Mapping Fire Scars in the Ecotone of Amazon Forest - Cerrado biome using Remote Sensing: a
study case in Mato Grosso, Brazil.
Gabriel Antunes Daldegan
PhD Student, Geography Department, UCSB
• Why is this important: Use of Fire in Deforestation, Greenhouse Gases emissions, Climate Change.
• Previous Studies: • INPE (Setzer, Shimabukuro) UMD (Justice, Schroeder), IMAZON (Souza Jr), SDSU
(Roy).• Landsat - Fire Scars • ASTER, AVHRR, GOES,MODIS, VIIRS, – Active Fire
• Climatology: Dry Season – June, July, August.
• Geographical Variation: • Transition between the two major Brazilian biomes (variation of physiognomies,
soil,climate, occupation)• Agricultural Frontier – Human Occupation / Deforestation arc (Cattle, Soybean,
Mining);
Problem:
Objective:
• to map the spatial distribution and the temporal permanence of fire
scars in the study area by identifying and delimitating burned areas
present on Landsat 5TM scenes - Path 226, Rows 68 and 69 - from
June, July and August of 2005.
Methods:• Identify annual driest months;
• Search for Landsat 5TM imagery that have higher probability to show fire scars in the area;
• Consolidate the satellite imagery database (Georrectify, DN to Radiance, Reflectance Retrieval, Haze Correction);
• Classification (Decision Tree with two classes -Fire Scar/ No Fire- 800 pixels average),• 3x3 median filter,• Export to Shapefile;
• Visual interpretation and edition of the polygons representing burned areas;
• Analysis (Spatial-Temporal Patterns);
Methods:
• Dataset: 6 LANDSAT 5TM scenes from 2005.
Image -Path/Row
Date
226/68
6/8/2005
7/10/2005
8/27/2005
226/69
6/8/2005
7/10/2005
8/11/2005
Methods:
Decision Tree 226 68 08 June 2005
Classification tree:
snip.tree(tree = treeJun08, nodes = 3L)
Variables actually used in tree construction:
[1] "V5" "V7" "V6" "V3"
Number of terminal nodes: 9
Residual mean deviance: 0.1779 = 285.6 / 1605
Misclassification error rate: 0.02788 = 45 / 1614
Results:
Results:
Image -
Path/RowDate
Classification -
ha
Visual Edition -
ha
Difference -
ha%
226/68
6/8/2005 99,144.44 43,212.85 55,931.59 43.59
7/10/2005 147,659.22 68,851.27 78,807.95 46.63
8/27/2005 158,486.13 121,728.06 36,758.07 76.81
226/69
6/8/2005 45,820.98 28,567.05 17,253.93 62.34
7/10/2005 91,356.93 40,420.74 50,936.19 44.24
8/11/2005 109,984.68 50,407.30 59,577.38 45.83
Results: Confusions
Results:
Image -
Path/RowPolygons - Count
Fire Scars Total
Area Mapped- ha
226/68 2,175 112,428.36
226/69 2,564 119,395.11
Total 4,739 231,823.47
Results:
Image -
Path/Row
Fire Scar
June-July
ha
%
Fire Scar
July-
August
ha
%
Fire Scar
June-
August
ha
%
226/68 32,116.43 74.32 50,431.56 73.2530,286.92 70.09
226/69 17,477.89 61.18 23,928.03 59.2016,753.22 58.65
Validation:
Total Burned Area Mapped -ha
231,823.47
Burned Area Overlappedwith Active Fire-ha 146,300.46
% 63.11
Conclusions
• Decision tree approach has over mapped fire scars• About 47% of the mapped area were discarded;
• The principal source of confusion was agriculture bare soil;
• About 66% of fire scars could still be detected after 1 month and 64% after two months;
• The median of the polygons size is 2.16 hectares, which could indicate that the majority of fire were managed fires and did not spread to wider areas.
Future Studies
• Further classifications to figure out patterns,• Other sites,
• Other dates,
• Cross the results with active fire and fire scars maps(MODIS, VIIRS and other studies)
• Cross the results with Land Cover Use.
Acknowledgments
• Prof. Dr. Dar Roberts – Department of Geography, UCSB.
• CAPES – Science Without Borders Fellowship.
• Jack and Laura Dangermond – Dangermond Travel Fellowship.
• SCGIS
THANK YOU!