Close Window

Mapping Burned Areas in the Brazilian Amazon Using Modis Data

Andre Lima, Instituto Nacional de Pesquisas Espaciais, (Presenting)
Yosio Edemir Shimabukuro, Instituto Nacional de Pesquisas,
Ramon Morais Freitas, Instituto Nacional de Pesquisas,
Luiz Eduardo Aragao, Oxford University,
Douglas Christopher Morton, University of Maryland,
Liana Oighenstein Anderson, Oxford University,
Bernardo Rudorff, Instituto Nacional de Pesquisas Espaciais,
Marcos Adami, Instituto Nacional de Pesquisas Espaciais,

Although global algorithms (e.g. Globscar, GBA2000, MCD45) were frequently used to identify burned areas in specific parts of the Amazon region, no basin-wide map has yet been carried out. The aim of this work is to map burned areas in the Brazilian Amazon using images from MODIS aboard Terra platform. The study area includes the States of Acre, Rondonia, Mato Grosso and Pará, sometimes referred to as the “arch of deforestation”. The image classification approach is based on INPE’s Digital PRODES project that applies image segmentation to the fraction images derived from a linear spectral mixture model. MODIS data (MOD09) at 250 m spatial resolution were acquired between July and November 2005. A dataset from the DETER (Detection of Deforested Areas in Real Time) project from year 2004 was used to classify the land cover types affected by the fires. Landsat-TM and CBERS-CCD images, acquired during 2005, along with field observations were used as reference data to validate the results derived from MODIS images. The map of burned areas was submitted to a visual interpretation to reduce commission and omission errors originated during the digital classification procedure. Results from this multitemporal MODIS images analysis showed that during the year of 2005 a total of 76,200 km2 (3% of study area) was burned in these four states. From this total, 27,400 km2 (36%) corresponded to areas deforested prior to 2004, 30,179 km2 (40%) occurred in the Cerrado biome, and 27,400 km2 (26%) was associated to new deforestations and/or burned forested areas. Cloud-free time series of multitemporal MODIS images are a viable option to map scars of burned areas in the Brazilian Amazon. Regional maps of burned areas, classified by land cover type, are an initial and crucial step to estimate carbon emissions from fires during the drought of 2005 in the Amazon region.

Science Theme:  LC (Land Use and Land Cover Change)

Presentation Type:  Poster

Abstract ID: 29

Close Window