Hidden Markov model segmentation of hydrological and enviromental time series

Abstract

Motivated by Hubert's segmentation procedure we discuss the application of hidden Markov models (HMM) to the segmentation of hydrological and enviromental time series. We use a HMM algorithm which segments time series of several hundred terms in a few seconds and is computationally feasible for even longer time series. The segmentation algorithm computes the Maximum Likelihood segmentation by use of an expectation / maximization iteration. We rigorously prove algorithm convergence and use numerical experiments, involving temperature and river discharge time series, to show that the algorithm usually converges to the globally optimal segmentation. The relation of the proposed algorithm to Hubert's segmentation procedure is also discussed.