FOREST FIRES TO HELP DETECT SATELLITE

In the late 20th century is over, fires in Indonesia has become
world spotlight both from neighboring countries in Southeast Asia and European countries
and America. Along with the large amount of assistance from donor countries to
cope with forest fires, as FFMP-JICA (Japan), GTZ (Germany), EU (EU
Europe), USDA Forest Service (United States), the use of sensing technology
far have also increased. Only remote sensing data sources used
in general have a low spatial resolution to detect the location of
points of fire and smoke. In this case the image used is the image of NOAA
AVHRR (belonging to the United States) and GMS (Japan's) which have a successive
spatial resolution of 1 km
2
and 5 km x 5 km (thermal band). Use of satellite imagery with
fine spatial resolution such as SPOT and Landsat Thematic Mapper (TM) intensively
the operational scale is still rarely done. With better spatial resolution
and different spectral resolutions required a technique that is typical (special).
The use of high spatial resolution satellite imagery such as Landsat TM for evaluation
post-fire forest condition is still rare. Its use is limited
the mapping of forest cover classes such as general, primary forest, the former
felling, scrub and bare ground. In this paper the authors evaluate
of several techniques to detect changes in forest cover and land due to
fires using Landsat TM imagery.

Image data and the data supporting
To achieve the expected goal, in this study used the image of
Landsat Thematic Mapper (TM) multiwaktu recorded on June 16, 1997 and
February 28, 1998 in Riau Province. For locations in South Sumatra Province,
image used was recorded on July 10, 1997 and January 18, 1998. Image
multiwaktu have size 550 x 800 pixels is expected to provide information about
conditions of forest cover / vegetation before and after the occurrence of forest fires and
land. Other data such as checking the results of the field, our earth and our way of work concession
the data supporting the crucial success of this study.
Post-classification comparison (Post-classification Comparison)
Based on the results of the evaluation as presented in Table 1, although the level
accuracy resulting from the combination of more bands (5 and 6 bands)
give higher yields, increased precision does not occur sharply.
Instead use the band less than 3, will result in accuracy
Inadequate (less than 90%). Accuracy of evaluation as summarized in
Table 1, the use of three bands, namely the combination of 3-4-5 and 2-4-7 for the 1997 image to image
year 1998 is the optimal combination, with an accuracy of 97.22% and 94.91%.
The 1997 image classification, accuracy produced by a combination of two bands (ie
band 3 & 4) generates only 86.92% accuracy, while the combination of three bands
can produce 97% accuracy. Accuracy with a combination of 4 or 5 band (1-2-4-7 & 1 -
2-3-4-7) only increases the accuracy of about 2.2%. The same trend occurred
in the 1998 image classification, where the combination with the 3 band (ie 2-4-7)
generate sufficient accuracy is 94.9%. Combination with 4 or 5 bands
only improve the accuracy of 3%.
Both the image of the 1997 and 1998, the accuracy of the 3 bands mentioned
former is able to distinguish eight classes with a good closure, with the separabilitas
between 1900 (good) s / d 2000 (excellent), as follows:
 Logged over forest / secondary forest land  empty
 plantations (oil palm, rubber, HTI)  meadow / grass
  shrub Shrubs
  Water Agency Cloud
 cloud shadow
Principal Component Analysis Method Multiwaktu (AKUM)
The results of the evaluation of the "eigenvector" indicates that the new axis of 12
resulting in the transformation, either by the method of 12-d 12-d UMPC and SMPC
only 4 axes that summarize the information changes or no changes
land cover (Table 2). Based on the evaluation of the accuracy and separabilitas two methods,
the 12-d 12-d SMPC and UMPC known that the latter method (12 d UMPC)
with a combination of PC1-PC2-PC4 (DB-SG-DG) gives better results, where
all grade changes can be detected by either the accuracy (Kappa) 98.19% and
separabilitas between classes (transformed divergence) is greater than 1900.
12-d method capable UMPC index summarizes the changes are complete, the DB,
DG, SB and SG, while the 12-d SMPC only summarize DB, DG, and SG.
12-d method SMPC has a slightly lower capacity, which of
the class being tested, one pair of classes has separabilitas by category
"Less" (1714) and one pair of classes, including the category of "pretty" (1809). Kappa accuracy
obtained from the combination of PC1-PC4-PC5 (DB-DG-SG) is 95.04%. With
another, the use of various matrix-peragam provide separabilitas class becomes more
fine. Standardization process was removing some pixels that information
declare the existence of forest and land cover change.
The classification method Multiwaktu Direct / KML (Direct Multidate Analysis / DMC)
At this KML method, we evaluated the combination band of the 4 band multiwaktu
up to 12 bands. From each of the band, then selected the best,
in terms of both accuracy and separabilitas as presented in Table 3.
From Table 2, it is known that the combination of 4 band multiwaktu (397-497-398-498) was
provide adequate results, namely 99.20% and the average separabilitas (TDavg) greater than 1900. This suggests that in order to detect changes
land cover and forest fire, the use of a long band multiwaktu
wave the red (TM band 3) and near infrared region (TM band 4) is sufficient
reliable way to produce high accuracy. It is understandable that these classes
changes made strongly associated with changes in green biomass (green biomass) and
the amount of chlorophyll