Bayesian modeling of hydrographs

Abstract

This article presents a new approach to model yearly hydrographs with daily or weekly streamflow measurements. The method considers yearly hydrographs as a sample of functions to be modeled nonparametrically in a Bayesian setting. The functional data analysis framework provides great flexibility to reproduce the features of yearly hydrographs, while the Bayesian probabilistic model ensures statistical coherence between the flood variables and the shapes of flood events. The proposed methodology is applied to two samples of hydrographs from two watersheds in the province of Quebec.