Using optical water types for satellite monitoring of brazilian inland waters

• Ano de defesa: 2020
• Autor(a) principal: Edson Filisbino Freire da Silva
• Orientador(a): Evlyn Márcia Leão de Moraes Novo, Felipe de Lucia Lobo
• Banca de defesa: Mauricio Almeida Noernberg, Maycira Costa
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Instituto Nacional de Pesquisas Espaciais (INPE)
• Programa de Pós-Graduação: Programa de Pós-Graduação do INPE em Sensoriamento Remoto
• País: BR
• Link de acesso: http://urlib.net/sid.inpe.br/mtc-m21c/2020/03.03.16.25

One of the key issues of monitoring inland water quality is spatial-temporal sampling because water quality can rapidly change due to natural and anthropogenic influence. Remote sensing of inland waters is a reliable tool for monitoring water quality in large areas and time-series. However, the traditional method of calibrating bio-optical algorithms for limnological parameters (e.g., Chlorophyll-a (Chl-a), Total Suspended Matter (TSM), and Colored Dissolved Organic Matter (CDOM)) is limited to bio-optical characteristics of the study sites used for algorithm calibration. Consequently, bio-optical algorithms are not suitable for monitoring inland waters on a macro-scale level. On the other hand, monitoring Optical Water Types (OWT) has shown a macro-scale application, while those OWTs also represent changes in Chl-a, TSM, and CDOM concentrations. Thus, monitoring Brazilian OWTs could be a useful tool for water management on a wide scale. The objective of this study is to create a method for monitoring the water quality of Brazilian inland waters using OWTs. The study is described in three chapters; the chapter 3 assesses the uncertainties related to the merging of spectra measurements obtained under different protocols of computing remote sensing reflectance (Rrs); the chapter 4 describes the identification of Brazilian OWTs using hyperspectral in situ Rrs, which was acquired for water bodies encompassing a wide range of optical characteristics in Brazil; the chapter 5 describes the training of classification algorithms for detecting the OWTs using satellite sensors. In the chapter 3, it is shown that Rrs computed on Kutsers method is lower than that of Mobleys in all water types, with bias reaching up to -100%. Both methods allow satisfactory calibration of biooptical algorithms when they are used apart, but there is a significant accuracy reduction when both methods are mixed in the same database. Furthermore, almost half of the samples are labeled with different clusters depending on the Rrs method. Hence, merge both methods for calibrating biooptical algorithms is viable when a validation dataset is used, but spectral clustering should be avoided. In the chapter 4, a total of eight OWTs are computed based on Rrs shape and magnitude, which represent different optical and limnological characteristics of Brazilian waters. The OWT 1 represents transparent waters with low TSM, Chl-a, and CDOM concentrations; the OWT 2 represents transparent waters with moderate CDOM and TSM; OWT 4 is characterized by waters with algae bloom in aquatic system with moderate TSM concentration; OWT 5 is characterized by waters with algae bloom in low TSM concentration; OWT 6 is composed by waters with severe algae bloom density; OWT 7 is characterized by waters with the highest CDOM concentration; OWT 8 is waters with high TSM concentration; OWT 9 is waters with the highest scattering and TSM concentration. In the chapter 5, classification algorithms are trained for detecting the OWTs in satellite images of Sentinel-2 MSI, Landsat-8 OLI, and Landsat-7 ETM+. Sentinel-2 MSI has the best spectral resolution for classifying OWTs and exhibited satisfactory accuracy (Recall from 0.77 to 0.99) in satellite images. On the other hand, Landsat-8 OLI and Landsat-7 ETM+ classifications are profoundly affected by the overestimation of nearx infrared bands, causing weak accuracy water bodies characterized by algae blooms (OWTs 4, 5, and 6). In conclusion, the proposed have many applications, such as i) support of sampling design and survey campaigns; ii) detection of water quality anomalies caused by abrupt changes such in sediment loading and onset of algal blooms; iii) it could also be used for a census of Brazilian surface waters and provide reliable data in a macroscale level; last, iv) It could be used for improving the accuracy and the scope of semi-analytical algorithms based on Rrs by using the OWT in the calibration and validation process.