Identification of sources and estimation of emission profiles from highly time-resolved pollutant measurements in Tampa, FL

ARTICLE IN PRESS

1352-2310/$ - se

doi:10.1016/j.at

�Correspondfax: +1301 314

E-mail addr1Presently at

Florida, Tampa

NC – 27711.

Atmospheric Environment 40 (2006) S467–S481

www.elsevier.com/locate/atmosenv

Identification of sources and estimation of emission profiles fromhighly time-resolved pollutant measurements in Tampa, FL

Joseph Patrick Pancrasa,1, John M. Ondova,�, Noreen Poorb,Matthew S. Landisc, Robert K. Stevensd

aDepartment of Chemistry and Biochemistry, University of Maryland, College Park, MD-207042, USAbCollege of Public Health, University of South Florida, 13201 Bruce B. Downs Blvd., Tampa, FL 33612, USA

cUS Environmental Protection Agency, Office of Research & Development, Research Triangle Park, NC – 27711, USAdFlorida Department of Environmental Protection on assignment to US EPA, Research Triangle Park, NC – 27711, USA

Received 30 July 2005; received in revised form 7 December 2005; accepted 15 December 2005

Abstract

Aerosol slurry samples were collected at 30-min intervals for sequential 1-month periods at each of two sites (Sydney and

‘‘Dairy’’) in the Tampa Bay area during the 2002 Bay Regional Atmospheric Chemistry Experiment using the University of

Maryland Semicontinuous Elements in Aerosol Sampler-II (SEAS-II). More than 500 samples, believed to be affected by

plumes from local utility and industrial sources, were selected for electrothermal atomic absorption spectrophotometric

analyses for elemental markers (Al, Fe, Cr, Cu, Mn, Pb, Se, As, Ni, Zn and Cd) based on excursions in SO2 and NOx

measurements. Correlation of short-term excursions in metals and SO2, and surface wind directions observed between May

23 and 26th, 2002, revealed the influence of an animal feed supplements production facility (AFS), 17 km upwind at a

station angle of 811, for which emissions had not previously been detected by standard monitoring methods. Emission

‘‘profiles’’ for this source were developed, separately, from the time series data and by using principle components analysis

(PCA) and positive matrix factorization (PMF). In addition, a local dust component was evident in Al and Fe

concentration profiles during periods of elevated wind speeds and was resolved by PCA/PMF. Similarly, large but brief

1.5-h excursions in Zn (maximum, 403 ngm�3), Cd, and Pb on May 17th were correlated with winds from the direction of

an incinerator (station angle, 2501) 17 km from Sydney. Lastly, large excursions in As concentrations (maximum,

86 ngm�3) observed (May 4th and 5th at Sydney and November 2nd and 3rd at the Dairy) were used to locate previously

unrecognized sources, tentatively associated with combustion/production of pressure-treated lumber. Profiles developed

directly from the time series data were in the range of those derived from PCA-PMF (AFS); and those for the incinerator,

with previously published values.

r 2006 Elsevier Ltd. All rights reserved.

Keywords: SEAS II; ETAAS; Time-resolved metals; PM1.2; BRACE

e front matter r 2006 Elsevier Ltd. All rights reserved

mosenv.2005.12.036

ing author. Tel.: +1301 405 1859;

9121.

ess: [email protected] (J.M. Ondov).

the College of Public Health, University of South

, FL, and on assignment to USEPA, ORD, RTP,

1. Introduction

While an association between exposure to parti-culate matter (PM) and human health effects hasbeen known for decades (Dockery and Pope, 1994;

.

ARTICLE IN PRESSJ.P. Pancras et al. / Atmospheric Environment 40 (2006) S467–S481S468

US EPA, 1996), recent epidemiological findingsimply that aerosol particles with diameters o2.5 mm(PM2.5) are more closely associated with dailymortality and morbidity than coarse particles(Schwartz et al., 1996; Klemm et al., 2000; Ladenet al., 2000). One casual hypothesis is that metalsmodulate the expression of pulmonary cytokines(e.g., Carter et al., 1997; Costa and Dreher, 1997;Wichmann et al., 2000; Mitcus, 2004; and referencestherein). Another suggests that Redox metals suchas Fe produce OH radical from H2O2 by the Fentonreaction, resulting in irreparable cellular damage(e.g., Kennedy et al., 1998; Mitcus, 2004). Severalmetals, including As, Hg, Sb, Pb, Cd, Cr, and Be,have been designated by EPA as ‘‘toxic airpollutants’’ owing to their toxicity as metabolicpoisons and/or carcinogens (Baird, 1995). Most ofthese metals are emitted in primary particles fromhigh-temperature combustion sources (Ondov andWexler, 1998), typically in discrete particle popula-tions with modal diameters near 0.1 mm, which growto between 0.3 and 1 mm by deposition of secondaryaerosol mass during atmospheric transport. As thecomposition of fuel and feed materials is reflected inthe aerosol particles emitted, elemental signatureshave long been used as the basis for identifying andapportioning the contributions of sources to ambi-ent air quality (Gordon, 1988). One method, i.e.,Chemical Mass Balance, has been approved by EPAfor use by the States for developing strategies (SIPS)for achieving compliance with air quality standardsfor particles (Gordon, 1988). For all these reasons,accurate measurement of the elemental compositionof ambient urban aerosol particles has been of greatinterest.

Much of this work is still done with integrated 24-h aerosol samples collected on filters. In fact, theEPA supports the operation of more than 250‘‘speciation’’ sites, at which aerosol samples arecollected for elemental analysis, specifically, to aidin source apportionment, largely, by employing theChemical Mass Balance (CMB) and factor analysis(FA) methods (Thurston and Spengler, 1985;Hopke, 1991; Paatero, 1997; Henry and Norris,2002). However, substantially improved resolutionof sources has been demonstrated by applyinghighly time-resolved sampling methods (Cahill etal., 1987; Rheingrover and Gordon, 1988; Kidwelland Ondov, 2001; Shutthanandan et al., 2001;Cahill et al., 2002; Kidwell and Ondov, 2004), bycorrelating transient concentration excursions ofmarker elements with wind directions and the

locations of known sources. For example, Kidwelland Ondov (2004) were able to identify the plume ofan individual oil-fired power plant with measure-ments made at 30-min intervals at College Park,MD, using the prototype University of MarylandSemi-continuous Elements in Aerosol Sampler(SEAS).

