MAPPING OF VEGETABLE COVERAGE FOR CARTOGRAPHIC UPDATE IN MARINGÁ/PR USING THE NDVI STATISTICAL APPROACH AND DECISION TREE

Abstract

This article aims to evaluate the use of orbital images from the MSI sensor, Sentinel 2A satellite, to map the vegetation in the area corresponding to the Maringá sheet (PR), chart SF-22-Y-D-II-3. The NDVI index was the spectral variable selected, and nine images were processed using Google Earth Engine between February 2021 and January 2022. The classifications were based on the statistical images of mean and standard deviation, instead of individual images. Two classification methods were tested, one by simple slicing, and the other method was the decision tree (algorithm J48), in which 413 reference points were used. The results showed that the decision tree classifier presented slightly better results than the simple slicing, with Kappa indexes equal to 0.893 and 0.877, respectively. The decision tree used the mean as the main variable, but when it was between 0.6678 and 0.7504 the pixels were also classified using the standard deviation. Simple slicing classified more areas as vegetation, while the decision tree classified less areas. While the first classifier would be more suitable for mapping conservation areas, regardless of the size of the vegetation, the second would be more suitable for mapping forest cover.