Analysis of multivariate observations from a monitoring station of a river basin

Abstract

Water quality evaluation provides important information to estimate water system status and to test for compliance with standards. The present work analyzes changes concerning the quality of surface water using a data set from a monitoring station of a Mediterranean river basin. The data set has a dependency structure that renders it multivariate. Principal component analysis is applied to characterize associations present in the multivariate measurements. The principal component scores exhibit temporal correlation. A combined Shewhart-CUSUM control chart is applied to the residuals of the scores time series model to detect changes in the mean level of the data set during the study period. This method detects an outlying observation in the study period due to an extreme value in magnesium concentration. The multivariate assessment of trend is performed using non-parametric tests. The covariance inversion test supported rejection of the hypothesis of no trend in the variables defined with each combination of water quality parameters and month. There is heterogeneity between the trends in the different combinations and an overall trend is not representative. The partial Mann-Kendall test is employed to analyze the trends of each physicochemical variable in the study months. Conductivity trends in two months (May and June) are significant and upward. In the same months calcium trend is also significant but downward. Sodium concentrations exhibit a significant decreasing trend in April. Magnesium levels significantly decrease in March but have an upward trend in June.