Measurement uncertainty in drinking water analysis

Problems with measurement uncertainty

According to the German drinking water regulation alle drinking water laboratories have to be accredited according to ISO/IEC 17025. This also implies that they have to know their measurement uncertainty and have to report under certain circumstances.

The laboratories are therefore urged to carry out uncertainty estimations. In order to propose a feasible procedure the NORDTEST approach on the estimation of measurement uncertainties was transformed into a German guideline (Deutsche Einheitsverfahren zur Wasser-, Abwasser- und Schlammuntersuchung A0-4 "Leitfaden zur Abschätzung der Messunsicherheit aus Validierungsdaten"). For more details about the uncertainty estimation see here. You will find more documents on this topic here.

From the customer's view the question comes up, how to handle this uncertainty statement and how to use it in conformity assessment. You may find our point of view here.

One could think that the measurement uncertainty reflects the quality of a laboratory. This might lead to the situation that customers are tempted to place contracts for laboratories on the basis of reported measurement uncertainties. This again would step by step lead to lower, unrealistic measurement uncertainties to be reported. Finally then there is no benefit from reported uncertainties. This would also contradict the fundamental idea of quality. The quality of a measurement result is sufficient, if the customer's requirements are fulfilled. We don't need more.

But how can customers, in the case of drinking water analyses the public health authorities, define the requirements? One could think that for drinking water analyses this is covered by the EU drinking water directive which specifies requirements for the analysis. Unfortunately the requirements laid down on precision, trueness and limit of detection cannot readily be converted into measurement uncertainties.

It would be benefitial to all involved if the requirements of the EU directive could uniformly be interpreted as customer's requirement on measurement uncertainty.

To support this, variance functions were calculated from the data of drinking water proficiency tests in Germany of the recent years. Data were taken form the following sources:

• Landesamt für Natur, Umwelt und Verbraucherschutz Nordrhein-Westfalen (http://www.lanuv.nrw.de)
• Institut für Hygiene und Umwelt der Freien und Hansestadt Hamburg (http://www.hamburg.de)
• Niedersächsiches Landesgesundheitsamt, Standort Aurich (http://www.nlga.niedersachsen.de)
• AQS Baden-Württemberg an der Universität Stuttgart

For the calculation of the variance function the relative reproducibility standard deviations (variation coefficients VC_R) are depicted versus the concentration of the analyte. The characteristic function, described by Thompson, was used as a regression function:

You will find more details on characteriustic functions here.

This function enables us to state an average reproducibility standard deviation at the limit as well as the presentation of the concentratrion dependence of this relative standard deviation. In the following table the data of all parameters analysed in the PTs are collected. For your own calculations we have also reported the coefficients α and β. Please note that in many cases the incerease of the variation coefficient at low concentrations has a relatively uncertainty

By clicking on the parameter name you will get a graphical display. Parameters are listed in alphabetical order except the biocides. You will find the biocides at the end of the table.