ARTICLE IN PRESS

1352-2310/$ - se

doi:10.1016/j.at

�Correspondmental Enginee

Houston, TX 7

fax: +713743 4

E-mail addr

Atmospheric Environment 40 (2006) 508–520

www.elsevier.com/locate/atmosenv

Lanthanum and lanthanides in atmospheric fine particles andtheir apportionment to refinery and petrochemical operations in

Houston, TX

Pranav Kulkarnia, Shankararaman Chellama,b,�, Matthew P. Fraserc

aDepartment of Civil and Environmental Engineering, University of Houston, 4800 Calhoun Road, Houston, TX 77204-4003, USAbDepartment of Chemical Engineering, University of Houston, Houston, TX 77204-4004, USA

cDepartment of Civil and Environmental Engineering, Rice University, Houston, TX 77005, USA

Received 5 March 2005; received in revised form 26 September 2005; accepted 26 September 2005

Abstract

A study was conducted in Houston, TX focusing on rare earth elements (REEs) in atmospheric fine particles and their

sources. PM2.5 samples were collected from an ambient air quality monitoring site (HRM3) located in the proximity of a

large number of oil refineries and petrochemical industries to estimate the potential contributions of emissions from

fluidized-bed catalytic cracking operations to ambient fine particulate matter. The elemental composition of ambient

PM2.5, several commercially available zeolite catalysts, and local soil was measured after microwave assisted acid digestion

using inductively coupled plasma—mass spectrometry. Source identification and apportionment was performed by

principal component factor analysis (PCFA) in combination with multiple linear regression. REE relative abundance

sequence, ratios of La to light REEs (Ce, Pr, Nd, and Sm), and enrichment factor analysis indicated that refining and

petrochemical cat cracking operations were predominantly responsible for REE enrichment in ambient fine particles.

PCFA yielded five physically meaningful PM2.5 sources: cat cracking operations, a source predominantly comprised of

crustal material, industrial high temperature operations, oil combustion, and sea spray. These five sources accounted for

82% of the total mass of atmospheric fine particles (less carbon and sulfate). Factor analysis confirmed that emissions from

cat cracking operations primarily contributed to REE enrichment in PM2.5 even though they comprised only 2.0% of the

apportioned mass. Results from this study demonstrate the need to characterize catalysts employed in the vicinity of the

sampling stations to accurately determine local sources of atmospheric REEs.

r 2005 Elsevier Ltd. All rights reserved.

Keywords: Rare earth elements; PM2.5; ICP-MS; Factor analysis; Catalyst; Industrial emissions

e front matter r 2005 Elsevier Ltd. All rights reserved

mosenv.2005.09.063

ing author. Department of Civil and Environ-

ring, University of Houston, 4800 Calhoun Road,

7204-4003, USA. Tel.: +713 743 4265;

260.

ess: chellam@uh.edu (S. Chellam).

1. Introduction

The Port of Houston is the second largest in theUnited States (and sixth in the world) in terms oftotal tonnage of both imports and exports, whichincludes petroleum/petroleum products, iron andsteel, organic/inorganic chemicals, plastics, cereals

.

ARTICLE IN PRESSP. Kulkarni et al. / Atmospheric Environment 40 (2006) 508–520 509

and cereal products, crude fertilizers, minerals, etc.The region around the Port of Houston and theHouston Ship Channel is home to an extensiveindustrial complex that contains 49% of thepetrochemical manufacturing capacity in the UnitedStates (Greater Houston Partnership, 2003). Directquantification of the PM2.5 emissions from Hous-ton’s petrochemical complex, which is also thesecond largest worldwide, has not yet been done.Prior work has focused on the organic compounds(Fraser et al., 2003; Zhao et al., 2004) as well aselements from the main groups and first threetransition series (Buzcu et al., 2003) for fine particlesource identification and apportionment in Hous-ton, TX. To focus on catalyst emissions specific topetroleum refining and petrochemical manufactur-ing, it is necessary to measure lanthanum andlanthanides (Olmez and Gordon, 1985; Dzubayet al., 1988; Olmez et al., 1988).

Fluidized-bed catalytic cracking (FCC) is awidely utilized operation during petroleum refiningand petrochemical manufacturing. FCC operationstypically employ zeolite catalysts enriched in rareearth elements (REEs), a portion of which mayescape to the atmosphere (Mizohata, 1986; Dzubayet al., 1988). The term ‘‘rare earth elements’’ refersto the elements Y, La, and the lanthanides (Ce–Lu)(Cotton et al., 1999). Kitto et al. (1992) quantifiedselected REEs (La, Ce, Nd, Sm, Eu, Gd, Tb, Yb,and Lu) in several catalysts and demonstrated adirect association between catalyst composition andFCC stack emissions. Given their high economicimpact, catalyst chemical characteristics are beingcontinuously changed to enhance their performancein terms of metal tolerance and product selectivity(Richardson, 1989). However, to our knowledge,the only detailed reports of catalysts compositionincluding their emissions as fine particle from FCCoperations are �15–20 years old (Mizohata, 1986;Kitto et al., 1992). To update the existing andpotentially outdated information on catalysts com-position, several currently employed FCC catalystswere analyzed to determine the concentrations of allREEs, including Y and the previously unreportedlanthanides (Pr, Dy, Ho, Er, and Tm).

In addition to petrochemical industries, emissionsfrom motor vehicles and entrainment of crustalmaterial can also potentially contribute to atmo-spheric concentrations of REEs. Hence, we alsoanalyzed local soil, PM2.5 collected from a localhighway tunnel, as well as an automobile catalyst toevaluate possible interferences.

The objectives of this research are to (1) providecomprehensive data on the elemental compositionof several currently employed zeolite catalysts,(2) quantify REEs present in trace (ngm�3) andultra trace (pgm�3) amounts in atmospheric fineparticulate matter, and (3) determine the relativecontribution of loss of catalyst from FCC opera-tions to PM2.5 in the Houston, TX area. Ambientfine particles were collected at an air qualitymonitoring site located adjacent to the HoustonShip Channel to identify and apportion all possiblelocal sources. Microwave assisted acid digestion athigh temperature and high pressure was firstemployed to dissolve PM2.5, catalysts, and soil.Next, concentrations of 42 elements were measuredusing inductively coupled plasma-mass spectrome-try (ICP-MS).

