Complex Markov models to simulate persistent streamflows

Abstract

Abstract Highly persistent dry periods of certain streams cannot be simulated by simple Markov models which use the observed values of the autocorrelation coefficients. Such streamflow series may, however, be modelled by assuming that the lag-one autocorrelation coefficient is a function of the flow magnitude, while preserving the observed overall value of the coefficient. In this study, run-length characteristics and transition probabilities of such models are investigated by simulation for three different types of autocorrelation functions. It is shown that these models can simulate persistent low flows as well as the phenomenon of differential persistence.