Using percentile analysis for determination of alarm values
The use of statistically derived alarm values for the minimisation of the risk of Cryptosporidium entering the water supply has been recommended in the report of the Bouchier expert group. Values derived using the mean plus standard deviations can be skewed by spikes, including those caused by filter washes. Indeed, theoretically, the use of standard deviations should only be used with normally distributed data. It is shown in this paper that turbidity and particle counting data is not normally distributed and that values derived by means of percentile analysis are more sensitive than using the mean plus standard deviations. Percentile analysis is able to produce alarms for data that is clearly different from normal operation, enabling smaller deviations from the usual to be identified. The values derived from one set of data are also shown to be appropriate for periods of around three months.