In May and November of 2002, we used thesecond-generation SEAS instrument (SEAS-II, Kid-well and Ondov, 2004) to collect aerosol samples attwo sites near Tampa, FL, as a part of the BayRegion Atmospheric Chemistry Experiment(BRACE) study. The objectives of our measure-ments were to (i) characterize background andelevated ranges of toxic elements, (ii) identify theirsources, and (iii) to explore the efficacy of usinghighly time resolved SEAS-II data to determinesource ‘‘profiles’’ needed for CMB modeling and tointerpret components resolved by FA methods.Herein, we focus on periods when prevailing windsfavored influence of plumes from an animal feedsupplements (AFS) plant, an incinerator, a sub-stantial source of arsenic, and a battery recyclingplant. As described below, abundance ‘‘profiles’’ fortwo of these sources are derived directly from thetime-series data and, additionally, using positivematrix factorization (PMF, Paatero, 1997), whereinthe number of factors was determined fromprinciple components analysis (PCA).

2. Methods

2.1. Sampling sites

Ambient aerosol samples were collected at tworural sites near Sydney, Florida, in May andNovember, 2002. The first site, herein referred toas ‘‘Sydney’’ (27.9653N, 82.2273W), was situated atthe intersection of Dover and Sydney Roads inValrico, Florida, about 15 km east of the Tampacity limit. The second site (Tower Dairy, 27.9208N,82.3831W) was located on a dairy farm in Brandon,FL, about 5 km south of the Tampa city limit, and17 km southwest of Sydney. As shown in Fig. 1,several large phosphate fertilizer production plantslie 24 km east-southeast of Sydney and another asimilar distance to the north-northeast (CF Indus-tries, at 221); and an AFS (Coronet plant, at astation angle of 801) lies about 17 km to the east.Some phosphate fertilizer is also produced at theCoronet facility. The Macintosh Power Plant lies36 km to the northeast. A large battery recycling

ARTICLE IN PRESS

Fig. 1. Map of the study site showing the location of the two sampling sites, industrial sources in the area and 2001 emissions data. Dark

circles indicate SO2 emitters (� 1–50, � 51–500, � 501–1000, and � 1001–100 00 tpy); triangles indicate fine PM emitters (D 1–20, D 21–75,

D 76–150, and D 151–400 tpy); and small dots represent facilities for which neither PM nor SO2 emissions are reported.

J.P. Pancras et al. / Atmospheric Environment 40 (2006) S467–S481 S469

plant is within 5 km north of the Dairy, severalpower plants are situated to the west of both sites,and several incinerators lie within 20 km west ofSydney. In addition, air quality at both sites isaffected by traffic on major freeways located asfollows: 7 km north and 2.5 km south of Sydney;and 2 km west, 4 km north, and 2 km east of theDairy site. Nearest roads are 0.5 km north of theDairy and 0.9 km east of Sydney.

2.2. Sampling and sample selection

Sampling was conducted between April 30th andMay 29th at Sydney, and between October 30th andNovember 16th at the Dairy, in 2002, using theSEAS-II instrument. Briefly, the SEAS-II consistsof a state-of-the-art dynamic aerosol concentrator

mated to an automated sample collector. In SEAS-II, particles as small as �75 nm are grown to a sizebetween 3 and 10 mm by condensation of watervapor using precisely metered steam injection. Aninlet impactor with a 1.2-mm cut-point was used.Ambient air was sampled at 90Lmin�1 and sampleswere collected automatically in polypropylene vials,every 30min. Collected samples were capped andstored in a cold room at �10 1C.

Additionally, 1-min wind speed and direction,and Federal Reference Method SO2 and NOx

measurements at Sydney were provided by theEnvironmental Protection Commission of Hillsbor-ough County (EPC-HC). NOy measurements werealso available (Williams et al., 1998; BRACE, 2002).Lastly, 24-h integrated PM2.5 samples were col-lected with a sequential dichotomous particulate

ARTICLE IN PRESS

Fig. 2. Atmospheric concentrations of metals measured simulta-

neously by 24-h filter and SEAS methods.

J.P. Pancras et al. / Atmospheric Environment 40 (2006) S467–S481S470

matter sampler (Ruprecht and Patashnik, NY) bythe EPC-HC for elemental analysis by the USEPAby energy dispersive X-ray fluorescence spectro-metry (XRF).

Approximately 1400 and 800 samples werecollected, respectively, at the Sydney and Dairysites. Herein, we discuss analyses for samplescollected (1) between May 22nd and 26th, whenwinds arrived at the Sydney at angles ranging from31 to 1001, wherein the major sources of PM2.5 arereported to be an AFS manufacturing plant and aphosphate fertilizer plant (see Fig. 1); (2) May 17th,when winds blew to Sydney from between 2201 and2601, wherein lies one of the Hillsborough Countyresource recovery incinerators (HCRR), a City ofTampa Resource Recovery incinerator, and theJanet and Charles Recycling facility (2421), and twohospitals; and (3) the periods May 4 and 5 (Sydney)and November 2 and 3 (Dairy), when both siteswere influenced by As sources. As discussed below,period 1 included two period winds from nearly 201(north-northeast), a continuous 10-h period ofrelatively swift (3 to 5m s�1) winds from 201 to501; and a 27-h period of winds between 701 and100o (easterly).

2.3. Slurry sample analysis

Elemental measurements were performed using aPerkin-Elmer SIMAA6000 electrothermal atomicabsorption as described in detail by Pancras et al.(2005). Commercial multi-element standard solu-tions from Perkin-Elmer and Claritas Centriprep(Metuchen, NJ) were used for instrument calibra-tions. Sample volume was determined gravimetri-cally. High-purity nitric acid was used to adjust thefinal acid concentration of the slurry samples. Afterultrasonic treatment, each sample was analyzed inthree replicates. All elemental concentrations werecorrected for laboratory blank values. The overalluncertainties in the concentration measurementswere obtained by propagating errors in samplingflow rates, sample volume determinations, labora-tory blanks, and replicate analyses.

