Landscape and Urban Planning 73 (2005) 49–58
Modeling urban environmental quality in a tropical city Janet Nichol∗ , Man Sing Wong Department of Land Surveying and Geo-Informatics, The Hong Kong Polytechnic University, Hunghom, Kowloon, Hong Kong, PR China Received 20 April 2004; received in revised form 6 August 2004; accepted 9 August 2004
Abstract Environmental quality is an abstract concept resulting from both human and natural factors operating at different spatial scales. In urban areas the local scale is dominated by individual buildings, streets and trees, but regional scale influences may include the whole city and beyond. This paper demonstrates the ability of current satellite-based sensing systems to depict parameters of urban environmental quality over large areas at detailed level, using 3D Virtual Reality models. A method is described for increasing the spatial detail and spectral accuracy of Landsat ETM+ thermal data, by fusion with an IKONOS image representing vegetation. Additionally, by depicting the complete radiating surface involved in energy exchange between the surface and atmosphere, including vertical walls, as well as the horizontal surfaces ‘seen’ by the satellite, a more accurate representation of the urban thermal environment is obtained. The models permit 3D visualization and fly-through animation to represent urban environmental quality, based on quantifiable image parameters, and assist the understanding of the complex and dynamic factors controlling urban environmental quality. © 2004 Elsevier B.V. All rights reserved. Keywords: Urban environmental quality; Remote sensing; Visualisation; Urban heat island; Biomass; 3D city models
1. Introduction There is an expanding market for 3D city models for a wide range of applications including land use/cover mapping, architectural applications, disaster and emergency assessment, facility mapping, real estate business, environmental studies and simulations, and utility provision. For many applications the data requirements are very specific in terms of structural and textural detail ∗ Corresponding author. Tel.: +852 2766 5952; fax: +852 2330 2994. E-mail address:
[email protected] (J. Nichol).
0169-2046/$20.00 © 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.landurbplan.2004.08.004
(Baltsavias and Gruen, 2003), and require sensors with high spatial, spectral and temporal resolution. Although spaceborne sensors are still unable to capture 3D structural details of the built environment (Baltsavias and Gruen, 2003), the resolution convergence of airborne and spaceborne systems permits the use of satellite image data for some thematic applications if the thematic data can be effectively combined with structural details derived from other sources such as photogrammetry and digitized maps. Following the SARS epidemic, the Hong Kong government is committed to improve the quality of the urban environment. Urban environmental quality is a complex and spatially variable parameter which is
50
J. Nichol, M.S. Wong / Landscape and Urban Planning 73 (2005) 49–58
a function of interrelated factors including the Urban heat island, the distribution of greenery, building density and geometry, and air quality. Current satellite sensors are unable to capture all of this complexity due mainly to their spatial and spectral inadequacies; for example, they are unable to detect air quality due to spectral limitations. However, the close positive relationship between this parameter and the Urban heat island (Dwyer et al., 1992; Klaus et al., 1999; Environmental Protection Department, 2001), and its inverse relationship with biomass (Gallo et al., 1993; Weng, 2001), is well documented. Surface temperature derived from thermal satellite images has been used to characterize urban heat islands (Roth et al., 1989; Nichol, 1994) and fine resolution sensors such as IKONOS are able to identify land cover, including biomass, at detailed level (Nichol and Lee, in press). Whereas some satellite-based studies have demonstrated strong relationships between the Urban heat island, thermal image data and biomass (Gallo et al., 1993) these have been at a generalized level. The present study is unique in its utilisation of both the satellite-derived parameters, temperature and biomass, as independent indicators of environmental quality and their application to urban structures at micro-scale. The overall objective of this study is to demonstrate an objective, satellite-based method for visualization of urban environmental quality which can be applied at both micro- and city-wide scales. This is achieved
by combining image fusion for increased spatial detail of Landsat ETM+ thermal images, with 3D Virtual Reality models. The models are thus able to depict the complete, actively radiating surface involved in urban climatology, at a detailed level. This permits interactive visualization of Environmental Quality over the whole city, or at the scale of the individual city block or building, and enables even small pockets of neglected green space, small parks or street trees in congested areas to be identified for their high Environmental Quality
2. The study area Like many hot tropical cities Hong Kong suffers from a degree of thermal discomfort for several months of the year. Its mountainous terrain resulting in a lack of building space, combined with its location adjacent to industrial cities on the Chinese mainland are other factors influencing its living environment. Thus many of Hong Kong’s urban areas are devoid of trees and greenery due to lack of space, and are congested, noisy, and several degrees warmer than rural areas (Planning Department, 2003). Air pollutants advected from mainland cities in winter due to northerly monsoon winds (Fig. 1) are trapped among high rise buildings and street canyons of the flat coastal plain, and their dispersion is blocked by steep mountain slopes. Air
Fig. 1. Location of urban areas in Hong Kong occupying the only flat land, and the dominant wind direction in winter.
