Avaliação de distribuições de probabilidades das vazões médias díarias máximas anuais do Brasil

• Ano de defesa: 2014
• Autor(a) principal: Karoline Tenório da Costa
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; UFMG
• Palavras-chave em Português: Meio Ambiente e Recursos Hídricos; Saneamento; Estações fluviométricas; Recursos hídricos Desenvolvimento; Engenharia sanitária
• Link de acesso: http://hdl.handle.net/1843/BUOS-9KHHCT

The frequency analysis of hydrologic and climate data has been the main tool used by engineers to estimate the environmental risk of civil works. One of the steps necessary for frequency analysis consists of choosing a probabilistic model to estimate flows for certain return periods based on observed data. However, this step has been performed subjectively in Brazil, which may compromise the use of this method. In this paper, several probabilistic models were evaluated to indicate which one was the best fit to Brazilian runoff data in order to stimulate discussion about the standardization of the methods used in the frequency analysis. The data were collected automatically from the Web Service of the National Water Agency (ANA), according to pre-defines criteria to eliminate years with outliers and big gaps. The adequacy of the 1943 samples was ensured through the application of criteria to eliminate stations with a large number of missing data and the Mann Whitney, Spearman and Pettit tests, to test the homogeneity, stationarity and shifts in the mean and variance, respectively. The Akaike information criterion (AIC), Bayesian information criterion (BIC) and a goodness-of-fit measure based on the Anderson Darling test, , the diagram of L-moments ratios, the method of Beard and the SEAF software were used to assess which probabilistic models are the most suitable to Brazilian data. The results obtained with the application of such procedures show that the most appropriate probability distributions for the Brazilian data are the two and three parameters log-normal distributions and that the only distribution among tested proved not adequate to the Brazilian data was the generalized Pareto. These methods were also applied to geographical regions and regions with similar characteristics defined by cluster analysis in this study. The results obtained from this analysis were more diversified and therefore were not as conclusive, although the log-normal distributions of two and three parameters were always among the most suitable in most regions. In the last sections of this work, the study was focused on the floods of records. Envelopes curves for maximum floods were developed for Brazil and for regions with similar characteristics defined previously in this work and their empirical equations were defined as a function of the drainage area. Moreover, the probability of exceedence of each record was estimated.