10 
0 =30°, 9 =30°, $ =180°, T =0.20, a =1.3 
s v v a a 
NDVI-0.9 NDVU0.6 NDVI-0.3 
REFLECTANCE IN CHANNEL 1 
Fig. 2: Solid lines, labeled NDVI, represent iso-lines of the actual NDVI and dashed lines, 
labeled 8, represent iso-lines of the difference between the actual NDVI and the NDVI 
measured at the top of an aerosol layer with an optical thickness of 0.20. The abscissa is the 
reflectance in band 1 (0.64 ¡am) and the y-axis, the reflectance in band 2. Results are presented for 
geometrical conditions 0 S =30°, 0 V =30°. The ozone content is 0.35 cm.atm. The two oval areas 
roughly correspond to bare soils (low NDVI) and high vegetated surfaces (high NDVI). (after 
Tame et al., 1992). 
3. DIRECT METHODS FOR CORRECTION OF THE AEROSOL EFFECT 
Direct methods for correcting the aerosol effect on satellite images are based on 
estimation of the aerosol loading and/or the optical properties and correction of the 
measured spectral radiances for the aerosol effect. In the following we shall discuss 
methods for estimation of the aerosol loading and optical effects from the image 
itself; estimates using aerosol climatology and estimates using measurements with 
ground-based instrumentation. Even if the atmospheric path radiance (solar light 
scattered back to space by aerosol particles and molecules before reaching the surface) 
is estimated from the image (see the following section), in order to use it in 
correction of the satellite data, the aerosol absorption and scattering phase function 
have to be assumed based on an aerosol climatology. Therefore the following three 
methods are not completely independent. 
3.1 Atmospheric correction using dark pixels to estimate the aerosol loading 
The aerosol effect is strongest for low values of the surface reflectance. Therefore, it is 
appropriate to use the darkest pixels in the image to estimate the aerosol loading and effect 
on remote sensing. Mainly the aerosol optical thickness and path radiance should be 
established. Remote sensing of aerosol and atmospheric correction stem from the 
relationship between the measured radiance at the top of the atmosphere p* (given in 
reflectance units) to the surface bidirectional reflectance properties p: 
P*(0,e o ,<t>)=Pa(e,0o,‘t))+Fd(eo)T(e)p(e,0o,<t))/(i-sp’) (i) 
where 0 is the vie 
scattered radiation 
radiance is attenual 
flux for zero surfai 
less than 1.0 due t< 
space. T(0) is the u 
For low val 
on p* is large mai 
and some soils) an 
But in order to use 
thickness, the surf; 
uncertainty of ±0.0i 
Kaufman ai 
forests the dark pi) 
with lowest reflect 
channel can be as 
Comparison betwi 
independent meas 
Application of the 
index, shows how 
hazy and a clear d 
Note that in order 
on the aerosol size 
have to be assum 
calculations. Since 
errors in the aeros 
c n 
LU 
X 
a. 
u. 
O 
OC 
LU 
CD 
£ 
3 
2 
0.0 
( 
Fig. 3: Histograms 
vegetation detectior 
1982) and a hazy 
ground-based sun j 
thickness and the 
surface reflectance