INCERTEZA NA ANÁLISE DE FREQUÊNCIA DE VAZÕES MÁXIMAS ANUAIS UTILIZANDO MODELOS ESTACIONÁRIO E NÃO ESTACIONÁRIO APLICADOS À REGIÃO SUL DO BRASIL

Código

END144

Autores

Debora Yumi de Oliveira, Daniel Bartiko, Pedro Luiz Borges Chaffe

Tema

1 - Previsão e mapeamento de inundações

Resumo

The detection of trends in annual maxima series has prompted the development and use of nonstationary models for flood frequency analysis. However, whether we should use (and how to use) nonstationary models for prediction is still under discussion. In this work, we compared the use of stationary and nonstationary models for flood frequency analysis using the lognormal distribution. This analysis was conducted using discharge records of fluviometric gauges located in Southern Brazil. The parameters of the lognormal distribution were inferred using Bayesian inference. We found that model selection based on information criteria or maximum a posteriori parameter estimates may lead to the selection of a nonstationary model for prediction, especially when a trend is detected in the fitting period. However, taking into account the uncertainty associated with the estimated parameters, the nonstationary model was never preferred. This result indicates that the higher uncertainty associated with nonstationary models is still a limitation for its extrapolation and use for predicting return periods of extreme fluviometric events.