Due to the aforementioned advantages of thermal remote sensing (see Section 2.4.1) compared to terrestrial measurements (spatial-completeness, homoge- neous coverage and repeatability (Miller and Small, 2003)), thermal Earth observed data have been used by a number of studies to quantify intra-urban variations in temperature in a spatially complete manner, that has not been attainable using weather stations (Matson et al., 1978; Dousset, 1989; Roth et al., 1989). Roth et al. (1989) used eight daytime and three night-time geo- rectified scenes from the Advanced Very High Resolution Radiometer (AVHRR) thermal band 4 (10.5-11.5μm) to examine the urban heat islands of three coastal cities of Western North America (Vancouver, Seattle and Los Ange- les). The 11 scenes were captured between the 29th November 1983 and 14th
February 1986, during atmospherically stable, cloud-free anti-cyclonic condi- tions. Of the 11 scenes only two (both daytime) covered Seattle and Los Ange- les, with the remaining nine scenes covering Vancouver, the primary focus of the study. Due to the requirements for information of atmosphere content (e.g. water vapour) and emissivity, Roth et al. (1989) did not perform atmospheric or emissivity corrections on the scenes and instead used the at-sensor bright- ness temperatures to quantitatively examine urban temperatures (Roth et al., 1989). Despite the limited number of scenes (potentially reducing the relia- bility of results), the study showed that in all scenes land use (derived from terrestrial mapping) appeared to correspond with urban temperatures. For example, the warmest locations in the images were found in areas of dense ur- ban or industrial development, although no quantitative correlation analysis between land-use and intra-urban temperatures was conducted (Roth et al., 1989).
Later analysis by Dousset and Gourmelon (2003) agreed with the initial find- ings of Roth et al. (1989) and showed that the industrial areas of Los Angeles were up to 7°C warmer than surrounding rural hinterlands. These findings were based on ESTs derived from 85 AVHRR scenes spanning the months July to August from 1984 and 1985, grouped by overpass time, and tempo- rally averaged over the time-series. Figure 2.8 (from Dousset and Gourmelon (2003)) shows the temporal averages of EST at selected locations for each of the local times of AVHRR overpass (the number of scenes used to create each average at the time of overpass is not presented in the text). Within the figure the diurnal cycle can clearly be seen, including temperatures in the city centre (“downtown”) exhibiting a > 20°C increase in the afternoon (14:50) compared to the early morning (04:25) (Dousset and Gourmelon, 2003). Rural areas outside of the city (“Chino fields”) exhibit a similar range of tempera- tures through the diurnal cycle and which are approximately 5°C cooler than urban areas (Figure 2.8).
Using a land cover classification derived from a SPOT-HRV image (Système Probatoire d’Observation de la Terre - high resolution visible) Dousset and Gourmelon (2003) showed that within Los Angeles large urban green areas
Figure 2.8: EST for Los Angeles and surrounding areas derived from a tempo- ral averages of 84 AVHRR scenes captured between July-August, 1984-1985 (source: Dousset and Gourmelon (2003)).
(total surface area not provided in text) created cool-islands ranging between 2.2-5.0°C compared to surrounding urban temperatures. However, no direct relationship between green area size and temperatures were derived in the study for Los Angeles.
Dousset and Gourmelon (2003) used a further 22 AVHRR scenes, processed to EST to examine intra-urban temperatures in Paris, and in particular to quan- tify the relationships between intra-urban temperatures, building density and vegetation cover. Such information is important to understand how dif- ferences in surface properties can influence EST so that future climate/heat- wave adaptation and mitigation strategies can account for the influence of ur- ban surface properties. Dousset and Gourmelon (2003) derived six land cover classes for the Paris area using a SPOT HRV image with unsupervised clas- sification; water, urban, densely built, sub-urban residential, light bare soil, and vegetation. Using the six classes a percentage built-up density within each AVHRR pixel (SPOT resolution is 20m) was then generated and used to investigate the correlation between urban built density and EST.
Figure 2.9 is an extract from Dousset and Gourmelon (2003) showing a cor- relation plot for building density and day/night ESTs derived from AVHRR scenes. No measure of correlation was derived, although based on the re- sults in Figure 2.9 , Dousset and Gourmelon (2003) state that “The nighttime distribution of LST [land surface temperature] is well correlated with increas- ing density of buildings from the suburbs to downtown...”. Furthermore, the study states that daytime temperatures are also correlated, but with higher variations due to large fluxes in temperatures during the daytime (as seen with the diurnal representation of temperatures in Los Angeles above). How- ever, without a measure of correlation, based on the results shown in Figure 2.9, the interpretation of these results by Dousset and Gourmelon (2003) is doubtful, and further work is required to prove that a statistically significant relationship exists.
