretation of the references to
colour in this ﬁgurelegend, the reader is referred to the web version of this article.)
264 H. Li et al. / Science of the Total Environment 624 (2018) 262–272
2013). Copernicus provides ISA products at both 20 m and 100 m resolu-
tion. The accuracy of the data was accessed based on a stratiﬁed system-
atic sampling approach using the EUROSTAT Land Use/Cover Area from
statistical Survey sampling frame (Sannier et al., 2016). The results sug-
gested that the Copernicus ISA data reached or even exceeded the re-
quired level of accuracy with b10–15% error for both omission and
commission errors. This accuracy is similar to the other studies (Kuang
et al., 2016a). To make ISA match with LST in grids, the nearest approach
was applied to re-sample the ISA data from the resolution of 100 m to
1000 m on the platform of ArcGIS.
2.4. Regionalization of ISA using a kernel density estimation method
Land surface temperature measurement over each pixel has a foot-
print due to the adjacency effect (Justice et al., 1998). The measurements
data are not only affected by the surface of corresponding pixel, but also
affected by the surrounding pixels. In order to take the inﬂuence of the
Rural Suburban Urban
Fig. 3. Schematic diagram of the quantiﬁcation of SUHII based on the linear regression
relationship between LST and ISA
mean the LST in the urban and rural
Fig. 4. Spatial distributions of the mean LST for Berlin in the year 2010.Three columns representthe mean values during different time periods: (left) wholeyear, (middle) summer,and
(right) winter. Rows (a), (b), (c), and (d) present the results at four observation times.
265H. Li et al. / Science of the Total Environment 624 (2018) 262–272
footprint on the LST measurement into account, the ISA was regionalized
using a Kernel Density Estimation (KDE) method. As one of the most
well-established spatial techniques, KDE method could calculate the
contribution of the surrounding points. The density (f
) of a speciﬁc
) was calculated as the sum of the weights of neighbor points
) within a circular neighbor areas as follows
where ris the kernel radius and controls the size of the circular neighbor-
hoods around x
.Kis the kernel function and controls the weight of the
In this paper, the KDE calculation was conducted on the platform of
ArcGIS. Firstly, the raster image of ISA was converted to point feature
using the function of ‘Raster to Point’under the toolbox ‘Conversion
Tools’. Then the function of ‘Kernel Density’under the toolbox of ‘Spatial
Analyst Tools’was used to calculate the density of each pointin a neigh-
borhood around each output raster cell with a resolution of 1000 m.
During this process, a smoothly curved surface (kernel surface) is ﬁtted
over a circular neighborhood of each point based on quartic kernelfunc-
tion (Silverman, 1986). The surface value is largest at the location of the
point and decreases with increasing distance fromthe point, reaching to
zero at the border of the circular neighborhood. The density at each out-
put raster cell is calculated byadding the values ofall the kernel surfaces
where they overlay the rastercell center. Then the f
to the range of 0 to 100% as follows
ISAKDE ¼fKDE−min fKDE
The kernel radius could affect the KDE calculation results (Anderson,
2009; Thakali et al., 2015). Based on the calculation process of the KDE
method, the kernel radius should be larger than the distance between
the centers of two neighbor pixels (value of the spatial resolution of
the pixels, 1000 m). Meanwhile, the largest radius should be b4.8
times (4800 m) the nadir resolution of MODIS LST data based on the
study of Wolfe et al. (2002). In order to ﬁnd outthe optimal kernel radi-
us, a sensitivity test was conducted using kernel radius ranging from
1500 m to 5000 m with an interval of 500 m. The LST shows best
relationship with KDE regionalized ISA using kernel radius of 3000 m.
The details of the sensitivity study are discussed in the Section 4.1.
2.5. Quantiﬁcation of SUHII
dominant reason for SUHI. Usually, the temperature increases with the
increase in ISA from rural to urban regions, presenting positive linear
trend (e.g. Yuan and Marvin, 2007; Schatz and Kucharik, 2014, 2015).
A linear regression function of LST can be ﬁtted using ISA (Fig. 3). The
slope of the ﬁtted function refers to the increase of LST versus ISA in-
creasing from 0% to 100%. The regression slope can be used to deﬁne
SUHII, if a good regression function is ﬁtted. In this study, the regional-
is chosen to ﬁt the linear function of LST. Urban surface has
high heterogeneity. LST may vary largely among the grids that have sim-
due to the difference in material, elements conﬁguration and
morphological characteristics (Oke, 2006; Wang et al., 2017). Here, the
Min-Max range of the mean LST (°C) for differentperiods and times of the day in Berlin.
Observation times Annual mean Summer mean Winter mean
MOD 10:20 9.20 11.09 7.91
MYD 11:50 11.24 12.76 8.88
MOD 20:50 6.32 6.45 5.97
MYD 01:20 8.52 8.35 5.40
Fig. 5. Spatial distributions of the (a) raw ISA and (b) KDE regionalized ISA using kernel
radius of 3000 m.