2. Methods

2.1. FCC catalysts

Samples of five different FCC zeolites (designatedas SMR1–SMR5) employed in a wide range ofrefining operations in the Houston area wereobtained from the world’s leading catalyst manu-facturer (Grace Davison Inc., Columbia, MD). Allfresh catalysts were odorless and white to brown incolor, in the form of a fine powder with bulkdensities between �0.45 and 1.00 g cm�3. During catcracking, catalysts get poisoned by metals such asNi, V, Cu, and Fe, due to high temperaturefractionation and also by coke deposition (Richard-son, 1989). In other words, catalyst compositionchanges upon usage. Hence, a sample of used zeolitecatalyst was also obtained from Shell Deer ParkRefining Company’s Fluid Catalytic Cracking Unit(FCCU), located in the Houston Ship Channelindustrial complex. This sample was obtained inJune 2003 and consisted of the spent catalystcollected from cyclone separators that had removedit from the FCCU product stream. The spentcatalyst was also an odorless fine powder but wasdark gray in color.

2.2. Ambient PM2.5

Ambient fine particulate matter was sampledfrom an air quality-monitoring site (HRM3) locatednear the Houston Ship Channel. A total of 25samples were obtained between 25 May 2001 and 4September 2001. PM2.5 samples were collected on

ARTICLE IN PRESSP. Kulkarni et al. / Atmospheric Environment 40 (2006) 508–520510

Teflon membrane filters at a flow rate of 10 Lmin�1

using a fine-particle sampler that used an AIHLcyclone to remove coarse particles, in-line filterholders, and critical orifice plates to control volu-metric airflows. Three parallel sampling channelswere used to collect equivalent samples for multipleanalytical procedures. After a 24 h collection period,filters were sealed inside sampling Petri dishes andfrozen in the laboratory until further analysis.Additional information on the geographic locationof HRM3, sampling methods, and equipmentemployed are given elsewhere (Buzcu et al., 2003).

2.3. Highway tunnel PM2.5 and automobile catalyst

Six PM2.5 samples were collected between 29August 2000 and 1 Sepetember 2000 in a localhighway tunnel (Washburn tunnel). Fine particulatematter samples were collected using a low-volumesampler, at a flow rate of 10Lmin�1 on Teflonmembrane filters. Detailed information on thetunnel, vehicular traffic, PM sampling procedures,etc. is available elsewhere (Chellam et al., 2005). Aused pellet type automobile catalyst (SRM 2556)was also purchased from National Institute ofStandards and Technology (NIST), Gaithersburg,MD.

2.4. Soil

Five soil samples were also collected using wellaccepted protocols (EPA, 1992) from an areaencompassing a 5-mile radius from the HRM3sampling station to establish the local crustalabundance pattern of REEs. After collection, thesoil samples were air dried for 15 days at roomtemperature, sieved (1mm), and ground to powderwith an agate ball mill after oven drying at 60 1C for1 day. The finely ground homogenized soil wasagain sieved (40 mm) and analyzed immediatelyafterwards.

2.5. Microwave acid digestion

NIST SRMs were used to develop and validatedigestion methodologies and analytical methods. Inthe absence of a zeolite FCC standard referencematerial, SRM 2556 (recycled pellet automobilecatalyst) and SRM 1547 (peach leaves containingtrace to ultra-trace amounts of REEs) were used assurrogates for FCC catalysts. SRM 1648 (urbanparticulate matter) was used as a surrogate for

PM2.5. Because the same procedure can be used fordigesting both soil particles and atmosphericparticulate matter (Wu et al., 1996), a separate soilSRM was not evaluated in this study.

Samples were digested in two-stages (Wu et al.,1996; Kulkarni et al., 2003) in Teflon PFA decom-position vessels in a microwave oven (MARS5,CEM Corp., Matthews, NC). In the first stage, acombination of 48% hydrofluoric acid (PPB/Teflongrade, Fluka, Milwaukee, WI) and 65% nitric acid(Suprapur grade, EM Science, Gibbstown, NJ) wasused to extract elements associated with siliceousmatter, which was present in significant amount(�30–50%) in all the samples. HF was employedin stoichiometric excess to ensure the completedissolution of the silicon matrix. Boric acid (Supra-pur grade, EM Science, Gibbstown, NJ) was addedduring the second digestion stage to mask anyexcess HF and dissolve fluoride precipitates. A20min dwell time for each stage and temperatureand pressure set points of 200 1C and 200 psigwas sufficient to achieve near 100% recoveriesof all certified and uncertified elements in each ofthese SRMs. Hence, these settings were alsoemployed for the actual catalyst, soil, and PM2.5

samples.Fifty milligrams of FCC catalyst, automobile

catalyst, and soil were digested using 5mL HNO3,0.4mL HF, and 2.4mL of 5% H3BO3. AmbientPM2.5 samples were digested using 2.5mL HNO3,3 mL HF, and 24 mL of 5% H3BO3. Highway tunnelfilters were analyzed using our previously developedmethod wherein HF was generated in situ byheating a mixture of the sample, nitric acid, andsodium fluoride (Kulkarni et al., 2003). Afterdigestion, all samples were diluted to achieve a finalHNO3 concentration of 0.4M before ICP-MSanalysis. Catalyst digestates were further diluted asnecessary to analyze all elements within the instru-ment’s dynamic range.

2.6. ICP-MS

Elemental analysis was performed using ICP-MS(Elan 6000, Perkin Elmer, Norwalk, CT). Thedigestate of each sample (i.e. FCC catalyst, auto-mobile catalyst, soil, as well as ambient and tunnelPM2.5 filters) was split into four aliquots; three wereused to establish the precision of the ICP-MSmeasurements. The fourth was spiked with knownconcentrations of all the elements and also analyzedby ICP-MS to determine matrix spike recoveries for

ARTICLE IN PRESSP. Kulkarni et al. / Atmospheric Environment 40 (2006) 508–520 511

both. REEs were analyzed separately from otherelements during each run.

