Métodos de sensoriamento remoto no mapeamento de veredas na APA Rio Pandeiros

Abstract

Researches aimed to the use of Remote Sensing in the characterization, delimitation or distinction of veredas from the others Cerrado phytophysiognomies are still scarce. Due to it, this work aimed to distinguish the vereda subsystem, in the Pandeiros River Environmental Protection Area (EPA), through products and techniques from remote sensing. For this, it resorted from a range of orbital products like RapidEye optical images REIS sensor (RapidEye Imaging System), Landsat 8 images OLI sensor (Operational Land Image) and data from SRTM model (Shuttle Radar Topographic Mission). These products were prepared, through techniques of Digital Images Processing (DIP) in order to receive automated operations. Therefore, it has become possible to generate a set of 18 attributes composed for: bands of OLI and REIS sensors, vegetation indexes (NDVI, SAVI, EVI, NDWI), spectral mixing components and fraction images resulted from transformations for principal components. It should be highlighted that, in the methodological framework, it was necessary carrying out the prior knowledge of the spatial reality of vereda subsystem with field visit. Given this situation, in laboratory, for helping in the distinction of vereda class, it has been constructed an influence zone (500m buffer) along the drainage network. Posteriorly, it was carried out the image classification through Maximum Likelihood Method and Decision Tree. The achieved results allowed the comparative verification of the discriminatory power of both classifiers in the vereda mapping. The classification accuracy was performed through the classification error matrix and Kappa coefficient. For so, with help of GPS receiver system, it was carried out three campaigns in the study area with the purpose of collecting the field truth and validate the results. The classification by Decision Tree method reached a more adjusted result in comparison to maximum likelihood method, pointing out a 93% total accuracy and 0.9190 kappa coefficient. In relation to the individual classes accuracies, the decision tree classifier showed a satisfactory result concerning to the identification of veredas. However, some areas mapped as vereda do not match fully with reality, presenting spatial errors, not compatible with kappa indicator. It is just the case of some regions around the main tributary of Pandeiros river which were included, but are not characteristic environments of the subsystem. However, the found errors do not derail the use of decision tree model proposed by this study. So, it is hoped that the methods and the found results on this study contribute effectively for the mapping, conservation and recovery of vereda subsystem, an environment of major ecological relevance.