Statistical Methodology

The classification of spatio-temporal data on water quality in rivers

The Scottish Environment Protection Agency (SEPA) monitors the quality of fresh waters in Scotland by conducting monthly sampling of various determinands at harmonised monitoring stations on 56 major rivers of Scotland. In these rivers, high concentrations of dissolved inorganic nitrogen (DIN) are indicative of poor water quality, because they can promote the eutrophication of surface or coastal waters and the contamination of groundwater. The DIN has three components (ammoniacal nitrogen, nitrite and nitrate) and can come from both agricultural and urban sources.

We have conducted a simultaneous analysis of all 56 trivariate time series of DIN over 10 years. The aim of this work has been to classify each month on each river into a small set of homogeneous groups, which represent different river states defined by the DIN concentrations. Our analysis was performed by means of hidden Markov models (HMMs), because they can define the states from the data, controlling the rates of transitions between states and taking into account both spatial and the temporal correlations. The fitted model accounts for data heterogeneity by allowing the three forms of DIN to have separate means, variances and correlations for each state of each river. High probabilities of transitions between states in a river are indicative of change-points occurring in the dynamics of DIN concentrations.

map of river sampling sites Map of the 56 river sampling sites, showing the closest recorded neighbours of each site.

Further details from:
Luigi Spezia

Article date 2013


Statistical Genomics and Bioinformatics

Process and Systems Modelling

Statistical Methodology

PhD Opportunities

Meetings & Seminars