Clustering and forecasting of dissolved oxygen concentration on a river basin

Abstract

The aim of this contribution is to combine statistical methodologies to geographically classify homogeneous groups of water quality monitoring sites based on similarities in the temporal dynamics of the dissolved oxygen (DO) concentration, in order to obtain accurate forecasts of this quality variable. Our methodology intends to classify the water quality monitoring sites into spatial homogeneous groups, based on the DO concentration, which has been selected and considered relevant to characterize the water quality. We apply clustering techniques based on Kullback Information, measures that are obtained in the state space modelling process. For each homogeneous group of water quality monitoring sites we model the DO concentration using linear and state space models, which incorporate tendency and seasonality components in different ways. Both approaches are compared by the mean squared error (MSE) of forecasts.