The ICP-MS was calibrated using multi-elementstandards consisting of 42 elements including allREEs (except Pm) (see Table 1). Pm cannot beanalyzed because no stable isotope of it occursnaturally (Cotton et al., 1999). An internal standard

Table 1

Elemental compositions for fluid cracking catalysts (5 fresh and 1 spen

Element Spent catalyst SMR 1 SMR 2

23Na 42407278 24677180 31897624Mg 50975 23472 2507427Al (%) 2571 2970.9 217128Si (%) 2073 2573 257439K (%) 2.170.2 0.670.1 1.170.445Sc 3772 25.770.3 28.070.47Ti 91097674 5132737 44887651V 44575 55.970.3 52.970.52Cr 133723 63720 6272455Mn 27.070.4 12.9670.00 22.170.57Fe 5984763 3559710 34817759Co 11772 3.4570.03 3.1570.60Ni 1094712 16.870.1 13.847063Cu 46.570.2 10.770.2 9.1870.68Zn 15372 71.070.9 927169Ga 6371 51.270.3 45.270.75As 6.3170.04 6.270.1 15.970.77Se 28.670.1 19.070.4 35.471.85Rb 2.5370.04 3.870.0 3.1670.88Sr 53.071.1 93.771.3 105.37089Y 1871 22.370.9 43.170.90Zr 66.370.4 6371 787195Mo 12.270.1 1.670.1 1.2870.111Cd o0.5a o0.5 o0.5133Cs 0.2270.06 0.2770.00 0.2470.137Ba 134.172.1 254.770.6 228.271139La 100197399 4597749 104527140Ce 770768 3120726 102657141Pr 502711 56675 169672144Nd 716710 96777 280674147Sm 15776 11873 23475153Eu 5.770.3 4.670.3 13.270.158Gd 493712 11973 329715159Tb 5.770.4 5.970.4 13.770.162Dy 2272 12.070.2 3070.4165Ho 1.370.1 1.270.1 2.370.2166Er 6.570.6 6.070.4 13.770.169Tm 0.470.0b 0.370.0 0.470.0172Yb 11.071.5 2.070.1 3.170.4175Lu o0.1 0.270.0 0.370.0208Pb 4871 35.070.3 31.970.238U 2.870.1 2.4470.01 2.4270.

All concentrations and respective standard deviations are given in mg gaValues after ‘o’ were below the minimum reporting level (MRL

experience and judgment of the analyst.bZero is accurate to one decimal place.

solution containing In (5 mgL�1) was used to correctinstrumental drift. To minimize isobaric interfer-ences, all instrument settings were adjusted, espe-cially the nebulizer gas flow rate (0.8–0.9 Lmin�1),to reduce formation of doubly charged ions andoxides even below the recommended values of0.01 for Ba++/Ba and 0.03 for CeO/Ce. The

t)

SMR 3 SMR 4 SMR 5

9 29567123 1208736 49357280

17470 23572 16471

1671 1870.6 1171

2773 2274 2572

1.470.2 2.3.70.2 2.170.3

8 26.070.9 23.670.8 19.471.1

5 38197187 89407195 6148789

4 34.870.1 7072 4073

52725 91732 81740

4 16.2670.05 11.470.2 9.170.1

7 2737719 4838741 286874

05 2.8870.07 4.870.1 3.270.3

.01 11.9870.08 2471 1571

02 6.4770.04 2271 8.570.4

59.470.1 7571 6771

4 3571 7173 4872

2 7.7770.03 1671 6.970.6

3 16.870.4 5672 14.970.3

04 2.1770.10 2.070.4 2.170.2

.8 6172 9573 3571

6 22.170.4 31.170.4 16.570.5

6072 4772 6472

01 1.1170.04 2.370.10 2.170.2

o0.5 o0.5 o0.5

05 o0.2 o0.2 o0.2

.0 18778 188711 8573

319 4782733 100507342 81857209

333 139276 61927122 1499714

6 49978 1183728 1402710

5 1278720 2553747 990716

10071 27174 8371

6 4.570.4 21.170.5 6.470.2

9772 1228724 6571

6 5.470.3 11.670.7 3.270.1

13.270.1 2370.6 8.270.1

1.170.1 2.070.2 0.770.2

4 6.770.4 13.270.4 4.370.1

0.270.0 0.470.1 0.170.0

1.670.1 14.970.8 0.870.1

0.270.0 0.670.1 o0.1

1 2271 4571 2471

02 1.6470.02 3.9870.02 2.770.5

�1 (except for Al, Si, and K, which are in % or g g�1� 100).

) calculated as 1–3 times the method detection limit based on

ARTICLE IN PRESSP. Kulkarni et al. / Atmospheric Environment 40 (2006) 508–520512

inter-element mass spectral overlaps for 152Sm, 164Dy,and 142Nd were eliminated by analyzing alternateisotopes, 147Sm, 162Dy, and 144Nd, respectively.Individual correction equations were derived forGd, Dy, and Eu, to reduce spectral interferencescaused by oxides of Ce, Nd, and Ba, respectively.Additional information regarding instrument tun-ing, calibration checks and parameter optimizationis given elsewhere (Kulkarni et al., 2003). Theseprocedures resulted in increased instrument sensi-tivity facilitating trace and ultra-trace REE mea-surements.

2.7. Source apportionment

Principal component factor analysis/absoluteprincipal component analysis (PCFA/APCA)(v. 13.0, SPSS Inc., Chicago, IL) was used toidentify sources of ambient PM2.5 and theirindividual contributions (Watson et al., 2002).PCFA produces several components each having astrong correlation with a specific group of elementsby capturing maximum variation in the datasetwithout a priori knowledge of source profiles(Hopke, 1999). While PCFA was used to investigatethe major sources of PM2.5 mass, APCA was usedto quantify and retrieve source contribution to eachmeasured element. The Varimax scheme was em-ployed to rotate factors and facilitate interpretation.Factors containing variance greater than unity afterVarimax rotation were retained and assigned tophysically meaningful sources by examining mostheavily weighted elements (high loadings). Moredetails of the PCFA/APCA methodology can befound elsewhere (Thurston and Spengler, 1985).

3. Results and discussion

3.1. Catalysts composition

The elemental composition of fresh FCC zeolitesand the spent catalyst is summarized in Table 1. Asexpected, Al and Si dominated all catalysts analyzedin this study, together comprising 36–54% of themass. Other important elements included K, Na, Ti,Fe, La, Ce, Pr, and Nd. The spent catalystcontained Ni, V, Co, Cu, and Mo in elevated levelscompared to the fresh catalysts consistent withcatalyst contamination or poisoning during crack-ing reactions. La and lighter lanthanides (Ce, Nd,Pr, Sm, and Gd) dominated the overall REE

composition (499%) in the catalysts with La(40–79%) and Ce (8–40%) being most dominant.

