SES IN
jparate the
ich visited
1l, in order
'ation Unit
g Landsat
on of the
the NDVI
iined 2000
howed the
n found in
'e analyses
rom —8 dB
seasonal
e Cerrado
shows the
3).
tential use
and Terra-
1ies within
razil. The
ect of the
rado) and
DA).
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
1.2. Methodology
Remote Sensing data have been frequently used to
classify vegetation all over the world. Most of the remote
sensing studies of vegetation are done with optical spectral
bands. In the recent years, radar images are becoming an useful
tool because their characteristics can be essential to efficiently
sense some vegetation parameters.
In thé particular case of the Cerrado biome, its
physiognomies may vary from campo cerrado, to cerrado s.s.,
and to cerradäo (the forest type). Other associated forest types,
such as riparian forest and SSForest, can also be found
(Mesquita Jr. 1998). Two aspect of the vegetation must be taken
into account: green leaves (which can be seasonal) and trucks
with branches (which can be permanent). Because of that, the
use of optical remote sensing can cause misclassification on the
forest physiognomies due to the fact that optical bands detect
predominantly the green leaves response. The optical spectral
response is directly proportional to the amount of phytomass
and the vegetation index, particularly the NDVI (normalised
difference vegetation index).
Although the NDVI has been shown useful in change
detection, land surface monitoring, and in estimating many
biophysical vegetation parameters, there is a history of
vegetation index research identifying limitations in the NDVI,
which may impact upon its utility in global studies which can be
simplified as follow:
- Canopy background contamination: background reflected
signal, soils, litter covers, snow, and surface wetness;
- Saturation with chlorophyll signal in densely vegetated
canopies; and
- Canopy structural effects associated with leaf angle
distributions, clumping and non-photosynthetically-active
components (woody, senesced, and dead plant materials).
There are several explanations for the NDVI saturation
problem over densely vegetated areas in which NDVI values no
longer respond to variations in green biomass. The NDVI has
been reported to be an insensitive to quantify LAI (leaf area
index) at values exceeding 2 or 3.
The atmosphere degrades the NDVI value by reducing the
contrast between the red and NIR reflected signals. The red
signal normally increases as a result of scattered, upwelling path
radiance contributions from the atmosphere, while the NIR
signal tends to decrease as a result of atmospheric attenuation
associated with scattering and water vapour absorption. The net
result is a drop in the NDVI signal and an underestimation of
the amount of vegetation at the surface. The degradation in
NDVI signal is dependent on the aerosol content of the
atmosphere, with the turbid atmospheres resulting in the lowest
NDVI signals (Huete et al., 1997 and 1999).
The MODIS NDVI images are being appointed as an
improvements over the current NOAA-AVHRR NDVI. Many
new indices have been proposed to further improve upon the
ability of the NDVI to estimate biophysical vegetation
parameters. However, the robustness and global implementation
of these indices have not been tested and one must be cautious
that new problems are not created by removing the ‘rationing’
properties of the NDVI.
545
The NDVI is a ‘normalized’ transformation of the NIR
(near infrared) to red reflectance ratio, D nir / f) red, designed
to standardise vegetation index values to between —1 and +1;
NDVI = {( P nir / P red) — 1}/ {( P nir / P red) + 1}
It is functionally equivalent to the NIR to red ratio and is
more commonly expressed as:
NDVI - (p nir - p red ) / ( nir * p red)
As a ratio, the NDVI has the advantage of minimising
certain types of band correlated noise (positively-correlated)
and influences attributed to variations in direct/diffuse
irradiance, clouds and cloud shadows, sun and view angles,
topography, and atmospheric attenuation. Rationing can also
reduce, to a certain extent, calibration and instrument-related
errors. The NDVI, as a ratio, can be computed from raw digital
counts, top-of-the-atmosphere radiances, apparent reflectances
(normalised radiances) and partially or total atmospheric
corrections. Although the units cancel out, the NDVI values
themselves change so one must be consistent in how the NDVI
is derived. The extent to which rationing can reduce noise is
dependent upon the correlation of noise between red and NIR
responses and the degree to which the surface exhibits
Lambertian behaviour (Huete ef al., 1999).
The NDVI is the only vegetation index currently adapted
to global processing and it is used extensively in global,
regional, and local monitoring studies. It has also been used on
a wide array of sensors and platforms. The MODIS NDVI
algorithm will utilise complete, atmospherically corrected,
surface reflectance inputs, avoiding atmosphere contaminants
such as water vapour. According to Huete and collaborator
(Huete et al., 1999) the MODIS NDVI can provide consistent,
spatial and temporal comparisons of global vegetation
conditions (structure and phenology).
The cerradäo and the SSForest vegetation differ from one
another, in the field, not only in species composition but also in
the structure (Batalha et al. 1997 and Batalha et al. 2001). The
cerraddo canopy is 10 to 15 meters high and has a regular
surface height geometry, whereas the SSForest canopy is 15 to
25 meters high and its geometry is relatively rough, mainly
because of the presence of emergent trees (highest trees in the
canopy). These emergent trees can be deciduous or
semideciduous, what difficult even more the use of optical
remote sensing (Mesquita 1998).
The microwaves radiation in the radar bands is
transmitted from antenna and, after that, it receives the reflected
signal from the earth surface. The sigma signal (c) value is the
ratio of the received backscattered energy over the emitted
energy. Usually o values are expressed in decibels (dB) units
which can be converted into digital numbers (DN) of a intensity
image (Roseqvist, 1997; Shimada, 2001).
Generally, the c values are dependent on the geometry of
the target on the ground and the wavelength. The JERS-1/SAR
signal interacts with earth surface roughness on a magnitude of
half of the wavelength A = 23 cm and mostly with objects
oriented according to the signal polarization VV vertical
emission — vertical reception (such as trunk and branch).
Some parameters are quite important to understand the
response of the target on the earth surface. They are: geometry
of satellite and antenna (satellite ephemeris and antenna angle)
in relation to surface and target (corner reflection and specular
