Measurement uncertainty for multiple measurands: characterization and comparison of uncertainty matrices

The 'uncertainty matrix', i.e. the variance/covariance matrix associated with a set of values obtained by measurement of multiple measurands, is discussed as the multivariate analogue of the standard uncertainty (or rather the variance) of a value obtained for a single measurand. The main topics are the characterization and comparison of uncertainty matrices with respect to the level of uncertainty of derived quantities and the level of correlation effects on the uncertainty of derived quantities. Examples include a comprehensive application study on the normalization of natural gas composition data. Supplementary technical material such as basic properties of uncertainty matrices and mathematical proofs are included in an appendix. [PUBLICATION ABSTRACT]