2.4. Source profile analyses

Relative concentration ratios (i.e., source profiles)were calculated after subtracting average back-ground concentrations observed before and afterexcursions. These were referenced to a markerelement because PM2.5 mass data were not made

available. Uncertainty estimates of the net concen-trations were propagated from the analyticaluncertainties in the measured values. In each case,the ratios were computed relative to the mostabundant or most definitive tracer species. Uncer-tainty-weighted means for these ratios are listed inTables 2 and 3.

3. Results and discussion

3.1. XRF filter vs GFAAZ-SEAS comparison

SEAS sample analysis encompassed two periods(May 13 and 23) for which 24 h average metalconcentrations could be calculated for comparisonwith those derived from XRF analysis of 24 hintegrated filter samples. Results are shown in Fig.2, wherein ‘‘error bars’’ represent analysis precision,expressed as one standard deviation (s) derivedfrom propagated uncertainties (SEAS-GFAAZ) oras propagated from counting statistics and measure-ment uncertainties (XRF). As shown in Fig. 2, goodprecision was obtained by both methods for Al, Fe,Cr, and Zn in at least one of the samples; however,uncertainties in the XRF measurements weretypically larger than those derived for GFAAZ.This is especially true for Cd, which was notroutinely detected by XRF. For all but Al and Fe,concentrations measured by the two methods werewell within 2s. Concentrations of Al, Fe, and Cr,which are often associated with supermicrometerdust particles, are generally lower as determined by

ARTICLE IN PRESSJ.P. Pancras et al. / Atmospheric Environment 40 (2006) S467–S481 S471

SEAS-GFAAZ. This is due in part to the larger cut-off size of the R&P inlet (2.5 vs 1.25 mm) and thefact that insoluble dust particles are inefficientlytransferred to the collection vials in SEAS-II(Kidwell and Ondov, 2004). Generally, efficiency ishigh when Al and Fe concentrations are notdominated by dust.

3.2. Summary statistics

A statistical summary, including median, 10th,and 90th percentile concentrations of elementsmeasured in all valid samples analyzed for the twosites is listed in Table 1. Higher blanks for Cu andZn were unavoidable, which resulted in poorermethod detection limits (MDL) for these twoelements in spite of their good atomic absorptionsensitivity. Abundant elements such as Al and Fewere measured in all samples, i.e., even at theirlowest background levels, whereas Cd, Cr, Cu, Niand Se were present above their MDL in only about50% of the samples. Thus, for the latter set ofelements, background concentrations could not beroutinely determined. Nevertheless, large concen-tration maxima were observed for As (87 ngm�3),Cd (20 ngm�3), and Zn (403 ngm�3) owing to theinfluence of discrete plumes from industrial orutility sources. The median Pb concentrationobserved at the Dairy was substantially greaterthan that for Sydney, likely owing to the proximityof the receptor site to a secondary lead recyclingplant (Golf Coast, 4 km to the North).

Table 1

Statistical summary of elemental concentration measurements

Al As Cd

Sydney sampling site

Total number of valid field samples analyzed 366 366 367

Minimum conc. measured (ngm�3) 3.17 o0.03 o0.04

10th percentile conc. (ngm�3) 6.93 0.04 0.07

Median conc. (ngm�3) 17.13 0.50 0.26

90th percentile conc. (ngm�3) 44.28 1.81 1.45

Maximum conc. (ngm�3) 133.22 86.60 19.58

% determinations above MDL 100 76 50

Tower Dairy sampling site

Total number of valid field samples analyzed 92 101 101

Minimum conc. measured (ngm�3) o3.00 0.35 0.06

10th percentile conc. (ngm�3) 3.65 0.43 0.13

Median conc. (ngm�3) 13.13 1.65 0.26

90th percentile conc. (ngm�3) 27.70 3.74 0.45

Maximum conc. (ngm�3) 59.81 10.76 0.78

% determinations above MDL 100 99 66

3.3. Concentration time series analysis

Pollutant concentrations are plotted against time-of-day for periods 1, 2, and 3 in Figs. 3, 6 and 7,respectively. Figs. 2–6 contain data only for Sydney,whereas Fig. 7 contains data for both Sydney andthe Dairy. In each of these figures, prominentexcursions are labeled with a number representingthe nth source tentatively identified, followed by aletter for each successive occurrence of an excursioninduced by that source.

In Tables 2 and 3, we list relative concentration‘‘profiles’’ for the sources and their station angles(as measured from north at the sampling site ofobservation) for the likely source of the excursion.The basis for source assignment is wind directionwith respect to station angle, wind speed and sourcedistance, and plume composition. The mean trans-port wind speed was also estimated, iteratively,based on the distance from the suspected source andtransport time at the average wind speed. The meantransport speed estimate and the average andstandard deviation of the wind angle over theestimated transport time are also listed.

3.3.1. Northeast quadrant (May 23, 24, 26; Sydney)

Prevailing winds during this period permittedobservation of the influences of sources from thenortheast and east of Sydney. As shown in Fig. 3,six substantial excursions in SO2 are clearly evident.The excursion at 6 AM on the 23rd (labeled 1a)was accompanied by Zn, Pb, and substantial

Cr Cu Fe Mn Ni Pb Se Zn

328 340 364 366 365 366 366 367

o0.05 o0.50 1.72 o0.14 o0.18 o0.31 o0.02 o3.50

o0.05 o0.50 4.38 o0.14 o0.18 o0.31 0.20 o3.50

0.20 0.33 11.31 0.53 0.41 0.78 0.48 3.62

1.38 1.30 33.15 1.40 1.48 3.22 1.36 12.10

11.73 5.20 350.10 7.36 6.02 9.18 3.42 403.30

62 37 100 83 31 66 49 74

92 91 92 92 101 101 101 101

o0.05 o0.50 4.11 0.28 o0.18 0.50 0.15 3.98

0.06 0.61 6.35 0.39 o0.18 1.08 0.39 6.15

0.11 1.39 11.32 0.69 0.22 3.16 0.67 10.36

0.24 2.54 26.83 1.37 0.94 5.88 1.32 21.57

0.51 4.81 59.54 2.82 1.24 56.18 2.74 70.89

32 91 100 100 19 100 80 100

ARTICLE IN PRESS

Fig. 3. Wind measurements and concentrations of SO2, NOy, and metals measured between May 23rd and 26th, at Sydney.

