Dados do Trabalho

Título

MODELANDO PRECIPITAÇAO EXTREMA NO BRASIL PELA TEORIA DOS VALORES EXTREMOS

Data de titulação

26/08/2016

Instituição de titulação

Universidade Estadual de Maringá

RESUMO (abstract)

The accurate modeling of extreme events is growing in relevance in the light of climate change. In particular, quantifying unusual rainfall is important for the design of hydraulic structures and for flood mapping and zoning. The Brazilian regulatory agency, Agência Nacional de Águas (ANA), has rainfall time series for 11,368 rain stations throughout Brazil, some of them initiated in the 19th century. The main goal of this work was to use the framework of extreme value theory to produce maps with reliable estimates of 25-year rainfall return levels for the regions covered by ANA. We incorporate previous information into a Bayesian setting and use a spatial latent variable model for the annual rainfall maxima. This model accounts for subtle patterns of spatial variation in the data and provides plausible return level estimates that depend on spatial location; in some cases, the mean annual precipitation explains some of the variation. In some regions extreme rainfall is associated with phenomena such as El Niño-Southern Oscillation. We also complement previous studies on biases from model misspecification or from small sample sizes when estimating the shape parameter of the extreme-value model, which seems to have a mean asymptotic value around 0.06.