Estimação indireta de quantis de enchentes extremas a partir de modelos chuva-vazão com emprego conjunto de um gerador estocástico de precipitação diária, análise bayesiana e distribuições limitadas superiormente

• Ano de defesa: 2015
• Autor(a) principal: Veber Afonso Figueiredo Costa
• Tipo de documento: Tese
• 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; Engenharia sanitária
• Link de acesso: http://hdl.handle.net/1843/BUBD-AC2GGJ

The estimation of extreme flood quantiles is absolutely necessary for the design of large hydraulic structures. However, their probabilistic description by means of conventional flood frequency analysis is a complex task, since the size of the available samples is usually small, which means that it is necessary the extrapolation to quantiles beyond acredible limit. An alternative to circumvent this problem is to proceed an indirect estimation of the flood quantiles with the use of rainfall-runoff models, since such an approach provides a more suitable characterization of the hydrological response of a catchment through simulation under distinct yet equally probable input conditions.In this research, it is proposed a Bayesian framework for this purpose. This framework comprises a new model for daily rainfall stochastic generation, which employs upperbounded distribution functions to describe extreme events, and a calibration method based on the generalized likelihood function, which synthetizes in a more accuratemanner the behavior of the parameters of the hydrological model. This structure of analysis has a physically based nature, which leads to more reliable estimates of extreme flood quantiles and their uncertainties.Applications of the proposed method were conducted for the Pará river basin, in the Brazilian state of Minas Gerais, and for the American river basin, in California. Results showed the efficiency of the stochastic generator in simulating the empirical characteristics of daily rainfall, as well as in the extrapolation of frequency curves to quantiles with very small exceedance probabilities. Furthermore, the rainfall-runoffmodel proved to be capable of reproducing even very large non-systematic floods, thus showing the suitability of the methodological framework proposed herein.