Methods for detecting change in hydrochemical time series in response to targeted pollutant mitigation in river catchments

Abstract

Detecting changes in catchment hydrochemistry driven by targeted pollutant mitigation is high on the scientific agenda, following the introduction of the European Union Water Framework Directive. Previous research has shown that understanding natural variability in hydrochemistry time series is vital if changes due to mitigation are to be detected. In order for change to be detected in a statistically robust manner, the data analysis methods need careful consideration. Previous work has shown that erroneous results have often been obtained when statistical analyses have been carried out despite the associated test assumptions not being met. This paper discusses the principal data issues which must be considered when analysing hydrochemical datasets, including non-normality and non-stationarity. A range of statistical techniques is discussed which could be used to detect gradual or abrupt changes in hydrochemistry, including parametric, non-parametric and signal decomposition methods. The statistical power of these techniques as well as their suitability for identifying change is discussed. Using the uniquely detailed hydrochemical datasets generated under the Demonstration Test Catchments programme in England, the efficacy and robustness of change detection methods for hydrochemical data series is assessed. A conceptual framework for choosing a change detection method is proposed, based on this analysis, in order to raise awareness of the types of questions a researcher should consider in order to perform robust statistical analyses, for informing river catchment management and policy support decisions.