J. Nichol, M.S. Wong / Landscape and Urban Planning 73 (2005) 49–58
pollution levels at this time are usually reported as ‘high’ (Environmental Protection Department, 2001). Thus the topographic setting combined with the highest concentration of high rise buildings in Hong Kong may explain why Causeway Bay, on Hong Kong Island consistently records the highest air pollution levels.
3. Urban environmental quality Urban environmental quality is difficult to quantify as it is a function of natural factors which vary over wide spatial scales, as well as local factors of the man-made city infrastructure. For example, Weng (2001) observed a spatial scale of 120 m for variations in land surface temperature derived from thermal satellite images, and Nichol (1966) has noted microclimatic surface and air temperature patterns which vary at the scale of the individual city block, building facet or tree canopy. Therefore, satellite images whose field of view can cover a whole city can be used to monitor urban environmental quality if the image data are sufficiently detailed. This combination of regional and local factors is shown in Fig. 2, a southward view of surface temperatures over parts of the densely urbanised area of Kowloon Peninsula. The model demonstrates the combined effect of regional topography and local building geometry on surface temperatures in this summertime scene. With a south-easterly solar azimuth at the image time, the opposite, shaded north-west slope of an inselberg, which also has a double row of 40 m tall buildings creating deep urban canyons, has a surface temperature 8 ◦ C cooler than similarly unvegetated areas exposed to the sun. On the other hand, local temperature patterns around an individual building can be visualized according to the building size and orientation in relation to the sun angle at the image time. Hong Kong’s National Stadium, with open stands surrounding artificial turf is the warmest distinct feature in the image (Fig. 3). Furthermore, the Landsat thermal image data are detailed enough to show the effects of shadows cast, even by low rise features; thus on this early morning scene the area shaded by the eastern row of stands is approximately 7 ◦ C cooler than the rest of the stadium. The nearby racecourse, having real grass (as opposed to artificial turf) remains cool with an average temperature of 33 ◦ C: approximately 6 ◦ C cooler than the National Stadium.
51
4. Thermal image data 4.1. Thermal image resolution Although thermal sensors are also restricted to low resolution, exemplified by the 60 m pixel size of the Landsat ETM+ thermal band, a method originally devised for emissivity correction in the conversion of thermal image data to surface temperature (Nichol, 1994) has the effect of decreasing the pixel size through image fusion, resulting in both spectral and spatial improvements. The image fusion is performed with a ratio between the 4 m resolution IKONOS multispectral image and the 60 m pixels of the ETM+ thermal image in the emissivity correction equation (Eq. (1)). At the same time the image is corrected for differences in emissivity between vegetated and non-vegetated areas. Since this would be the main source of error in the derivation of surface temperature from thermal images, the correction is necessary. The correction, using Planck’s constant (Eq. (1)) (Artis and Carnahan, 1982) effectively fuses the thermal image with the 4 m resolution IKONOS image which has been classified into vegetated and non-vegetated areas. The 4 m pixels in this binary mask image are then allocated emissivity values according to whether they represent vegetation or non-vegetation: Ts =
Tb [1 + (λT/α) ln ε]
(1)
where Ts is the surface temperature (K), Tb the black body temperature (K), λ the wavelength of emitted radiance, α = hc/K (1.438 × 10−2 m K), h the Planck’s constant (6.626 × 10−34 J s), c the velocity of light (2.998 × 108 m/s) and K the Boltzman constant (1.38 × 10−3 J/K). The emissivity values (ε) for vegetated and nonvegetated surfaces were 0.96 and 0.92, respectively (Nichol, 1994). After this correction, although the pixel size is 4 m, this is not equivalent to increasing the spatial resolution of the thermal image from 60 to 4 m, but the spatial and spectral accuracy are enhanced to the extent that emissivity differences between vegetated and unvegetated areas affect the emitted, remotely sensed temperature. For example, vegetated and concrete surfaces having emissivities of 0.96 and 0.92, respectively, both with an actual temperature of 27 ◦ C would have very different radiant (uncorrected