Fig. 6. Longitudinal proﬁles of (a–d)the mean LST and (e)the ISA
acrossthe city within
the latitude extent between 52.34° N and 52.68° N (the administrative border of Berlin).
The mean LST of the whole year (black curve), summer (red curve), and winter (green
curve) were calculated, separately. The blue bars indicate the grids that strongly affected
by water bodies. (For interpretation of the references to colour in this ﬁgure legend, the
reader is referred to the web version of this article.)
266 H. Li et al. / Science of the Total Environment 624 (2018) 262–272
zonal analysis method was applied referring to Yuan and Marvin (2007).
All the pixels within the study areas were divided into 50 parts with 2%
interval of the ISA
. Given the sub-interval variation of LST caused by
the heterogeneity of urban surface and different ecosystems of the
rural surface, the mean values of LST and ISA
were calculated within
each interval to ﬁt the function of LST. Least square method was applied
to ﬁt the linear regression function. The coefﬁcient of determination (R
was used to evaluate the accuracy of the ﬁtted function. Water bodies
have strong thermal properties, which may skew the trend of LST with
the change in ISA (Hu and Brunsell, 2015). Given the large areas of
water bodies in the study areas (6.7% based on Corine land cover data),
the pixels with more than one-quarter of water bodies areas were ex-
cluded during the calculation process.
3.1. Spatial patterns of the LST
Fig. 4 shows the spatial variations of the mean LST at the four obser-
vation times for the different periods in 2010 for Berlin. The LST within
the city center is obviously higher than those within the surrounding
rural regions, presenting pronounced surface urban heat island effect.
The distribution of high LST areas is corresponding with the urban land
cover distribution shown in Fig. 1. The max-min ranges of the annual
mean LST are up to 9.20, 11.24, 6.32 and 8.52 °C at the four observation
times (Table 1). The LST shows larger urban-rural difference in daytime
and summer than in nighttime and winter.
3.2. Spatial patterns of the ISA
Fig. 5 shows the spatial distributions of the raw ISA and KDE region-
alized ISA (ISA
) using the kernel radius of 3000 m for Berlin. Most of
the high ISA areas are located in the central regions of the city. In the sur-
rounding rural regions, the pixels with large and small values crossly dis-
tribute. The spatial pattern of the ISA is very similar to that of the LST in
Fig. 4. The ISA
shows a smoother and more continuous distribution,
compared with the raw ISA. The Standard Deviation (STD) of the ISA
(20.91%) is smaller than that of the raw ISA (25.74%).
3.3. Longitudinal variations of the LST versus the ISA
In order to clearly show the spatial conﬁguration of LST-ISA
mean longitudinal variations of the LST and ISA
are presented in Fig.
6. It shows that as the areas change from rural to inner-city and the
0 204060801000 20406080100
Mean LST (°C)
Annual mean Summer mean Winter mean
Mean LST (°C)Mean LST (°C)
Mean LST (°C)
Fig. 7. Variationsof the (left) annul mean LST, (middle)summer mean LST and (right) winter mean LST versusISA
. Red lines are theﬁtted linear regressionfunctions. The valuesof the
SUHII (K) were calculated using the regression slope. Rows (a), (b), (c), and (d) present the results at four observation times. All the correlations are signiﬁcant here (p b0.01). (For
interpretation of thereferences to colour in this ﬁgure legend, the reader is referred to the web version of this article.)
267H. Li et al. / Science of the Total Environment 624 (2018) 262–272
land covers change from natural to urban, both the LST and ISA
crease. In the city center, the mean ISA
are close to 80%, and the tem-
peratures maintain high values, presenting a typical urban heat island
phenomenon. Then back from the city center towards rural regions,
the LST and ISA
decrease with the land covers changing from urban
to natural. The longitudinal variation of the LST shows good agreement
with the ISA
. Daytime and summer show more signiﬁcant urban-
rural variation in LST than nighttime and winter. The grids that are affect-
ed by water show lower LST during the day and higher LST at night than
other land cover types. These pixels are ﬁltered out when calculating the
ﬁtted functions of LST and calculated SUHII in annual scale
Fig. 7 shows the variations of the mean LST versus ISA
for the dif-
ferent time periods and observation times. In general, the LST increases
smoothly with the increase in ISA
, showing a good agreement. ISA
accounts for most of the variation in LST. The regression functions of LST
were well ﬁtted,withhighR
for all seasons and times of the day. The
values of R
for the ﬁtted function of the annual mean LST are up to
0.96, 0.97, 0.97 and 0.98, respectively at the four observation times. The
regression functions of LST ﬁtted by ISA
can be used to quantify
SUHII. The slopes of the ﬁtted functions of LST are deﬁned as the SUHII,
and are also shown in Fig. 7. At an annual scale, the values of the SUHII
are 5.43, 5.39, 3.58 and 3.95 K at 10:20, 11:50, 20:50 and 01:20, respec-
tively. The SUHII in summer and daytime are higher than that in winter
and nighttime. This is consistent with the max-min range of the LST
within the study areas in Table 1.
