Selection of methods for the detection and estimation of trends in water quality

Abstract

One result of increased scientific and public interest in water quality over the past few decades has been the gradual accumulation of reliable long-term water quality data records and an interest in examining these data for long-term trends. This paper summarizes and examines some of the major issues and choices involved in detecting and estimating the magnitude of temporal trends in measures of stream water quality. The first issue is the type of trend hypothesis to examine: step trends versus monotonic trend. The second relates to the general category of statistical methods to employ: parametric versus nonparametric. The third issue relates to the kind of data to analyze: concentration data versus flux data. The fourth relates to issues of data manipulation to achieve the best results from the trend analysis. These issues include the use of mathematical transformations of the data and the removal of natural sources of variability in water quality due to seasonal and stream discharge variations. The final issue relates to the choice of a trend technique for the analysis of data records with censored or “less than” values. The authors' experiences during the past decade with the development of several trend detection techniques and application of these techniques to a large number of water quality records provide insight into the issues related to a choice of a statistical test for trend in water quality.