47°20' 
47° 11' 
  
  
IS. 
  
6 22°03 
LEGEND 
1- LV 
2-LR 
3- PV 
4- Hi 
5- TE 
6-LE 
7 -AQ 
== Main River 
—~< Tributaries 
Q Soil Limites 
Escale 
0 1 2Km 
et 
ESTATE OF SAO PAULO 
  
52° 48° 44° 
  
  
  
22°10' 
  
  
Fig. 2- Study area as mapped in 1982 by field work of the IAC- Instituto Agronômico de Campinas, 
published at 1:100.000. 
and reflected infrared radiation. The 
images were from orbit 220/75D for the 
month of December. The topographic 
quadrilateral of Araras (1:50,000) by the 
IBGE was used as the cartographic base 
for plotting information. 
The digital images were processed 
with the SITIM-150 system on a 
microcomputer with a PROCON 
projector/enlarger. The work sequence is 
presented in Figure 3. 
3.Analysis of the images 
For selecting the subgroups of 
bands for generation of the color 
composites, the Jeffreys-Matusita 
distance method was used, as discussed by 
Swain and King (1973). The JM distance 
is an appropriate technique to measure 
the average separability between spectral 
classes, calculated as functions of 
probability density. The following 
researchers implemented or applied the JM 
method: Bendat and Piersol (1986), 
Andrade (1985), and Paradella (1984). 
This technique is a convenient 
alternative for selecting the best color 
composite images. The series of 
interband statistical measurements of the 
JM method results in a reduction of the 
dimensionality, processing and redundance 
of data. 
In general, each class of 
interest (e.g. Typic Eutrorthox) in an 
image can be characterized by a function 
298 
of density of probability Pl (x) that 
gives the values of probability densities 
that the pixels x belong to a -class in 
function of x. For two classes wj and 
Wy, the JM distance is defined as: 
2 
JMij = fx {es ce) V2. -« [p309] 1/2} ax 
where: 
JMjj = JM distance between classes wj and 
wa: 
J? 
Pj (x) = probability density of the pixels 
belonging to class Wi; 
Pix). = probability density of the pixels 
belonging to class wj; 
X = range of interest for the X values 
The software implemented on the 
SITIM system calculates the JM distances 
between classes selected by the user for 
all possible combinations of bands. The 
output includes subsets that maximize the 
JM average and minimum distance criteria. 
As the method to classify the 
scene, a  multi-variate analysis was 
applied that offers the advantage of 
working with both parametric and non- 
parametric data. Cluster analysis was 
adopted in order to work with a group of 
units characterized by diverse variables. 
The result is the separation of existing 
groups characterized by homogeneity