Desempenho de índices de vegetação e atributos texturais na análise espaço-temporal e classificação do uso e cobertura da terra em áreas de enclaves da Caatinga

Abstract

The study of the performance of vegetation indices and textural attributes in Caatinga enclave areas allows a more precise analysis of land use and land cover dynamics. The spatiotemporal approach, using remote sensing data, allows the identification of variations in vegetation over time. In this context, analyzing the performance of vegetation indices and textural attributes for the discrimination and classification of land use and land cover classes in an enclave area of the Caatinga allows for the evaluation of different analyses on the biome's response to anthropogenic and climatic pressures, contributing to a more detailed and efficient mapping of these areas, which historically lack more accurate environmental information. For this research, two scientific studies were developed, in the form of articles, with the municipality of Mucugê, in Chapada Diamantina, Bahia, as the study area. The first study aimed to evaluate the performance of four spectral bands and three vegetation indices derived from PlanetScope monthly mosaics in discriminating different phytophysiognomies of Caatinga enclaves and agricultural crops in the municipality. The second study aimed to analyze the potential of monthly mosaics from the PlanetScope satellite in classifying temporary and permanent agricultural crops. In the development of the articles, the monthly mosaics from the PlanetScope satellite, the vegetation indices NDVI, EVI and NGRDI and textural attributes were used. In addition, field data on the different phytophysiognomies and agricultural crops present in the study area and non-parametric supervised classifiers, such as CART and Random Forest, were used. The results obtained in the first article demonstrated the sensitivity of vegetation indices and bands to natural vegetation areas, even during the dry season. In temporary crops, the IV profiles exhibited high amplitude and a short crop cycle, while in permanent crops, such as coffee, there were variations in the profiles due to different agricultural practices. In the second article, the results obtained showed high overall accuracy, especially during the dry period, reaching up to 92.03% (AO), with an estimated error of 5.08%, as well as the importance of textural attributes derived from the four available spectral bands (R, G, B, NIR) particularly the mean and variance in the classification.