Detection of deforestation and land conversion in Rondônia, Brazil using change detection techniques

Abstract

Fires associated with tropical deforestation, land conversion and land use greatly contribute to emissions as well as the depletion of carbon and nutrient pools. The objective of this research was to compare change detection techniques for identifying deforestation and cattle pasture formation during a period of early colonization and agricultural expansion in the vicinity of Jamari, Rondônia. Multi-date Landsat Thematic Mapper (TM) data between 1984 and 1992 were examined in a 94 370 ha area of active deforestation to map land cover change. The tasselled cap (TC) transformation was used to enhance the contrast between forest, cleared areas and regrowth. TC images were stacked into a composite multi-date TC and used in a principal components (PC) transformation to identify change components. In addition, consecutive TC image pairs were differenced and stacked into a composite multi-date differenced image. A maximum likelihood classification of each image composite was compared for identification of land cover change. The multi-date TC composite classification had the best accuracy of 0.78 (kappa). By 1984, only 5% of the study area had been cleared, but by 1992, 11% of the area had been deforested, primarily for pasture, and 7% lost due to hydroelectric dam flooding. Finally, discrimination of pasture versus cultivation was improved due to the ability to detect land under sustained clearing opposed to land exhibiting regrowth with infrequent clearing.

Acknowledgments

The authors would like to acknowledge funding for this research by the NASA Ames Graduate Students Researchers Program and data made available by the EROS Data Center DAAC. Additionally, we greatly appreciate scientific guidance from Beverly Law, Dan Edge, George Stankey and Christine Hlavka, and reviews of this manuscript and valuable input from Louisa Beck, Jennifer Dungan and Dave Peterson.