Modelos para predição de sólidos dissolvidos totais, sólidos totais e turbidez no reservatório de Jaguara a partir dos dados espectrais dos satélites sentinel-2

• Ano de defesa: 2025
• Autor(a) principal: Silva, Willian Geraldo da
• Orientador(a): Fia, Ronaldo
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Lavras; Escola de Engenharia (EENG)
• Programa de Pós-Graduação: Programa de Pós-Graduação em Recursos Hídricos
• País: brasil
• Palavras-chave em Português: Sensoriamento remoto; MSI/Sentinel-2; Qualidade da água; Aprendizado de máquina; Modelagem preditiva; Remote sensing; Water quality; Machine learning; Predictive modeling
• Área do conhecimento CNPq: ENGENHARIAS::ENGENHARIA SANITARIA::RECURSOS HIDRICOS
• Link de acesso: https://repositorio.ufla.br/handle/1/60347

The objective of this study is to develop models to estimate water quality parameters such as Total Dissolved Solids (TDS), Total Solids (TS), and Turbidity in the UHE Jaguara reservoir. To achieve this, reflectance data from the MSI/Sentinel-2 sensor were used in conjunction with in situ measurements, aiming to provide more efficient and accurate water quality monitoring in this important reservoir. The input data were obtained from 21 sampling campaigns conducted between 2018 and 2023, evaluated in scenarios with and without outlier removal. The results indicated that machine learning models performed better in predicting the three parameters. The Random Forest model using band 2186 was the most effective for TDS (r = 0.84), while the 559/864 band ratio combined with the Decision Tree model stood out for TS (r = 0.51). For turbidity, the Random Forest model using all bands achieved r = 0.70. The analysis of the predictive maps confirmed the field data, showing that the highest concentrations of TDS, TS, and turbidity occurred during the rainy season, particularly in the water inflow areas of the reservoir. In addition to rainfall, factors such as effluent discharges and land use changes also influenced water quality. The combination of remote sensing data and machine learning techniques proved effective in predicting water quality parameters. This enabled more accurate estimates and continuous visualization of TS, TDS, and turbidity concentrations, representing an advancement over traditional sampling methodologies. The application of this approach in the UHE Jaguara reservoir shows potential for expansion to other reservoirs, particularly in areas with varying hydrological conditions.