NONPARAMETRIC MARKOV MODEL FOR DAILY RIVER FLOW

This paper presents to an audience of research hydrologists what is believed to be a significant new development in time series modeling. The model class is the class of (not necessarily finite state) Markov chains. The basic advantage of this class is that in comparison to parametric models (such as autoregressive moving average) it is a very rich class, and the value of the statistical method described herein is that, as proven elsewhere, it provides convergence over this large class. The technique is applied to Cheyenne River data, and discussion is provided on how to incorporate prior statistical and geological information into the model. Also, comparisons are made between the nonparametric Markov analysis provided here and the currently popular streamflow models and statistical techniques. Copyright 1979 by the American Geophysical Union.