J.P. Pancras et al. / Atmospheric Environment 40 (2006) S467–S481S472

concentrations of NOy, but relatively small amountsof Se (0.5–1.0 ngm�3). During the previous 5.5 h,the wind angle was consistently between 71 and 121(1075) at a mean speed of 1.3m s�1. The presenceof Se suggests a coal combustion source, for whichZn:Se ratios are near 1.8 (Park et al., 2005);however, the relatively large Zn:Se ratio (5:1) andthe presence of a concurrent excursion in ammonia(21.8 ppb; Al-Horr et al., 2003) suggest influencefrom a phosphate fertilizer plant (which in a priorepisode, Zn:Se was found to be between 5 and 9).

A similar pattern of concentrations followed at9:30 AM (1b), when the mean wind speed and anglefor the preceding 2 h were 3m s�1 and 51715,

respectively, as the direction shifted from 17711,3 h earlier. Because the wind was shifting, plumeswere likely following more curved and broadlydispersed paths and winds aloft may have beenrotated somewhat, precluding association with aspecific source.

Excursions occurring between noon on the 23rdand noon on the 24th (labeled 2a–c) and again onthe 26th (2d) occurred under predominately easterlywinds. The first of these (2a) was small (maximumof 5 ppb SO2) and accompanied by substantialexcursions in Zn, Pb, Cr, Cd, and some As, but verylittle Se and NOy. This peak occurred at 8:30 PM onthe 23rd, when the mean wind direction for the

ARTICLE IN PRESS

Fig. 4. Source contribution metrics derived from PMF for samples collected between midnight May 22nd and noon on the 24th.

J.P. Pancras et al. / Atmospheric Environment 40 (2006) S467–S481 S473

previous 1.8 h was 78731 and corresponding meansurface speed, 2.6m s�1. This corresponds to thedirection of an AFS plant (operated by CoronetIndustries) 17 km upwind at a station angle of 801.In fact, if the transit speed were twice the meansurface speed, the plume transit time would be 1 hand the mean direction during that hour would be80� 31. This factor of 2 between wind speedaverage measured at 10m and the plume transittime estimated from the distance of the likely sourceis consistent in many of the periods describedherein.

The second peak (2b, maxima at 11:30 PM on the23rd) is of similar composition, and occurred whenmean wind direction for the previous 2.5 h (84731)remained from the direction of AFS plant. How-ever, the corresponding mean surface wind speed(1.8m s�1) was substantially slower than that for theprevious case, and concentrations were substantiallygreater (e.g., the maxima for SO2 and Cd were 17.5ppb and 19.6 ngm�3, respectively). The AFS facilityhad been under scrutiny by the EPC-HillsboroughCounty for quite some time after local residentscomplained of suspicious atmospheric discharges

from it, despite the fact that neither SO2 nor metaldischarge to the atmosphere had been reported.Consequently, on the 23rd, CALPUFF air qualitydispersion modeling results (Poor, 2003) usingmeasured SO2 emission rates, available at 8:00 and11:30 PM for the SO2 plumes from the McIntoshpower plant (36 km from Sydney at 681), suggestthat the plume from McIntosh was very weak atSydney at 8:00 PM and lay just north of Sydney at11:30 PM. Moreover, the plume from CoronetIndustries AFS plant would have been directly overSydney at 11:30, i.e., in agreement with ourhypothesis that the AFS plant was the major sourceof this excursion.

The third SO2 excursion (2c, maxima at 6:30 AMon the 24th) was also accompanied by excursions inZn, Pb, Cr, Cd, and As, but with larger proportionsof NOy and Se. As indicated in Fig. 3, surface windspeed was declining (from 1.7 to a minimum of0.1m s�1) between 12:30 AM on the 24th and thetime of observation of the SO2 maximum, and winddirections changed from near 801 to at least 931before speeds became too low to accurately measuredirection. Because of the low wind speeds, transport

ARTICLE IN PRESS

Fig. 5. Measured concentrations of Al, Fe, and Mn and those predicted by the PMF road-dust component.

Fig. 6. Wind and elemental concentration data for samples collected on May 17th when prevailing winds were aligned with an incinerator.

J.P. Pancras et al. / Atmospheric Environment 40 (2006) S467–S481S474

directions cannot be reliably estimated. The pre-sence of Se and a large excess of NOy suggestinfluence from a power plant. The nearest powerplants, McIntosh, a coal-fired plant located 32 km

away at a station angle of 681, and the Larsonpower plant (31 km at 741) are, therefore, crediblecandidate sources. Continuous emission monitor(CEM) data showed that Larson operated through

ARTICLE IN PRESS

Fig. 7. An Arsenic plume observed at Sydney on May 4th (A) and arsenic and lead observed at the Dairy on November 2nd (B).

J.P. Pancras et al. / Atmospheric Environment 40 (2006) S467–S481 S475

most of the evening on May 23, but as this is a gas-fired plant, the SO2 emissions would be very lowand NOy emissions high. The combined influence of

both these plants would explain the concurrence ofSe and high NOy concentrations in 2c. Finally, thefourth SO2 peak (2d, at 9:30 PM on the 26th) was

ARTICLE IN PRESS

Table 2

Source profile for animal feed supplements (AFS) company

Species/SO2 (ng mg�1)