3.5. Calculated daily SUHII and its temporal variation
The linear regression functions of the daily LST were ﬁtted for each
valid image in 2010 using ISA
. Most of the LST images show good re-
lationships with ISA
. More than three-quarters of the R
of the ﬁtted
functions are larger than 0.90. Wet surface caused by the precipitation
(including rain and snow) before or during the observation time should
50 100 150 200 250 300 350
50 100 150 200 250 300 350
50 100 150 200 250 300 350
(c) MOD 20:50
(a) MOD 10:20 (b) MYD 11:50
(d) MYD 01:20
50 100 150 200 250 300 350
Fig. 8. Seasonalvariationsof the daily SUHII (red dots)and its correspondingR
(blue dots)at the four observationtimes in 2010 for Berlin.Subplots (a), (b), (c),and (d) present the results
at four observation times, respectively. (For interpretation of the references to colour in this ﬁgure legend, the reader is referred to the web version of this article.)
Statistics of daily SUHII (K)calculated using the valid regression functions of LST, with R
within the outlier boundary. The upper boundary is 1.5 times of the Interquartile Range
(IQR) largerthan the 3rd interquartile, while thelower boundary is 1.5 timesof IOR lower
than the 1st interquartile.
MOD 10:20 10.16 7.07 5.88 3.87 1.93 5.73
MYD 11:50 10.33 6.57 5.25 3.49 1.67 5.38
MOD 20:50 8.22 4.95 4.04 3.29 1.35 4.09
MYD 01:20 7.18 5.60 4.56 3.62 1.42 4.61
1500 2000 2500 3000 3500 4000 4500 5000
Year Summer Winter
Search radius (m)
Fig. 9. Variations of (a) the R
of the ﬁtted functions and (b) the calculated SUHII versus
kernel radius. The LST used here are the mean values of all valid images for different
268 H. Li et al. / Science of the Total Environment 624 (2018) 262–272
be the main reason for the weak LST-ISA
relationships in some days,
and is further discussed in Section 4.4. In general, the ISA
can be used
to quantify the daily SUHII well when the disturbance of precipitation is
removed. The well ﬁtted linear regression functions of LST with R
er than the lower outlier boundary (0.82) were used to calculate daily
SUHII. Seasonal variations of the daily SUHII are shown in Fig. 8.Gener-
ally, the daily SUHIIs displayhigher values in summer andlower values
in winter. It is consistent with the results at annual scale shown in Fig. 7.
The maximum value of the daily SUHII reaches up to10.32 K happening
at 11:50 on Julian day 156, while minimum value is 1.35 K happening at
01:20 on Julian day 274. The mean values are 5.73, 5.38, 4.09 and 4.61 K,
respectively, at the four observation times (Table 2). Daytime always
shows higher SUHII than nighttime.
4.1. Sensitivity of the quantiﬁcation of SUHII to kernel radius
In this study, the kernel radius represents the extent of the footprint
of the LST measurement over each pixel. The kernel radius could affect
the calculation results of KDE method, and then further affect the LST-
relationship. In order to obtain the optimal kernel radius for the
SUHII calculation, a sensitivity test at annual scale was conducted. The
linear functions of the mean LST during different periods were ﬁtted
using the kernel radius ranging from 1500 to 5000 m with an interval
of 500 m. The R
of the ﬁtted functions and the calculated SUHII versus
kernel radius are present in Fig. 9. The variations of the R
with kernel radius show similar trend for the whole year, summer, and
winter. The best regression functions of LST with highest R
at radius of 3000 m (3.0 times of the resolution). Either smaller or larger
kernel radius would weaken the LST-ISA
relationship. That is because
when the radius is too small, the KDE results could not contain enough
information inside the footprint, while when the radius is too large, the
KDE results would contain some noise information and outliers outside
the footprint (Thakali et al., 2015). The calculated SUHII is also affected
by kernel radius. Larger radius tends to produce higher SUHII when the
value was b2000 m. When the radius further increase, the sensitivity
of the calculated SUHII to kernel radius disappear.