Catalysts composition provided by Kitto et al.(1992) and those shown in Table 1 differ in twomain respects. First, the relative abundance of a fewelements (K, Gd, Sm, and Zr) is different. K and Gdare significantly elevated (p ¼ 0:10 and 0.05, respec-tively) whereas Zr is significantly depleted (p ¼ 0:05)in catalysts analyzed in this study. Thus, in contrastto earlier results, Gd is generally present in higherlevels than Sm in catalysts reported in Table 1.Secondly, we measured Y, Pr, and the heavylanthanides (Dy, Ho, Er, and Tm) in FCC catalysts,which have not been quantified to date. Note thatone of the previously unreported REEs, Pr, wasmeasured in relatively high concentrations whichwere comparable to Ce and Nd in some of thecatalysts suggesting that it can be potentiallyemitted in significant levels from FCC operations.Hence, data given in Table 1 represents animportant addition to existing information oncatalyst composition in terms of (1) 6 additionalelements, (2) new catalyst formulations resultingfrom research and development over the past�15–20 years, and (3) inclusion of a spent catalyst.

3.2. Ambient PM2.5 composition

The composition of ambient PM2.5 in terms ofLa, 7 lanthanides (Ce, Pr, Nd, Sm, Eu, Gd, and Dy),and 15 other elements (Na, Mg, Al, Si, V, Cr, Mn,Fe, Co, Ni, Cu, Zn, As, Ba, and Pb) is shown inFig. 1. All other elements were consistently mea-sured below their respective minimum reportinglevels. The substantial variation in elemental com-position of PM2.5 is probably caused by theinherently high variability associated with local finePM sources in the Houston area (Russell et al.,2004). Similar to the catalysts shown in Table 1, Laand the light lanthanides (Ce, Pr, and Nd) were thedominant REEs in ambient PM2.5. Trace signaturesof Sm, Eu, Gd, and Dy were also measured, whereasthe heavier REEs Er, Tb, Yb, Ho, Tm, Lu were notpresent in PM2.5. Inspection of Fig. 1 reveals thatthe relative abundance of REEs in atmospheric fineparticles was in the order: La4Ce4Nd4Pr4Gd4Sm4Dy\Eu.

To our knowledge, this is the first report of Prand Dy in PM2.5 in an urban environment. We usethe measured Pr values to calculate enrichmentfactors (Section 3.3) and La/Pr ratios (Table 3)to demonstrate the contributions of FCC and

ARTICLE IN PRESS

NaMg Al Si V Cr Mn Fe Co Ni Cu Zn As Ba La Ce Pr NdSmEu Gd Dy Pb

1E-3

0.01

0.1

1

10

100

1,000

Co

nce

ntr

atio

n (

ng

/m3)

Fig. 1. Non-parametric depiction of the elemental composition of 25 samples of ambient fine particles. Heavier lanthanides Er, Yb, Ho,

Tb, Tm, and Lu were not detected in PM2.5. The box encompasses the 25th and 75th percentiles and the whiskers are determined by the 5th

and 95th percentiles. The horizontal line inside the box is the median and the open square is the average. Crosses (� ) denote 1st and 99th

percentiles and the maximum and minimum values are represented by dashes (–).

Table 2

Enrichment factors of Sc, Y, La–Lu present in catalysts relative

to soil using Fe as the reference

Element Spent catalyst SMR1 SMR2 SMR3 SMR4 SMR5

45Sc 9 11 12 14 7 1089Y 2 5 10 7 5 5139La 1291 986 2293 1334 1586 2179140Ce 63 432 1454 251 631 258141Pr 33 62 190 71 95 190144Nd 251 570 1692 980 1108 725147Sm 78 99 201 109 167 86153Eu 3 4 11 5 13 6158Gd 1293 524 1483 558 3986 358159Tb 3 5 11 5 7 3164Dy 14 13 33 18 18 11165Ho 1 1 3 2 2 1166Er 22 34 81 50 56 31169Tm 0 1 1 0 1 0172Yb 82 25 39 26 137 13175Lu 1 0 1 0 1 0

P. Kulkarni et al. / Atmospheric Environment 40 (2006) 508–520 513

automobile catalysts as well as the local soil. Thepresence of Dy in urban fine PM indicates that itsinclusion might strengthen source contributionestimations.

3.3. Enrichment factors

Table 2 summarizes enrichment factors of in-dividual elements (X) in catalysts with respect totheir crustal abundances calculated using iron as thereference element:

Enrichment factor ðX Þ ¼½X �catalyst=½Fe�catalyst½X �soil=½Fe�soil

.

Fe was chosen as the reference instead of Al or Sibecause it is highly abundant in soil compared tocatalysts whereas both soil and catalysts containlarge amounts of Al and Si. Enrichment factors nearunity suggest a similar concentration pattern in thecatalyst and soil (Watson et al., 2002). As observedin Table 2, all REEs except three heavy lanthanides(Ho, Tm, and Lu) were enriched in the catalystscompared to their crustal abundances. Generally,La and the lighter lanthanides (Ce, Pr, Nd, Sm, andGd) were greatly enriched compared to the heavierREEs (Dy, Eu, Er, Tb, and Yb), Sc, and Y. Largedifferences in REE composition of crustal material

and catalysts demonstrate that the contribution ofthese two sources to PM2.5 can be clearly distin-guished.

Light REEs enrichment factors for ambientPM2.5 relative to catalysts were also calculated withrespect to Gd. Gd was chosen as the reference

ARTICLE IN PRESS

0.1

1

10

100

1000

10000

Con

cent

ratio

n (µ

g/g)

La Ce Nd Pr Gd Sm Dy Eu Er Yb Ho Tb Tm Lu

Fig. 2. Concentrations of La and lanthanides in fluid cracking catalysts (box plot) and soil (filled circles). Results shown are for 5 fresh

catalysts and one spent catalyst and 5 soil samples. The box construction is the same as in Fig. 1.

P. Kulkarni et al. / Atmospheric Environment 40 (2006) 508–520514

because it was present at smaller levels in soil ascompared to the catalysts (see Fig. 2 and Section3.4). Enrichment factors of La, Ce, Nd, Pr, and Smwere all close to unity (median values were 0.6, 1.2,0.9, 0.5 and 0.7, respectively), which indicatesthat FCC emissions were the primary source oflight REEs to ambient fine particulate matter inHouston.