Time-series-derived ratios PMF-derived ratios duration

including peaks 2a–cPeak 2a Peak 2b Peak 2c Peak 2d

05/23/02 05/23/02 05/24/02 5/26/02 evening

n ¼ 3 n ¼ 6 (main

peak)

n ¼ 3 n ¼ 3 n ¼ 72

Mean transport wind angle 8073 8473 84713 8272 —

Mean surface wind speed (m s�1) 2.3 1.76 0.8 2.4 —

Mean transport wind speed (m s�1) 4.6 2.7 1.6 5.6

Al 0.8570.24 0.5870.13 0.3770.09 0.6170.16 0.3370.06

As 0.1270.04 0.1370.02 0.270.05 0.1570.04 0.1370.01

Cd 0.3670.1 0.5370.08 0.4670.06 0.3470.08 0.5470.05

Cr 0.370.08 0.2970.05 0.4370.06 0.2770.07 0.3570.03

Cu 0.1470.01 0.0670.01 0.0270.02 0.0570.02 0.0170.01

Fe 0.6670.26 0.5270.12 0.7070.24 0.3470.1 0.2070.04

Mn 0.0470.01 0.0470.02 0.0170.01 070

Ni 0.0170.02 0.0370.08 0.0870.05 070.01

Pb 0.2670.09 0.1470.03 0.1370.05 0.5370.14 0.1570.02

Se 0.0570.05 0.0370.01 0.0970.04 0.0570.02 0.0170.01

Zn 0.7870.19 0.9770.17 1.4770.18 1.1570.29 0.7870.10

SO2 170.29 170.13 170.11 170.21 170.07

NOy 0.1670.06 0.0770.01 0.9170.16 0.170.02 070.02

Table 3

Comparison of incinerator profiles with stack emission studies

Element to zinc ratio (mg of element/g of Zn)

Ambient sub-hourly

metals by SEAS (2002)aPrevious study (suspended particles with diameter p2mm)

RRI-FL (1995)b MW-FL (1995)c MWI-VA (1978)d SWRC (1978)e

Al 3574.6 22.5714.5 40.6 56

As 16.571.8 3.670.9 0.470.1 1.8 2

Cd 14.272.2 2.7871.3 20.1722.3 9.5 13.6

Cr 1.570.7 1.4 4.4

Cu 8.571.1 13.7371.9 45.476.7 16.5 10.3

Fe 22.876.8 11.371.6 81.0711.6 30.9 14.7

Mn 2.170.4 0.670.1 2.770.8 7.5 2

Ni 2.271.6 1.370.3 1.770.7

Pb 14.373.1 118716 159724 844 644

Se Not detected 0.1 0.2

aWeighted average and standard deviation.bResource Recovery Incinerator—Miami, FL; stack sampling (Shrock et al., 2002).cMedical Waste Incinerator—Miami, FL; stack sampling (Shrock et al., 2002).dMunicipal Incinerator—Alexandria, VA; stack sampling (Greenberg et al., 1978).eSolid Waste Reduction Center—Alexandria, VA; stack sampling (Greenberg et al., 1978).

J.P. Pancras et al. / Atmospheric Environment 40 (2006) S467–S481S476

again accompanied by Zn, Pb, Cr, Cd, and As, andvery little NOy. The mean surface wind direction forthe preceding 0.9 h was 82721 at a mean surfacespeed of 2.4m s�1, and the mean transport speed isestimated to be 5.6m s�1.

As shown in Table 2, composition profiles derivedfrom peaks 2a, b, and d were in good agreement forthe important components (As, Cd, Cr, Zn, andSO2), and well within the limits of their uncertain-ties, hence, are tentatively assigned to the AFS

ARTICLE IN PRESSJ.P. Pancras et al. / Atmospheric Environment 40 (2006) S467–S481 S477

plume. It is noteworthy that EPC-HC also mea-sured airborne metals in and around the AFSfacility (Price, 2003) using standard ambient mon-itoring methods, but were unable to detect the AFSplant emissions. In fact, one of the largest 24-haverage Cd concentration calculated from the SEASII measurements of our limited sampling intensiveat Sydney (5/23/02 7AM to 5/24/02 7AM) was2.8770.13 ngm�3 but was reported to be1.4071.39 ngm�3 by the standard method (i.e,XRF analysis of the 24 h integrated filter samples).Moreover, this belies the fact that the 30-minmaximum concentration was 19.470.3 ngm�3, i.e.,only slightly less than the inhalation referenceconcentration (20 ngm�3) set by the USEPA Officeof Air Quality and Standards.

3.3.1.1. Factor analysis. Because peaks 1a, 1b, and2c each appeared to contain contributions from twosources, factor analysis methods were applied inattempt to better resolve their profiles. For thisanalysis, 1-min ambient SO2 and NOy concentra-tions were averaged to 30-min corresponding to theSEAS sampling periods, and converted to mgm�3

and included in the data matrix along with the 11elements measured between May 23rd and noon onthe 25th. A varimax rotated PCA of the final72� 13 data matrix yielded four principle compo-nents with eigen values greater than 1. The firstfactor accounted for 44% of the total variance, andhad high loadings for Cd, Cr, As, Pb, Zn and SO2,which corresponds to the excursions induced by theAFS facility as identified above. Factor 2 explained18% of the total variance, and contained highloadings for Mn, Cu, and Al, which we attribute tourban dust contaminated by metals from tire (Mn)and break (Cu) wear. The third factor explained14% of variance and, weighed predominantly by Seand NOy, is identified as the coal-combustioncomponent. Eleven percent of the variance wasexplained by the fourth factor, which was weighedby Fe and Ni. As shown below, the excursioncorresponding to this factor is accompanied by SO2

and Se (between 4:00 and 6:00 AM on the 23rd),which is resolved as a second power plant by PMF.The quantitative contribution of each source groupto the aerosol burden was then attempted by theprocedure suggested by Thurston and Spengler(1985). However, negative absolute factor scoreswere obtained for the first source category exceptduring the peaking periods of the respective highloading elements.

To obtain positive solutions, PMF (Hopke 1991;Paatero, 1997) was applied to the data set. In PMFthe factor analysis equation is solved by a leastsquares method in which Chi2 is weighted usingerror estimates for the individual measurements andsolutions are constrained to be positive. Criticalsteps in PMF analysis are determining the numberof factors and removing rotational ambiguity(optimizing FPEAK). One method of choosing thenumber of factors is to assess the fit variable Q (Q inPMF is the sum of the weighted Chi2s, Hopke,1991). For our case, a four-source model was run inaccordance with our expectations of three sourcetypes identified above from the time-series analysis(i.e., phosphate plant, Coronet AFS, and coal-combustion) and dust, based on the PCA result.Uncertainties used in the analysis are those origin-ally derived from error propagation. A smooth andbroad shaped parabolic curve was obtained whenFPEAK was varied between �2 and +2. Therefore,0 was assumed to be optimum. PMF predictsabundance ratios (F matrix) and a matrix (G) ofconcentrations, which in our case represent the sumof the masses of species loaded in the various factorsfor each measurement interval and is, thus, anabsolute pollutant mass metric in ngm�3 (but afterscaling SO2 to mgm�3). Actual elemental concentra-tions (in mgm�3) predicted from the products of F

