channels and in the 2.1 and 3.9 pm channels present on MODIS. The presence of 
vegetation decreases the reflection in the 0.41, 0.47 and 0.66 channels due to chlorophyll 
absorption and in the 2.1 and 3.9 pm channels due to liquid water absorption. Wet soil 
has a lower reflectance in the visible channels due to light trapping capability, and a 
lower reflectance in the 2.1 and 3.9 pm channels due to liquid water absorption 
(Kaufman and Remer, 1993). 
Using these principles the following procedure for remote sensing of aerosol and 
subsequent atmospheric correction is suggested: 
Thresholds based on the reflective part of the 3.9 pm channel, p 39 , (derived from the 
radiance at 3.9 pm after correction for the emissive part) and the 2.1 pm channel, p 
are used on a 50x50 km 2 grid to determine cloud free pixels with vegetation cover or 
otherwise low reflection in the 0.41, 0.47 and 0.66 pm bands (Kaufman and Remer, 
1993). The thresholds are first corrected for the water vapor absorption using the total 
precipitable water vapor derived from MODIS. 
expansion of the w 
amount of data ani 
latitude, longitudi 
physical aerosol pr 
the optical propert 
scattering albedo, 
specific aerosols t] 
(Hoppel et al., 199( 
and Andrea, 199 
measurements, m 
measurements do 
ambient instrumei 
Sulfate Cloud An 
distribution varies 
based sun/sky rai 
climatology with , 
describe the prope: 
For example: p 39 < 0.025 or p 2J < 0.05 
The 2.1 pm channel is used to estimate the reflectance in the 0.41, 0.47 and 0.66 pm 
channels (p,_, p,_ p„) of these pixels. 
r 41 r 47 r 66 r 
For example: p 4J = p 4? = 0.7p 66 = 0.33p n 
A lookup table is generated using radiative transfer computations for the 0.41, 0.47 and 
0.66 pm channels, for the range of viewing and illumination conditions of MODIS and 
for several values of the aerosol optical thickness. The aerosol model used in these 
computations will be varied from a rural model for low aerosol optical thickness to a 
sulfate dominated model for high optical thickness. The models are based on the AFGL 
aerosol models (Shettle and Fenn, 1978) but modified using observations. Ground based 
remote sensing of the size distribution and scattering phase function of the ambient 
aerosol are used to modify these models (Kaufman, 1993; Kaufman et al., 1994). The 
lookup table is used to infer the aerosol optical thickness from the measured radiances. 
Correction for ozone absorption and residual water vapor is applied. With analytical 
equation adjust the lookup table for the box elevation (correct the Rayleigh scattering) 
the ozone and water vapor absorption. The transformation are described by Fraser et al. 
(1992), Tanre et al. (1992). 
The derived aerosol optical thicknesses are averaged over a grid of 50x50 km (to exclude 
extreme values an average between the 10th percentile and the 40th percentile is used) 
to give a map of the aerosol loading on this resolution. For grid boxes were no values 
are available, the average of its immediate neighbors is used. If no such values are 
available (e.g. in bright deserts) climatological values will be used and the correction 
will not be very effective. 
Atmospheric corrections are applied to every pixel, using the inverse of eq. 1 and a look 
up table described before. The aerosol optical thickness used for each pixel is a result of 
interpolation on the derived optical thicknesses on the 50x50 km grid. An aerosol 
model is assumed based on aerosol climatology discussed in the next section. 
3.2 Aerosol climatology 
At a global scale, aerosol climatology was summarized by d'Almeida et al. (1991) as an 
There may be < 
that can reduce th 
aerosol concentrai 
important in suppl 
refractive index) th 
the dark target app 
3.3 Ground basée 
The aerosol pa: 
from in-situ measi 
measurements (Vo 
carried on a netw 
performed a simil 
instruments, Holbt 
1986 using 15 mor 
size distribution a 
almucantar measui 
from the collectio: 
atmospheric bounc 
determined from ; 
1979; Wang and C 
optical properties i 
which atmospheric 
Sunphotomete 
(BAPMoN) has be 
Meteorological Or£ 
a recent report (1 
measurements: po< 
quality. Measurem 
conducted recently 
1994); in Africa b 
Eastern US (Y. Kai 
(e.g. sunphotometi 
properties are the 
total loading and 1 
high data quality 
14