¡HHmmnaHMMMMi 
12 
Fig. 4: Histograms of the vegetation index 
derived from Landsat MSS before and after 
the atmospheric correction using the 
optical thicknesses shown in Fig. 3. (a): 
northern subarea, original radiances; (b) 
southern subarea, corrected radiances. 
Dashed lines represent vegetation index for 
a clear day (20 August 1982) and solid lines 
for a hazy day (2 August 1982). Note the 
excellent agreement between the vegetation 
index histograms after the correction except 
of the zero and negative values derived 
over the ocean. The correction algorithm 
here assumed a lambertian surface, 
assumption that was reasonable for the 
nadir view over the land but wrong over the 
water (After Kaufman and Sendra, 1988). 
0.0 0.4 0.8 -0.8 -0.4 0.0 0.4 0.8 
VEGETATION INDEX [ (L 3 -L 2 >/(L 3 » L 2 )] 
Maryland/Delaware Pennsylvania 
reflectance (0.64 pm) reflectance (0.64pm) 
Fig. 5: Scatter diagrams of the relationship between the apparent surface reflectance in the red 
(0.64 pm) and the mid IR (3.75 pm) AVHRR channels, (after Kaufman and Remer, 1993) 
Determination of dark pixels using the vegetation index is not very satisfactory, since 
only dense vegetation pixels can be determined, not dark soils, and the vegetation index 
itself is affected by the aerosol, a feedback circle that causes the method to be applied only to 
images for which it is known a priori that dense vegetation pixels are present in the image. 
An alternative technique suggests to use longer wavelengths (2.1 or 3.7 pm) that are less 
affected by aerosol scattering (since they are much lager than most aerosol particles) and are 
still sensitive to surface characteristics. Such wavelengths can be used to find pixels that are 
dark in the visible channels (Holben et al, 1992). Analysis of 40 AVHRR images over the 
Eastern US shows that the AVHRR 3.75 pm channel is not sensitive to the presence of 
smoke or pollution aerosol (Kaufman et al., 1990) but is very sensitive to the presence of 
forest pixels and can be used for its determination (Kaufman and Remer, 1993) (see Fig. 5). 
To use the reflective part of the 3.75 pm channel it has to be corrected for thermal emission. 
This introduces u 
shorter wavelengtl 
dark pixels. Fig. 6 
pm and that at 0.<; 
DC and the Chesa 
the visible channe 
blue channels resp 
6 
=1 
CD 15 - 
T-H 
rd 
tN 
-C io 
u 
Ol 
u 
G 
03 . 
J-i 5 
u 
Ol 
CG 
01 
' / 
/ 
/ 
reflectarte 
Fig. 6: Example of 
and 0.64 pm, deri< 
Chesapeake Bay are 
of different reflecta 
image. All data 
reflectance at 0.86 
plotted. The uncert 
pm channel is ±0.0i 
pm <0.1. The da 
based on the aeroso 
An algorithm 
monitoring of aero 
pixels that are expe 
0.47 and 0.66 pm) 
bands. The procedi 
Except for du: 
1993). Therefoi 
The aerosol e: 
sunlight and 1 
dominates wh: 
smaller effect ( 
the range 0.2 < 
The surface rt 
usually a smi 
correlation b< 
wavelength sj: