Atmospheric Research 94 (2009) 276–284

Contents lists available at ScienceDirect

Atmospheric Research
j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / a t m o s

Urban heat island diagnosis using ASTER satellite images and ‘in situ’
air temperature
Janet E. Nichol a,⁎, Wing Yee Fung b, Ka-se Lam b, Man Sing Wong a
a
Department of Land Surveying and Geo-Informatics, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong
b
Department of Civil and Structural Engineering, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong

a r t i c l e i n f o a b s t r a c t

Article history: This study demonstrates that thermal satellite images combined with ‘in situ’ ground data can
Received 7 December 2008 be used to examine models of heat island genesis and thus identify the main causes of urban
Received in revised form 27 March 2009 heat islands (UHIs). The models, although proposed over 30 years ago, have not been
Accepted 7 June 2009
thoroughly evaluated due to a combination of inadequate ground data and the low resolution of
thermal satellite data. Also there has been limited understanding of the relevance of satellite-
derived surface temperatures to local and regional scale air temperatures. A cloud-free ASTER
thermal image of urban and rural areas of Hong Kong was obtained on a winter night with a
well-developed heat island, accompanied by a 148 km vehicle traverse of air temperatures.
Over the whole traverse a high R2 of 0.80 was observed between surface and air temperatures,
with the two datasets showing a similar amplitude and general trend, but with the surface
exhibiting much higher local variability than air temperature. Gradients in both surface and air
temperature could be related to differences in land cover, with little evidence of large scale
advection, thus supporting the population/physical structure model of UHI causation, rather
than the advection model. However, the much higher surface and air temperatures observed
over the largest urban area, Kowloon, than over any smaller urban centre with similar physical
structure in the New Territories, would seem more indicative of the advection model. The
image and ground data suggest that Kowloon's urban canopy layer climate is mainly influenced
by local city structure, but it is also modified by a strongly developed, regional scale urban
boundary layer which has developed over the largest urban centre of Kowloon, and reinforces
heating from both above and below.
© 2009 Published by Elsevier B.V.

1. Introduction the accumulation of heat as rural air moves inwards (Oke,

1976; Summers, 1964) and (ii) the population/physical
The sustainability of tropical and sub-tropical cities such as structure model whereby population is a surrogate for the
Hong Kong where dense and high-rise buildings are accom- intensity of urbanization. Thus factors such as building size,
panied by an intense urban heat island (UHI) effect may be anthropogenic heat inputs, heat capacity and thermal inertia
threatened if predictions of global warming are realized. increase toward the city centre, and sky view factor and
Two different models have been proposed to explain the vegetation are reduced (Oke, 1976). These models are relevant
extent and magnitude of the UHI. These are (i) the advection because they may indicate future planning policies for heat
model which relates heat island magnitude (ΔT(u − r)) to the island mitigation. For example according to the advection
distance of fetch from the rural–urban boundary, and thus to model, the magnitude of ΔT(u − r) increases with distance
from the rural boundary Therefore this model may suggest
that structures built at the edge of the city, irrespective of their
⁎ Corresponding author. Tel.: +852 2766 5952. height, building density, street geometry and building mate-
E-mail address: lsjanet@polyu.edu.hk (J.E. Nichol). rial used, would exhibit a minimal heat island effect, but

0169-8095/$ – see front matter © 2009 Published by Elsevier B.V.

doi:10.1016/j.atmosres.2009.06.011
 J.E. Nichol et al. / Atmospheric Research 94 (2009) 276–284 277