52
J. Nichol, M.S. Wong / Landscape and Urban Planning 73 (2005) 49–58
Fig. 2. Thermal model with southward view over Kowloon Peninsula. (a) Model overlaid with IKONOS image: buildings on sunny and shady side of inselberg extruded to actual height, with color according to image-derived temperatures. (b) Model overlaid with image-derived surface temperature: temperatures in ◦ C. Low rise buildings on the flat coastal plain exposed to the sun are 6 ◦ C warmer than parallel rows of high rise buildings with shady canyons on the shady side of a wooded inselberg: the cooling due to the combined effect of topography, vegetation and building shadow.
J. Nichol, M.S. Wong / Landscape and Urban Planning 73 (2005) 49–58
53
Fig. 3. Model showing the temperature difference between the National Stadium, having a cover of artificial turf, and the Hong Kong racecourse with real grass cover. The image is of sufficient resolution to depict the cooler shadowed area within the stadium due to the low sun angle on this early morning image.
image) temperatures, of 24 and 20.8 ◦ C. After the correction, a double row of street trees as seen in the color air photograph (Fig. 4a), which is not detectable on the uncorrected thermal image (Fig. 4b), is visible as a cool corridor on the corrected image (Fig. 4c). This treed
street is now seen as a distinct linear feature, being approximately 5 ◦ C cooler than an untreed street which joins it obliquely. Even larger surface temperature differences due to tree cover are common in other tropical regions, and Nichol, working in Singapore
Fig. 4. The effect of image fusion for emissivity correction on the spatial and spectral properties of the Landsat ETM+ thermal waveband. (a) Color air photograph of a major road intersection, showing one tree-lined street. (b) Uncorrected/unfused thermal image of the same area. (c) Emissivity corrected (fused) image showing the treed street as a distinct cooler feature, and greater spatial detail.
54
J. Nichol, M.S. Wong / Landscape and Urban Planning 73 (2005) 49–58
(1996) recorded a mean difference, over the course of several days, of 8 ◦ C in surface temperatures between the ground below tree canopies and open ground. The corresponding mean air temperature differences were in the order of 2 ◦ C. Thus, using the method described above, the corrected thermal data are detailed enough to be related to individual buildings, streets, or even single trees. 4.2. Representing the complete urban surface Conventional satellite derived heat islands measure the temperature of only the horizontal ‘seen’ surface, which, in high rise areas, is considerably smaller than the complete surface (Voogt and Oke, 1996; Nichol, 1998; Weng, 2001). Yet all urban surfaces both natural and man-made, are active radiating surfaces and are heat sources and sinks for the adjacent atmosphere. They also play a role in providing shade, and act as barriers and funnels for fresh rural air on the one hand, or stale urban air, thus all surfaces collectively influence urban air quality, including the dynamics of the Urban heat island. The under-representation of the active surface by satellite image data is accentuated in tropical cities such as Hong Kong due to the high sun angle: thus horizontal surfaces ‘seen’ by the satellite, such as roofs and tree canopies may be significantly hotter than the mean temperature of the active surface. Thus the 3D model provides a method for compensating for the systematic error of anisotropy associated with nadir viewing of an incomplete urban surface at city scale. In high rise areas of Hong Kong, at Causeway Bay, for example, where building density is 45% and the average building height is 50 m, the active radiating surface is 2.67 times the planimetric surface ‘seen’ by the satellite. This constitutes an image error of +1.5 ◦ C for the satellite ‘seen’ surface in Causeway Bay. 4.3. Temperatures of building facets The 3D models were constructed from digital data of building outlines from the Hong Kong Lands Department using 3D Studio Max software, and ArcGIS. Building heights were estimated by multiplying the number of floors, with the 3 m height of each floor. In order to visualize the complete active surface it was necessary to add vertical surfaces as 3D facets whose temperature could ideally be varied according to the
