nd 
of 
cs 
al 
to 
as 
to 
ny 
  
2m 
ly 
le. 
70) 
er 
“he 
ve 
gh 
ite 
that the significance of iron oxides in 
the reflectance increases with the 
increase of the wavelength in the 
electromagnetic spectrum, especially in 
the regions of visible light and near- 
infrared. 
Cipra et alii (1976) compared 
spectral-radiometer measurements of 
exposed soils with digital data from 
Landsat-1. Results showed that Landsat 
radiance and spectroradiometer 
reflectance values were highly correlated 
for all wavelength bands. 
Lund et alii (1980) and Harrison 
and Johnson (1982) concluded that the use 
of spectral maps derived from Landsat 
data improved accuracy and/or quality of 
map unit delineations. More recently, 
Coleman and Montgomery (1987) and Everitt 
et alii (1989) studied the same question. 
The former encountered great 
interdependence between the moisture 
content and the reflectance of the 
respective soils. The latter studied the 
moisture content, organic mater and the 
level of iron in alfisols and vertisols, 
finding high correlations between 
reflectance and the studied variables. 
In both cases the soils were separated 
and accurately mapped on the basis of 
spectral, physical and chemical 
properties. 
Agbu and Nizeyiama (1991) 
compared soil maps from SPOT spectral 
data with maps produced in the field. 
Although the maps based on field work 
were found to be better than those based 
on spectral analyses, the differences did 
not attain statistical significance at 
the 0.05 level, using the Kappa 
statistic. 
Visual analysis of spectral 
differences are insufficient for the 
required studies. Although only a few (4 
to 12) bands are selected from the 
continuum of electromagnetic energy, each 
band contains a continuum of extremely 
small variation of the intensity of 
reflectance in that band. Therefore, the 
number of combinations of bands (colors) 
is extremely large. Consequently, 
research that deals with such spectral 
behavior of targets in images from space 
satellites require digital analyses via 
computer processing. These capabilities 
exist in image analysis systems such as 
ERDAS or, in Brazil, SITIM (Sistema de 
Tratamento de Imagem) from the Space 
Research National Institute, INPE. 
Among the methods for spectral 
image analysis, of particular note are 
those that are based on the statistical 
distance between probability densities 
that characterize the standard classes. 
These methods include divergence, 
transformed divergence, Bhattacharyia's 
distance and Jeffreys-Matusita (JM) 
distance  (Swain and King, 1973, and 
Richards, 1986). 
METHODS 
1.Description of the study area 
The study area is approximately 
10 x 10 minutes of latitude and longitude 
297 
(220 square kilometers) on the Araras 
topographic sheet in the state of Sao 
Paulo, Brazil (see Figure 2). The area 
is tropical, being 130 kilometers north 
of the Tropic of Capricorn. The maximum 
and minimum elevations are 560 and 680 
meters above sea level. The prominent 
relief is a slightly rolling landscape. 
Only ten percent of the area has 
limitations that prevent mechanized 
agriculture. 
According to the maps of the IGC 
(1982), the geology of the area includes 
rocks from the Tubaräo Group, the Irati 
and Corumbatai (siltstone and shales) 
formations of the Passa-Dois Group, basic 
intrusives, sandstones from the Botucatu- 
Pirambóia formation, and the Cenozoic. 
In the Koppen system of climatic 
classification, the climate of the area 
is mesothermic with dry winter, type Ewa. 
The winter dryness extends from April to 
September; the rains for summer occur 
from October to March. June-July 
temperatures average 182 C (649  F), 
rising to 229 .C (729 F) in January- 
February.  Frosts do not occur. 
The natural vegetation is 
classified as subtropical forest. Today 
the area is used for sugar cane, citrus, 
cotton and corn agriculture. Pastures 
and  reforestation are found in the 
steeper areas. Keeping in mind the 
methodological considerations of the 
research, we selected an area 
predominately occupied with annual crops 
and obtained images from the period prior 
to planting. The major part (85$) of the 
area was free of vegetation. 
The soils of the area, according 
to Oliveira et alii (1982), are listed 
below, in order of highest to lowest 
occurrences. Their approximate 
distribution, according to the  pre- 
existing map at 1:100.000, is shown in 
Figure 2. 
LV - Latossolo Vermelho Amarelo 
(USA) - Quartzipsammentic Haplorthox 
. LR - Latossolo Roxo - eutrófico 
(USA) - Typic Eutrorthox 
PV - Podzólico Vermelho Amarelo 
(USA) - Typic Paleudult 
. TE - Terra roxa Estruturada - eutrófica 
e distrófica 
(USA) - Rhodic Paleudalf + Rhodic 
Paleudult 
AQ - Areias Quartzosas 
(USA) - Typic Quartzipasamment 
. Hi - Solos Hidromórficos 
(USA) - Hydromorphic soils 
LE - Latossolo Vermelho Escuro 
(USA) - Typic Haplorthox 
2. Characteristics of the images 
and equipment used 
Analogue (1:100,000) and digital 
images from Landsat Thematic Mapper (TM) 
were obtained for six bands of visible 
ee rT 
EE UE