Premium
Modeling multivariate extremes of precipitation series in northern Moravia
Author(s) -
Jarušková Daniela
Publication year - 2009
Publication title -
environmetrics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.68
H-Index - 58
eISSN - 1099-095X
pISSN - 1180-4009
DOI - 10.1002/env.979
Subject(s) - multivariate statistics , series (stratigraphy) , precipitation , mathematics , statistics , monte carlo method , maximum likelihood , nested set model , multivariate normal distribution , econometrics , statistical physics , meteorology , geography , computer science , physics , geology , paleontology , database , relational database
Four trivariate models are considered to model tail dependence among daily precipitation amounts measured at different meteorological stations. Models are described, it is shown how to generate realizations from them by Monte Carlo method and their properties are compared. It appears that one of the models that belongs to the class of nested logistic models is the best model for all studied triples of stations when values of the maximum likelihood function serve as a criterion. The models are applied to estimate the probability that daily precipitations at several stations exceed some high values as well as the probability that the total of precipitations exceeds some high level. It is interesting that the differences between estimated probabilities obtained from different models are very small but the exceedance probabilities calculated from the considered models are always slightly larger than the corresponding observed frequencies. It seems that for our data the extremal multivariate distributions are good, but far from being perfect, as is often the case in applications. Copyright © 2009 John Wiley & Sons, Ltd.