Historical maximum rainfall data are used to forecast extreme rainfall, which is important to elaborate agricultural and hydraulic engineering projects. Generalized Extreme Value Distribution (GEV) has been applied in such type of studies. Since those values are extracted from the upper (or lower) tail of the original distribution, a scarce amount of data is obtained in most cases, which may be a problem acquiring reliable estimates about some measure of interest. An alternative to overcome this potential problem would be the use of information available from experts in the area. Therefore, this paper intended to analyze the application of the Bayesian Inference using a priori distribution based on extreme quantiles, which facilitates the incorporation of the information supplied by the experts in order to determine the punctual and the 95% upper limit estimates of the probable maximum precipitation for return periods of 10 and 20 years, yearly and monthly in Jaboticabal, São Paulo State, Brazil. Markov Chain Monte Carlo (MCMC) methods were used to a posterior inference of each parameter. Bayesian inference yielded more suitable and accurate results in the estimation of the parameters of the GEV distribution as well as in the determination of the values of the probable maximum precipitation estimates for Jaboticabal. It turned out as an interesting way of incorporating prior knowledge to the study of extreme data.
Bayesian Inference; prior distribution; prior knowledge; extreme quantiles Markov Chain Monte Carlo method; Maximum rainfall