3.4. Concentration patterns of elements La– Lu in

soil, air, and catalysts

La and lanthanide concentrations in catalysts andsoil are depicted in Fig. 2. Catalyst composition isdepicted in a non-parametric fashion because of thelarge differences in elemental concentrations of thesix samples. In contrast, only the average soilconcentrations are shown because its variabilitywas very low. As observed from Fig. 2 and Table 1,the abundance sequence in all catalysts wasLa4Ce4Nd4Pr4Gd4Sm4Dy4Eu�Er�Tb4Yb4Ho4Tm�Lu. Comparison of Figs. 1 and 2reveals that the abundance sequence of REEsdetected in ambient fine particles was identical tothat in catalysts. In contrast, soil REE concentra-tions followed the sequence; Pr4Ce4La4Nd4Tb�Eu�Sm4Dy�Ho4Tm�Lu4Gd�Er4Yb,which is different from that observed in bothcatalysts and PM2.5. Identical sequence of REE

concentrations in catalysts and ambient PM2.5

further suggests that REE enrichment in fineparticulate matter was primarily caused by emis-sions from FCC operations.

3.5. La to light lanthanides ratios

Ratios of La to Ce, Nd, Pr, and Sm for FCCcatalysts, soil, ambient PM2.5, automobile catalyst,and Washburn tunnel PM2.5 are given in Table 3.Note that Sm concentrations in the Washburntunnel PM2.5 samples was below the minimumreporting level because a higher dilution factor wasemployed (Kulkarni et al., 2003; Chellam et al.,2005). As seen, each of these ratios was very similarin both FCC catalysts and atmospheric fine particlesbut was considerably different than that measuredin local soils, automobile catalyst, and PM2.5

collected in the Washburn tunnel. Additionally,similar to previous observations (Olmez and Gor-don, 1985; Kitto et al., 1992), the La/Sm ratio wasapproximately conserved between the FCC catalystsand PM2.5 further suggesting that FCC emission isthe primary source of REEs in PM2.5.

La/Ce ratios in Table 3 for ambient PM2.5

generally agree with previous investigations inurban areas (Kitto et al., 1992; Dillner et al.,2005). However, the mean La/Sm ratio for bothPM2.5 and catalysts was �54 (see Table 2) which is

ARTICLE IN PRESS

Table 3

La to light lanthanides ratios

Ambient PM2.5 FCC catalyst Soil Auto catalyst SRM 2556 Washburn tunnel PM2.5

Mean sa Median Mean s Median Mean s Median Mean s Median Mean s Median

La/Ce 2.9 1.8 2.4 4.3 4.6 2.5 0.65 0.04 0.7 0.7 0.0b 0.7 0.2 0.1 0.3

La/Pr 11.0 5.2 10.3 9.7 5.2 8.3 0.72 0.04 0.7 9.4 0.1 9.4 95.6 44.5 96.1

La/Nd 5.4 3.8 4.5 6.4 4.1 4.3 2.1 0.1 2.1 2.9 0.0b 2.9 36.3 19.8 32.4

La/Sm 53.7 36.1 49.1 55.2 23.3 46.2 3.9 0.2 3.9 200.7 1.1 200.7 Not determinedc

Number of samples ¼ 25, 6, 5, and 6 for ambient PM2.5, fluid cracking catalysts, soil, and Washburn tunnel PM2.5 respectively. Three

separate samples of SRM 2556 were digested and each digestate was analyzed in triplicate by ICP-MS.as is the standard deviation.bZero is accurate to one decimal place.cNot determined because Sm was less than the minimum reporting levels (0.003 ngm�3) because of higher dilution factor (Kulkarni et

al., 2003; Chellam et al., 2005).

P. Kulkarni et al. / Atmospheric Environment 40 (2006) 508–520 515

higher than that previously reported values in therange 17–25 in Philadelphia, PA and Osaka, Japan(Olmez and Gordon, 1985; Mizohata, 1986; Dzubayet al., 1988; Kitto et al., 1992). This is possiblycaused by differences in the composition of catalystsemployed locally as well as improvements incatalyst formulation over the last �15–20 years(see Section 3.1). These results demonstrate the needto characterize catalysts employed in the vicinity ofthe sampling station, rather than using literaturedata, in order to accurately identify local PM2.5

sources.Even though La/Ce ratios were not altered to a

statistically significantly degree (p ¼ 0:05) betweenFCC catalysts and ambient PM2.5, a slight depletionin the mean values from 4.3 to 2.9 was observed (seeTable 3). Note that the automobile catalyst SRM2556 was highly enriched in Ce rather than La(La=Ce ¼ 0:70) consistent with the previous auto-mobile emission study (Silva and Prather, 1997).Another NIST certified auto catalyst (SRM 2557,used monolith type auto catalyst) has an even lowerLa/Ce ratio of 0.05 (NIST). Similar to a previousinvestigation of vehicular emissions (Huang et al.,1994), the highway tunnel PM2.5 also revealed avery low La/Ce ratio (0.2). Further, motor vehiclesemit REEs primarily as coarse particles (Huanget al., 1994). The slight depletion of the La/Ce ratiofrom FCC catalyst to ambient PM2.5 indicates asmall contribution of these REEs from automobileemissions and soils. Therefore, as reported earlier(Olmez and Gordon, 1985), light REEs in PM2.5 arethe signature of petrochemical and refining opera-tions. Hence, La, Ce, Pr, Nd, and Sm can be used asmarkers in source apportionment calculations toisolate and determine FCC contributions to PM2.5.

3.6. Source apportionment using PCFA

Elemental composition of PM2.5 shown in Fig. 1was subjected to PCFA with Varimax rotation. Toisolate FCC emissions, La, Ce, Sm, and Nd thatwere enriched both in catalysts and atmospheric fineparticles were included in the factor analysis alongwith 15 other non-REE elements. Communalitiesfor Cr, Cu, and Zn were o0.6 suggesting that asubstantial fraction of their concentrations couldnot be apportioned to factors whose Eigen valueswere 41. Additionally, because PM2.5 sourcespreviously identified in the same site (Buzcu et al.,2003) were not enriched in Cr, Cu, and Zn, theseelements (and Pr) were later omitted to increase thedegrees of the freedom before performing factoranalysis. This resulted in 9 degrees of freedom,which is less than the recommended threshold of 20(Thurston and Spengler, 1985). However, consistentresults can be obtained even with lower degrees offreedom (Choi et al., 2003; Senaratne and Shooter,2004). Sensitivity tests were also performed in whichthe degrees of freedom were further decreased byincluding Cr, Cu, and Zn. Even though theseelements were all present in significant quantitiesin PM2.5, their inclusion did not alter the loadingmatrix and yielded identical factors. Importantly,several factors obtained in this work have also beenidentified as PM2.5 sources in previous studiesperformed in Houston Ship Channel area (Buzcuet al., 2003; Dillner et al., 2005). Hence, factoranalysis was able to estimate the sources in a stableand robust manner.

