Optimally smoothed maps of pollution source potential via particle back-trajectories and filtered kriging

Abstract

For over 20years, the Potential Source Contribution Function (PSCF) has been used by the aerosol research community to identify areas around an air-quality receptor location that are associated with high levels of pollutant emissions. PSCF uses particle back trajectories associated with multiple times as well as measured levels of a pollutant at each time, linking high-measurement days with specific back trajectories. For a given rectangular area (s) on a map, the probability p(s) that the area contains an important source of the pollutant of interest is estimated with pˆs=Xs/ns, where n(s) is the number of back trajectories that trace back through that area and X(s) is the number of those back trajectories that are associated with a high day for the pollutant. Results are generally illustrated with a PSCF plot in which p(s) is plotted at each area (or pixel) on the map. However, the PSCF exhibits high pixelwise volatility and strong spatial discontinuity, particularly for high resolutions. We propose a modified potential source map that exploits both prior knowledge about p(s) and a filtered kriging approach that accounts the heterogeneous measurement error variances. Results are illustrated using air quality data from the EPA St. Louis–Midwest Supersite.