Araujo, Luciana Spinelli
All this behavior (Figure 2) may be explained by
higher proportion of shade in the primary forest area,
due to the structural variation of the canopy, causing oss
lower spectral response in relation to the areas in the a
secondary succession process, specifically in band 4 0,45 4
of TM (near infrared portion). A different
photosynthetic capacity among several individuals, à oss. T
which compose the several stratum of primary e
forest, has different effects in the spectral response >
in band 3 (red portion) of TM, when compared to 9 025
secondary vegetation cover.
0,15-
Concerning to savanna areas, it is observed that is :
impossible to distinguish intra-classes, with standard ips
deviation values occupying the same attributes space, Pr . 7 7
either for grass/shrub savanna and park savanna. The sr PS ses
charact e ri sti co f p a rk sav a i s com p ose d by few PF -Primary ForestSF -Secondary Forest PS -Park Savanna SGS -ShrubGrass Savanna
arboreal individuals sparsely distributed over a grass ,.. o i
stratum, which causes a certain similarity in the Figure 2. Mean digital values and standard deviation extracted
spectral response with the grass/shrub savanna, in from SAVI 0.5 image according to the vegetation cover types.
this kind of synthetic image.
The construction of simple regression model, with a linear function, permitted to observe the high relationship of SAVI
values with biomass of forest formation, which model is expressed as: Biomass = 419.64 - 797.27 (SAVI). It is
important register that the presented model represents the condition of undisturbed primary forest, as well as with some
degradation, identified in the samples inventoried in the fieldwork. In these forested areas (Figure 3 a), the significant
correlation coefficient (r* = 0.80) demonstrates that the most part of biomass variable can be explained by the SAVI
values, on the contrary of savanna formation (Figure 3 b) with low correlation coefficient (r? ^ 0.16), represented by a
following expression: Biomass — 0.2531 + 38.8 (SAVI).
(a) (b)
200 Br
"s Biomass = 419.64 - 797.27 (SAVI) jo | Biomass = 0.2531+38.8(SAVD — *,
= = = 2= 0.1622
S 150 | . 2= 0.8001 = 2= 0.
8 6 8 9|
S S
= 100 6
a %
5 S d
3 91 = >.
= © =
0 u T T T 0 T ud T T T
0,3 0,35 04 0,45 0,5 0,55 005 0,075 0,1 0,125 015 . 0175 0,2
Savi_0.5 Savi 0.5
€ Primary Forest O Secondary Forest € Park Savanna © Grass/shrub Savanna
Figure 3. Scatterplot relating SAVI values with: (a) forest biomass and (b) savanna biomass.
The analysis of the confusion matrix corresponding to the classification of SAVI (L — 0.5) image demonstrates that
there is a more difficulty for discriminating between classes of savanna. For instance, 33% of the classified areas as
grass/shrub savanna occupy in the synthetic image, the same attributes space of park savanna. The primary and
secondary forest areas are better discriminated, presenting a classification error no greater than 8%. The analysis of
classification performance was performed using 30 test samples, spatially distributed in the several vegetation cover
types found within the study area. Firstly, with the identification of all 8 thematic classes (primary forest, regrowth,
park savanna, grass/shrub savanna, pasture, bare soil, water, and unclassified) a kappa coefficient of 0.74 was found for
the thematic classification of the study area. The kappa coefficient increased to 0.88 due to the combination of both
park savanna and grass/shrub savanna in just one class (Figure 4)
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 79