conversely they should intensify ΔT(u − r) in the city centre by sufficient ‘in situ’ air and surface temperatures for statistical
increasing the distance from the countryside. On the other analysis of climatological processes. Indeed, most studies
hand, according to the population/physical structure model, if which were able to collect ‘in situ’ ground temperatures at the
city centres were less densely built (though not necessarily image time have used the low resolution sensors AVHRR and
less tall) with higher sky view factors, using material with MODIS, due to their high (12-hourly) temporal coverage
lower heat capacities and thermal inertia, ΔT(u − r) would be coinciding with fixed station data. At the ca. 1 km resolution
significantly reduced. Furthermore, adding new buildings at of these sensors it is easy to understand why Roth et al. (1989)
the periphery and thus increasing the distance from the warn against assuming any direct relationship between
periphery to the centre, as for example in Hong Kong's harbour satellite-derived surface temperature and regional air tem-
reclamation schemes, would not affect the overall UHI perature, and Balling and Brazel (1988, 1989) and Gallo et al.
magnitude. These questions are important because of high- (1993), also working with AVHRR found an insignificant, or
rise trends in modern cities due to considerations of energy low correlation between Surface Temperature (Ts) derived
and transport efficiency. Thus by 2030 China is predicted to from AVHRR at 1.1 km resolution, and Air Temperature (Ta)
have 15 high-rise mega-cities each with 25 m people. from fixed ground stations. This may be because as the sensor
Studies conducted by night-time mobile traverse resolution decreases, the temperature difference between
(Goldreich, 1985; Eliasson, 1992) support the physical land uses also decreases, implying a decrease in the observed
structure model with, for example, cooler air temperature at range of surface temperatures (Stoll and Brazel, 1992).
open street intersections (Eliasson, 1992). On the other hand Unfortunately, for urban studies, the highest thermal satellite
Stoll and Brazel (1992) and Spronken-Smith and Oke (1998) image resolution currently available is 90 m from ASTER, and
observe that although the thermal properties of surfaces and there are no reported studies of ASTER images supported by
their radiative geometry are dominant factors in heat island extensive ‘in situ’ ground data collection over urban areas.
formation, the correlation between the surface and air is
affected by advection from adjacent land uses, thereby 3.2. Relevance of satellite-derived Surface Temperature (Ts)
supporting Summers' (1964) advection model. Thus both
models may be relevant, but their influence may vary for The relationship between Ts and Ta remains the greatest
different cities and climates. Traditional methods of UHI unknown in remotely sensed studies of heat islands. Although
analysis, the use of fixed stations and/or vehicle traverse have there is a continuous feedback of energy exchange between
been unable to address such basic conceptual questions because the surface and the adjacent air, the scales of operation appear
the data collected are spatially incomplete. Although the UHI to be different (Voogt and Oke, 2003; Nichol, 2009). This has
has been observed from thermal satellite images for four given rise to a distinction in remote sensing studies between
decades, the studies have not contributed substantially to surface urban heat islands (SUHIs) which are identifiable at
understanding of the causes of urban heat islands, and the micro-scale of individual streets and building blocks (e.g.
inferentially, their mitigation (Roth et al., 1989; Arnfield, Nichol, 1994, 1996a,b), and atmospheric heat islands which
2003; Voogt and Oke, 2003). are recognizable over a whole city at meso-scale. However, the
distinction has not assisted the application of remote sensing
2. Hong Kong and its urban heat island technology to heat island analysis because of a scarcity of
two types of data (i) ‘in situ’ surface temperature data
At 22°N Hong Kong experiences very hot humid summers collected at the time of image acquisition to verify satellite-
and warm dry winters. Air temperature typically reaches derived surface temperatures, and (ii) substantial ‘in situ’ air
34 °C on the hottest summer days, cooling to ca. 28 °C in temperature data: usually only a few points from fixed clima-
urban, and 24 °C in rural areas at night. Relative humidity is tic stations are available to represent a whole city. A better
very high averaging 81% in July and August, with low mean understanding of the Ts/Ta relationship is essential to
wind speeds of 2.8 m/s. Kowloon, with a population of over optimize the application of remote sensing to heat island
2 million is the largest continuous urban area and is separated analysis, since Oke's (1976) physical structure model dis-
from the urbanized northern shore of Hong Kong island by cussed in Section 1, assumes a direct causative relationship
the 1 km wide harbour and from small towns in the New between Ts and Ta. The relationship between surface and air
Territories by mountain ranges. These geographical con- temperatures increases closer to the surface (1996a), with
straints result in extremely high population densities of lower wind speeds (Stoll and Brazel, 1992), and also at night
over 52,000 persons km2 in some areas. Reduced ventilation, when micro-scale advection is weaker.
high temperatures and the blocking of sea breezes by tall The present study attempts to advance the understanding
buildings on coastal reclaimed land, the so called ‘wall effect’ of surface and air temperature relationships over an urban
are contentious issues. Planners urgently require empirical heat island, using a night-time ASTER image of Hong Kong,
and spatially comprehensive environmental data to support accompanied by over 14,000 paired image and air tempera-
structure plans and policies. ture points recorded at the image time as well as 18 ground
points representing contact Surface Temperature.
3. Satellite sensors for urban heat island analysis
4. Image data used
3.1. Spatial resolution
The study used a night-time image of Hong Kong at
There is no comprehensive study to date, of the urban heat 10.42 pm on 31.01.07 (Fig. 1). This was approximately 4 h after
island using medium resolution sensors supported by the evening thermal crossover time, when ΔT(u − r) had
278 J.E. Nichol et al. / Atmospheric Research 94 (2009) 276–284

