Utilização de inteligência artificial e sensoriamento remoto para estimativa de biomassa, carbono e CO2 equivalente acima do solo em sistema agroflorestal ; Use of artificial intelligence and remote sensing to estimate biomass, carbon and CO2 equivalent above ground in agroforestry system

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 ...