effects of sun angle and azimuth at the image time. For horizontal surfaces, temperatures were derived by overlaying the image data with building outlines after the method of Nichol (1998). The temperature of each building was based on the weighted mean temperature of image pixels intersecting the outline. Temperatures for vertical facets were determined according to relationships between horizontal and vertical surfaces from fieldwork conducted in September 2002 at the same season and time of day as the image. A total of 82 paired readings of horizontal and vertical surfaces were obtained in differently oriented street canyons in urban Hong Kong. Thus the temperature adjustments as a departure from horizontal ground temperatures, for vertical surfaces were −2.5 ◦ C for shaded and −1 ◦ C for sunny surfaces, respectively. These were applied to the image data. These facet temperatures would vary throughout the day, and for different times of year, according to the angle at which the sun strikes the surface and the constructed models apply only to the image time. The models provide animation in Microsoft Video Interlaced format for fly-through simulation. 4.4. Building geometry as a control on urban climate The models show that where vegetation is absent, local factors of topography and building geometry in relation to sun angle and azimuth appear to be the most influential temperature control. Thus within the most densely urbanized parts of Kowloon which are almost completely unvegetated, Shamshuipo, having streets and buildings oriented NW–SE, parallel to the direction of sunlight (solar azimuth 127◦ ), is up to 6 ◦ C warmer at the image time, 9.30 a.m., than Mongkok whose streets run roughly north–south (Fig. 5). The solar azimuth would change to more southerly toward midday, and approaching noon Shamshuipo streets would become shadowed and Mongkok streets would experience a period of direct solar illumination. The model in its present form cannot represent these dynamic changes.
5. Vegetation mapping using IKONOS Vegetation is a major parameter in urban environmental quality for many reasons, including aesthetic considerations, temperature control due to evapotran-
J. Nichol, M.S. Wong / Landscape and Urban Planning 73 (2005) 49–58
55
Fig. 5. Effect of street and building geometry on surface temperature: (a) Map showing street orientations in Shamshuipo where solar azimuth is parallel to the streets, and Mongkok, where it is not. (b) Thermal model showing that surface temperature is up to 6 ◦ C higher where solar azimuth is parallel to the street direction.
spiration and shading, the filtering and recycling of pollutants and as urban wildlife habitats (Dwyer et al., 1992; Akbari et al., 1990, 2003). In paved areas with no vegetation, evapotranspiration may be zero, thus most incoming radiation is transferred to the urban atmosphere as heat. In parks and irrigated grassy areas such as the Hong Kong racecourse (Fig. 3), where latent heat loss may exceed net incoming radiation, the result is a marked cooling effect. A vegetation map was produced from the IKONOS image in order to obtain the vegetation mask used in emissivity correction. The map was derived from the chlorophyll index (a ratio of the green and red IKONOS wavebands) based on 41 sample ground quadrats representing Vegetation Density (a measure of biomass amount) regressed against the image data (Eq. (2)). An R2 value of 0.8 was obtained (Nichol and Lee, in press). Vegetation density = 1.6 × 1000 × −404.4
g r
g r
+ 34.8
(2)
where g is the IKONOS 0.52–0.6 m waveband and r the IKONOS 0.63–0.69 m waveband.
