Urban heat island (UHI) phenomenon, as an important environmental issue affecting cities, has received much attentions in the past decades. The spatial structure of UHI, however, is not easy to verify in detail because of the complexity of its mechanism and cost of observations. Many researchers have studied this issue with satellite sensed thermal images, based on the assumption that the remotely sensed surface temperature and the ground air temperature are correlative to a large extent.
This study, focusing on Tokyo Metropolitan Area, examined the statistical and spatial correlation between remotely observed brightness temperature and ground air temperature observation. Five scenes of thermal images of Landsat TM, including two daytime ones and three nighttime ones, and about 100 stationary meteorological observations have been used as data sources.
The ground air temperature extracted from the meteorological records is compared separately to the satellite-estimated surface temperature for each data set and each radius from 100 m to 1000 m around meteorological stations. As a result, the following factors are concluded.
The correlation coefficient between ground-observed air temperature and satellite derived surface temperature could reach 0.9 under good observation conditions. However, the magnitude of the coefficients is affected by weather conditions, time and date of data acquisitions as well as spatial scale of data aggregation.
Wind speed not only affects the intensity of UHI but also the correlation coefficients of surface temperature and air temperature. The weaker the wind is, the stronger the UHI is, and the larger the coefficient is between the two measurements.
The correlation coefficients between the two are weak on daytime images but strong on nocturnal images. This suggests that nocturnal thermal images prefer to capture the same structure of a surface heat island with air heat island at night. Emissivity correction of surface temperature can enhance the correlation, but the intensity of improvement is uncertain. The gap of correlations between nocturnal images and daytime images could not be closed up sufficiently with an emissivity correction.
The surface temperature from the radiative power is mainly influenced by the urban canopy layer composed by micro land features. However, the ground air temperature is also affected by the local or mesoscale urban boundary layer. This essential difference causes the correlation of surface temperature and air temperature spatial-dependently. The maximal correlation is within 600 m around a meteorological station. Therefore, the range of 400-600 m may be a reasonable distance for planners to discuss mitigation policy. This could also be a reference to decide the spatial resolution in a simulation model.
The simple linear regressive equation in this paper could not be applied to other data and other cities. However, this study successfully verified the possibility of transferring satellite derived surface temperature to air temperature. This is meaningful for researchers and planners to explore the structure of heat islands or to discuss mitigation policies in microscale. Further research is in progress to identify the correlations between air temperature and surface temperature with urban landforms and environmental factors.