Fig. 1. Route of mobile traverse (shown as black or white line) over the thermal image of Hong Kong, on the night of 31st January 2007. The 18 field data points were
located in urban and rural areas along the traverse route.

reached 80% of its full development (Fig. 2). The atmospheric which exaggerate ΔT(u − r) (Roth et al., 1989) because
heat island is best developed a few hours after sunset when differences in surface temperature are largest during the
the rural surface cools rapidly and heat is extracted from the daytime.
adjacent air, thus differences in air temperature are greatest at Band 13 (10.25–10.95 µm) of ASTER's five thermal bands
night. However large majority of satellite-based studies of the was selected for processing since bands 10 and 14, being close
urban heat island have been carried out using daytime images to the edge of the atmospheric window are most affected by

Fig. 2. Urban and rural air temperatures for 48 h around the image time from Kings Park (urban) and Ta Kwu Ling (rural) climate stations. Source: Hong Kong
Observatory.
 J.E. Nichol et al. / Atmospheric Research 94 (2009) 276–284 279

atmospheric effects (Gillespie et al., 1998) and the sensitivity 5.2. Field data collection
of band 12 is said to be decreasing (ERSDAC, 2005). The image
was obtained as Level 1B Radiance at Sensor data (AST_L1B), Continuous air temperature measurements were taken
uncorrected for emissivity or atmospheric effects. The noise- from two vehicles covering approximately 148 km in both
equivalent temperature difference (NEΔT) of the ASTER urban and rural areas across Kowloon and the New Territories
thermal bands is given as 0.3 °C (ERSDAC, 2005). (Fig. 1), within 90 min of the image time, on a clear winter
In addition, the ASTER level 2B Surface Temperature night. The thermometers were placed on the roof of the
product, AST2B3 was obtained for the study. This image, vehicles and covered by a paper box to reduce wind chill.
equivalent to Ts, is an automated product of the Temperature Many slot openings on two sides ensured adequate aspiration
Emissivity Separation (TES) algorithm (Gillespie et al., 1999) of the sensors. Air temperatures measured by the vehicle
and is atmospherically and emissivity corrected. were constantly checked by manual ‘off vehicle’ measure-
ments. The accuracy of the sensor is 0.1 °C. Elevations along
the route were near sea level in Kowloon, and around the
4.1. Spatial enhancement and conversion to Surface
small towns of Yuen Long, Shatin and Tai Po, whereas other
Temperature (Ts)
sections covered rugged terrain up to 600 masl. Additionally,
paired surface and air temperatures, the latter at 1 m offset
A simple method to enhance the spatial resolution of
from the ground surface, were taken at 18 points, using Testo
thermal images which may be termed Emissivity Modulation
720 portable digital thermometers with air and contact
(EM) (Nichol, 1994, 2009) was applied to the level 1B image.
surface probes. All sensors were calibrated with an Omega
The differential emissivity of land cover types within a large
calibrator CL730A, and a certified mercury thermometer NIST
thermal image pixel may be represented by applying known
SRM943 was used as a standard reference. Field-based air and
emissivity values to land cover types classified from a higher
surface temperatures were adjusted to the precise image time
resolution visible band image using an equation derived from
by fitting to a polynomial curve derived from 24-hourly air
the Stefan Boltzmann law (Eq. (1)) (Sabins, 1997). The
temperatures from the Hong Kong Observatory (for air
equation ratios the larger pixels of the blackbody tempera-
temperatures) and 1-minute interval readings from contact
ture (Tb) image with the emissivity values (ε) of the higher
thermistors located near the centre of the study area. Low
resolution dataset, and the resulting emissivity corrected
wind speeds of ca. 1 m/s, north–north easterly, and low
image (Ts) has a pixel size corresponding to that of the
relative humidity of 55% prevailed at the image time and a
higher resolution emissivity data.
temperature inversion was observed at 600 m altitude during
the vehicle traverse. At the image time surface temperature
1=4 was only slightly warmer than the air with a Mean Absolute
Ts = Tb = ε : ð1Þ
Difference (MAD) of 1.6 °C (1c, rows 2 and 3). In summary,
the timing of the image at night, the advanced stage of heat
island development, the low wind speed and clear sky with
temperature inversion provided ideal conditions for heat
5. Methods island development (Kawashima et al., 2000).

