摘要
The control of chemical and physicochemical properties of water bodies is essential to guarantee a proper environment, not only for the species of a certain habitat, but also for human health. In this sense, the study of pollution and its variability can be assimilated to the detection of outliers (anomalous values that indicate the deviation of the measured variable compared to the defined objectives). In this paper, two methods have been used and compared for detecting outliers: statistical process control and functional data analysis (FDA). These techniques have been tested on three key water quality variables: turbidity, electrical conductivity and dissolved oxygen. Data were continuously recorded in 2008 using an automatic monitoring station located in the Ebro River (NE Spain). The results of the research show FDA as a powerful tool for this kind of study, since it takes into account the time correlation structure of the data. This research is focused on an essential natural resource, water. The study intends to provide analysts with a methodology for detecting anomalous values (possible pollution episodes) within a large dataset of measurements. The collection of information of natural resources is a basic task for their management and control, but the analysis of gathered data is frequently difficult due to the vast number of measurements taken in the field. This work is focused on water, but the methodologies presented here can be applied to other natural data. Thus, geoscientists and geoengineers might find this research useful for interpreting many kinds of data and detecting anomalous episodes among them with an objective approach. Keywords Functional data analysis Statistical process control Water quality Outlier Water quality monitoring