and G matrix elements are plotted in Fig. 5, for oneof the sources (road dust).As shown in Fig. 4, the first factor (panel A), was

clearly loaded with Cd, Cr, Pb, As, Zn and SO2,which corresponds nicely to peaks 2 a–c (see Fig. 3),which we ascribed to the Coronet facility based onwind direction and composition. A second factor(Fig 4, panel B) contained Al, Fe, Mn, Cu and Zn(see Fig. 5). This factor nicely follows the rise andfall of the wind speed during this period, and istherefore readily identified as dust. Ambient con-centrations of Fe, Al, Mn, Cu, and Zn are well fit bythis component, except as expected, during periodsof influence of the other sources.The third factor contains (Fig. 4, panel C) SO2,

NOy and Se, which we ascribe to the influence ofpower plants, inferred from the time-series analysisdiscussed above. Moreover, excursions in thisfactor’s concentrations correspond to those ob-served in 1a, 1b, and 2c, as we had anticipated.(Note that small power plant contributions are alsopredicted for peaks 2a and 2b). However, winddirections were much too different for these to bethe same coal-fired power plant. The data of the

ARTICLE IN PRESSJ.P. Pancras et al. / Atmospheric Environment 40 (2006) S467–S481S478

26th (2d) suggest that some Se is associated with thissource, which is likely due to the presence ofphosphate fertilizer production. However, it ap-pears that Se has been removed from this factor byPMF and, instead, loaded onto the power plantfactor. Nevertheless, PMF does resolve what appearto be these two power plants. As shown in panel D,the fourth factor is nearly exclusively associatedwith peak 2c, during which time (6:00 AM on May24th) winds were from the east. This factor containsNOy and As but not Se, which PMF attributes tothe Coronet Industry factor (panel A). The Asapportioned for source 4 is small (�2.5 ngm�3)which may simply be some residual belonging tofactor 1. On this basis, we ascribe the fourth factor(panel D) to the influence of a natural gas-firedpower plant as inferred from the time seriesanalysis, above. Thus, the influence of two differentpower plants are predicted by PMF, in agreementwith the time series analysis.

3.3.1.2. AFS abundance profile. Composition pro-files (in this case ratios of species to SO2) derived forthe AFS plant from the time series analysis andfrom PMF are listed in Table 2. With the exceptionof Pb, agreement between ratios is excellent for thetwo major peaks (2b and 2d) ascribed to theCoronet AFS plant. As mentioned above, both thetime series and PMF analyses indicated that peak 2cwas also influenced by coal- and gas-fired powerplants. As shown in Table 2, ratios for 2c weresomewhat larger for NOy, and perhaps marginallyso for As, Cr, and Zn. Ratios derived from PMFencompass all three occurrences of this source andgenerally agreed well with those for the main peak(2b) for species measured well above background.

3.3.2. Period 2: May 17 (incinerator plume;

Sydney)

As shown in Fig. 6, an extremely large Znconcentration (403 ngm�3, i.e., about 50-fold great-er than its background concentrations during theentire study period) was observed between 4:30 and5:00 PM on May 17. Arsenic, Cd, Ni, Pb and Cuconcentrations were also elevated (7.9, 4.5, 2.4, 9.2,and 5.2 ngm�3, respectively). In contrast, SO2 andNOy were at their background levels. Incineratorparticles contain up to 40% ZnCl2 (by mass) and�1% levels of Cd, Pb, Cr, and Sb (Greenberg et al.,1978; Ondov and Wexler, 1998) and are often amajor source of atmospheric Zn in urban areas(Gordon, 1988). Here, Cd and Pb concentrations by

mass (of ZnCl2) are �0.5 and 1%, respectively, ingood agreement with published values (see Table 3),including those for two incinerators in Miami. Asshown in Table 3, concentrations relative to Zn areremarkably similar to those reported by Greenberget al. (1978) for the Alexandria-VA Municipalincinerator and the Washington DC Solid WasteReduction Center nearly 30 years ago, and byShrock et al. (2002) for the Municipal waste andmedical waste incinerators in Miami, FL, in 1995.Notable exceptions are those for As, which is 8-foldgreater than Greenberg et al.’s ratio; and Pb, �50-fold less. We suspect the high As:Zn ratio in ourstudy reflects burning of pressure treated lumber,which is used abundantly in the damp Tampa Bayclimate. As:Zn ratios for the Miami incineratorswere also substantially less than that derived fromthe ambient SEAS measurements. However, As isoften volatile at stack sampling temperatures andmight have been inefficiently collected. The lowerPb:Zn ratio, here, probably reflects reduction of Pbuse in paints and newspaper print. This ratio was,nonetheless, 10-fold greater for the Miami incin-erator particles.

During this period, surface wind arrived almoststeadily from an angle of 2401. We estimate themean transport wind angle to have been 238781 ata mean surface speed of 2.8m s�1. The meantransport speed is estimated to be 3.1m s�1. Thereare three Hillsborough County municipal incinera-tors (HCMI), a hospital incinerator, and an aircurtain incinerator at a waste recycling facility(Janet and Charles Recycling: JCR), located alongthis westerly wind trajectory. According to EPC-HC, the HCMI located 17 km from the Sydney at astation angle of 2501 and the JCR (18 km at 2461)were non-operational during our study althoughthese facilities had valid permits. The nearbyBrandon Regional Hospital (5.6 km at 2331) hasemissions from a boiler but we were unable todetermine if trash is also burned in the boiler.However, the City of Tampa (�25 km at 2651) andthe Faulkenberg road (�10 km at 2651) incineratorswere operational. Given the proximity of theFaulkenberg Rd HCRI to the sampling site, wesuspect that an edge of its plume could haveimpacted the Sydney site on May 17th.

