Rationalization of a regional network designed for trend detection of lake water quality in presence of spatial correlation

While studying the changes in lacustine water quality at a regional scale, a network of monitoring stations consisting of several lakes in a given region can be established to accelerate the detection of trends using the compounded information. However, the use of several lakes in a region may produce spatially correlated observations if lakes are not chosen according to a probability-based sampling scheme. In a single-stage cluster sampling, the use of lakes too close to one another induces spatially correlated observations resulting in an underestimation of the variance, and causing too many trends to be detected. In this paper, we use data from water quality samples collected on small lakes in the Algoma region of Ontario, Canada, between 1988 and 1996 to illustrate that a network rationalization based on both information content from water samples and spatial correlation considerations can result in a reduction of 50 per cent or more of the sample size, with only a slight decrease in trend detection power. Copyright © 2001 John Wiley & Sons, Ltd.