5.1. Image processing

6. Results
The image radiance values were converted to blackbody
temperature (Tb) using the gain and offset values from the 6.1. Accuracy of image-derived Ts
image metadata with the Planck function. Since no visible
wavebands were available for this night-time scene, this The image-derived Ts at 10 m resolution was found to give
90 m Tb image was emissivity corrected according to land the best representation of ground measured surface tem-
cover type, using a classified SPOT 5 image of 10 m peratures, when compared with the 18 field sampling points,
resolution and acquired within one month of the ASTER having the lowest Mean Absolute Difference (MAD) of only
image. A classification accuracy of 92% was obtained. 1.0 °C (Table 1), as well as the highest correlation (R2 = 0.71)
Emissivity values from the MODIS emissivity library (Sup- with the field data (Fig. 4). For urban areas, the EM image
plementary Table 1 in Appendix A) were allocated to the values were even closer to the field measured Ts with an MAD
classes. Different emissivities for trees which remain green,
and grass which senesces, during the dry season were used.
The Surface Temperature (Ts) image produced from Eq. (1) Table 1
resulted in an output pixel size of 10 m. Finally the Ts image Coefficient of Determination (R2) and Mean Absolute Difference (MAD)
was atmospherically corrected using an average of four Sea between 18 paired field points of contact surface temperature, against image
Ts and air temperature.
Surface Temperature (SST) measurements obtained by the
Hong Kong Observatory at the image time, from easterly and Field data Image processing method R2 MAD
westerly waters off Hong Kong. Surface field 10 m Ts (EM method) 0.71 1.0
The EM corrected Ts image at 10 m resolution, the 90 m Ts Surface field 90 m Ts 0.58 1.4
image (emissivity corrected at 90 m resolution) and the Surface field Level 2B 90 m (AST2B3) 0.49 2.6
Air field Surface field 0.81
ASTER level 2B Surface Temperature image (AST2B3) were Air field 10 m Ts (EM method) 0.81
validated against 18 field Ts measurements.
280 J.E. Nichol et al. / Atmospheric Research 94 (2009) 276–284

of only 0.5 °C. Since the 10 m emissivity modulated image is Table 2

closest to the field measured surface temperatures, further Surface and air temperatures (°C) along the mobile traverse route, with
surface temperature derived at two resolutions.
reference to Ts will imply the EM image, unless otherwise
stated. a. Urban (Kowloon) Mean SD Min Max Range
Air temperature along traverse 18.7 0.5 17.9 19.8 1.9
6.2. Observed temperature relationships in the study area Surface temperature 10 m 19.7 0.8 16 21.2 5.2
along traverse
Surface temperature 90 m 19.6 1.2 15.1 21.3 6.2
Surface temperatures at the 18 field data points were still along traverse
ca. 2 °C higher than the adjacent air temperatures at the Urban surface temperature 16.6 2.3 7.8 21.3 13.5
image time 10.42 pm, suggesting incomplete UHI develop- (whole image) 10 m
ment. Nevertheless, air temperatures from the vehicle
b. Rural (New Territories) Mean SD Min Max Range
traverse (Fig. 3) indicate a well-developed heat island of
Air temperature along traverse 12.9 1.12 10.4 15.0 4.6
over 9 °C with a range of 10.4–19.8 °C between the core urban
Surface temperature 10 m 15.2 1.3 10.8 19 8.2
area and the most rural sites. However within the urban area along traverse
along the vehicle traverse, only small air temperature Surface temperature 90 m 15.8 1.6 11.3 18.4 7.1
variations (between ca. 18 and 20 °C) were observed, and along traverse
these corresponded to significantly greater image Ts varia- Rural surface temperature 13.6 1.4 7.3 18.4 11.1
(whole image) 10 m
tions of ca. 16–21 °C (Table 2a). Over the whole image, Ts
values indicate a much greater range, with an SUHI of 14 °C c. All Mean SD Min Max Range
and with lowest Ts values of ca. 7 °C corresponding to barren Air temperature along traverse 15.1 2.5 10.4 19.8 9.4
areas near sea level. Surface temperature 10 m 16.7 2.2 10.8 21.2 10.4
An increase of approximately 1 °C from sea level to 600 m along traverse
Surface temperature 90 m 17.4 2.1 10.8 21.3 10.5
was observed along the traverse, which corresponds with a
along traverse
temperature inversion recorded by the Hong Kong Observa- Surface temperature 7.3 21.3 14.0
tory at 8 pm. However, it is remarkable that both the mobile whole image
traverse (Ta) and image data (Ts) indicate much higher air
and surface temperatures in Kowloon (ca. 4 °C higher) than in
the smaller towns of Yuen Long, Shatin and Tai Po, with
populations of approximately one quarter that of Kowloon's 6.3. Scales of Surface Heat Island (SUHI)
roughly 2 million. For example the lowest Ta value in Kowloon
is 17.5 °C, which is higher than the highest Ta value (16.3 °C) Fig. 4 represents the general distribution of the SUHI over
in Yuen Long. This is surprising since the core areas of these an area of ca. 40 km2.of the Kowloon Peninsula), and
towns are compact and densely built, with similar land cover represents an SUHI magnitude of 7–8 °C. The core of the
types to Kowloon. Also, within Kowloon, high-rise residential SUHI is aligned north–south, corresponding to medium rise
developments on the urban periphery appear cooler than and very high density commercial districts along the main
similar high-rise developments elsewhere (H on Fig. 4). shopping street, Nathan Road, with a few smaller outliers

