Positive Matrix Factorization (PMF) and Data Quality Assessment of EPA's PM_(2.5) Chemical Speciation Network (CSN) Derived from Six Collocated CSN Sites for the Period 2010 - 2013

摘要

Reconstructed mass using the SDVAT method underestimates measured mass (≈70-87%) and there is no attempt to determine organic material (OM) from OC. The results are not blank corrected. Primary and collocated RCM seem to agree well, but are influenced primarily by "major" constituents. The results are not a true material balance. The "best" fits with PMF occur with 4-6 factors (in robust mode with no rotations) and account for ≈90-98% of the measured mass. There are reasonably consistent results between primary and collocated samplers and one has the same factors, with dust for the collocated sampler at Rubidoux an exception. The sources seem highly mixed, presenting challenges for PMF to separate them. Coal + oil are roughly equal for the Roxbury primary and collocated samplers, but for the primary sampler coal is ~1/3 of the combination while for the collocated sampler it is ~2/3. This may be the result of uncertainty in the "tracer" elements in the factor (e.g., Se for coal and V for oil) and are one of the consequences of having many sample parameter values at or below MDLs and uncertainties. These considerations also have implications for trying to optimize control strategy abatement costs. DRI hopes to resolve carbon metadata issues and redo some of the analyses with more data, but is likely that the PMF results will still have generalized factors, a consequence of the large uncertainties and low values for many of the more significant "tracer" species.