Fig. 3 shows elemental loadings resulting fromVarimax rotation. Loadings are the coefficients ofcorrelation between individual factors and the

ARTICLE IN PRESSP. Kulkarni et al. / Atmospheric Environment 40 (2006) 508–520516

elements. A loading close to one indicates that theelement is characteristic of that particular source.As observed in Fig. 3, five factors summing to 82%of total variation in the dataset were obtained.Communalities of all elements in the dataset were inthe range 0.60–0.95 indicating that each element wassatisfactorily apportioned to the identified factors.Each factor or source was physically interpreted byits association with strongly loaded marker ele-ments, typically emitted from that source.

The first factor showed very high loadings forlight REEs La, Ce, Sm, and Nd (40.80) and Al(0.70) and a moderate loading for Si (0.39). Theenrichment of highly loaded elements (light REEsand Al) in catalyst samples signifies that this factoris associated with petrochemical and refiningoperations. In this factor, significant loadings werenot obtained for V and Ni commonly associatedwith heavy fuel combustion suggesting a uniqueassociation with cat cracking operations. This factorwas not resolved by us earlier (Buzcu et al., 2003),because amongst the petrochemical markers, onlyLa was measured and incorporated into the crustalcontribution. However, in this earlier work thelanthanides (Ce–Lu) were not measured.

0.00

0.25

0.50

0.75

1.00Factor 1: cat cracking (41%)

0.00

0.25

0.50

0.75

0.00

0.25

0.50

0.75 Factor 3: High temp. operations (10%)

Fac

tor

Load

ings

0.00

0.25

0.50

0.75

Na Mg Al Si V Mn Fe Co0.00

0.25

0.50

0.75

Fig. 3. Loadings obtained from Varimax rotation of PM2.5 composit

o0.30) have not been depicted for the sake of clarity.

The second factor explaining 19% variationshowed highest loadings for Mg, Mn, Fe, and Al(0.84, 0.85, 0.78, and 0.58, respectively) andmoderate loadings for La, Sm, and Nd (0.36, 0.43and 0.41, respectively) consistent with resuspensionof crustal material. Note that higher loadings ofgeological marker elements (Mg, Mn, and Fe) in thesecond factor as compared to the first one, isconsistent with their higher concentrations in soilthan catalysts. Moreover, enrichment factors usingMn as the reference were close to unity for Al, Fe,and Mg. The mean Al/Na and Fe/Mg ratios were1.5 and 3.2, respectively (see Fig. 1), which weresimilar to the ratios measured in local soil (1.2 and2.6) and the Taylor abundance patterns (Taylor andMcLennan, 1985). Moderate loadings for La, Sm,and Nd are consistent with their presence in thelocal soil (see Fig. 2). Lower loading of Al in thesecond factor is the result of its reduced concentra-tion in soil (6.8%) compared to catalysts. Hence, thesecond factor predominantly accounts for thecontributions of crustal material and road dust toPM2.5 but other sources may also contribute to thisfactor. For example, high-temperature sources suchas coal combustion can also emit geological marker

Factor 2: Predominantly crustal (19%)

Factor 4: oil combustion (7%)

Ni As Ba Pb La Ce Sm Nd

Factor 5: sea spray (5%)

ion data for the five identified sources. Minor factors (loadings

ARTICLE IN PRESSP. Kulkarni et al. / Atmospheric Environment 40 (2006) 508–520 517

elements such as Mg, Al, Si, Mn, and Fe (Pacyna,1998; Dillner et al., 2005). Additionally, La, Sm,and Nd can also be emitted in automobile exhaust(see Table 3) and may be unresolved from crustalmaterial which vehicular traffic resuspends to theatmosphere.

The third factor explains almost the entirevariance of Pb and As in PM2.5. Because samplingwas performed in a highly industrialized area, theseelements were most likely emitted from localindustrial high-temperature combustion sourcessuch as incineration, oil combustion, coal combus-tion, and smelters/metal works (Dvonch et al., 1999;Morawska and Zhang, 2002; Dillner et al., 2005).Moderate loading for Si (0.30) was also obtained inthis factor, which could be explained by contribu-tion from coal-fired boilers (Morawska and Zhang,2002; Watson et al., 2002).

A fourth factor highly loaded with V, Ni, and Cowas also isolated. The most probable sources ofthese fine particles are oil combustion emissionsfrom local industries and ships traversing the Gulfof Mexico. (Note that over 6300 ships called atHouston, TX in the year 2001 alone (GreaterHouston Partnership, 2003)). REEs can also bereleased from oil-fired boilers due to contaminationof catalyst material in fuel petroleum products(Olmez and Gordon, 1985; Kitto et al., 1992).However, REEs were very strongly and positivelycorrelated with cat cracking operations (factor 1)and crustal material (factor 2) suppressing theirpresence in this oil combustion factor.

The fifth factor was assigned to sea spray becauseit had a very high loading for Na. Sea spray hasbeen previously identified as an important source offine particles in the Houston area (Buzcu et al.,2003) due to inland movement of air masses fromthe Gulf of Mexico.

Consistent with other studies (references withinWatson et al., 2002) PCFA apportioned 480% ofPM2.5 mass while yielding the above five sources. Inaddition to these five sources, recent studies (Buzcuet al., 2003; Dillner et al., 2005) reveal thatautomobiles and wood burning can also contributeto local PM2.5 concentrations in the Houstonarea. It is possible that the episodic nature ofwildfires affecting air quality in Southeast Texasprecluded samples analyzed in this study beingimpacted by this source. This is confirmed by thewood smoke marker, potassium, being below themethod detection limit (MDL ¼ 56 ngm�3) in mostsamples.

