Modelagem de precipitação máxima na bacia do rio São Francisco: uma comparação entre diferentes métodos de estimativa de parâmetros da função generalizada de valores extremos

Resumen

Studies of extreme rainfall have great relevance for water resource management. The Generalized Extreme Value (GEV) density function has assisted in the modeling of extreme natural events. There are several methods for inferring the parameters of the GEV function, such as the Maximum Likelihood Estimation (MV) and L-Moments (ML) methods. Inference Bayesian (IB) constitutes an alternative to the frequentist approach in that it allows combining the information given by random samples of maximum precipitation with prior information about the parameters in the estimation of models. However, few works use the Bayesian approach. The aim of this work was to present the modeling of the GEV function, with parameters estimated by the following methods: MV, ML and IB. With goodness-of-fit tests, the most suitable probabilistic model was defined, and prediction of daily maximum precipitation values for return levels 10, 25, 50, 100, 200, 500, and 1000 years was performed. The study area was the São Francisco river basin, and the data used in this study were obtained from the National Water and Sanitation Agency database. Using the AD test, it was verified that the  and IB methods were considered the most suitable for studies of annual daily maximum precipitation probability. It was found that the stations with less asymmetrical series, the IB method obtained better results while the most asymmetrical the ML method was better. It can be inferred that the IB method showed greater uncertainties for predicting rains with return time above 100 years.