Water quality sampling: Some statistical considerations

Typically, a state, regional, or national water quality (WQ) monitoring system has many objectives. Examples include (1) providing a system‐wide synopsis of WQ; (2) determining whether selected WQ parameters show gradual changes over time; (3a) detecting actual or potential WQ problems, (3b) determining the specific causes of actual problems, and (3c) assessing the effect of any corrective action; and (4) enforcing the law. While these purposes have different data requirements, they are interrelated, and data collected for any one can have value for others; for example, data collected for the listed reasons could be used for future long‐range planning. Each objective makes certain demands of a WQ sampling network, and these in turn have implications concerning sampling design and statistical analysis of resulting data. In a sampling scheme flexible enough to encompass multiple objectives, data sets collected for answering questions arising under points 1–4 need not all have similar characteristics; consequently, a variety of statistical methods are needed. A good sampling design is important for achieving the first two objectives. A working definition of ‘water quality’ is given initially, and this is followed by the definition of a sampling population. The set of all segments of the drainage network serves as the population for several sampling plans. In particular, the use of probability sampling to design a network of synoptic sampling stations is discussed in detail. Finally, sequential statistical procedures are applied to a specific problem of type 3c. In terms of the amount of data required, sequential methods are quite efficient for detecting changes of a specified magnitude in some WQ parameter.