3.7. Estimation of source contributions

Absolute principal component scores (APCS)were calculated from rotated principal componentsbecause they are proportional to identified sources(Thurston and Spengler, 1985). Total PM2.5 mass(less elemental carbon, organic carbon, and sulfate)were regressed on APCS using multiple linearregression to extract source contributions. TotalPM2.5 mass less carbon and sulfate was chosen asthe basis for apportionment because more accuratemethods are available for apportionment of carbonmass and sulfate mass is dominated by secondaryformation rather than primary emissions. Forexample, carbon apportionment by means oforganic tracers has been widely used to apportionthose primary sources rich in organic particulatematter (Fraser et al., 2003). Also, there arenumerous carbon sources of similar composition,precluding their resolution by factor analysis.Organic carbon mass unscaled for other elementsassociated with organic species was subtracted fromgravimetrically determined PM2.5 mass becausethere is general ambiguity about the appropriatescaling factor for organic carbon mass to mass oforganic species. Nitrate concentration was not asubstantial fraction of the measured PM2.5 mass andtherefore not removed from model calculations. Tominimize the impact of particle bound water, filterswere desiccated in a low relative humidity (RH)environment prior to the mass determination as isstandard procedure in measuring PM2.5 massconcentrations as noted in (Buzcu et al., 2003 andreferences therein). Regression coefficients werethen used to convert APCS for each element intotheir quantitative contribution for 25 samples fromeach of five sources. The error bars in Fig. 4represent the standard deviations of 25 PM2.5 massestimates. Our source contribution calculationswere robust and reliable because (1) the vastmajority of APCS was positive, (2) even the veryfew negative APCS were close to 0, (3) regressioncoefficients were all positive, and (4) 82% of thePM2.5 mass (less C and sulfate) was apportioned.

Fig. 4 depicts the average source contributionsalong with the standard deviation of each of fivesources. The second factor, which is dominated bycrustal material (but may have had contributionsfrom high temperature sources and automobiles),was responsible for 62.5% of the apportioned PM2.5

mass. Two previous Houston area studies have alsorevealed that resuspension of crustal materials and

ARTICLE IN PRESS

0

1

2

3

4

5

Sea spray

Oilcombustion

High temp.operations

Predominantlycrustal

Sou

rce

cont

ribut

ion

(µg/

m3 )

Cat cracking

Fig. 4. Source apportionment (less C and sulfate) of PM2.5 samples collected at HRM3, Houston, TX using PCFA/APCA. Average

source contributions (mgm�3) from 5 sources (n ¼ 25) and their respective standard deviations are shown.

P. Kulkarni et al. / Atmospheric Environment 40 (2006) 508–520518

road dust is the highest contributor of PM2.5 massat this sampling site (Buzcu et al., 2003; Fraseret al., 2003). (Buzcu et al., 2003) found that �50%PM2.5 (less C and sulfate) mass was from crustalmaterial and road dust. Fraser et al. (2003)attributed �44% of apportioned PM2.5 to thissource. A possible explanation for this result is thepresence of a large area of unpaved surfaces in thevicinity of the Houston Ship Channel. The highercontribution in our study (62.5%) in comparisonwith previous results (50% and 44%) implies thatother sources (such as coal combustion, cat crackingoperations and automobile emissions) that emit thesame elements may be unresolved from the crustalfactor. Separately, the crustal contribution was alsocalculated by scaling the elemental concentrationsof the five common crustal elements, Si, Al, Mg, Ca,and Fe to their most common oxide form yielding acontribution of 2.971.5 mgm�3. (Since Ca was notquantified in this study, its concentrations wereobtained from Buzcu et al. (2003)). The meancrustal contribution values determined by thisapproach was statistically not different at 99%confidence with that obtained by factor analysis(3.671.0 mgm�3) indicating the validity of oursource apportionment calculations.

Emissions from high temperature sources, oilcombustion, and sea spray accounted for 13.5%,3.3%, and 18.7% of the apportioned PM2.5 mass,

respectively. Consistent with a previous study inPhiladelphia, PA (Dzubay et al., 1988), loss of FCCcatalyst from local petrochemical facilities contrib-uted very little to the apportioned mass (2.0% or0.12 mgm�3), but they are primarily responsible forREE enrichment in the atmosphere.

4. Conclusions

Because La and lanthanide composition patternsin both fresh and spent cat cracking catalysts arepreserved in the atmosphere, emissions from refin-ing and petrochemical sources can be uniquelyidentified by including ambient REE concentrationsdata (especially light lanthanides) in source appor-tionment calculations. However, because catalystsare poisoned by several transition metals, careshould be taken to only use the elemental composi-tion of spent catalysts as FCC source profiles (e.g. inchemical mass balance modeling). REE catalystsignature ratios obtained in this study were higherthan in Philadelphia, PA in the early 1990s (Kittoet al., 1992) and Osaka, Japan in the mid 1980s(Mizohata, 1986) demonstrating the need to mea-sure composition of catalysts employed locally andcurrently.

Factor analysis in combination with multiplelinear regression was able to separate and quantifyfive major sources of fine particles (cat cracking

ARTICLE IN PRESSP. Kulkarni et al. / Atmospheric Environment 40 (2006) 508–520 519

operations, a source predominantly comprisedof crustal material, industrial high-temperatureoperations, oil combustion, and sea spray) account-ing for 82% of PM2.5 (less C and sulfate) in thevicinity of Houston’s petrochemical industry. Re-suspension of soil and road dust from unpavedsurfaces was the chief contributor to PM2.5 mass(less carbon and sulfate), suggesting that moreattention should be given to reducing crustalemissions in order to improve Houston’s air quality.Eventhough catalytic cracking operations wereresponsible for only a small fraction of fine particlemass, REE enrichment in the Houston atmospherecould be solely attributed to cat-cracking opera-tions. Additionally, because cat-cracking can be asignificant source of atmospheric La, its enrichmentin PM2.5 should not be solely attributed to crustalmaterial or road dust especially in the vicinity ofpetrochemical industries.

Acknowledgments

This project has been funded entirely with fundsfrom the State of Texas as part of the program ofthe Texas Air Research Center. The contents do notnecessarily reflect the views and policies of thesponsor nor does the mention of trade names orcommercial products constitute endorsement orrecommendation for use. We also thank Karl Loosof Shell, and Tom Habib and Larry McDorman ofGrace Davison for providing catalyst samples aswell as two anonymous referees for their valuablecomments.