3.3.3. May 4– 5, November 2– 3 (arsenic and lead;

Sydney and Dairy sites)

A dramatic elevation in As was measured atSydney at 9:00 PM on May 4th at which time its

ARTICLE IN PRESSJ.P. Pancras et al. / Atmospheric Environment 40 (2006) S467–S481 S479

concentration increased from background levels of1.3 to 87 ngm�3 (Fig. 7A). This excursion wasaccompanied by considerable excursions in Zn(51 ngm�3) and to a minor extent, Pb (5 ngm�3),but not substantially by any of the other metalsmeasured, including Se. This occurred as windspeeds dropped precipitously from the 5:30 PMhigh of 2.5 to 0.4m s�1 at 9:00 PM, and themeasured surface wind direction, which had beenconsistently near 2401 between 4:30 and 6:00 PM,became highly variable and more northerly(3001–3201). Arsenic concentrations remained ele-vated until at least 7:00 AM the next morning undercontinuing stagnation conditions. We argue thatsince the surface wind direction abruptly changedfrom southwesterly to a mean of 318731 at 1m s�1,o2 h before the arrival of the excursion, themaximum transport distance would likely havebeen 7 km at the mean surface speed, and 14 km ata more likely transport speed of 2m s�1 (i.e., twicethe surface speed). This fits nicely with the locationof a construction waste incineration site 15 kmfrom Sidney at 3151. Pressure treated lumbercontaining an arsenate preservative is used exten-sively in Florida and is probably burned in thisfacility. Such material contains as much as1.4 kgAsm�3 wood (based on retention of6.4 kgCuCrArsenatem�3 wood, a value suitablefor pressure treated wood prepared to be in contactwith the ground; US EPA, 2003), i.e. as much as35 g As per single 12 foot 200 � 600 piece of pressuretreated lumber. During the Sydney study, Park et al.(2005) estimated meteorological dispersion factorsof 10�6–10�5 sm�3 for two small stationary sources(stack heights both 46m) located 20 and 15 km fromSydney. Application of these to the maximumobserved As concentration (87 ngm�3) sited abovesuggests that this concentration could be induced byburning only two 120 � 200 � 600 pieces of pressuretreated lumber per hour, if half (Green, 2000) the Aswould escape into the atmosphere!

A much smaller As excursion (10.8 ngm�3) wasobserved at the Dairy at 7:00 PM on November 2nd(Fig. 7B) when mean surface winds for the preced-ing 1.5 h were 3.5m s�1 from 3357221, i.e., from thedirection of a pressure treated lumber manufactur-ing facility (Robbins Wood) 13 km from the Dairy.Elevated As concentrations continued for another2 h as winds from this direction persisted.

At 9 PM on the 2nd, Pb concentration increasedsharply to 56 ngm�3 as winds came from 339751 at3.8m s�1. The concurrence of SO2 along with Pb

and Se during this event is consistent with likelyGolf Coast emissions (a secondary lead recyclingplant, 4 km north at a station angle of �21 of theDairy site) as this facility operates two coal-firedblast furnaces.

Observance of sources in the same wind quadrantwithin a time interval of 2 h demonstrates theresolving power of our time-resolved tracer study.However, additional measurements and sophisti-cated modeling approach, e.g., Pseudo-Determinis-tic Receptor Model, are advisable, to verify theseinitial source apportionment assignments morecomprehensively.

4. Conclusion

We conclude that highly time resolved measure-ments for criteria gases and particle-borne metalswere extremely useful for revealing the short-termvariability in their concentrations and developinginsights into their sources. For the metals studied,atmospheric concentrations were 5–100 times great-er than their background levels when discreteplumes of local sources influenced air at thesampling sites. Agreement between abundance‘‘profiles’’ ascribed to an incinerator plume suggestthat ground-level sampling for ‘‘source profiles’’ isfeasible. Moreover, we detected and characterizedelevated emissions which we ascribe to an animalfeed supplements facility which were not detected bystandard ambient monitoring methods. Althoughnot discussed herein, the AFS emissions were laterverified by the EPC-FL through emission pointsampling (Price, 2005).

Acknowledgments

We thank Tom Atkeson (FLDEP), the BRACEprogram coordinator for his vision and support; hisstaff for criteria gases, speciation and geographicdata; Ben Hartsell (Atmospheric Research &Analysis, Inc.) for NOy and Purnendu Dasgupta(TTU) for ammonia data. Funding for this researchwas provided by FLDEP and the US Environ-mental Protection Agency Office of Research andDevelopment (EPA). This study has not beensubjected to the FLDEP peer or policy reviewand, therefore, does not necessarily reflect the viewsof the FLDEP and no official endorsementshould be inferred. However, it has been subjectedto EPA review and approved for publication.

ARTICLE IN PRESSJ.P. Pancras et al. / Atmospheric Environment 40 (2006) S467–S481S480

Mention of trade names or commercial productsdoes not constitute an endorsement or recommen-dation for use.

References

Al-Horr, R., Samanta, G., Dasgupta, P.K., 2003. A continuous

analyzer for soluble anionic constituents and ammonium in

atmospheric particulate matter. Environmental Science and

Technology 37, 5711–5720.

Baird, C., 1995. Environmental Chemistry, second ed. W.H.

Freeman and Co., pp. 347–381.

BRACE, 2002. Bay Region Atmospheric Chemistry Experiment,

/http://hsc.usf.edu/publichealth/EOH/BRACE/S.

Cahill, T.A., Feeny, P.J., Eldred, R.A., 1987. Size-time composi-

tion profile of aerosols using the drum sampler. Fourth

International PIXE Conference. Tallahassee, FL, June 9–13,

Nuclear Instruments and Methods in Physics Research B22,

344–348.

Cahill, T.A., et al., 2002. Analysis of size, time, and composi-

tionally resolved aerosols in Detroit, Michigan, February

25–April 10, 2002, /www.ladco.org/reports/rpo/MWRPO-

projects/Monitoring/UC_DavisReport.pdfS.

Carter, J.D., Gjop, A.J., Samet, J.M., Devlin, R.B., 1997.

Cytokine production by human airway epithelial cells after

exposure to an air pollution particle is metal-dependent.

Toxicology and Applied Pharmacology 46 (2), 180–188.

Costa, D.L., Dreher, K.L., 1997. Bioavailable transition metals in

particulate matter mediate cardiopulmonary injury in healthy

and compromised animal models. Environmental Health

Perspectives 105 (S5), 1053–1160.

