2000-07-14_REVISION - M1977378 - Laserfiche WebLink

Home Browse Search

Monthly sampling over a five year period would result, on average, three points or false alarms plotting out of control. Wider or narrower control limits would result in fewer or more "false alarms," respectively. No information is gained if the limits are so wide that no points plot out of control, and conversely if the limits are too narrow, many false alarms will be generated. The proposed sampling frequency for the Animas River is a single sample per month. A sample standard deviation cannot be calculated for a single sample, so a range chart cannot be constructed. Increasing the sampling frequency to weekly yielding a sample size of four, may increase monitoring costs to an unacceptable level. Averaging monthly samples for quarterly points results in four points a year, but it would require 15 years to yield, on average, three out of control points. Narrow control limits are recommended because of the relatively low sampling effort. Comments on the SGC Control Chart The SGC control chart methodology does not have adequate sensitivity for the expected sampling intensity (frequency and number of samples). The criteria for identifying an out of control condition and corrective actions to follow also need to be expanded upon. I suggest that 1) three separate charts proposed by SGC be combined into a single chart to increase the statistical power, thereby increasing its sensitivity; 2) that the control limit be based on 1.645 standard deviations which is more suitable for the proposed monitoring frequency; and 3) that additional sensitizing criteria be applied to the chart which include a warning limit placed at one standard deviation and a runs test based on a consecutive number of points above the center line. Consideration should also be given to including an EWMA chart, because the proposed sampling effort does not provide for an range chart. The EWMA chart does not necessarily require the same sensitizing criteria as the Shewart chart. A single chart provides a continuous water quality record and the random variation of single measurements is distributed among a larder number of samples which increases the statistical power to identify a shift in the mean water quality. The three control charts proposed by SGC can be combined into one by removing known causes of variation in the zinc concentration, i. e. streamflow, and charting the deviations (residuals) rather than the raw data values. This is accomplished through a variance reduction procedure, i.e. regression. The residuals, arranged in time series, plot randomly about the mean and are normally distributed which satisfies the condition 4 on page 2 of the SGC Technical Memorandum.