Efficient data quality evaluation in automated water quality measurement stations

In this paper, software tools for automatic data quality assessment with apractical orientation are proposed. Two different approaches are presented that usetime series information. First, univariate methods based on autoregressive modelsare applied for data correction (outliers detection for data replacement). Faults aredetected by defining acceptable thresholds to data features and to the residuals’standard deviation (RSD). Second, multivariate statistical methods based onPrincipal Components Analysis are used to extract correlations between variablesfrom data sets and performing fault detection using the T2 and Q statistics. Theproposed tools are successfully tested on river water quality time series obtainedfrom in situ monitoring stations collecting a large amount of physical and chemicalvariables.