788 
Table 1,-Error in the emissivity correction, 8T e (in K),for different values of and different errors in the 
emissivity estimation, Seand 8Ae. We have considered that 8T e =j (40Sef+((IbAef ]l/2. 
mi 
8e; SAe 
50 
100 
200 
300 
0.005; 0.005 
0.4 
0.6 
1.0 
1.1 
0.01; 0.005 
0.6 
0.7 
1.1 
1.2 
0.01; 0.01 
0.7 
1.1 
1.3 
1.4 
Table 2,- Error in LST, 8T, obtained from the atmospheric correction error, 8T a , and the emissivity correction, 
8l' e . We have considered that 81= [(81' a f +(8T e f j 112 . 
8T e (K) 
ST a (K) 
0.5 
0.7 
0.9 
0.5 
0.7 
0.9 
1.0 
1.0 
1.1 
1.2 
1.3 
1.5 
1.6 
1.7 
1.8 
2.2. Mapping Surface Emissivity 
Due to the large impact of surface emisivity in the split-window algorithm, it seems worthwhile to develop 
methodologies for determining and mapping surface emissivities at the AVHRR spatial and spectral resolutions. 
We have developed methods for both the mean emissivity, e, and the spectral difference, Ac. 
2.2.1. Mean Emissivity. The emissivity of varied land surfaces can be easily performed in the field by 
using the box method (Sobrino and Caselles, 1993). The emissivity measured with radiometers operating in the 
10.5-12.5 pirn waveband can be used as good estimates of the mean emissivity in the window channels, 
e=(E 4 +e 5 )/ 2 . In the 10.5-12.5 pirn region, the emissivity of land materials is close to unity, with values ranging 
from 0.95 for bare soils to 0.99 for fully vegetated surfaces. At the AVHRR pixel scale, several surface types 
can exist. Thus, the effective pixel emissivity will depend on the soil composition, vegetation cover, etc. A 
simple model for defining effective emissivity for partially vegetated surfaces is (Caselles and Sobrino, 1989): 
e = e v P v + £ S (1-P V ) + de ( 6 ) 
where E v is the emissivity of vegetation, e s is the emissivity of bare soil, and P v is the proportion of vegetation 
observed by the radiometer, and d£ is an additional term which takes into account the reflections between the 
different parts of the system. An equation similar to ( 6 ) can be written for the normalized vegetation index 
(NDVI, referred here as i for abbreviation) of partially vegetated areas 
i = i v P v + i s (l-P v ) + di (7) 
where i v and i s are, respectively the normalized vegetation indices for vegetation and bare soil, and di is an 
additional term which takes into account that the vegetation index does not satisfy the associative property for 
area measurements (Price, 1990). From equation (7) we can obtain the proportion of the vegetated area as 
where i s and i v can be approximated by the the minimum and maximum values of the NDVI image of the study 
area, respectively. Then, emissivity images can be produced by using measured values of e v and e s , and applying 
equations ( 6 ) and ( 8 ). The validity of the method is restricted to the homogeneity in soil emissivity. In the 10.5- 
12.5 (im waveband most soils have close emissivities, then e s will represent the average value of soil 
emissivities present in the study area. 
2.2.2. Spectral emissivity difference. The measurement of the spectral emissivity of natural surfaces is 
much more complicated, so that only laboratory measurements of soils or plant components are usually 
available (Elddvidge, 1988; Salisbury and D’Aria, 1992). Thus, it is difficult to obtain the effective spectral 
signature at the AVHRR pixel scale. In the recent years, different approaches have been developed for retrieving 
the spectral variation of emissivity from satellite data (Becker and Li, 1990; Vidal, 1991; Li and Becker, 1992).