Noernberg, Mauricio Almeida
package used is an integrated system which allows for the processing, classification and analysis of
monoespectral SAR data. It is based on the statistical properties of the data. Besides the basic functions
necessary for the classification, the software computes the descriptive statistics, performs the goodness of fit test
to investigate the appropriateness of given distributions, performs digital classification, and provides an
evaluation of the results based on the Kappa statistics.
4 RESULTS
The p-value for all the classes, in all polarizations, was zero, thus leading to the rejection of the null
hypothesis for all statistical distributions tested, at any level of significance. In other words, the probability of
erroneously rejecting the null hypothesis was almost null, indicating that none of the assumed distributions
provided an adequate fit to the observed data. A similar result was obtained by Dutra et al. (1993) when testing
the fit of several statistical distributions to pasture and vegetation regrowth data. Hence, the class-statistical
distribution association was based solely on visual analysis, associating to each class the statistical distribution
that provided the best visual fit. The distribution associated to each class, at each polarization, as well as the
histograms with the curve fitting, are displayed in Figures 1, 2, 3, 4, 5, and 6.
From Figures 1, 2, 3, 4, 5, and 6 it can be seen that the two most frequently distributions associated with
the different classes at the distinctive polarizations were the Amplitude GO (12 times) and the Log-Normal (7
times). The good performance of the Amplitude GO distribution (which is a particular case of the Amplitude G
distribution) suggests the usefulness of its application to SAR data (Vieira, 1996).
For each class, the statistical distribution associated to the data at different polarizations (HH, VV, HV)
were the same, indicating that the backscatter variation due to polarization does not affect the distribution of the
data from different classes of aquatic plants. The only exception was the Pistia class, which displayed a different
behavior at polarization VV.
Apparently, there is no relationship between the type and structure of the backscatering elements of the
aquatic plants with the corresponding statistical distribution, since different types of distributions were
associated to the data from classes with similar morphological characteristics (such as Salvinia and Pistia).
EICHHORNIA
HH (Log-Normal) VV (Log-Normal)
0.030 - o 8 : j 0.030 F : j
$ e S 5 > * e
0.020 - > 7 0.020 [- j
0.010 = 0.010} :
0.0D0 And 1 L^ 0.0D0 L 1 i
20 40 80 80 100 20 40 80 80 100
HV (Log-Normal) VH (Log-Normal)
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 1013