Fig. 3. Comparison of Ta from mobile traverse from two vehicles across urban and rural areas, with image-derived Ts. Breaks in the data represent driving through
tunnels. The combined distance is 148 km, with 14,814 air temperature points corresponding to the nearest image pixel.
 J.E. Nichol et al. / Atmospheric Research 94 (2009) 276–284 281

Fig. 4. General distribution of the SUHI over Kowloon Peninsula, showing a north–south alignment of the heat island core along Nathan Road, the main commercial
and shopping district.

of similar structure and function to the north and east. ture is available along the 148 km route of the mobile air
Interruption of the SUHI by cooler parks, and lower density temperature traverse (Figs. 1 and 3, and Table 2). Over the
residential districts is also evident. At this meso-scale, the whole traverse a high R2 value of 0.80 between image Ts and
SUHI forms a coherent pattern, with a progressive and mobile Ta was observed, and visually the two datasets trend
gradual transition from cooler peripheral areas to highest in the same direction, with apparent positive covariance at a
temperatures at the dense urban core, with the appearance of general level. Although the amplitude of variations for Ts is
an isothermal map. only slightly greater than for Ta, the frequency of Ts variation
Fig. 5a shows micro-scale patterns of Ts within a 0.8 km2 is much greater. Furthermore, if the Ts/Ta relationship is
area where more specific detail corresponding to individual examined for urban and rural areas separately the correla-
large buildings, sports pitches (S on Fig. 5a) or small belts of tions are lower, though still significant and weakly significant,
trees (T) can be observed. Such detail is not evident on the with R2 = 0.42 and R2 = 0.09, considering n N 6000 for each of
90 m resolution images (Fig. 5c). The 10 m pixel size used in urban and rural areas respectively. This is partly due to the
this study is able to indicate a close correspondence between reduced absolute range of the data values within urban and
image Ts and the surface characteristics (Fig. 5a) with open rural areas individually, making the micro-scale variations in
spaces (sports pitches (S)) at the top of the extract being the Ts more pronounced compared with the meso-scale varia-
coolest areas, followed by belts of tree vegetation such as tions in Ta. Although the 90 m Ts image is of lower frequency
trees (T) the next coolest. The meso-scale thermal pattern than the 10 m image its amplitude is actually greater, and due
referred to above is less evident, but can be observed, with the to lower spatial correspondence in the derivation of Ts values,
edge of the SUHI core at bottom left, giving way to more open lower correlations with Ta are obtained (Table 3).
suburban areas in centre and right. Thus both micro- and The high frequency Ts variations along the mobile traverse
meso-scales of SUHI, having steep and shallow temperature route (Fig. 3) correspond to the micro-scale variations
gradients respectively, can be observed in such complex areas. observed in the detailed image (Fig. 5a), whereas the lower
frequency trends in Ts can be seen in the meso-scale
6.4. Relationship between Ts and Ta isothermal pattern of Fig. 4, with less steep temperature
gradients. This meso-scale pattern seen on both the traverse
The high correlation of R2 = 0.81 between field measured (Fig. 3) and image (Fig. 4) depicts a gradual rise in
surface and air temperatures at the 18 ‘in situ’ data points temperature toward the urban core areas i.e. the classical
(row 5, Table 1) suggests that the image-derived Ts may also heat island situation. The correspondence between these
show a strong correlation with air temperature at these points micro-scale surface patterns and meso-scale trends in Ta is
and in fact an R2 of 0.81 is observed. However, a more illustrated by Fig. 6, which represents a 2.5 km section of the
comprehensive comparison of the surface and air tempera- mobile traverse along Hong Kong's main shopping street,
282 J.E. Nichol et al. / Atmospheric Research 94 (2009) 276–284

