5a. The data fall onto lines of near constant angular difference because of the ERS -1 orbit 
and scan geometry of the ATSR. Considerable variation in temperature difference (both at 11 
/rm and 12 fxm) is apparent at each of these angular differences. These variations are due to 
atmospheric variations. There are some points at each of the lines of constant angular difference 
that have quite large temperature differences: these all occur during the daytime (see Fig. 
5c). There is only a slight indication that the temperature difference increases with angular 
difference. Fig. 5b shows the variation of temperature difference with surface temperature 
(in situ ) for three different combinations of data. The triangles indicate differences obtained 
from the nadir split-window, and are similar to results obtainable from the AVHRR/ 2 . The 
solid circles and squares show the angular temperature difference and angular/split-window 
temperature difference respectively. In both cases the temperature differences are larger than 
those obtained form the split-window only. The extra temperature difference observed is due to 
a combination of the atmospheric and emissivity effects; although we will argue that the former 
effect is much greater than the latter. Fig. 5d shows more clearly the greater temperature 
difference observed due to angle over that observed due to spectral (split-window) effects. 
By using the emissivity model (11) with an assumed, but constant emissivity at nadir, (12) 
has been used to derive LSTs for the 31 “clear-sky” coincidences of the ATSR and in situ 
data. Constant values of i7=1.5 g cm -2 , &n=0.12 cm 2 g _1 , &i 2 =0.20 cm 2 g _1 , e 1 i(0)=0.962 and 
ei2(0)=0.964 were used throughout. The result is very good with a bias of 0.02 °C and an rms 
difference of 1.20 °C. 
To investigate the effects of angular variations, (12) was also used without any variation 
of emissivity with angle and with transmittances of Tn( 0 i) = exp{ — kuU} and t 12 ( 0 2 ) = 
exp { — 1.743A: 12 C/}. When the angular variation of emissivity is absent, Ae=0 and the trans 
mittance variation also has no effect. The results are presented in Table 1, in which biases 
and rms differences are shown for the three estimates. These results show that the angular 
Table 1: Biases (surface temperature-satellite estimate) and rms differ 
ences for three different LST algorithms. Algorithm A is equation (12). 
Algorithm B is (12) with Ae=0 and e,\( 0 )=e.\(O). Algorithm C is (12) 
with constant values for the transmittances. 
Alogorithm bias (°C) 
rms (°C) 
11 fim 
channel 
A 
+ 0.02 
1.20 
B 
-0.60 
1.21 
C 
- 0.20 
1.20 
12 urn 
channel 
A 
+0.04 
1.40 
B 
-0.40 
1.41 
C 
- 0.02 
1.41 
variation of emissivity is an important source of error (bias) for LST determination. The actual 
effect is not too large however, and this is probably due to the fact that the surface has a cover 
of growing vegetation. The impact of the transmittance is much less because, for dual angle 
algorithms in low water vapour conditions (U <3.0 g cm -2 ) the angular variation is very pr e ' 
dictable (it depends simply on geometry) and is accounted for by the factor 7 $ in the algorithms. 
3.3 TIMS 
The TIMS instrument is a 6 channel infrared scanning radiometer with two on-board calibra-