Measurements recorded over monitoring networks often possess spatial and temporal correlation inducing redundancies in the information provided. For river water quality monitoring in particular, flow-connected sites may likely provide similar information. This paper proposes a novel approach to principal components analysis to investigate reducing dimensionality for spatiotemporal flow-connected network data in order to identify common spatiotemporal patterns. The method is illustrated using monthly observations of total oxidized nitrogen for the Trent catchment area in England. Common patterns are revealed that are hidden when the river network structure and temporal correlation are not accounted for. Such patterns provide valuable information for the design of future sampling strategies.
Keywords: PCA; connected monitoring networks; flow direction.