Nathan Road. Surface temperatures rise from 18 °C (green- Table 3

yellow) at left, where the route adjoins a large park, to above Correlations (R2) between Ts and Ta along mobile traverse route.

21 °C (white) at the intensely built and congested shopping Urban, Rural, Whole route,
district, Mongkok. Although Ta is lower than Ts, the trend n = 8714 n = 6100 n = 14,814
from cooler to warmer is similar, and class intervals are of Ts 10 m image with Ta 0.42 0.09 0.80
similar magnitude. A sudden fall in Ta at an intersection (T on Ts 90 m image with Ta 0.26 0.05 0.62
Fig. 6) corresponds to a fall in Ts (V) where sea breezes are NB. All values are significant at 1% level.

admitted by a street intersection open to the west coast (top

of image). However, in the central section a fall in Ts over a
small localised area corresponding to a treed park (P) is not
matched by a fall in Ta.

7. Discussion

A strong predictive power of the satellite Ts for urban heat

island analysis is suggested by the study. Both the satellite Ts
and air temperatures correspond to a classical heat island
situation with a ‘cliff-like’ temperature gradient at the urban–
rural boundary, and a gradual rise in temperatures toward the
urban core (Oke, 1976; Goldreich, 1985). The fact that the Ts/
Ta correlation is higher for urban and rural areas combined
than for each individually due to the high frequency Ts
variations being submerged in the meso-scale trends,
indicates two distinct scales of temperature variation. Thus
small green areas and open spaces such as the small tree park
(P on Fig. 6) which can be seen as high frequency cool areas
on the image are not matched by a corresponding fall in air
temperatures, resulting in only moderate correlations
between Ts and Ta. This problem indicates that the mismatch
referred to earlier, between satellite-derived Ts from low
spatial resolution (e.g. 1 km) sensors and Ta at fixed ground
stations (Stoll and Brazel, 1992; Gallo et al., 1993) may be
attributable not to deficiencies in image resolution, but to a
distinct difference in scale of variation of surface and air
temperatures. Indeed the present study suggests that an
image resolution of 10 m may not be necessary, and even the
90 m resolution of ASTER thermal images is much finer than
the apparent variations in air temperature which characterise
the UHI (see also Schmid et al., 1991; Stoll and Brazel, 1992;
Kawashima et al., 2000; Nichol, 2009). The higher detail
afforded by the 10 m EM image however, has the advantage of
greater accuracy in surface temperature retrieval (Table 2b)
and higher correlations with air temperature (Table 3). It
additionally permits a more complete inventory of radiative
entities forming the canopy layer heat island (Fig. 5) albeit
still incomplete, since vertical surfaces not ‘seen’ by the
satellite, are absent (Voogt and Oke, 2003). This higher detail
on the 10 m EM image indicates, for example, that open areas
such as sports pitches (S on Fig. 5a) are the coolest areas at
night, and even cooler than treed areas and waterbodies. This
supports the findings of Spronken-Smith and Oke (1998)
using a helicopter-borne thermal scanner with 2 m resolu-
tion, that treed parks may not be particularly cool.
The larger spatial scale of air temperature variation is
attributed to microadvection from adjacent land uses driven
by turbulent fluxes at the surface, since under the meteor-
ological conditions of the study (low winds of 1 ms, and
Fig. 5. The SUHI at detailed level showing steep temperature gradients night-time temperature inversion) regional scale advection
between land cover types (legend as for Fig. 4). would be low. Temperature gradients in both Ts and Ta can be
 J.E. Nichol et al. / Atmospheric Research 94 (2009) 276–284 283

