Trend analyses for ambient water quality data
by
Gordon Smyth
Walter & Eliza Hall Institute of Medical Research

The Queensland Environmental Protection Agency began a large scale water quality monitoring program of streams and estuaries in Queensland in 1992 and there is now a strong need to assess any trends from the accumulated data. Trend analyses should incorporate seasonal patterns and other factors such as stream flow or temperature which are likely to impact on water quality. More general questions include assessment of regional trends, assessment of relationships between indicators, setting of water quality guidelines for impacted streams and the design of future sampling schemes. This article describes methods and software developed for this problem. A censored regression strategy is used to accommodate arbitrary detection limits for the indicator variables. Use of a heavy-tailed response distribution gives a high degree of insensitivity to outliers. Harmonic terms are used to model periodic seasonal effects. Regression splines are used to model nonlinear long-term trends. The use of regression methodology allows covariates such as flow rate, temperature or tidal hight to be incorporated into the model.

Date received: August 31, 2001