i 2004
————
'esence
ly also
| Of the
ify the
e scale
s: The
visual
m the
It from
pectral
r cloud
the dry
he fire
in the
(VIL)
for the
ociated
hanges
ed fire
x to the
due to
m new
ination
ated by
rved in
trasted
>, until
renced
1996),
period.
mation
ed out.
ial fire
of fire
y each
ciated,
nt).
d was
pixels
n year.
"Tegate
renced
d from
ge Was
done
ites.
-annual
site by
urned"
on the
h year)
in the
| pixel
e area
99 and
fied as
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
‘Burned” in the aggregate image 1996 (AI 1996),
*Background" in the AI 1997, 'Burned" in AI 1998, and
"Background" in Als 1999 and 2000. There are 32 possible
combinations of "Bs'or 'Gs" in a five-digit representation,
indicating the status of any specific pixel in the aggregate
images. From these 32 sequences, only 26 are of interest, since
the sequence GGGGOG is discarded (no fires), as well as the five
sequences which do not characterize recurrence (presence of a
single 'B"in the sequence, such as BGGGO). The 26 remaining
sequences were cross-referenced with the vegetation maps, so
that the recurrence area could be estimated for each type of
vegetation physiognomy.
3.24. Permanence of the Fire Scars: The permanence of the
fire. scars. was defined, «a priori, in periods of 16 days,
corresponding to the revisit time of the TM/Landsat satellite.
This definition implied in the elimination of year 1999 from the
analysis, since in this year both TM or ETM images were
included, and the time difference between acquisitions by
Landsat 5 and 7 was only eight days. Theoretically, in order to
cover the entire dry season (burning season), images acquired in
times t+16, t+32, t+48, t+64, t+80, t+96, t+112, t+128, t+144
and t+160 (in days) should be available. However, during the
period considered, only partial sequences were obtained, as
indicated in Table 1.
Chapada dos Veadeiros site
Year Sequences of revisit (days)
i6 | 33. [48 | &4 | $0 | 96 [112 [128 [144 [ 160
1996 | X X X X
1997 | X X X X X X
1998 X X X X X
2000 | X ¥ IX [7 X
Araguaia site
1996 | X X X X X
1997 X X X X
1998 | X | X T= X
OIX Ix Ix Ix [x X
Table 1. Availability of Landsat 5 imagery during the dry
season.
The permanence of the fire scars was estimated on a pixel by
pixel basis, by analyzing the spectral-textural variability of the
fire scars along the dry season of each year. This was carried
out by associating to each pixel in every image of a given year,
the label *Burned" (B) or " Background" (G), if the interpreter
could still identify the fire scar in the image, or not,
respectively. In a year where five images were available, the
sequence BBGGOG associated to a given pixel would indicate
that the fire scar was spectrally visible for a period of two
consecutive dates, and spectrally not discernible in the last three
acquisition dates. Areas covered by clouds and shadows were
disregarded from the analysis, as well as those associated to ‘B”
in the first date of the sequence.
This was due to the impossibility to determine the number of
days between the occurrence of the fire and the satellite
passage, thus creating uncertainties in the estimation. Hence,
the observation of the fire scars started from the second image
available (reference date).
245
4. RESULTS AND DISCUSSION
4.1. Area Burnt
The estimates of the burnt area indicated, for both Chapada dos
Veadeiros (CV) and Araguaia (A) sites, a systematic decrease
from 1996 to 2000, from approximately 4.186 km’ and 4,752
km” in 1996 to 1,327 km? and 2,138 km? in 2000, respectively.
The only exception to this pattern was the year 1998, with
increases of 214% and 246% in relation to 1997, for CV and A
sites, respectively. This increase may be due to the fact that
1998 was possibly the warmest year of the decade (IPCC,
2001), resulting in diminished soil moisture and water content
in the vegetation. In the CV site, the physiognomies most
affected by biomass burning were the campo cerrado and
parque de cerrado (which correspond to approximately 8196 of
this study area, being responsible for approximately 98% of the
total area burned). These physiognomies were also the most
affected ones in A site, covering approximately 61% of the arca
of this site and concentrating approximately 95% of the total
area burned.
4.2. Recurrence of Burns
Approximately 1696 and 1896 (4,703 km? and 5,373 km?) of the
total area of the cerrado, in the CV and A sites, were affected by
the recurrence of fires from 1996 to 2000, respectively. The
sequences characterized by a single recurrence, independently
of the year of the fire, were the most common (77% in CV site
and 6296 in A site) and in both areas, the single recurrence,
spaced by one year (ie., sequences of the type BGBGG,
GBGBG or GGBGB), represented approximately 68% in CV
site and 59% in A site. The second most frequent type of
sequence was that characterized by two recurrences (19% in CV
site and 31% in A site). It is interesting to note that in CV site,
approximately 896 of this type of recurrence was sequential (i.e.,
BBBGG, GBBBG, or GGBBB), whereas in A site, the most
frequent sequence (approximately 12%) was characterized by
intercalated burns (i.e., BGBGB). The sequence characterized
by four recurrences (BBBBB) was very minor (0.35% in CV
site and 0.79% in A site).
4.3. Mean Permanence of the Fire Scars
The analysis of the data indicated that the mean permanence of
the burnt scars is subject to significant annual variations. For the
Chapada dos Veadeiros, the mean permanence for the period
1996-1997-1998-2000 were 79, 55, 102 and 47 days.
respectively, whereas the corresponding data for Araguaia were
64, 96, 82 and 57 days. Regardless of the differences (in
absolute terms) of the mean permanence for each year in the
period analyzed, the physiognomy campo was the one that
consistently presented the longest mean permanence of the fire
scars, followed by the parque de cerrado and campo cerrado.
5. CONCLUSIONS
Although this study was carried out in two specific areas of the
Brazilian cerrado, it introduces a methodology that can be
applied to larger areas, being also adequate for use with data
from other types of sensors, such as the MODIS. In this case,
small burnt areas may not be detected due to the coarser spatial
resolution of the MODIS (250 metres) relative to the TM (30
metres). The range of mean permanence from one year to
another in a same area indicates the difficulties to consistently
characterize the permanence of the fire scars.