Dockery, D.W., Pope, C.A., 1994. Acute respiratory effects of

particulate air pollution. Annual Review of Public Health 15,

107–132.

Green, A.E.S., 2000. Thermal disposal of CCA treated wood,

Florida Center for Solid and Hazardous Waste Management,

Report 00-07, University of Florida, June 2000.

Greenberg, R.R., Zoller, W.H., Gordon, G.E., 1978. Composi-

tion and size distributions of particles released in refuse

incineration. Environmental Science and Technology 12,

566–573.

Gordon, G.E., 1988. Receptor models. Environmental Science

and Technology 22 (10), 1132–1142.

Henry, R.C., Norris, G.A., 2002. EPA Unmix 2.3 User guide.

Hopke, P.K., 1991. Receptor Modeling for Air Quality Manage-

ment. Elsevier, Amsterdam.

Kennedy, T., Ghio, A.J., Reed, W., Samet, J., Zagorski, J., Quay,

J., Carter, J., Dailey, L., Hoidal, J.R., Devlin, R.B., 1998.

Copper-dependent inflammation and nuclear factor-kB acti-

vation by particulate air pollution. American Journal of

Respiratory Cell and Molecular Biology 19 (3), 366–378.

Kidwell, C.B., Ondov, J.M., 2001. Development and evaluation

of a prototype system for collecting sub-hourly ambient

aerosol for chemical analysis. Aerosol Science and Technol-

ogy 35, 596–601.

Kidwell, C.B., Ondov, J.M., 2004. Elemental analysis of sub-

hourly ambient aerosol collections. Aerosol Science and

Technology 38, 1–14.

Klemm, R.J., Mason, R.M., Heilig, C.M., Neas, L.M., Dockery,

D.W., 2000. Is daily mortality associated specifically with fine

particles? Data reconstruction and replication of analyses.

Journal of the Air and Waste Management Association 50 (7),

1215–1222.

Laden, F., Neas, L.M., Dockery, D.W., 2000. Association of fine

particulate matter from different sources with daily mortality

in six U.S. cities. Environmental Health Perspective 108 (10),

941–947.

Mitcus, R., 2004. Analysis of the role of zinc, a major component

of ambient Baltimore fine particulate matter, in eliciting

cytokine and chemokine release and disrupting cellular tight

junctions in vitro, Ph.D. Dissertation, University of Mary-

land, Baltimore County, MD.

Ondov, J.M., Wexler, A.S., 1998. Where do particulate toxins

reside? An improved paradigm for the structure and dynamics

of the urban mid-Atlantic aerosol. Environmental Science and

Technology 32, 2547–2555.

Paatero, P., 1997. Least square formulation of robust non-

negative factor analysis. Chemometrics and Intelligent La-

boratory Systems 37, 23–35.

Pancras, J.P., Ondov, J.M., Zeisler, R., 2005. Multi-element

electrothermal AAS determination of 11 marker elements in

fine ambient aerosol slurry samples collected with SEAS-II.

Analytica Chimica Acta 538, 303–312.

Park, S.S., Pancras, P.J., Ondov, J.M., Poor, N., 2005. A new

pseudo-deterministic multivariate receptor model for accurate

individual source apportionment using highly time-resolved

ambient concentrations. Journal of Geophysical Research 110

(D7) (citation no. D07S15).

Poor, N., 2003. CALPUFF modeling of sulfur dioxide and

nitrogen transformation products, unpublished data.

Price, D.J., 2003. Air sampling around Coronet junction, August

22–September 12, 2003. Memorandum submitted to the

Director of the Hillsborough County department of Health

and Florida Department of Health, October 8.

Rheingrover, S.W., Gordon, G.E., 1988. Wind-trajectory

method for determining compositions of particles from major

air pollution sources. Aerosol Science and Technology 8,

29–61.

Schwartz, J., Dockery, D.W., Neas, L.M., 1996. Is daily

mortality associated specifically with fine particles? Journal

of the Air and Waste Management Association 46 (10),

927–939.

Shrock, J., Bowser, J., Mayhew, W., Stevens, R.K., 2002. South

Florida mercury monitoring and modeling pilot study. EPA

report 600/R-00/102.

Shutthanandan, V., Thevuthasan, S., Disselkamp, R., Stroud, A.,

Cavanaugh, A., Adams, E.M., Baer, D.R., Barrie, L., Cliff,

S.S., Cahill, T.A., 2001. Development of PIXE, PESA and

transmission ion microscopy capability to measure

aerosols by size and time. Nuclear Instruments and Methods

in Physics Research, B: Beam Interactions with Materials and

Atoms.

Thurston, G.D., Spengler, J.D., 1985. A quantitative assessment

of source contributions to inhalable particulate matter

pollution in metropolitan Boston. Atmospheric Environment

19 (1), 9–25.

US EPA (Environmental Protection Agency), 1996. Air Quality

Criteria for Particulate Matter, RTP, NC, EPA Office of

Research and Development, National Center for Environ-

mental Assessment. EPA600-P-95-0001aF-cF.3v.

US EPA (Environmental Protection Agency), 2003. CCA Case

Review, Document OPP-2003-0250-0003, /http://docket.

ARTICLE IN PRESSJ.P. Pancras et al. / Atmospheric Environment 40 (2006) S467–S481 S481

epa.gov/edkpub/do/EDKStaffCollectionDetailView?objectId=

0b0007d480197894SWichmann, H.E., Spix, C., Tuch, T., Wolke, G., Peters, A.,

Heinrich, J., Kreyling, W.G., Heyder, J., 2000. Daily

mortality and fine and ultrafine particles in Erfurt, Germany.

Part 1: Role of particle number and particle mass. Research

Report 98. Health Effects Institute, Cambridge, MA.

Williams, E.J., Baumann, K., Roberts, J.M., Bertman, S.B.,

Norton, R.B., Fehsenfeld, F.C., Springston, S.R., Nunner-

macker, L.J., Newman, L., Olszyna, K., Meagher, J., Hartsell,

B., Edgerton, E., Pearson, J.R., Rodgers, M.O., 1998.

Intercomparison of ground-based NOy measurement techni-

ques. Journal of Geophysical Research—Atmospheres 103

(D17), 22261–22280.