ARTICLE pubs.acs.org/est
Application of EPA Unmix and Nonparametric Wind Regression on High Time Resolution Trace Elements and Speciated Mercury in Tampa, Florida Aerosol Joseph Patrick Pancras* Alion Science and Technology, P.O. Box 12313, Research Triangle Park, North Carolina 27709, United States
Ram Vedantham, Matthew S. Landis, and Gary A. Norris U.S. Environmental Protection Agency, Office of Research and Development, Research Triangle Park, North Carolina 27711, United States
John M. Ondov Department of Chemistry, University of Maryland, College Park, Maryland 20742, United States
bS Supporting Information ABSTRACT: Intensive ambient air sampling was conducted in Tampa, FL, during October and November of 2002. Fine particulate matter (PM2.5) was collected at 30 min resolution using the Semicontinuous Elements in Aerosol Sampler II (SEAS-II) and analyzed off-line for up to 45 trace elements by high-resolution ICPMS (HR-ICPMS). Divalent reactive gaseous mercury and particulate bound mercury were also measured semicontinuously (2 h). Application of the United States Environmental Protection Agency’s (EPA) Unmix receptor model on the 30 min resolution trace metals data set identified eight possible sources: residual oil combustion, lead recycling, coal combustion, a Cd-rich source, biomass burning, marine aerosol, general industrial, and coarse dust contamination. The source contribution estimates from EPA Unmix were then run in a nonparametric wind regression (NWR) model, which convincingly identified plausible source origins. When the 30 min ambient concentrations of trace elements were time integrated (2 h) and combined with speciated mercury concentrations, the model identified only four sources, some of which appeared to be merged source profiles that were identified as separate sources by using the 30 min resolution data. This work demonstrates that source signatures that can be captured at 30 min resolution may be lost when sampling for longer durations.
’ INTRODUCTION Both source- and receptor-oriented models use trace elements extensively to apportion air pollution sources in particulate matter (PM) samples collected on filter substrates over a period of time, typically 24 h.1,2 Transient emissions from local sources may be obscured in this integrated filter sampling approach, resulting in either overestimation of background concentrations or attribution to composite sources. Continuous or semicontinuous shortduration sampling has been shown to preserve the dynamic variation in chemical concentration and composition of ambient aerosol.3 The significance of time-resolved PM measurements has been demonstrated in the Supersite program sponsored by the United States Environmental Protection Agency (EPA).3,4 The University of Maryland designed a high time resolution aerosol sampler, the Semicontinuous Elements in Aerosol Sampler (SEAS), for sampling air particulate matter at subhourly intervals in a form of slurry.5,6 A modified version of it, SEAS-II, was operated in the EPA Supersites program and the Bay Regional Atmospheric Chemistry Experiment (BRACE) project sponsored by the Florida Department of Environmental Protection (FLDEP) between 2001 and 2003.7,8 Offline analysis of subhourly samples collected by SEAS-II for trace elements has r 2011 American Chemical Society
revealed improved resolution of sources on the time-series concentrations.7-9 Short-term changes in concentrations measured in semicontinuous samples, in combination with particle size distribution, criteria gases, inorganic PM anion data, and meteorological data, have been used to identify and apportion unique point sources impacting a receptor site.4,8-11 The objective of the research presented here was to explore and apportion highly time-resolved trace metals and speciated mercury measurements data collected in Tampa, FL, during BRACE using a receptor modeling tool. While there are many approaches to source apportionment,12 the EPA Unmix receptor model is particularly well suited for highly time-resolved monitoring data for the following reason. The model constructs “edges” based on the observational data where at least one source has negligible or zero contribution, and these instances are more evident in high time resolution data than in time-integrated data typically collected over 24 h intervals. Transient plume events Received: October 7, 2010 Accepted: February 9, 2011 Revised: January 27, 2011 Published: March 14, 2011 3511
dx.doi.org/10.1021/es103400h | Environ. Sci. Technol. 2011, 45, 3511–3518
Environmental Science & Technology from local sources can occur on short time scales and will otherwise be mixed with other sources in time-integrated data. We considered using a model such as Positive Matrix Factorization (PMF), which requires sample uncertainty data. The procedure to generate sample uncertainty for PMF, however, is not fully understood 13 and has not been evaluated for newer instruments such as SEAS-II and automated mercury speciation instrumentation that generate high time resolution data. This article presents the first attempt to apportion highly timeresolved trace elements and speciated mercury data using the EPA Unmix model and nonparametric wind regression (NWR), implemented through the EPA Air Pollution Transport to Receptor (APTR) platform. While EPA Unmix helps identify source profiles, NWR uses surface meteorology data along with either the sampled data or source contributions to identify contributing transport sectors.14 The work presented in this article also shows loss of source resolution as the sampling interval increases.
