Métodos de sensoriamento remoto orbital para o diagnóstico e monitoramento dos níveis de degradação de pastagens tropicais: uma contribuição ao plano de agricultura de baixa emissão de carbono

• Autor(a) principal: Almeida, Mateus Benchimol Ferreira de
• Data de Publicação: 2022
• Outros Autores: [e-mail]
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações da UERJ
• Texto Completo: http://www.bdtd.uerj.br/handle/1/20404
• topic: Vegetation indices; Degradação ambiental - Valença (RJ); Vegetação - Valença (RJ); Senoriamento remoto - Valença (RJ); Cerrados - Brasil; Sentinel-2; Sentinel-1; Índices de vegetação; NCI; CIENCIAS EXATAS E DA TERRA::GEOCIENCIAS::GEOFISICA

The present scientific research aimed to develop a methodology for mapping and monitoring the different pastures degradation levels (L1, L2L3 and L4) in the Atlantic Forest (specifically Valença/RJ municipality and surroundings) and in the Cerrado biome, based on techniques of Remote Sensing and Digital Image Processing from Sentinel-2 and Sentinel-1 satellites, in addition to other satellite products, as well as georeferenced field points with their respective assigned classes. This dataset was classified by a supervised classification method called random forest, which in turn is an artificial intelligence and machine learning algorithm that uses different decision trees for the labeling of pixels in the classification process. In addition, random forest in conjunction with Google Earth Engine is even more powerful, due to parallel computing and cloud computing provided by the Google Earth Engine platform. The main data processing was performed in Google Earth Engine, but the auxiliary processing, which greatly favored the main processing, was performed in R. The most important vegetation index was NCI, however, by itself, it didn’t guarantee a satisfactory accuracy, being necessary to add other remote sensing products for this purpose. The present scientific research also aimed to estimate the proportions of the elements present in the pastures composition at their different degradation levels (L1, L2L3 and L4), based on Spectral Mixture Analysis (SMA) and on NPV (Non-Photosynthetic Vegetation), PV (Photosynthetic Vegetation), BS (Bare Soil) and Shade fraction images from the SMA for both the wet and dry seasons. Based on these fractional images, it was possible to perform a random forest classification of the different pastures degradation levels in the Cerrado biome. For estimation of NPV, PV, BS and Shadow proportions, it was necessary to apply the Principal Component Analysis (PCA) before the SMA in order to reduce the correlation among the pure bands of Sentinel-2, even so, it was only possible to estimate the proportions in the wet season. The results obtained based on the proposed methodologies and on the statistical control of the data (accuracy) allow us to deduce that both are robust, efficient and effective in separating the different pastures degradation levels (L1, L2L3 and L4) both in the Atlantic Forest (Valença/RJ Municipality and surroundings) and the Cerrado biome