Fig. 6. Mobile traverse (thick black line on image) from south to north along Kowloon's main shopping street Nathan Rd, showing lower temperatures alongside
Kowloon park at southern end (left centre, immediately above traverse), rising to higher temperatures northwards (right) at heat island core. The graph includes
five colour classes of Ts, which can be related to the class boundaries in the image.

related to land cover changes, and give little evidence of large New Territories by a mountain range of approximately 500 m
scale advection, except where a locally strong low level flow is elevation, coupled with the observed inversion at ca. 600 m
produced along ventilation corridors, as at the intersection and low wind speeds. These suggest that at the image time a
shown on Fig. 6, with a fall of 2 °C and 0.5 °C for Ts and Ta strong warm urban boundary layer (UBL) heat island (Oke,
respectively. 1976) had developed over Kowloon. Warm air would be
The close relationship between Ts and Ta and their unable to penetrate north of the mountains, especially since
observed correspondence with land cover appears to support the wind at the image time was light (1 m/s), north–north
the population/physical structure model of UHI causation easterly. Likewise, cold air drainage from the rural New
rather than the advection model (Oke, 1976). However, higher Territories would be blocked by the mountain range coupled
temperatures near the urban centre, as well as the much with the inversion. Thus the regional climate of Kowloon
higher surface and air temperatures observed in Kowloon would be dominated by separate boundary layer conditions,
than in four smaller urban centres with similar built structure and its canopy layer heat island would be reinforced by
in the New Territories, would seem more indicative of city size heating from both above and below. If this theory is correct,
in terms of its areal extent and distance of fetch from the rural the advection model is also a surrogate for city size, but in this
boundary, thus supporting the advection model. The higher case measured by spatial extent (not structure, as with the
temperatures in the largest urban area of Kowloon may be population/physical structure model). Thus larger urban
explained by the separation of Kowloon from the largely rural areas are less cooled by ventilation because the distance of
284 J.E. Nichol et al. / Atmospheric Research 94 (2009) 276–284

