Data quality in airborne particulate matter measurements

Environmental measurements are complicated by uncontrollable natural variations in the environment, which cannot be reproduced in the laboratory. These variations affect the measurement uncertainty and detection capabilities -- two measures of data quality. Variations in a measurement series that arise from uncertainty in the measurements should not be interpreted as variations in the environment. Accurate estimates of measurement uncertainty are thus important inputs to data analyses. Collocated (duplicate) measurements are the most direct approach to characterizing uncertainty and detection capabilities because the observed differences reflect the actual measurement performance under the natural environmental variability. This dissertation uses collocated measurements of airborne particulate matter chemical speciation collected by the Interagency Monitoring of Protected Visual Environments (IMPROVE) and Speciation Trends Network (STN) to explore data quality issues.;In addition to the complications introduced by uncontrollable environmental factors, the concepts of measurement precision and detection capabilities are often complicated by incomplete and inconsistent definitions. In this dissertation, collocated IMPROVE data are used to illustrate different formulations for precision and their ability to fit the observed differences. Collocated IMPROVE data are also used to show that measurement precision is typically better at concentrations well above the detection limit, when the analysis is performed on the whole filter instead of just a fraction of the filter, and for species predominantly in the smaller size fractions. For most species, the collocated differences are worse than the differences predicted by the current uncertainty model, suggesting that some sources of uncertainty are not accounted for or have been underestimated in the model. In addition, collocated measurement differences are shown to be correlated among several species. In both IMPROVE and STN, obvious correlations exist among differences in elements associated with soil dust, which are dominated by particles with diameters > 1 mum. These correlations suggest the current model is missing significant sampling errors associated with the size discrimination operation. Measurement uncertainty generally increases as concentrations approach the detection limit. This dissertation introduces an empirical approach for estimating detection limits using collocated IMPROVE and STN data that accounts for the natural variations in the environment.