Figure 2.9: Joint distribution of percentage of built-up density for Paris and surrounding area with average night-time and daytime AVHRR EST values (source: Dousset and Gourmelon (2003)).
In an attempt to better understand the broad-scale relationships shown by Roth et al. (1989) and Dousset and Gourmelon (2003), thermal satellite sen- sors with higher spatial resolution (<500m) have been used to examine the micro-climatic effects of urban morphology (e.g. Nichol, 2005). Nichol (2005) used a daytime Landsat ETM+ scene (60m spatial resolution) and a night- time ASTER scene (90m spatial resolution), both processed to EST, to exam- ine micro-climatic temperature variations across the Western New Territo- ries Region of Hong Kong (Nichol, 2005). The daytime EST from the Landsat ETM+ scene showed variations of up to 8°C across the urban area. Using an- cillary land use classification data including building and road outlines and an aerial image, the study showed that daytime ESTs appeared to be related to sky view factor (SVF). The study showed that areas with low SVF (SVF val- ues not presented in text) had lower temperatures (~35°C) than areas with high SVF (~37-38°C) (Nichol, 2005). Furthermore, Nichol (2005) showed that locations shaded by buildings at the time of overpass were up to 4°C cooler
than adjacent non-shaded areas.
In contrast to the daytime EST values in Nichol (2005), the night-time ESTs from the ASTER scene overpass (21:40) showed almost complete uniformity across the urban area of the Western New Territories Region, with intra- urban temperature variations of less than 2°C. Nichol (2005) states that the marked difference in intra-urban temperature variation between day EST (Landsat) and night-time EST (ASTER) indicate that micro-climatic varia- tions are more pronounced during the day. This suggests that urban morpho- logically derived variation in temperatures is driven primarily by exposure of the urban surface to solar input during the day. Further, Nichol (2005) hy- pothesis that an ASTER scene captured later in the night would have shown greater variations in temperature, as urban surfaces cool at different rates de- pending on their thermal capacity and exposure to solar infrared and thermal radiation during the day.
Nichol (2005) also examined the influence of the sea on cooling of the urban areas, which has previously been found to be a key driver of intra-urban tem- peratures (Roth et al., 1989; Dousset, 1989; Dousset and Gourmelon, 2003; Eliasson and Svensson, 2003), in terms of air temperatures (Eliasson and Svensson, 2003) and satellite Earth observed ESTs (Roth et al., 1989; Dous- set, 1989; Dousset and Gourmelon, 2003). However, in contrast to previous studies Nichol (2005) found that the proximity to the sea appeared to have no influence on night time EST, despite temperature differences between sea EST and coastal ESTs of up to 10°C (derived from the night-time ASTER scene). Equally, day-time ESTs from the Landsat ETM+ scene showed little variation between sea and land EST (<3°C).
In addition to studies investigating the influence of urban built density on intra-urban temperatures (e.g. Nichol (1996); Dousset and Gourmelon (2003); Nichol (2005)), urban green areas are often cited as exhibiting lower tempera- tures than surrounding areas. For example Nichol (2005) found daytime ESTs up to 8°C cooler in large urban green spaces (e.g. parks) than the surround- ing urban areas. Similarly, Owen et al. (1998) and Dousset and Gourmelon
(2003) both found strong negative correlations between vegetation and EST in Pennsylvania and Paris respectively. Owen et al. (1998) showed that in EST pixels which were at least 25% urbanised, the vegetation-EST relation- ship was significant at the 95% confidence level (results of statistical testing not presented in the text). Owen et al. (1998) hypothesised that this relation- ship was the result of latent heat loss via evapotranspiration from vegetation, creating a cooling effect. However, Nichol (1994) and Nichol (2005) note that strong correlations between EST and vegetation are common. In studies cov- ering both Singapore and Hong Kong (Nichol, 1994, 2005) using Landsat TM and ETM+ data, they found no meso-scale advective influence from vegeta- tion on neighbouring ESTs. These results suggest that whilst urban green spaces are clearly cooler, they have limited influence on surrounding urban surfaces (Nichol, 1994, 2005), potentially limiting the utility of urban green- ing programs such as those currently employed in Singapore (Nichol, 1994) and New Jersey (Solecki et al., 2005).