’ EXPERIMENTAL SECTION Receptor Site. The Tower Dairy receptor site (27°540 48.200
N, 82°220 30.700 W) was located on a dairy farm in Tampa, FL. The geographic location was suitable for studying local point sources, marine contribution, and mobile sources. Many of the region’s major industrial facilities, including coal-fired electric power plants (CFPPs), oil-fired electric power plants (OFPPs), fertilizer production plants, recycling centers, and other manufacturing facilities, are located within 25 km of the receptor site. Traffic from major interstates and highways located less than 2 km from the sampling site also had an impact at the receptor site. Instrumentation and Measurement Frequency. A suite of automated gas and particle analyzers was operated in a temperature-controlled mobile laboratory. Ambient concentrations of SO2 were continuously measured at 1 min intervals using ThermoEnvironmental (Franklin, MA) model 43A equipment. Meteorological data such as surface wind direction and wind speed (15 m above ground level) were also measured during the same time interval using RM Young (Traverse City, MI) instrumentation. An automated Tekran Instruments Corp. (Knoxville, TN) mercury speciation system (models 2537A, 1130, and 1135) was operated between October 27 and November 16, 2002. The Tekran system was configured to collect 2 h integrated divalent reactive gaseous mercury (RGM) and particulate-bound mercury (Hg(p)) samples at 10 L min-1. During the 2 h RGM and Hg(p) collection period, 5 min integrated gaseous elemental mercury (GEM) samples were quantified. After the 2 h sample interval, RGM (500 °C) and Hg(p) (800 °C) were thermally desorbed in turn and quantified as GEM, resulting in 2 h collection and 1 h of analysis. During the analysis period, the Tekran system was offline and not collecting ambient samples.15 A SEAS-II was set up and operated between October 31 and November 16, 2002, by the University of Maryland. The SEAS-II collects 30 min integrated PM2.5 samples as a slurry (suspension of PM2.5 in water) for off-line wet chemical analysis. The SEAS-II was configured with a PM2.5 glass impactor inlet and sampled at a flow rate of 92 L min-1. Operation of the SEAS-II instrument was described elsewhere.5,6 SEAS-II Sample Analysis. Elemental concentration measurements were carried out using a ThermoFinnigan (Bremen, Germany) Element2 high-resolution magnetic sector field inductively coupled plasma-mass spectrometer (HR-ICPMS), which is housed in an EPA Office of Research and Development
ARTICLE
Class 100 clean laboratory in Research Triangle Park, NC. This instrument has predefined nominal resolution settings (m/Δm at 10% valley definition) of 300 (low), 4000 (medium), and 10 000 (high). The operating conditions of the Element2 are given in Supporting Information Table S1. External calibrations were performed daily with multielement standards from High Purity Standards (Charleston, SC). A standard curve was deemed acceptable if the coefficient of determination values (r2) were e0.99. An internal standard (2 ppb rhodium) solution was added to all samples online to account for analytical signal drifts. Analytical accuracy was ensured daily by analyzing 10% (v/v) NIST SRM 1640 and/or SRM 1643 standards, and instrument response was verified to be between 90% and 110% of the certified values. SEAS samples were acidified with concentrated ultrapure nitric acid (to 0.2% v/v) and placed in an ultrasonic bath for 2 h. Samples were then left to leach for 30 days before they were analyzed. SEAS-II Atmospheric Concentrations. Atmospheric concentrations were calculated by multiplying elemental concentration in solution by sample volume and dividing the resulting absolute elemental mass by the SEAS-II system air sample volume. During this study, blank samples were collected by sampling particle-free air at a supersaturation condition (saturation ratio of 15) that facilitates homogeneous condensational growth. Laboratory data from later studies showed that a heterogeneous condensational growth condition (saturation ratio of 1.3 or target RH of