logarithm two times in the two sides of Equation 1 one can 
obtain the following fitting equation: 
In (-In (g))=In k+b.Inuw+c.Inm (2) 
The 5S Model calculates the value of tg for given values of uw 
and m. The triple [tg; ww; m] is used then to fit the Equation 2 
to obtain the values of k, b and c. The following equation was 
obtained using 216 fitting triples [tg; uw; m] (18 values for m 
and 12 (0.5, 1.0, 1.5, … , 6.0) for uw) with r = 0.999952: 
tg = exp[-0.6767 . uw°°°%3 | 05175] — (3) 
The linear interpolation between the direct solar radiation at 
870nm (Rdirs7o) and 1026nm (Rdir1026) gives the direct solar 
radiation at 948nm (Rdiross) for the theoretical case of an 
atmosphere without water vapor. The gaseous transmission fg 
at 948nm is equal to the ratio between the measured (Rdiro;5) 
and the estimated (Rdiro4s) values for the direct solar radiation 
at 948nm: 
tg = Rdiross / Rdiross (4) 
Applying this value of tg into Equation 3 (our case) or 
Equation 1 (general situation), it is possible to obtain the value 
of the water vapor contents uw for a given air mass m. 
The INRA/Avignon develops a similar method to use with the 
CIMEL sunphotometer. 
This method will be useful to intercalibrate the ground level 
measurements of water vapor content with the TOVS data, 
which is very interesting to correct images with spatial 
variability of atmospheric parameters. The estimate of water 
vapor contents is very useful also to process thermal infrared 
images. 
3.1.2. Ozone. The ozone content is obtained in the table 
proposed by London et al. (1976) since its values are in a good 
agreement with our tropical conditions (Lazutim, 1993). 
The UNICAMP has a Russian ozonometer that measures the 
ozone contents three times a day since July 1993. Lazutim et 
allii (1994) presents some results of these daily measurements. 
This equipment will be very useful to intercalibrate the TOVS 
data that can be used to correct images with spatial variability 
of atmospheric parameters. 
3.2. Aerosol concentration 
We calculate the aerosol concentration using the Beer-Bouguer 
Law and the spectral direct solar radiation measured in the 
ground level by a LI1800/LICOR spectroradiometer adapted by 
us. 
Another possibility is to use the global solar radiation in a 
method presented by Zullo et allii (1994). It is possible to 
estimate the aerosol concentration with a good precision 
knowing the type of the aerosols. 
There are some big and clear rivers and a part of the Atlantic 
Ocean in the state of Säo Paulo that can also be used to 
calculate the aerosol concentration from the image. It is very 
interesting to correct Noaa images, for example. 
3.3. Aerosol type 
There are three main regions in the state of Säo Paulo 
according to the predominant type of aerosol present in each 
one: the coast (where there are only maritime aerosols), the big 
cities (where the aerosol is predominantly urban) and the 
interior (with continental aerosols). The maritime aerosols no 
penetrate in the interior of the state because the "Serra do Mar" 
(sea mountain ridge) is a sufficient natural obstacle. 
4. APPLICATIONS AND IMPORTANCE 
The calculation of the vegetation index NDVI (Rouse et allii, 
1971) and the automatic pattern classification are two typical 
uses of the satellite images for agricultural purposes. These 
treatments illustrate very well the practical importance of the 
atmospheric correction. 
The results were obtained using a Landsat-TM image with 512 
pixels by 512 rows, acquired on August 6th, 1992 at 
12:27GMT, orbit-point 219.76, quadrant A, whose center is on 
22°49’S and 47°03'W. The atmospheric optical thickness was 
equal to 0.283 for continental aerosols. The water vapor and 
ozone contents were equal to 3.08g.cm? and 0.31cm.atm, 
respectively. 
The Table 1 shows the NDVI values of three typical surfaces 
existing in the test-image (forest, sugar cane and water) 
calculated from images whose grey levels correspond to 
radiance and reflectance. 
  
  
  
  
  
  
  
  
  
  
  
  
Surface Radiance Reflectance 
Original | Corrected |Original | Corrected 
Forest 0.317 0.553 0.434 0.648 
Sugar Cane 0.363 0.530 0.476 0.621 
Water -0.306 -0.433 | -0.251 -0.475 
Table 1. NDVI 
The difference between the NDVI calculated from the original 
image and the corrected image is near 0.2 for forest and sugar 
cane. This value is very important and significant considering 
that the agronomic parameters are usually expressed by 
exponential curves. This can leave to estimating errors greater 
than 100% for the fresh biomass, for example, if we use the 
models presented by Tucker (1979). 
The importance of the atmospheric correction in the automatic 
pattern classification is presented here using a self-organized 
method based on the algorithm proposed by Pao (1989). This 
method, by its time, is based on the Adaptive Resonance 
Theory described by Carpenter & Grossberg (1987). The 
original method proposed by Pao (1989) was implemented with 
some additional resources and controls to improve its 
efficiency considering its application to classify satellite 
images. 
833 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996