NON-STATIONARY HYDROLOGIC FREQUENCY ANALYSIS USING TIME DEPENDENT PARAMETERS AND ITS MODEL SELECTION

Abstract

 A model selection method is discussed for non-stationary hydrologic frequency models introducing time dependent parameters. For nested models, the likelihood ratio test is commonly used to choose an appropriate non-stationary hydrologic frequency model. The method, however, cannot be used to select a model among non-nested models. In this study, the Takeuchi information criterion, TIC and the Akaike information criterion, AIC are analytically confirmed for model selection criteria for non-stationary sequences. Then, non-stationary hydrologic frequency models with time dependent parameters, a non-stationary GEV model, a non-stationary Gumbel model, a non-stationary SQRT-ET model and a non-stationary lognormal model are applied to the observed annual maximum daily rainfall series in Japan and the best model are selected to examine the change of the annual maximum series. The non-stationary four models are also applied to GCM rainfall time series to examine future change of extreme rainfall intensity.