Content area
The design and implementation of ground water quality monitoring systems may result in data errors which produce uncertainty in the information used for making resource management decisions. The sources of data error in ground water quality monitoring systems are identified and may be divided into: sampling errors, which result from the monitoring network design step (station location, sample frequency, and variable selection), and non-sampling errors, which result from operational activities of the monitoring system (sample collection, laboratory analysis, and data handling). Methods for estimating non-sampling data error are presented and applied to case study examples to show the usefulness of quantifying data uncertainty.
The effect of non-sampling errors on trend detection, standard violations, and design of monitoring systems are examined. The effect of non-sampling errors on detecting trends is to increase the alpha level at which a hypothesis may be rejected and to decrease the power of the test to detect differences for a fixed alpha level. When variable or positive bias non-sampling errors exist, the estimated probability that a water quality standard will be exceeded is larger than the true probability, while for negative bias non-sampling errors the estimated probability will be smaller than the true probability. The estimated sample size required when non-sampling errors exist as compared to when non-sampling errors do not exist is smaller for trend detection when bias errors cause the desired difference between the two means to decrease, and for standard violation detection when positive bias errors or variable errors exist. However, the sample size is larger for trend detection when bias errors cause the desired difference between the two means to increase or when variable errors exist, and for standard violation detection when negative bias errors exist.
Case study ground water quality variables were statistically analyzed and the characteristics observed are large horizontal spatial variations, vertical changes with aquifer depth, seasonality, autocorrelation in quarterly samples, normal or skewed right frequency distributions, and correlations among variables.