4.2. Inﬂuence of the regionalization of ISA on the quantiﬁcation of SUHII
In order to evaluate the advantage of the newly developed ISA
calculating SUHII, a comparison study between the linear functions of
LST ﬁtted by ISA
and raw ISA was conducted. Tables 3 and 4 present
of the ﬁtted functions and calculated SUHII using these two
indicators. In general, the functions of LST were better ﬁtted by ISA
with higher R
for all time periods. ISA only reﬂects the surface property,
and does not include the inﬂuence of neighboring pixels on the LST mea-
surements within the footprint. So it cannot perform as well as ISA
terms of the ﬁtting of LST functions. The SUHII calculated using ISA
shows larger values than those calculated using raw ISA. Compared
with the map of the ISA
, the ISA clearly detects the distributions of
the peak values among pixels (Fig. 5a). The spatial variation of the ISA
is smoother than that of the raw ISA with lower STD. The urban-rural dif-
ference in the ISA
is smaller than that of the raw ISA, leading to higher
4.3. Validation of the new method using Landsat data
MODIS data can be used to study the synoptic overview and the tem-
poralvariationofurbanareas(Pu et al., 2006). In this study, MODIS LST
achieved good relationships with ISA
, and the diurnal and seasonal
patterns of the SUHII are analyzed. However, the coarse spatial resolu-
tion MODIS data cannot distinguish the ﬁne-scale variation of urban sur-
face, limiting the accurate detection of complex urban thermal
environment in detail. Urban surface temperature varies largely in both
the city and street scales due to the high heterogeneity of urban surface.
Strong sub-pixel variations of temperature even exist within the coarse
MODIS pixels. An more accurate study of SUHI demands higher resolu-
tion thermal remote sensing data. In order to further examine the reli-
ability of the new method, Landsat data at a 30 m resolution and
Copernicus ISA data version 2012 at a 20 m resolution were used to cal-
culate the SUHII. Two Landsat 7 ETM+ images (Fig. 10a, b) were collect-
ed from the USGS website (https://earthexplorer.usgs.gov/). The LST was
calculated using a mono-window algorithm (Zhang et al., 2006). The ISA
data was resampled to 30 m ﬁrstly, and then further regionalized using
KDE method with a kernel radius of 90 m (3 times resolution of Landset
images referring to the sensitivity analysis results above). The ISA
shows a smoother spatial pattern than the raw ISA (Fig. 10c, d). The func-
tions of LST are ﬁtted using the raw ISA and ISA
, respectively, and the
slopes of the functions are used to calculate the SUHII. Table 5 shows the
statistics of the ﬁtted functions and the calculated SUHII. The R
ﬁtted functions using ISA
is larger than those using the raw ISA. Com-
pared with the raw ISA, the ISA
better reﬂects the spatial variation of
the Landsat LST. Meanwhile, the STD of ISA
(27.67%) is smaller than
the raw ISA (31.82%), which leads to larger values of the calculated
SUHII using ISA
. The difference between the ISA
and the raw ISA
in ﬁtting the functions of LST and the calculated SUHII using the Landsat
data are consistent with those using MODIS data as shown in Tables 3
4.4. Inﬂuence of precipitation on the new method application
Precipitation could alter surface moisture, and decrease the urban-
rural difference in both albedo and water availability, weakening the
relationship. The foundation of the hypothesis of this new
method shown in the Fig. 3 would be broken by precipitation. Fig. 11
shows the box plots of the R
of the ﬁtted functions of daily LST under
rainy and non-rainy conditions. When precipitation occurs, the R
shows a lot of small values (Fig. 11a). The 25th percentile and lower
boundary of outliers are only 0.77 and 0.48. When no precipitation oc-
curs, the R
shows high values, with the 25th percentile and the lower
boundary of outliers of 0.91 and 0.84, respectively. Here only the precip-
itation happened within 12 h before observation times are taken into
account. Strong precipitation has a longer term impact on the ground
moisture. Most of the outliers in Fig. 11b happen due to the strong pre-
cipitation before 12h of the observation times. Inaddition, the evapora-
tion and themeltingof snow in winter is very slow (Li et al., 2016). The
snow covering could last long time and weaken the UHI effect, which
leads to parts of the outliers in Fig. 11b.
Comparison of the R
for the ﬁtted functions of LST using ISA
and raw ISA.
Observation time ISA
Year Summer Winter Year Summer Winter
MOD 10:20 0.96 0.96 0.95 0.92 0.91 0.91
MYD 11:50 0.97 0.96 0.97 0.90 0.90 0.89
MOD 20:50 0.97 0.96 0.96 0.83 0.84 0.79
MYD 01:20 0.98 0.98 0.96 0.84 0.86 079
Comparison of the SUHII calculated using the slope of the LST functions ﬁtted by ISA
and raw ISA.
