Utilização de inteligência artificial e sensoriamento remoto para estimativa de biomassa, carbono e CO2 equivalente acima do solo em sistema agroflorestal

• Ano de defesa: 2023
• Autor(a) principal: Stangarlin, Maurício Pegoraro
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Santa Maria; Brasil; Recursos Florestais e Engenharia Florestal; UFSM; Programa de Pós-Graduação em Engenharia Florestal; Centro de Ciências Rurais
• Palavras-chave em Português: Redes Neurais Artificiais; Sensoriamento remoto; Agroflorestas; Artificial Neural Networks; Remote sensing; Agroforestry; CNPQ::CIENCIAS AGRARIAS::RECURSOS FLORESTAIS E ENGENHARIA FLORESTAL
• Link de acesso: http://repositorio.ufsm.br/handle/1/31145

On the North Coast of Rio Grande do Sul, there are silvi-banana agroforestry systems (SAF), based on the cultivation of banana trees (Musa sp), palm trees (Euterpe edulis Mart.) and tree species from the Atlantic Forest. The objective of this work was to obtain models, adjusted by Artificial Neural Networks (ANNs), that estimate carbon dioxide equivalent (CO2eq), using remote sensing data (SR), being vegetation indices (IV) and reflectance data, obtained with different sensors. A 2.37 ha SAF located in Morrinhos do Sul was used as an experimental area. In August 2022, a scheduled flight was carried out with a remotely piloted aircraft (ARP) Phantom 4 Advanced Plus (DJI) containing an RGB sensor, to obtain reflectance data and vegetation indices, the processing was carried out with the Agisoft Metashape software. A forest inventory was carried out, with 32 sampling units (ua) installed using the Prodan sixth tree method, georeferenced with a GNSS navigation receiver. In these waters, tree and palm species were identified, CAP (cm) and height (m) were measured, and CAP (cm) was measured in banana trees. CAP data were transformed into DAP (cm) and together with height (m) were used to calculate biomass to estimate CO2eq above ground, using specific allometric equations for each species. To complement the SR data, free images, free of cloud cover, were acquired at the SAF location under study, from the MSI Sentinel 2A sensor and the WPM CBERS 4A sensor. In QGIS, the IVs were calculated and images of the spectral bands (RGB, NIR and RedEdge) and ARP were inserted, the limits of each ua were generated, and using the zonal statistical tool (EZ) the minimum, average and maximum values were obtained. for all available SR data. Multivariate Factor Analysis (FA) was used to understand the relationships between the variables, using the Statistica12 software. To estimate CO2eq, ANNs were used, aiming to obtain a simple and practical model with the lowest possible residues. AF showed no significant relationships in CO2eq and SR data, factors 1 and 2 were defined by the Sentinel 2A and CBRS 4-A sensors, indicating difficulty in modeling from optical SR data, therefore, the use of ANN, if justifies. Through ANNs, a complete model was generated with all SR variables and a reduced one using the SR data that presented importance above 70% in the complete model, the models presented RMSE of 22.92 and 50.20, Bias of -2, 11 and -1.85 and a coefficient of determination in cross-validation of 0.99 and 0.95 respectively, indicating precision in the estimated CO2eq results in relation to the CO2eq value obtained by the indirect method. It was concluded that the models are efficient to estimate CO2eq using SR data in SAFs in the study region and even though the complete model presented better statistics, the reduced model is still accurate and offers an effective and practical tool for using a smaller number of variables.