The chlorophyll index was used in preference to the more commonly used normalised difference vegetation index, which is a ratio of the near infra-red and red wavebands. The green/red ratio was found to be more highly correlated with vegetation amount, due to insufficient difference in the near infra-red band between vegetation and the light colored reflective urban surfaces which typify many tropical cities such as Hong Kong. When the derived image of vegetation density is overlaid, and compared independently with the image of surface temperature, a strong negative relationship would be expected due to shade and evapotranspiration of the tree canopy and this was observed in the present study, with an R2 of 0.82 between surface temperature and vegetation density. Due to the multiple benefits of vegetation in urban environmental quality in addition to temperature control, it is suggested that both imagederived parameters, surface temperature and vegetation density, should be present in order to designate areas of high urban environmental quality based on image data. Thus the image data can be queried using standard conditional statements applied to these two quantifiable image parameters, to depict variable levels of urban environmental quality. Pixels which do not sat-
56
J. Nichol, M.S. Wong / Landscape and Urban Planning 73 (2005) 49–58
isfy thresholds on both conditions can be isolated using standard deviation and flagged as having low urban environmental quality. These queries return areas which are: (i) warm and unvegetated, (ii) warm but vegetated, (iii) cool but unvegetated. For example, vegetation density >1.5S.D. below mean, and surface temperature >1.5S.D. below mean return pixels of type (iii). The concept is shown to be robust at detailed level, as small pockets of greenery within densely urbanized areas are depicted as having high environmental quality, being both vegetated and cool. Fig. 6 shows the result of such a query. It is a small public park of only 0.15 ha within the densely urbanized and congested streets of Mongkok. Fig. 6a is a binary image of pixels representing both low temperature and high biomass. Surrounding buildings approximately 30 m high cast shadows over street canyons at the image time, in a similar direction as those on the digital color air photograph (Fig. 6b). However, Fig. 6c shows that the dominant cooling effect is not shadow but vegetation, as the surface temperature of the park is 6–7 ◦ C cooler than surrounding areas which include shady street canyons. Such small islands of greenery are especially important for their higher environmental quality and many people use them. Thus the combination of image fusion and 3D representation promotes better understanding
of environmental relationships at detailed, as well as regional levels.
6. Topographic setting and fresh air corridors The model in Fig. 7 represents a 9.30 a.m. image, of which the inset shows the high rise business and commercial district of Causeway Bay. This area is cool relative to built-up areas of Kowloon (Fig. 5). This is due to two reasons: (i) the taller buildings in Causeway Bay create deep shady street canyons in the early morning and (ii) the high albedo of the reflective glass and tile surfaces of Causeway Bay’s modern office buildings (albedo = 0.6), compared with concrete (albedo = 0.1–0.3) in the older mixed residential and commercial districts of Mongkok and Shamshuipo in Kowloon. These factors may operate interactively to create what appears to be a ‘heat sink’ in Causeway Bay at the image time. However, the dominant factor cannot be confirmed since, although the phenomenon of a daytime heat sink has been noted for a temperate zone city in the USA (Carnahan and Larson, 1990) and a tropical city in Nigeria (Nichol, 2003), there is no research on the influence of reflective building surfaces on heat island magnitude at city scale. The east–west oriented street and building geometry on Hong Kong island parallel to the footslope contour (O in Fig. 7) reinforces the trapping of polluted air at the mountain foot by temperature inversions, as well
Fig. 6. Thermal model used to analyse the result of a conditional query on image data (i.e. areas which are both cool and vegetated). (a) Result of query showing pixels of high urban environmental quality as white, overlaid with street outlines. (b) Model with texture derived from color orthophotograph showing that the area is a small park within the densely built Mongkok district. (c) Thermal model showing the park 6–7 ◦ C cooler than surrounding areas.
J. Nichol, M.S. Wong / Landscape and Urban Planning 73 (2005) 49–58
57
Fig. 7. Thermal model of the north shore of Hong Kong island backed by mountainous terrain, showing the dominant street direction parallel to the footslope (O). The inset represents the high rise business district of Causeway Bay and indicates particular buildings representing obstructions to air flow corridors (B).
as blocking fresh air corridors to and from vegetated mountain valleys. The model facilitates visualization of this process and shows specific buildings in Causeway Bay which appear to block potential air flow corridors preventing dispersion of the urban pollution plume (Fig. 7). The two buildings marked B in Fig. 7 lie across the street at right angles to the dominant wind direction in winter, when air pollution is most severe.
