Discriminação da Castanheira (Bertholletia excelsa) com Análise de Mistura Espectral com Múltiplos Membros-finais aplicada a dados multiespectrais WorldView-2
Bertholletia excelsa is one of the most important species of Amazon. Its nut has a particular economic, social and environmental importance for the Brazil northern region, since it is one of the main product of exportation and the unique non-timber product completely extract from native forests by traditional communities. The aim of this research is to test the use of Multiple Endmember Spectral Mixture Analysis on high-spatial-resolution multispectral image from WorldView 2 sensor, in order to map the Bertholletia excelsa fraction, in a per-pixel basis, in Tefé, state of Amazonas, Brazil. The image had its continuum removed, and methods for endmember selection, as EAR-MASA-CoB (EMC) and Iterative Endmember Selection (IES), were tested. Fraction thresholds were also tested. By using field samples and pan- sharpening regions of interest were selected on the Bertholletia excelsa and other vegetation type’s crowns. They were randomly divided into training and validation samples. The best result was obtained on the continuum removed image, using endmembers selected through EMC, with fraction threshold between – 0,01 and 1,01, and RMSE ≤ 0,025. When defined a threshold of 56% for Bertholletia excelsa fraction, 65% of its validation crowns were rightly classified and 9,5% of the validation crowns from other vegetation types were erroneously classified. This result shows the good performance of the visible- near infrared data for mapping and discussing the spatial distribution of the target species.