Home > Geospatial Application Papers > Environment > Climate Change




A comparison of daytime and night-time thermal satellite images of Hong Kong for urban climate studies


Due to the different types of thermal image data available (Table 1), different processing techniques were applied for derivation of the required parameter; Surface (or Kinetic) Temperature (Ts).

Table 1. Image details and climatic data
  LANDSAT ETM+ ASTER
DATE 17.09.01 06.10.01
TIME 09.15am 21.40pm
SPATIAL RESOLUTION (THERMAL) 60m 90m
RADIOMETRIC RESOLUTION (K) 0.5 0.3
WAVEBANDS (mm) 10.4-12.5 (band 6:1) 8.125-8.457 (band10)
8.457-8.825 (band 11)
8.925-9.275 (band 12)
10.25-10.95 (band 13)
10.95-11.65 (band 14)
IMAGE DATA AVAILABLE Digital number (DN) Kinetic Temperature (Ts)
(based on ratios of 5 thermal bands)
MEAN 24-HOUR AIR TEMPERATURE ON IMAGE DATE (C) 29 27
MEAN SEA SURFACE TEMP. (C) 29 27
SUN ELEVATION(), AZIMUTH() 58, 127 0, 0
MEAN WIND SPEED (kmph), DIRECTION 2.9, 230 32.6, 80

  1. LANDSAT ETM+


  2. The low gain thermal band (band 6:1) of ETM+ was converted to Ts in four steps, as follows:

    1. Digital Number (DN) to Spectral Radiance (Ll)


    2. Conversion of the image DN values to spectral radiance is carried out using the gain and offset values given in the image header file (Eq. (i)). Thus

      Ll = ((LMAX-LMIN)/(QCALMAX-QCALMIN)) * (QCAL-QCALMIN) + LMIN -------------(i)

      Where:
      QCALMIN = 1, QCALMAX = 255, and QCAL = Digital Number
      LMIN and LMAX = spectral radiance for band 6:1 at DN 0 and 255

    3. Spectral Radiance to Black Body Temperature


    4. The ETM+ thermal band data can be converted from spectral radiance Black Body Temperature (BBT) which assumes surface emissivity=1 (Eq. (ii))

      T = K2 / ln(K1/ Ll+ 1) -------------------(ii)

      Where:
      T = Effective at-satellite temperature in Kelvin
      K1 = Calibration constant 1 (W.m2.sr-1)(666.09)
      K2 = Calibration constant 2 in K (1282.7)
      Ll = Spectral radiance in (W.m2.sr-1)

    5. Emissivity Correction


    6. The visible wavelength bands of ETM+ image were classified into three main land cover classes: vegetation, non-vegetation and water using a supervised classification. Corrections for emissivity differences were carried out by land cover type by ratioing the BBT image with the classified image in which the pixel values for the land cover class were replaced with the corresponding emissivity value. Thus the emissivity corrected surface temperature (Ts) is derived by Equation (iii). (Artis and Carnahan 1982).

      Ts = T / [1 + (lT / a) ln e] ---------------------(iii)

      l = Wavelength of emitted radiance,
      a = hc/K (1.438 10-2 mK),
      h = Planck's constant (6.26 10-34 J.sec),
      c = velocity of light (2.998 108 m/sec),
      K = Stefan Bolzmann's Constant (1.38 10-23 J/K).


      LANDSAT daytime image (Figure 3)

      * High rise areas are relatively cool ("37C) if buildings are closely spaced (Figure 5)
      * Densely-built low rise areas are relatively warm ("39-40) (Figure 6)
      * Car parks and open spaces are the hottest areas (Tuen Mun bus station forecourt "45C)
      * Aspect of buildings in relation to sun elevation and azimuth has significant influence on Ts (Figure 7)

      The emissivity correction procedure has the effect of increasing the spatial resolution of the thermal data due to fusion with the land cover map based on the 30 metre pixels of ETM+ visible wavebands.

    7. Atmospheric correction


    8. The Ts image was atmospherically corrected by comparison with Sea Surface Temperature data from the Hong Kong Observatory, which appeared to be stable both spatially and temporally over several days (at 29C) at the image date. The addition of 10 DN values before conversion to radiance had the effect of raising Ts values by 5C, resulting in Ts values for the sea surface, of 29C, and all other image values were adjusted accordingly. The above procedures resulted in the majority of image Ts values between 34 and 45C. Since air temperature at the image time (Figure 2) was 28.7C and had been rising since 7am, and surface heating usually precedes increases in air temperature, usually exceeding it by several degrees, then these values appear acceptable.


      Figure 2. Air temperatures for 24-hours on image dates (source: Hong Kong Observatory)

  3. Aster


  4. An ASTER image of Kinetic Temperature in C (equivalent to Surface Temperature) was available. This image is a higher level product from EDC, derived from data collected in the five ASTER thermal wavebands (Table 1) and corrected for emissivity and atmospheric effects using image ratios and external data (Gillespie et al., 1996; Schmugge et al., 1995). Most image values for the study area ranged from 22-28C, and this compares well with Sea Surface Temperature in-situ data of 27C at Waglan Island (image value also 27C at this point) and air temperature at the image time (9.40pm), of 26.6C, which had been falling since 4pm (Figure 2). Such a situation, with Ts values on land lower than air and sea surface temperatures, is expected, especially on cloud-free nights due to long-wave radiant heat flux. The images were not de-striped, as the difference between adjacent scan lines was less than 0.3for this 16-bit data.


    Figure 3. ETM+ image of Surface Temperature (Ts) with building outlines.

  5. Geometric correction, Overlay and Visualisation


  6. The ASTER and LANDSAT Ts images were geometrically corrected to the Hong Kong 1980 Grid using the polynomial method and approximately eight ground control points resulting in accuracies of better than 0.3 pixel. Image Ts data were overlaid with road and building outlines in vector format (Figures 3 and 4) in order to examine micro- and meso-scale climatic patterns in and surrounding the two urban areas studied, and to compare the day and night situation.
Page 2 of 3
| Previous | Next |