fetch from the urban–rural boundary is greater, thereby References

contributing to a stronger UBL. It was observed in Goteborg
(Haeger-Eugensson and Holmer, 1999) that city centre cool- Arnfield, A.J., 2003. Two decades of urban climate research: a review of
ing by an Urban Heat Island Circulation (UHIC) consisting of turbulence, exchanges of energy and water, and the urban heat island.
International Journal of Climatology 23, 1–26.
advective flow along valleys, sets in following the attainment Balling, R.C., Brazel, S.W., 1988. High resolution surface temperature patterns
of maximum ΔT(u − r), and when ΔT(u − r) reaches 2 °C. In in complex urban terrain. Photogrammetric Engineering and Remote
the present study, although ΔT(u − r) exceeded 2 °C, it was Sensing 54, 1289–1293.
Balling, R.C. and Brazel, S.W. (1989), High-resolution night-time temperature
only 80% developed (Fig. 2) and additionally any potential
patterns in Phoenix, Journal of the Arizona–Nevada Academy of Science,
UHIC from cool rural areas to the north of Kowloon (Fig. 1) 23, 75–81. International Journal of Climatology 23, 1–26.
would be blocked by mountains higher than the inversion. Eliasson, I., 1992. Infrared thermography and urban temperature patterns.
The observed ventilation corridor from the coast (Fig. 6) may International Journal of Remote Sensing 13, 869–879.
Ersdac, SNM, 2005. ASTER Users Guide, Part 1 General, Version 4. http://
offer only weak UHIC potential due to the relative warmth of www.science.aster.ersdac.or.jp/en/documnts/users_guide/part1/pdf/
the sea on a winter night. Part1_4E.pdf, last accessed, August, 2007.
The study suggests that both the physical structure and Gallo, K.P., McNab, A.L., Karl, T.R., Brown, J.F., Hood, J.J., Tarpley, J.D., 1993. The
use of NOAA AVHRR data for assessment of the urban heat island effect.
advection models are relevant in the measurement of heat Applied Meterorology 32, 899–908.
island intensity, and demonstrate the usefulness of remotely Gillespie, A.R., Rokugawa, S., Matsunaga, S., Hook, T., Kahle, A.B., 1998. A
sensed UHIs in combination with synchronous ground data temperature and emissivity separation algorithm for advanced space-
borne thermal emission and reflection radiometer (ASTER) images. IEEE
for UHI diagnosis. For example, resulting from the present Transactions on Geoscience and Remote Sensing 36, 1113–1126.
study, planning strategies for Hong Kong may include the Gillespie, A.R., Rokugawa, S., Hook, S., Matsunaga, T., Kahle, A.B., 1999.
following. Development of peripheral areas by harbour Temperature/emissivity separation algorithm theoretical basis document,
version 2.1. NASA/GSFC, Greenbelt, MD.
reclamation is to be discouraged since the increased distance Goldreich, V., 1985. The structure of the ground level heat island in a Central
from the urban boundary to the core may increase the overall Business District. Journal of Climatology and Applied Meteorology 2,
ΔT(u − r) of the city, increase the extent of the heat island 1237–1244.
Haeger-Eugensson, Holmer, 1999. Advection caused by the urban heat island
core, and prolong the duration of critically hot periods.
circulation as a regulating factor on the nocturnal urban heat island.
Building at lower density to increase ventilation and the SVF, Journal of Climatology 19, 975–988.
even at the expense of constructing taller buildings, would be Kawashima, S., Ishida, T., Minomura, Miwa, M., 2000. Relations between
an effective measure, and high-rise buildings with wide surface temperature and air temperature on a local scale during winter
nights. Journal of Applied Meteorology 19, 1570–1579.
spacing on the periphery as well as in central areas could New York State Energy Research and Development Authority (NYSERDA),
accommodate equal activities and floor area. Such measures 2006. Mitigating New York City's heat island with urban forestry, living
should also produce beneficial feedback effects to reduce the roofs and light surfaces. Report 06-06. 173 pp.
Nichol, J.E., 1994. A GIS based approach to microclimate monitoring in
boundary layer heat island. Singapore's high rise housing estates. Photogrammetric Engineering &
The high resolution depiction of a well-developed urban Remote Sensing, U.S.A 60, 1225–1232.
heat island over a whole city, surrounding rural areas and Nichol, J.E.,1996a. High resolution surface temperature patterns related to urban
morphology in a tropical city: a satellite-based study. Journal of Applied
small towns at a single point in time is able to suggest that Meteorology 35, 135–146.
both a city's morphological structure as well as the influence Nichol, J.E., 1996b. Analysis of the urban thermal environment of Singapore
of its areal extent on the overlying boundary layer are major using LANDSAT data. Environment & Planning B: Planning & Design 23,
733–747.
determinants of the canopy heat island. More cloud-free Nichol, J.E., 2009. An emissivity modulation method for spatial enhancement
thermal images are required to confirm such observations of thermal satellite images in urban heat island analysis. Photogram-
under different climatic conditions and for different seasons, metric Engineering and Remote Sensing.
Oke, T.R., 1976. The distinction between canopy and boundary-layer heat
since New York State Energy Research and Development
islands. Atmosphere 14, 268–277.
Authority (NYSERDA) (2006) as well as current work in Hong Roth, M., Oke, T.R., Emery, W.J., 1989. Satellite derived urban heat islands from
Kong, suggest similar patterns within any one season. three coastal cities and the utilisation of such data in urban climatology.
International Journal of Remote Sensing 10, 1699–1720.
Sabins Jr., F.F., 1997. Remote sensing principles and interpretation. W.H.
Freeman and Co, New York. 404 pp.
Acknowledgement Schmid, H.P., Cleugh, H.A., Grimmond, C.S.B., Oke, T.R., 1991. Spatial variability
of energy fluxes in suburban terrain. Boundary Layer Meteorology 54,
The work was funded by CERG projects B-Q01Q and B- 249–276.
Spronken-Smith, R.A., Oke, T.R., 1998. The thermal regime of urban parks in
Q13A. two cities with different summer climates. International Journal of
Remote Sensing 19, 2085–2104.
Stoll, M.J., Brazel, A.J., 1992. Surface–air temperature relationships in the
Appendix A. Supplementary data urban environment of Phoenix. Arizona, Physical Geography 13, 160–179.
Summers, P.W. (1964), An urban ventilation model applied to Montreal.
Unpublished PhD thesis, McGill University, Montreal.
Supplementary data associated with this article can be found, Voogt, J.A., Oke, T.R., 2003. Thermal remote sensing of urban climates.
in the on line version, at doi:10.1016/j.atmosres.2009.06.011. Remote Sensing of Environment 86, 370–384.