7. Conclusion To take advantage of improvements in temporal, spatial and spectral resolution of satellite platforms,
there is a need for improvements in data analysis techniques especially in data storage structures, in order utilize the full dimensionality of data available from remote sensing platforms in the context of visualization. The objective is to permit automated interpolation of multitemporal satellite data into three-dimensional, and ultimately, four-dimensional models, allowing for the variable sun angle and azimuth according to seasonal and diurnal cycles e.g. linkages of database facets to image patches (Baltsavias and Gruen, 2003). Such linkages were created semi-manually in this model following the method of Nichol (1998) which used ArcCad. The relationships between topography, the Urban heat island, biomass and air pollution within
58
J. Nichol, M.S. Wong / Landscape and Urban Planning 73 (2005) 49–58
the urban boundary layer are complex and dynamic and only superficially understood. Improved objectoriented storage structures which facilitate 3D visualization in the context of each city’s unique topographic setting, would improve this understanding.
Acknowledgement The authors would like to acknowledge funding from The Hong Kong Government CERG grant no. B-Q611.
References Akbari, H., Rosenfeld, A.H., Taha, H., 1990. Summer heat islands, urban trees and white surfaces. ASHRAE Trans. 96 (1), 1381–1388. Akbari, H., Shea Rose, L., Haider, T., 2003. Analyzing the land cover of an urban environment using high resolution orthophotos. Landsc. Urban Plann. 63, 1–14. Artis, D.A., Carnahan, W.H., 1982. Survey of emissivity variability in thermography of urban areas. Rem. Sens. Environ. 12, 313–329. Baltsavias, E.P., Gruen, A., 2003. Resolution convergence: a comparison of aerial photographs, LIDAR and IKONOS for monitoring cities. In: Mesev, V. (Ed.), Remotely Sensed Cities. Taylor and Francis, London, pp. 474–482. Carnahan, W.H., Larson, C.H., 1990. An analysis of a heat sink. Rem. Sens. Environ. 33, 65–71. Dwyer, J.F., MacPherson, G.E., Schroeder, H.W., Rowntree, R.A., 1992. Assessing the benefits and costs of the urban forest. J. Arboricult. 18 (5), 227–234.
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. J. Appl. Meteorol. 32, 899–908. Environmental Protection Department, 2001. Air quality in Hong Kong 2001. The Government of Hong Kong, 22 pp. Klaus, D., Jauregui, E., Poth, A., Stein, G., Voss, M., 1999. Regular circulation structures in the tropical basin of Mexico City as a consequence of the heat island effect. Erdkunde 53, 231–243. Nichol, J.E., 1994. A GIS-based approach to microclimate monitoring in Singapore’s high rise housing estates. Photogramm. Eng. Rem. Sens. 60, 1225–1232. Nichol, J.E., 1966. High resolution surface temperature patterns related to urban morphology in a tropical city: a satellite-based study. J. Appl. Meteorol., U.S.A. 35 (1), 135–146. Nichol, J.E., 1998. Visualisation of urban surface temperatures derived from satellite images. Int. J. Remote Sens. 19 (9), 1639–1649. Nichol, J.E., 2003. Heat island studies in third world cities using GIS and remote sensing. In: Mesev, V. (Ed.), Remotely Sensed Cities. Taylor and Francis, London, pp. 243–264. Nichol, J.E., Lee, M.H., in press. Urban vegetation monitoring in Hong Kong using high resolution multispectral images. Int. J. Remote Sens. Planning Department, 2003. Hong Kong Planning Standards and Guidelines, November 2003. Government of Hong Kong. Roth, M., Oke, T.R., Emery, W.J., 1989. Satellite derived urban heat islands from three coastal cities and the utilisation of such data in urban climatology. Int. J. Remote Sens. 10 (11), 1699– 1720. Voogt, J.A., Oke, T.R., 1996. Complete surface temperatures. In: Proceedings of the 12th Conference on Biometeorology and Aerobiology. American Meteorological Society, Atlanta, pp. 438– 441. Weng, Q., 2001. A remote sensing-GIS evaluation of urban expansion and its impact on surface temperature in the Zhujiang Delta, China. Int. J. Remote Sens. 22 (10), 1999–2014.