Observation times ISA
Year Summer Winter Year Summer Winter
MOD 10:20 5.43 6.55 3.81 3.23 3.90 2.27
MYD 11:50 5.30 6.55 3.81 3.01 3.79 2.11
MOD 20:50 3.58 3.91 3.05 1.74 2.00 1.38
MYD 01:20 3.95 4.74 2.89 1.83 2.25 1.26
269H. Li et al. / Science of the Total Environment 624 (2018) 262–272
5. Conclusions and outlook
This paper proposed a new approach to quantify the SUHII by ﬁtting
the linear functions of LST using ISA. Given the footprintof LST measure-
ment, the ISA was regionalized using a Kernel Density Estimation ap-
proach. The spatial variation of the LST in Berlin region displayed
pronounced SUHI characteristic. The spatial patterns of the LST and
were similar. The linear functions of LST were well ﬁtted using
in both annual and daily scale. The slope of the linear regression
was deﬁned as SUHII. The annual mean SUHII were 5.72, 5.38, 4.09and
4.61 K at UTC time 10:20, 11:50, 20:50and 01:20, respectively. The LST
showed higher correlation with ISA
than with raw ISA across all time
periods. The reliability of the new method was further veriﬁed using
ﬁne resolution Landset data. Precipitation could weaken the depen-
dence of LST on surface imperviousness and then inﬂuence the calcula-
tion of SUHII. The practical application of the new method should avoid
The well ﬁtted linear functions of LST using ISA
provide a promis-
ing approach to quantify the SUHII using remote sensing data. Compared
with the traditional approach of calculating the deﬁcit of measurements
at urban and rural stations or pixels, the new approach could avoid the
bias caused by the choice of the stations or pixels. This method can be
easily applied in other cities, which makes the comparisons of SUHI
among different cities possible. Netherless, it should be noted that the
hypothesis of the new method is that LST increases along with ISA. This
hypothesis is true for most cities in biomes dominated by forests and
grasslands. However, for the cities in desert environments, the LST's re-
sponse to ISA presented U-shaped horizontal gradient (e.g. Imhoff et
al., 2010; Zhang et al., 2010). So this method does not work for the cities
in the arid or desert environment. Compared with SUHII, UHII calculated
by air temperature is more concerned in term of heat stress. Usually,
SUHII is larger than UHII. This study only focuses on the quantiﬁcation
of SUHII. As for the feasibility of this method for quantifying UHII, a fur-
ther study is needed in the next step. In addition, there is a slight differ-
ence in the acquired time among the pixels within the images of daily
MODIS LST products. This may affect the result to some extent and can
be investigated in future studies.
This research is supported by the China Scholarship Council, FUBright
Mobility Allowances for Research Stays promoted by the German Aca-
demic Exchange Service (DAAD)-Dahlem Research School of Freie
Universität Berlin, Mobility Allowance for Junior Research Stays from
University Alliance for Sustainability promoted by Freie Universität Ber-
lin. The authors thank the Beijing Ofﬁce of Freie Universität Berlin. Spe-
cial thanks go to Dr. Hamid Taheri Shahraiyni for help in processing
Fig. 10.Spatial patternsof (a and b) the Landsat LSTand (c and d) ISA. The (a)and (b) show the retrieved LST attime 9:53 UTC, 28th April and 9:55 UTC,9th July 2010, respectively.The (c)
and (d) show the raw ISA and regionalized ISA
Statistics of the ﬁtted functions of the Landsat LST and the calculated SUHII.
Image Indicator R
28th, April ISA 0.84 2.23
9th, July ISA 0.92 4.13
Fig. 11. Box plots of the R
for the ﬁtted functions under (a) rainy and (b) non-rainy
conditions within the 12 h before the observation times. Top and bottom of the blue
boxes represent the 75th and 25th percentile and red horizontal line within the box
indicates the median. Short top and bottom bars outside the boxes are the boundaries of
upper and lower outlier deﬁned by 1.5 IQR, green crosses are the outliers. (For
interpretation of the references to colour in this ﬁgure legend, the reader is referred to
the web version of this article.)
270 H. Li et al. / Science of the Total Environment 624 (2018) 262–272
remote sensing data and Patricia Margerison for proofreading this
Anderson,T.K., 2009. Kernel density estimation and K-means clustering to proﬁle road ac-
cident hotspots. Accid. Anal. Prev. 41 (3), 359–364.
Arnﬁeld, A.J., 2003. Two decades of urban climate research: a review of turbulence,
exchanges of energy and water, and the urban heat island. Int. J. Climatol. 23,
Azevedo, J.A., Chapman, L., Muller, C.L., 2016. Quantifying the daytime and night-time
urban heat island in Birmingham, UK: a comparison of satellite derived land surface
temperature and high resolution air temperature observations. Remote Sens. 8:153.
Buyantuyev, A., Wu, J.G., 2010. Urban heat islands and landscape heterogeneity: linking
spatiotemporal variations in surface temperatures to land-cover and socioeconomic
patterns. Landsc. Ecol. 25 (1), 17–33.
Cadenasso,M.L.,Pickett,S.T.,Schwarz,K.,2007.Spatial heterogeneity in urban ecosystems:
reconceptualizing land cover and a framework for classiﬁcation. Front. Ecol. Environ. 5
Campagnolo, M.L., Montano, E.L., 2014. Estimation ofeffective resolution for daily MODIS
gridded surface reﬂectance products. IEEE Trans. Geosci. Remote Sens. 52 (9),
Cheval, S., Dumitrescu, A., 2015. The summer surface urban heat island of Bucharest (Ro-
mania) retrieved from MODIS images. Theor. Appl. Climatol. 121, 631–640.