References

Buzcu, B., Fraser, M.P., Kulkarni, P., Chellam, S., 2003. Source

identification and apportionment of fine particulate matter in

Houston, TX using positive matrix factorization. Environ-

mental Engineering Science 20, 533–545.

Chellam, S., Kulkarni, P., Fraser, M.P., 2005. Emissions of

organic compounds and trace metals in fine particulate matter

from motor vehicles: a tunnel study in Houston, TX. Journal

of the Air and Waste Management Association 55, 60–72.

Choi, Y., Elliott, S., Simpson, I.J., Blake, D.R., Colman, J.J.,

Dubey, M.K., Meinardi, S., Rowland, F.S., Shirai, T., Smith,

F.A., 2003. Survey of whole air data from the second airborne

biomass burning and lightning experiment using principal

component analysis. Journal of Geophysical Research 108

(D5), ACH3-1–ACH3-10.

Cotton, F.A., Wilkinson, G., Murillo, C.A., Bochmann, M.,

1999. Advanced Inorganic Chemistry. Wiley, New York.

Dillner, A.M., Schauer, J.J., Christensen, W.F., Cass, G.R., 2005.

A quantitative method for clustering size distributions of

elements. Atmospheric Environment 39, 1525–1537.

Dvonch, J.T., Graney, J.R., Keeler, G.J., Stevens, R.K., 1999.

Use of elemental tracers to source apportion mercury in south

Florida precipitation. Environmental Science and Technology

33, 4522–4527.

Dzubay, T.G., Stevens, R.K., Gordon, G.E., Olmez, I., Sheffield,

A.E., Courtney, W.J., 1988. A composite receptor method

applied to Philadelphia aerosol. Environmental Science and

Technology 22, 46–52.

EPA, 1992. Preparation of Soil Sampling Protocols: Sampling

Techniques and Strategies. Office of Research and Develop-

ment, EPA/600/R-92/128, Las Vegas, NV.

Fraser, M.P., Yue, Z.W., Buzcu, B., 2003. Source apportion-

ment of fine particulate matter in Houston, TX, using

organic molecular markers. Atmospheric Environment 37,

2117–2123.

Greater, Houston Partnership, 2003. Houston facts. A publica-

tion of Greater Houston Partnership Research Department,

Houston, TX.

Hopke, P.K. (Ed.), 1999. An introduction to source receptor

modeling. In: Landsberger, S., Creatchman, M. (Eds.),

Elemental Analysis of Airborne particles. Gordan and Breach

Science, Amsterdam, The Netherlands, pp. 273–315.

Huang, X., Olmez, I., Aras, N.K., 1994. Emission of trace

elements from motor vehicles: potential marker elements and

source composition profile. Atmospheric Environment 28,

1385–1391.

Kitto, M.E., Anderson, D.L., Gordon, G.E., Olmez, I., 1992.

Rare earth distributions in catalysts and airborne particles.

Environmental Science and Technology 26, 1368–1375.

Kulkarni, P., Chellam, S., Ghurye, G., Fraser, M.P., 2003. In situ

generation of hydrofluoric acid during microwave digestion of

atmospheric particulate matter prior to trace element analysis

using inductively coupled plasma mass spectrometry. Envir-

onmental Engineering Science 20, 517–531.

Mizohata, A., 1986. Rare earth element on airborne particles in

motor-vehicle emissions. Journal of Aerosol Research 1,

274–283 (in Japanese).

Morawska, L., Zhang, J., 2002. Combustion sources of particles.

1. Health relevance and source signatures. Chemosphere 49,

1045–1058.

NIST. Certificate of Analysis (www.nist.gov/SRM).

Olmez, I., Gordon, G.E., 1985. Rare earths: atmospheric

signatures for oil-fired plants and refineries. Science 229,

966–968.

Olmez, I., Sheffield, A.E., Gordon, G.E., Houck, J.E., Pritchett,

L.C., Cooper, J.A., Dzubay, T.G., Bennett, R.L., 1988.

Compositions of particles from selected sources in Philadel-

phia for receptor modeling applications. Journal of Air

Pollution and Control Association 38, 1392–1401.

Pacyna, J.M., 1998. Source inventories for atmospheric trace

metals. In: Harrison, R.M., Van Grieken, R. (Eds.), Atmo-

spheric Particles. Wiley, Chichester, pp. 385–421.

Richardson, J.T., 1989. Principles of Catalyst Development.

Plenum Press, New York.

Russell, M., Allen, D.T., Collins, D.R., Fraser, M.P., 2004.

Daily, seasonal, and spatial trends in PM2.5 mass and

composition in Southeast Texas. Aerosol Science and

Technology 38, 14–26.

ARTICLE IN PRESSP. Kulkarni et al. / Atmospheric Environment 40 (2006) 508–520520

Senaratne, I., Shooter, D., 2004. Elemental composition in source

identification of brown haze in Auckland, New Zealand.

Atmospheric Environment 38, 3049–3059.

Silva, P.J., Prather, K.A., 1997. On-line characterization of

individual particles from automobile emissions. Environmen-

tal Science and Technology 31, 3074–3080.

Taylor, S.R., McLennan, S.M., 1985. The Continental Crust: Its

Composition and Evolution. An Examination of the Geo-

chemical Record Preserved in Sedimentary Rocks. Blackwell

Scientific, Oxford.

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

source contributions to inhalable particulate matter pollution in

metropolitan Boston. Atmospheric Environment 19, 9–25.

Watson, J.G., Zhu, T., Chow, J.C., Engelbrecht, J., Fujita, E.M.,

Wilson, W.E., 2002. Receptor modeling application frame-

work for particle source apportionment. Chemosphere 49,

1093–1136.

Wu, S., Zhao, Y.-H., Feng, X., Wittmeier, A., 1996. Application

of inductively coupled plasma mass spectrometry for total

metal determination in silicon-containing solid samples using

the microwave-assisted nitric acid–hydrofluoric acid–hydro-

gen peroxide–boric acid digestion system. Journal of Analy-

tical Atomic Spectrometry 11, 287–296.

Zhao, W., Hopke, P.K., Karl, T., 2004. Source identification of

volatile organic compounds in Houston, Texas. Environ-

mental Science and Technology 38, 1338–1347.