Classificação supervisionada de imagens Sentinel-2 para identificação de floresta no parque natural de Montesinho

Abstract

Currently, with the advancement of technologies and the ease of obtaining satellite images, the monitoring and spatial analysis of terrestrial vegetation, with reduced costs and efficiency, make it possible to assemble profiles of multitemporal and seasonal images of the earth's surface. This work aims to identify the best time for classification and calculation of forest vegetation indices in the Montesinho Natural Park (PNM) and to locate its forests of greatest natural value holm oak (Quercus rotundifolia), black oak (Quercus pyrenaica Willd. ), maritime pine (Pinus pinaster Aiton) and chestnut (Castanea sativa Miller) using remote sensing and geoprocessing techniques. All data processing was performed in QGIS 3.18.1 software using SCP (Semi-Automatic Classification Plugin). Vegetation indices and the supervised Random Forest (RF) classification were used. For the production of maps, training samples (ROI) were collected, based on the support material: land use and occupation charts (COS), and the national forest inventory (IFN) of 2015, made available by the ICNF, where it was possible to remotely identify the forests of the studied species. To validate the results, the Kappa Index and GlobalAccuracy were calculated for the three months studied: the month with the highest index was July (63%), followed by February (41%) and October (15%). We can relate these results with the season of the year of the images, being those of winter and autumn, where the plants are in vegetative rest. It is concluded that the classification made by the RF for the identification of the species of interest fits better with the real (IFN) in the month of July. However, more study applied to the area is needed.