Coll, C., Wan, Z., Galve, J.M., 2009. Temperature-based and radiance-based validations of
the V5 MODIS land surface temperature product. J. Geophys. Res. Atmos. 114 (20).
Constantinescu, D., Cheval, S., Caracaş, G., Dumitrescu, A., 2016. Effective monitoring and
warning of Urban Heat Island effect on the indoor thermal risk in Bucharest (Roma-
nia). Energ. Buildings 127, 452–468.
Dou, J., Wang, Y., Bornstein, R., Miao, S., 2015. Observed spatial characteristics of Beijing
urban climate impacts on summer thunderstorms. J. Appl. Meteorol. Climatol. 54 (1),
Du, H., Wang, D., Wang, Y., Zhao, X., Qin, F., Jiang, H., Cai, Y., 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,
Earl, N., Simmonds, I., Tapper, N., 2016. Weekly cycles in peak time temperatures and urban
heat island intensity. Environ. Res. Lett. 11 (7), 074003.
Feranec, J., Hazeu, G., Christensen, S., Jaffrain, G., 2007. Corine land cover change detection
in Europe (case studies of the Netherlands and Slovakia). Land Use Policy 24 (1),
Grimm, N.B., Faeth, S.H., Golubiewski, N.E.,Redman, C.L., Wu, J., Bai, X., Briggs, J.M., 2008.
Global change and the ecology of cities. Science 319, 756–760.
Hu, L., Brunsell, N.A., 2015. A new perspective to assess the urban heat island through re-
motely sensed atmospheric proﬁles. Remote Sens. Environ. 158, 393–406.
Hu, L., Monaghan, A., Voogt, J.A., Barlage, M., 2016. Aﬁrst satellite-based observational as-
sessment of urban thermal anisotropy. Remote Sens. Environ. 181, 111–121.
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 (3),
Inouye, D.W., 2015. The next century of ecology. Science 349, 565.
Privette, J.L., Riggs, G., Strahler, A., 1998. The moderate resolution imaging
spectroradiometer (MODIS): land remote sensing for global change research. IEEE
Trans. Geosci. Remote Sens. 36, 1228–1249.
Kottek, M.,Grieser, J., Beck, C.,Rudolf, B., Rubel, F.,2006. World map of the Köppen-Geiger
climate classiﬁcation updated. Meteorol. Z. 15 (3), 259–263.
Kuang, W.,Liu, J., Zhang, Z., Lu,D., Xiang, B., 2013.Spatiotemporal dynamics of impervious
surface areas across China during the early 21st century. Chin. Sci. Bull. 58 (14),
Kuang, W., Liu, Y., Dou, Y., Chi, W., Chen, G., Gao, C., Yang, T., Liu, J., Zhang, R.,2015a. What
are hot and what are not in an urban landscape: quantifying and explaining the land
surface temperature pattern in Beijing, China. Landsc. Ecol. 30 (2), 357–373.
heat ﬂux regulation of metropolitan land-use/land-cover components by coupling re-
mote sensing-modelling with in situ measurement. J. Geophys. Res. Atmos. 120 (1),
Kuang, W., Chen, L., Liu, J., Xiang, W., Chi, W., Lu, D., Yang, T., Pan, T., Liu, A., 2016a. Remote
sensing-based artiﬁcial surface cover classiﬁcation in Asia and spatial pattern analysis.
Sci. China Earth Sci. 9 (59), 1720–1737.
Kuang, W., Liu, J., Dong, J., Chi, W., Zhang, C., 2016b. The rapid and massive urban and in-
dustrial land expansions in China between 1990 and 2010: a CLUD-based analysis of
their trajectories, patterns and drivers. Landsc. Urban Plan. 145, 21–33.
Kuang, W.,Yang, T., Liu, A., Zhang, C., Lu, D., Chi, W., 2017.An EcoCity model for regulating
urban land cover structure and thermal environment: taking Beijing as an example.
Sci. China Earth Sci. 60 (6), 1098–1109.
Langanke, T., 2013. GIO Land (GMES/Copernicus Initial OperationsLand) High Resolution
Layers (HRLs)—Summary of Product Speciﬁcations. European Environment Agency.
Lee, D.S., Storey, J.C., Choate, M.J., Hayes, R.W., 2004. Fouryears of Landsat-7 on-orbit geo-
metric calibration and performance. IEEE Trans. Geosci. Remote Sens. 42 (12),
Zhang, Q., Zhao, M., Zhou, M., 2015. Priorities for boundary layer meteorology research
in China. Bull. Am. Meteorol. Soc. 96 (9), 149–151.
Lefebvre,A.,Beaugendre,N.,Pennec,A.,Sannier,C.,Corpetti,T.,2013.Using data fu-
sion to update built-up areas of the 2012 European High-Resolution Layer Im-
perviousness. Proceedings of the 33rd EARSeL Symposium Conference, Matera,
Italy. 36, p. 321328.
