Comparação de técnicas de extensão de séries hidrológicas

• Ano de defesa: 2016
• Autor(a) principal: Lucas Filipe Lucena Jesus
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; UFMG
• Palavras-chave em Português: Eventos extremos; Hidrologia modelos matemáticos; Hietogramas de projeto; Distribuição temporal de precipitação; Recursos hídricos Desenvolvimento; Engenharia sanitária; Hidrologia Modelos matematicos
• Link de acesso: http://hdl.handle.net/1843/BUOS-AV2NF4

Hydrologic frequency analysis is a extensively studied research topic in the literature, due to the relevant role it plays in the practice of water resources engineering and to several uncertainties inherent to the traditional methodology. One of the sources of these uncertainties is the short extension of the reduced hydrologic series used as input to frequency analysis. Several record extension techniques have been developed since the middle of the last century, such as the group named Maintenance of Variance Extension (MOVE1 to MOVE4) and the techniques KTRL2 (Kendall-Theil Robust Line 2) and RLOC (Robust Line of Organic Correlation). Along with regression models OLS (Ordinary Least Squares), KTRL and GLM (Generalized Linear Models), such techniques have been studied with the objective to extend reduced hydrologic series in order to obtain more accurate time series than the original ones with respect to the estimation of the population moments, leading to improvements in the estimates of frequency of extreme hydrologic events. All models studied are based on a reference time series (rainfall or streamflow records) used to extrapolate the observations of the short time series. The methodology consisted of Monte Carlo simulations encompassing several scenarios of short and long series extension, and the linear correlation between them. Also, scenarios covering the distribution of the data (bivariate Normal, Gumbel and Pearson 3 with positive and negative skewness) were simulated. Results evaluated were bias, variance and error in the estimation of the population mean, variance, skewness coefficient and quantiles, along with the error and the bias in the estimation of the individual records. These results were used for the definition of criteria for the practical application of record extension techniques, based on the estimators theory. The criteria developed pooled the techniques that generated extended records with superior performance than the short series with respect to the estimation of population moments. Results show that KTRL and GLM regression models are not suitable for the extension of the hydrologic time series. In the estimation of individual records, OLS outperformed the other techniques. In the other aspects, the MOVE techniques performed best. The developed criteria were tested by means of the application on real data. It was concluded that the record extension techniques are able to generate extended records with better estimates of the population descriptive measures than the short series