Acessibilidade / Reportar erro

Criteria For Choosing Probability Distributions in Studies of Extreme Precipitation Events

Abstract

This study aimed to establish a criterion on which good of fit test should be preferred in the choice of probability distribution functions (pdf). For this, the Gumbel (GUM), Generalized Extreme Value (GEV) and 2-parameter Log-Normal (LN2) were fit Moments, L-moments and maximum likelihood methods, in annual maximum daily precipitation series of 11 rain gauges located in the hydrographic basin of the Sapucaí river. The adherence of these fdp to the data was made by the Kolmogorov-Smirnov (KS), Chi-square (χ2), Filliben (Fi) and Anderson-Darling (AD) tests. It was verified which good of fit tests are more rigorous in the selection of pdf and also the good of fit test that converge the results to the choice of the best performing fdp in the uncertainty analysis and/or statistics performance. The most stringent good of fit test on the adherence of fdp to the data are the Fi and AD tests. The Fi test, followed by the χ2 test showed best performance in terms of goodness of fit and convergence therefore, they should be preferred to others tests. The GUM and GEV are great pfd in representing the annual maximum precipitation data.

Keywords:
goodness of fit test; probabilistic modeling; intense rainfall; statistical inference

Sociedade Brasileira de Meteorologia Rua. Do México - Centro - Rio de Janeiro - RJ - Brasil, +55(83)981340757 - São Paulo - SP - Brazil
E-mail: sbmet@sbmet.org.br