Evaluation of water quality using acceptance sampling by variables

Under section 303(d) of the Clean Water Act, states must identify water segments where loads of pollutants are violating numeric water quality standards. Consequences of misidentification are quite important. A decision that water quality is impaired initiates the total maximum daily load or TMDL planning requirement. Falsely concluding that a water segment is impaired results in unnecessary TMDL planning and pollution control implementation costs. On the other hand, falsely concluding that a segment is not impaired may pose a risk to human health or to the services of the aquatic environment. Because of the consequences, a method is desired that minimizes or controls the error rates. The most commonly applied approach is to use the Environmental Protection Agency (EPA)'s raw score approach in which a stream segment is listed as impaired when greater than 10 per cent of the measurements of water quality conditions exceed a numeric criteria. An alternative to the EPA approach is the binomial test that the proportion exceeding the standard is 0.10 or less. This approach uses the number of samples exceeding the criteria as a test statistic along with the binomial distribution for evaluation and estimation of error rates. Both approaches treat measurements as binary; the values either exceed or do not exceed the standard. An alternative approach is to use the actual numerical values to evaluate standard. This method is referred to as variables acceptance sampling in quality control literature. The methods are compared on the basis of error rates. If certain assumptions are met then the variables acceptance method is superior in the sense that the variables acceptance method requires smaller sample sizes to achieve the same error rates as the raw score method or the binomial method. Issues associated with potential problems with environmental measurements and adjustments for their effects are discussed. Copyright © 2003 John Wiley & Sons, Ltd.