Li, C.F., Yin,J., 2013. A study on urban thermalﬁeld of Shanghai using multi-source remote
sensing data. J. Indian Soc. Remote Sens. 41 (4), 1009–1019.
Li, J., Song, C., Cao, L.,Zhu, F., Meng, X., Wu,J., 2011. Impacts of landscape structure on sur-
face urban heat islands: a case study of Shanghai, China. Remote Sens. Environ. 115
Li, H., Wang, A., Guan, D., Jin, C., Wu, J., Yuan, F., Shi, T., 2016. Empirical model development
for ground snow sublimation beneath a temperate mixed forest in Changbai Moun-
tain. J. Hydrol. Eng.:21(11) https://doi.org/10.1061/(ASCE)HE.1943-5584.0001415.
Li, H., Wolter, M., Wang,X., Sodoudi, S., 2017. Impact of land cover data on the simulation
of urban heat island for Berlin using WRF coupled with bulk approach of Noah-LSM.
Theor. Appl. Climatol. https://doi.org/10.1007/s00704-017-2253-z.
McDonnell, M.J., MacGregor-Fors, I., 2016. The ecological future of cities. Science 352,
Mills, G., 2008. Luke Howard and the climate of London. Weather 63 (6), 153–157.
Morris, C., Simmonds, I., Plummer, N., 2001. Quantiﬁcation of the inﬂuences of wind and
cloud on the nocturnal urban heat island of a large city. J. Appl. Meteorol. 40, 169–182.
Oke, T.R., 1982. The energetic basis of the urban heat island. Q. J. R. Meteorol. Soc. 108,
Oke, T.R., 1988. The urban energy balance. Prog. Phys. Geogr. 12, 471–508.
Oke, T.R., 2006. Instruments and observing methods: Report No. 81: initial guidance to ob-
tain representative meteorological observations at urban sites. World Meteorological
Organization, WMO/TD (1250).
Peng,J.,Liu,Q.,Wang,L.,Liu,Q.,Fan,W.,Lu,M.,Wen,J.,2015.Characterizing the pixel foot-
print of satellite albedo products derived from MODIS reﬂectance in the Heihe River
Basin, China. Remote Sens. 7 (6), 6886–6907.
Pu, R., Gong, P., Michishita, R., Sasagawa, T., 2006. Assessment of multi-resolution and
multi-sensor data for urban surface temperature retrieval. Remote Sens. Environ. 104
Rajasekar, U., Weng, Q.H., 2009. Urban heat island monitoring and analysis using a non-
parametric model: a case study of Indianapolis. ISPRS J. Photogramm. Remote Sens.
64 (1), 86–96.
Rannik,Ü.,Aubinet,M.,Kurbanmuradov,O.,Sabelfeld, K.K., Markkanen, T., Vesala, T., 2000.
Footprint analysis for measurements over a heterogeneous forest. Bound.-Layer
Rigo, G., Parlow, E., Oesch, D., 2006. Validation of satellite observed thermal emission with
in-situ measurements over an urban surface. Remote Sens. Environ. 104 (2), 201–210.
Rizwan, A.M., Dennis, L.Y.C., Liu, C., 2008. A review on the generation, determination and
mitigation of Urban Heat Island. J. Environ. Sci. 20, 120–128.
Roth, M., Oke, T.R., Emery, W.J., 1989. Satellite-derived urban heat islands from three coast-
al cities and the utilization of such data in urban climatology. Int. J. Remote Sens. 10
Copernicus high resolution layer on imperviousness degree for 2006, 2009 and 2012.
Proceedings of the International Symposium on Spatial Accuracy Assessment in Natu-
ral Resources and Environmental Sciences, Montpellier, France, pp. 5–8.
Schatz, J., Kucharik,C.J., 2014. Seasonality of the urban heat island effect in Madison, Wis-
consin. J. Appl. Meteorol. Climatol. 53 (10), 2371–2386.
Schatz, J., Kucharik, C.J., 2015. Urban climate effectson extreme temperaturesin Madison,
Wisconsin, USA. Environ. Res. Lett. 10 (9), 094024.
Schwarz, N., Lautenbach, S., Seppelt, R., 2011. Exploring indicators for quantifying surface
urban heat islands of Europeancities with MODIS land surface temperatures. Remote
Sens. Environ. 115 (12), 3175–3186.
Silverman, B.W., 1986. Density Estimation for Statistics and Data Analysis. Chapman and
Hall, New York.
Stewart, I.D., 2011. Asystematicreviewandscientiﬁc critique of methodology in modern
urban heat island literature. Int. J. Climatol. 31 (2), 200–217.
Stewart, I.D., Oke, T.R., 2012.Local climate zones for urban temperature studies. Bull. Am.
Meteorol. Soc. 93, 1879–1900.
