Process identification by principal component analysis of river water-quality data

Time series of nutrient concentrations and related water quality parameters taken at several locations along the River Elbe were subjected to multivariate statistical analysis. The main question underlying this study is concerned with whether known interactions between water quality variables can be recovered as statistically significant covariance patterns. For this purpose, the standard technique of principal component analysis (PCA) was applied. Raw data and deviations from an estimated seasonal cycle were analysed. In both cases, two leading patterns of covariance was obtained, one discharge-dependent and the other related to biological activities. Linear regression modelling based on discharge and temperature was used to approximately eliminate the impact of meteorological forcing; this led to a large reduction of the seasonal component. The remaining partial variance of water-quality variables could be shown to be dominated by biological activities for which temperature is of secondary importance. Amplitudes of the pattern related to biological processes are much less correlated between different stations than those of the pattern induced by spatially homogenous discharge. The analysed covariance patterns agree well with general knowledge about basic dynamical processes in the river. Therefore, multivariate statistical analysis offers an objective method to estimate the observed strengths of the given processes that involve simultaneous changes of several water-quality parameters. Such an assessment is a prerequisite when observations are to be compared with corresponding results from process-oriented numerical models in order to increase the knowledge about the nutrient system. A related application would be to use it to identify the number of degrees of freedom needed to appropriately describe the nutrient system's variability.