Biomathematics & Statistics Scotland
An important question in water chemistry is whether rivers and streams can be thought of as alternating between a finite set of states, characterised by different sets of physical characteristics. As a first step in the spatio-temporal analysis of data from Scotland’s major rivers, a fully Bayesian inference procedure has been used to fit hidden Markov models to multivariate time series of ammoniacal nitrogen, nitrite, and nitrate measured in single rivers. Reversible jump Markov chain Monte Carlo methodology has been used to infer the number of hidden states. Interpretation of the sequences of underlying states estimated by this methodology aids our understanding of hydrological processes. Work is ongoing to generalise the analysis to multiple rivers using a hidden Markov random field approach, in which a measure of the distance between neighbouring rivers is taken into account in an inhomogeneous, anisotropic Potts model.
De-seasonalized time series of log-concentrations of dissolved inorganic nitrogens in the
River Garnock, from SEP’s Harmonised Monitoring Scheme, together with the marginal
posterior modes of the five states of the hidden Markov model.
Further details from:
Luigi Spezia, Mark Brewer or Chris Glasbey
