MAPA DAS FITOFISIONOMIAS DO PARQUE ESTADUAL SERRA DE CALDAS NOVAS

Abstract

Classifying the Cerrado's phytophysiognomies is a challenge, due to the difficulty in clearly discerning phytophysiognomy boundaries and because of the spatial variability, spectral similarity, frequency and intensity of impacts. Monitoring the dynamics of land cover through remote sensing has shown considerable promise in identifying spatial and temporal patterns of phytophysiognomic diversity. In this context, the aim of this study was to identify and obtain parameters related to Cerrado plant formations using geoprocessing, remote sensing and machine learning techniques, with a focus on classifying the phytophysiognomies of the Serra de Caldas-PESCAN State Park. For this purpose, the Sentinel-2A satellite image was obtained for the dry season and Radiometric indices were extracted. Spatial polygons of the Gallery Forest, Vereda, Cerradão, Cerrado Típico, Cerrado Rupestre, Campo Rupestre, Campo Sujo and Solo Exposto phytophysiognomies were sampled in loco to generate the classification dataset. Cluster analysis and principal component analysis were carried out in order to select indices that were representative of the groups formed and not redundant. The selected indices were used as predictors in a supervised classification model, using the Random Forest classifier. The performance of the classifier and the quality of the data generated were evaluated by evaluating the overall accuracy, the Kappa Index and the confounding matrix with the test data. Kappa was 0.87 and accuracy of 90.5%. The confusions occurred mainly between phytophysiognomies of the same group of plant formation and in the transition areas. The classification allowed the preparation of a map with a higher level of detail in terms of vegetation classes than those currently available and used, such as those of MapBiomas and TerraBrasilis. It is estimated the predominance of Cerrado Rupestre (40%) and Cerrado Típico (37%). The map, also available in HTML format, can be used as a management tool for the park. The trained model can be extended to other regions. The technical product developed represents a significant contribution to the conservation of the biome, as well as to research on the quantification of species diversity and to the analysis of changes in plant formations.