Sensoriamento remoto de parâmetros hidráulicos geométricos para estimativas da vazão de um rio de médio porte

• Ano de defesa: 2020
• Autor(a) principal: Évelyn Márcia Pôssa
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; Brasil; IGC - DEPARTAMENTO DE GEOGRAFIA; Programa de Pós-Graduação em Geografia; UFMG
• Palavras-chave em Português: Sensoriamento remoto; Vazão; Rio de médio porte; Largura fluvial; Imagem de satélite; Altimetria radar; Extração de superfície de água; Sensoriamento remoto – São Francisco, Rio; Vazante – São Francisco, Rio; Imagens de sensoriamento remoto – São Francisco, Rio; Hidráulica – Modelos; Nível da água – São Francisco, Rio
• Link de acesso: http://hdl.handle.net/1843/34343; https://orcid.org/0000-0003-2617-9136

In view of the increasing number of Earth observing missions, remote sensing stands ever more as an alternate means of acquiring hydrological data in regions where these data are difficult to obtain or infrequent because of logistical, political or economical reasons. The fluvial regime of a river can now be monitored and measured from a number of remote sensing techniques and instruments (imaging and altimetry) in diverse spectral regions and spatial, radiometric and temporal resolutions. So far the technology and parameters of these sensors still offer limited use for smaller rivers (width < 800 m) and this represents many research challenges for such hydrological applications. Within this context, the central objective of this thesis circles around the investigation of inferring river discharge from time series of hydraulic geometry variables extracted from satellite multi-mission and multi-sensor data. A number of methodological approaches are suggested to reach this objective. The first phase of the thesis consists in invertigating the level of accuracy attainable in delineating the water surface using image data from the Sentinel-1, Sentinel-2 and Landsat-8 missions and a batery of classification algorithms (k-means, Expectation Maximization, Random Forest, k-nearest neighbors, Maximum Likelihood Classification, Support Vector Machine, Mahalanobis ). The results show that accuracies better than 90% were reached with the three types of image data and with relatively simple classification algorithms. In the second phase, the method aims at estimating the discharge of a medium-sized river by combining time series of river width (extracted from Sentinel and Landsat images) and power functions from Hydraulic Geometry Theory. About 70% of the models generated by the approach based on the concept of virtual station, produced NSE values of 0.7 or better. In the third phase, the width time series was complemented with water stage time series from satellite radar altimetry missions to estimate river discharge using a modified version of the Manning equation. Here, the challenge resided in evaluating the contribution of altimetry considering some of its inherent limitation in continental waters, especially medium-sized rivers. The best of these models reached NSE values between 0.5 and 0.77. Inserting altimetry data had a positive effect in reducing prediction errors for discharges situation with flow permanence of 57% of more when compared with using only the river width. Overall, the study was well succeeded in using multi-mission, multi-sensor remote sensing data to estimate discharge in rivers of medium width. In addition some attention was given on the effect of the morphology of the river channel of the discharge estimates.