Noernberg, Mauricio Almeida 
  
STATISTICAL ANALYSIS OF BACKSCATTER DATA FROM DIFFERENT GENUS OF AQUATIC 
PLANTS 
Mauricio NOERNBERG*, Thelma KRUG**, Evlyn NOVO** 
Federal University of Paranä, Brazil 
Center for Marine Studies 
— mauricio@cem.ufpr.br 
** National Institute for Space Research, Brazil 
thelma@ltid.inpe.br, evlyn@ltid.inpe.br 
KEY WORDS: SAR, Multipolarization, Aquatic Plants, Statistical Distribution. 
ABSTRACT 
This is a study of the statistical properties of a multipolarimetric SAR data to discriminate aquatic 
plants. Here was tested the goodness-of-fit of data from different genus of aquatic plants to five statistical 
distributions, as a support to the development of new digital classifiers. 
1 INTRODUCTION 
Visual analysis of radar imagery acquired at different polarizations indicate differences amongst the 
distinct genus of aquatic plants. These differences are more evident in the color composite generated from 
images at different polarizations, suggesting the possibility to perform digital classification of these classes of 
aquatic plants. However, the results from digital classification to discriminate amongst the different genus are 
gq still very poor, possibly due to the non-conformity of the backscatter data to the normality assumption of some 
gai conventional classifiers. This normality assumption is not too unrealistic for data obtained from optical sensors 
operating in the visible part of the spectrum. The interaction processes between the eletromagnetic radiation 
(EMR) and the target, in the microwave portion of the spectrum, are dominated by the tri-dimensional 
distribution of the scatter elements and by the organization of the water molecules within the target. This causes 
a completely random behavior of the EMR return. Hence, the appropriate modeling to image processing requires 
an accurate knowledge of the statistical properties of the SAR data (Ulaby and Dobson, 1989: Yanasse et al., 
1994 and 1995). Several researchers (Dutra et al., 1993; Yanasse et al., 1994 and 1995; Frery, 1993) have been 
studying the statistical properties of SAR data to discriminate and classify land cover. Vieira (1996) obtained 
better classification results when applying a contextual Markovian classifier to the data than when using 
conventional methods that rely on the distribution of the data. The objective here is to test the goodness-of-fit of 
data from different genus of aquatic plants to five statistical distributions, as a support to the development of 
281 | new digital classifiers. 
2 TEST AREA 
The study area comprises the Tucuruí reservoir, which was built in the Tocantins river, Para state, 
Brazil. The Tocantins river basin is limited by latitude 2°S and 18°S and longitude 46°W and 55°W. The total 
basin drainage area is around 767,000 km^. The high water season ranges from F ebruary to April and low water 
season from September and October. The minimum water level is around 58 m and the maximum water level 
reaches 72 m above sea level. The reservoir's water volume is around 45 billion cubic meters, flooding an area 
of approximately 2,875 km?. The maximum depth is 72 m; the reservoir average depth is 18 m. The reservoir 
surface is highly dentritic, expressed by a perimeter of 7,700 km. The Pucurui inlet was selected as test site due 
to the high concentration and variety of aquatic plants. In addition, this area is very sensitive to water level 
fluctuation along the year. These features make it difficult to delimit the reservoir/water boundary in this region. 
  
3 | DATA ACQUISITION 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 1011