Development of a method for the classification of geochemical data quality

This report describes the development of a method for the classification of geochemical data quality. The project had four tasks: (1) review of quality information; (2) development of Evidence Support Logic (ESL) models; (3) development of rules for choosing parameter values; (4) classification of existing data. During H15, JNC developed a preliminary system for classifying groundwater chemical data according to its quality. This system gives a very general indication of data quality, but has a number of limitations. Notably, it is based on adding together scores for individual quality indicators, so that high scores given by some indicators tend to compensate for low scores given by other indicators. Additionally, the system does not distinguish between cases where data quality is poor and cases where data quality is unknown. A further limitation is that the system is based on only a small number of the quality indicators that could be used. By using ESL to develop a new system for classifying geochemical data quality, these limitations can be avoided. This methodology involves weighing evidence for and against a particular hypothesis being true or reliable. Varied evidence, which may be quantitative or qualitative, can be evaluated in an integrated fashion. A process model is constructed to link a hypothesis of interest to evidence corresponding to observations and quantitative data, usually via intermediate processes. An arithmetical approach is then used to propagate evidence through the model. Thus, the approach does not rely on simple addition of quality scores for individual parameters. In this project, evidence for and evidence against the hypothesis that groundwater chemical data represent in-situ conditions are evaluated independently. Both kinds of evidence are represented using numerical scales from 0 to 1. Lack of information about data quality is then represented by 1 - evidence for - evidence against. In this way the situation where the available information indicates low data quality is distinguished from the situation in which there is no quality information. Separate process models have been constructed to evaluate the quality of each of pH measurements, Eh measurements, redox-sensitive trace element analyses and analyses of species of inorganic carbon. The model for evaluating the quality of pH data could also be applied to evaluate the quality of major cations and anions and non-redox sensitive trace elements. These process models have been applied to data from boreholes. The process models provide a visualisation of data quality judgments that may be appraised rapidly. The models can be revised readily as and when additional quality information becomes available, or to reflect the differing opinions of different experts concerning data quality. It is suggested that the process models should be reviewed and revised as necessary by different experts, so as to build a consensus about what levels of data quality are desirable and attainable. (author)