International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
Despite the usefulness of satellite images, classification
problems do exist, due to problems associated with the
discrimination amongst thematic classes. For instance, fire scars
in grassland physiognomies may not be discriminated in the
satellite images, since they present similar spectral values to
other classes (such as clouds, shadows, water bodies). Hence, it
is important to understand the process associated with the
permanence of the fire scars, to guide the sampling of the
images to be used to estimate the annual area burnt in the
cerrado. Permanence, in this context, is defined as the time lag
between the appearance of a fire scar in a Landsat image and its
posterior disappearance, resulting from vegetation regeneration.
Finally, the knowledge of the time elapsed between fires
(recurrence) is important, for many reasons. For one, Gillon
(1983) indicates that if grass fires occur infrequently, they are
far less destructive, due to the amount of heat released, than if
they take place annually. In addition, the recurrence of fires in
the Brazilian cerrado influences ecosystem structure, nutrient
dynamics, and is a significant source of atmospheric carbon,
trace gases and particulates (Kauffman et al, 1994). According
to Mueller-Dombois and Goldammer (1990), repeated burning
destroys organic matter in the upper few centimeters of the
mineral soil, nitrogen is removed in gaseous form, and repeated
grass fires could result in skeletal soils highly impoverished of
nutrients.
In view of the above, this paper sets up as objectives:
e To discriminate, in a sample of Landsat imagery, fire
scars associated with the different types of the cerrado
vegetation, and to quantify the corresponding areas;
e To study the recurrence of fires in the different types
of physiognomies of the cerrado, during the period
1996-2000; and
* To estimate the permanence of the fire scars in the
different physiognomies.
2. STUDY AREA
The areas encompassed by TM/Landsat image path/row 221/69
(henceforth referred to as Chapada dos Veadeiros site) and
path/row 223/67 (herein referred to as Araguaia site) were
selected for this study. Both areas are located in the core area of
the Brazilian cerrado, and encompass areas of approximately
30,344 km” and 30,568 km”, respectively.
According to RADAMBRASIL (1981), the most extensive
vegetation cover in the Chapada dos Veadeiros site is campo
cerrado (approximately 65%), followed by parque de cerrado
(17%), forest (4%). campo (0.11%) and cerradäo (0.02%). The
dominant cover in the Araguaia site is campo cerrado (33%),
followed by parque de cerrado (2896), cerradáo (8%) and
forest (896). The physiognomy campo is not present in the
Araguaia site. The remaining area is occupied by agriculture
and water bodies.
3. MATERIALS AND METHODS
3.1 Materials
Sixty-three digital colour composite images of bands 3 (0.63-
0.69 m), 4 (0.76-0.90i M) and 5 (1.55-1.75i m À TML and
5 or ETM+/Landsat 7 were used. At least one scene from each
month of the dry period for each year investigated (1996 to
244
2000) was selected, based on the image quality and the presence
of clouds. In addition to the satellite images, the study also
included topographic charts, to guide the geo-referencing of the
Landsat imagery, and vegetation maps, necessary to identify the
distinct physiognomies affected by the burning, both at the scale
1:250,000.
3.2 Methods
3.2.1. Visual Interpretation of the Landsat Images: The
identification of the fire scars was conducted through visual
analysis of geo-referenced digital images, directly from the
screen, to minimize the errors of inclusion that could result from
digital classification. In the latter, areas which present a spectral
signature similar to the fire scars, such as water bodies or cloud
shadows, could otherwise be included under fire scars.
Initially, all the images associated with the first date of the dry
period, for each year investigated, were analyzed and the fire
scars mapped (those identifiable at the scale 1:50,000 in the
Landsat images). Vectorial Information Layers (VIL)
containing the boundaries of the fire scars were created for the
first dates of each year, and layed over the images associated
with the second dates. From the spectral-textural changes
observed in the fire scars in the second date, the mapped fire
scars from the first date were altered or deleted, according to the
following criteria: (1) decrease in size of the fire scar, due to
vegetation regeneration; (2) increase in size, resulting from new
burns contiguous to the previously burnt areas; (3) elimination
of the fire scar, when its signature could not be discriminated by
the interpreter; (4) inclusion of new fire scars, not observed in
the previous date. The result of this procedure was contrasted
with the subsequent date, according to the criteria above, until
the entire sequence of images was analyzed.
This process was carried out using SPRING 3.6 (Georeferenced
Information Processing System) (Camara et al, 1996),
generating at least one VIL for each month of the dry period.
The resulting VILs were transformed into Matrix Information
Layers (MILs) to allow algebraic operations to be carried out.
These operations were necessary to estimate the annual fire
recurrence, the total burned area and the permanence of fire
scars. Each MIL consisted of a binary image where, for each
pixel, a label “Burned” or “Background” was associated,
depending on the condition of that pixel (burnt or not burnt).
3.2.2. Annual Area Burnt: The annual area burned was
estimated by aggregating into a new MIL all the pixels
classified as “Burned” in the MILs associated to any given year.
Each one of the five MIL images (referred to as aggregate
image 1996, 1997, 1998, 1999 and 2000), was cross-referenced
with the vegetation maps, which were annually updated from
the Landsat imagery. Hence, each pixel in the MIL image was
associated to a vegetation physiognomy. This was done
independently for Chapada dos Veadeiros and Araguaia sites.
3.2.3. Inter-annual Recurrence of Fires: The inter-annual
recurrence of fires (1996 to 2000) was estimated for each site by
aggregating into a new MIL all the pixels classified as "Burned"
in at least one of the five aggregate images. Each pixel on the
new MIL was associated to a 5-digit label (one for cach year)
using the symbol “B” if the pixel was labeled “Burned” in the
aggregate image, or “G”, otherwise. For example, a pixel
labeled BBBBB in the new MIL indicated that the area
corresponding to that pixel was burnt in 1997, 1998, 1999 and
2000; or BGBGG, indicating that the pixel was classified a5
Internatio
nl
‘Burned”
‘Backgrot
‘Backgrou
combinati
indicating
images. F
the sequei
sequences
single 'B"
sequences
that the r
vegetatior
3.2.4. Per
fire scars
correspon
This defin
analysıs,
included,
Landsat 5
cover the ‘
times t+1(
and t+160
period co
indicated 1
Year
16
1996 | X
1997 | X
1998
2000 | X
1996 | X
1997
1998 | X
2000 | X
Table 1.
season.
The perm:
pixel basis
fire scars
out by ass
the label “
could stil
respectivel
sequence |
that the fi
consecutiv
acquisition
disregarde
in the first
This was «
days betw
passage, th
the observ:
available (/