Tan, J., Zheng, Y., Tang, X., Guo, C., Li, L., Song, G., Zhen, X., Yuan, D., Kalkstein, A.J., Li, F.,
Chen, H., 2010. The urbanheat island and its impact on heat wavesand human health
in Shanghai. Int. J. Biometeorol. 54 (1), 75–84.
Thakali,Lalita, Kwon, T.J., Fu,L., 2015. Identiﬁcation of crash hotspots using kerneldensity
estimation and kriging methods: a comparison. J. Mod. Transp. 23 (2), 93–106.
Townshend, J.R.G., Huang, C., Kalluri, S.N.V., Defries, R.S., Liang, S., Yang, K., 2000. Beware
of per-pixel characterization of land cover. Int. J. Remote Sens. 21, 839–843.
Wan, Z., 2008. New reﬁnements and validation of the MODIS land-surface temperature/
emissivity products. Remote Sens. Environ. 112 (1), 59–74.
Wang,K.,Jiang,S.,Wang,J.,Zhou,C.,Wang,X.,Lee,X.,2017.Comparing the diurnal
and seasonal variabilities of atmospheric and surface urban heat islands based
on the Beijing urban meteorological network.J.Geophys.Res.Atmos.122(4),
Williamson, S.N., Hik, D.S., Gamon, J.A., Kavanaugh, J.L., Koh, S., 2013. Evaluating cloud
contamination in clear-sky MODIS Terra daytime land surface temperatures using
ground-based meteorology station observations. J. Clim. 26 (5), 1551–1560.
Wolfe, R.E., Nishihama, M., Fleig, A.J., Kuyper, J.A., Roy, D.P., Storey, J.C., Patt, F.S., 2002.
Achieving sub-pixel geolocation accuracy in support of MODIS land science. Remote
Sens. Environ. 83, 31–49.
Yang, P., Ren, G., Liu, W., 2013. Spatial and temporal characteristics of Beijing urban heat
island intensity. J. Appl. Meteorol. Climatol. 52, 1803–1816.
Yang, X., Leung, L.R., Zhao, N., Zhao, C., Qian, Y., Hu,K., Liu, X., Chen, B., 2017. Contribution
of urbanization to the increase of extreme heat events in an urban agglomeration in
east China. Geophys. Res. Lett. https://doi.org/10.1002/2017GL074084.
271H. Li et al. / Science of the Total Environment 624 (2018) 262–272
Yuan, F.,Marvin, E.B., 2007. Comparison of impervious surface area and normalized differ-
ence vegetation index as indicators of surface urban heatisland effects in Landsatim-
agery. Remote Sens. Environ. 106 (3), 375–386.
Zhang, J., Wang, Y., Li, Y., 2006. A C++ program for retrieving land surface temperature
from the data of Landsat TM/ETM+ band6. Comput. Geosci. 32 (10), 1796–1805.
Zhang, P., Imhoff, M.L., Wolfe, R.E., Bounoua, L., 2010. Characterizing urban heat islands of
global settlements using MODIS and nighttime lights products. Can. J. Remote. Sens.
36 (3), 185–196.
Zhang,P.,Imhoff,M.L.,Bounoua,L.,Wolfe,R.E.,2012.Exploring the inﬂuence of impervious
surface density and shape on urban heat islands in the northeast United States using
MODIS and Landsat. Can. J. Remote. Sens. 38 (4), 441–451.
Zhang, P., Bounoua, L., Imhoff, M.L., Wolfe, R.E., Thome, K.J., 2014. Comparison of MODIS
land surface temperature and air temperature over the continental USA meteorologi-
cal stations. Can. J. Remote. Sens. 40 (2), 110–122.
Zhou, Y., Gurney, K., 2010. A new methodology for quantifying on-site residential and
commercial fossil fuel CO
emissions at the building spatial scale and hourly time
scale. Carbon Manage. 1 (1), 45–56.
Zhou, Y., Weng, Q., Gurney, K.R., Shuai, Y., Hu, X., 2012. Estimation of the relationship be-
tween remotely sensed anthropogenic heat discharge and building energyuse. ISPRS
J. Photogramm. Remote Sens. 67, 65–72.
Zhou, Y., Smith, S.J., Elvidge, C.D., Zhao, K., Thomson, A., Imhoff, M., 2014. A cluster-based
method to map urban area from DMSP/OLS nightlights. Remote Sens. Environ. 147,
Zhou, Y., Smith, S.J., Zhao, K., Imhoff, M., Thomson, A., Bond-Lamberty, B., Asrar, G.R.,
Zhang, X., He, C., Elvidge, C.D., 2015. A global map of urban extent from nightlights.
Environ. Res. Lett. 10 (5), 054011.
Zhou, B., Rybski, D., Kropp, J.P., 2017. The role of city size and urban form in the surface
urban heat island. Sci. Rep. 7, 4791.
272 H. Li et al. / Science of the Total Environment 624 (2018) 262–272