Monitoramento remoto de formigas cortadeiras com o radar sentinel-1 em plantios de eucalipto

• Ano de defesa: 2025
• Autor(a) principal: Oliveira, Ricardo Jair de
• Outros Autores: https://orcid.org/0000-0003-1060-9701
• Orientador(a): Santos, Alexandre dos
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Lavras; Escola de Ciências Agrárias de Lavras (ESAL)
• Programa de Pós-Graduação: Programa de Pós-Graduação em Entomologia
• País: brasil
• Palavras-chave em Português: Monitoramento florestal; Pragas florestais; Radar de abertura sintética; Sensoriamento remoto; Silvicultura de precisão; Aeronaves remotamente pilotadas (RPA); Forest monitoring; Forest pests; Synthetic Aperture Radar (SAR); Remote sensing; Precision forestry; Remotely Piloted Aircraft (RPA)
• Área do conhecimento CNPq: Ciências Agrárias
• Link de acesso: https://repositorio.ufla.br/handle/1/60649

Brazil has more than 9.93 million hectares of eucalyptus plantations and ranks among the world’s largest producers of pulp and paper. In this context, leaf-cutting ants constitute one of the main challenges for the sector, as their intense defoliation leads to loss of leaf area, reduced photosynthetic capacity, and decreased tree growth. Although the management of these pests depends on continuous monitoring, traditional methods are labor-intensive, costly, and exhibit low scalability, underscoring the need for more efficient operational approaches. This study evaluated the use of Sentinel-1 radar data and machine learning techniques to estimate canopy cover as an indirect indicator of defoliation caused by leaf-cutting ants, integrating SAR indices (RVI, RATIO, and RFDI) previously selected through collinearity analyses. Three supervised algorithms were compared: Neural Networks, Random Forest, and Support Vector Machine (SVM). SVMs particularly those with polynomial and radial kernels showed the best performance, with mean absolute errors close to 0.23 and high stability between training and validation. The results demonstrated that, even at a 10 m resolution, Sentinel-1 can capture structural changes in the canopy associated with herbivory, coherently reproducing patterns observed in high-resolution imagery (3 cm). The approach enabled the identification of defoliation gradients and the discrimination of healthy, moderately affected, and severely damaged areas, revealing strong potential for operational application in Integrated Pest Management. The development of this spatial monitoring process, based on big data and cloudbased virtual machine processing, supports early decision-making, reduces costs, and decreases the need for chemical inputs. Furthermore, it aligns with the requirements of environmental certifications (FSC, Cerflor) and the sustainable management guidelines proposed by the FAO, offering a technological solution for large-scale precision forestry.