Identifying Sources of Uncertainty from the Inter-Species Covariance of Measurement Errors

Abstract

A standard metric of measurement precision in environmental monitoring is the variance of differences between duplicate (collocated) samples. With duplicate measurements of multiple species, we can extend this variance analysis to include the interspecies covariance of differences between duplicate samples; these covariances can provide clues about the sources of error. We illustrate the potential of such an analysis with atmospheric aerosol measurements from two national air quality monitoring networks: Interagency Monitoring of Protected Visual Environments (IMPROVE) and Speciation Trends Network (STN). These aerosol "speciation" networks provide the multivariate data sets needed to characterize error covariance by operating duplicate samplers at several of their monitoring locations and analyzing both the collected aerosol samples for multiple species. We observe covariance among the measurement differences for multiple species in both networks. The covariance among measurement differences for soil-derived elements suggests an error associated with the particle size discrimination step in sampling, which is not currently included in either network's uncertainty estimates. The multivariate statistical analyses of aerosol speciation data performed by standard source apportionment models assume that measurement errors in different species are independent of each other; the present analysis invalidates this assumption for several species measured by IMPROVE and STN.