Analysis of air quality data using positive matrix factorization

Positive matrix factorization (PMF) was applied to air quality and temperature data collected as part of the Program for Research on Oxidants: Photochemistry, Emissions, and Transport 1997 summer measurement campaign. Unlike more conventional methods of factor analysis such as principal component analysis, PMF produces non-negative factors, aiding factor interpretation, and utilizes error estimates of the data matrix. This work uses PMF as a means of source identification and apportionment, important steps in the development of air pollution control strategies. Measurements of carbon monoxide, particulate matter, peroxyactyl nitrate (PAN), isoprene, temperature, and ozone were taken from a 31 m tower in rural northern Michigan and analyzed in this study. PMF resulted in three physically interpretable factors: an isoprene-dominated factor, a local source factor, and a long-range transport factor. The isoprene-dominated and local source factors exhibited strong and weak diurnal signals, respectively. Factor strengths for the long-range transport factor were relatively high during periods of south and southwesterly flow. The average contribution of the three factors was determined for each pollutant, enabling the modeled matrix to be compared to the data matrix. Good agreement between the fitted and data matrix was achieved for all parameters with the exception of coarse particulate matter. The PMF model explained at least 75% of variation for all species analyzed.