Spatial deforestation patterns and the accuracy of deforestation mapping for the Brazilian legal amazon

Abstract

Since 1988, the Brazilian National Institute for Space Research (Instituto Nacional de Pesquisas Espaciais - INPE) has been executing the annual inventory of forest cover loss in the Legal Amazon using satellite data under the Program for Deforestation Monitoring in the Brazilian Legal Amazon (PRODES). This survey comprises mapping more than four million squared kilometers and the produced information are employed by the Brazilian government to evaluate and establish public policies related to the control and counter deforestation. This data has been produced and published annually since 1988. However, its accuracy is not known. Therefore, this work objective is to develop a methodology to estimate the accuracy of the deforested areas mapped by PRODES for the year 2014 using stratified sampling of the deforestation patterns mapped in 50 by 50 Km cells. Mapping these patterns was accomplished by establishing and using a typology of deforestation patterns, landscape metrics and data mining techniques. Typical sample points of each deforestation pattern of were randomly drawn and analyzed visually by independent experts. This evaluation results established that the global accuracy level of the mapping under study is estimated to be 93%, with omission and commission indices estimates being 7% and 1.5%, respectively. Patterns such as fishbone, multidirectional and consolidated ones, which are considered the most complexes, present the lowest indexes of correctness, showing coherence and indicating that they should be mapped with more rigor. The presented results are consistent in a general way, indicating that the developed methodology can be applied to similar mappings.