Sources and source variations for aerosol at Mace Head, Ireland

Atmospheric Environment 35 (2001) 1421}1437 Sources and source variations for aerosol at Mace Head, Ireland Suilou Huang!,*,1, Richard Arimoto", Kenneth A. Rahn! !Center for Atmospheric Chemistry Studies, Graduate School of Oceanography, University of Rhode Island, Narragansett, RI 02882, USA "Carlsbad Environmental Monitoring & Research Center, New Mexico State University, 1400 University Drive, Carlsbad, NM 88220, USA Received 15 November 1999; accepted 5 July 2000 Abstract The sources and source variations for aerosol at Mace Head, Ireland, were studied by applying positive matrix factorization (PMF), a variant of factor analysis, to a 5-yr data set for bulk aerosol. Signals for the following six sources were evident year round: (1) mineral dust, (2) sea salt, (3) general pollution, (4) a secondary SO2~}Se signal that is 4 composed of both natural (marine) and pollution (coal) components, (5) ferrous industries, (6) and a second marine (possibly biogenic) source. Analyses of seasonally strati"ed data suggested additional sources for iodine and oil emissions but these were present only in certain seasons, respectively. The marine signal is particularly strong in winter. The main pollution transport from Europe to Mace Head occurs in May, but the in#uence of continental European emissions is evident throughout the year. Mineral aerosol evidently follows a transport pathway similar to that of pollution aerosol, i.e., recirculation via the westerlies brings pollutants mixed with dust to the site from nearby land, i.e., Ireland, the United Kingdom, and the Belgium, Netherlands, and Luxemburg (Benelux) region, with some inputs from Scandinavia, Western Europe, Eastern Europe, and even the Mediterranean region. Compared with Bermuda, aerosol at Mace Head has stronger marine sources (especially marine-derived secondary SO2~ and Se) but weaker crustal and oil signals. Transport 4 across the North Atlantic, especially in winter, cannot be ruled out. ( 2001 Elsevier Science Ltd. All rights reserved. Keywords: Aerosol; Sources; Trace elements; Multivariate; Factor analysis 1. Introduction Aerosol at Mace Head (Ireland) can provide important information on marine sources, emissions from Europe, and perhaps long-range transport across the ocean, yet few studies have been conducted there to date. One prior study there showed that maritime air brought coarser particles to the Island than continental air from Europe does (Jennings et al., 1991), and in a second study, Franc7 ois et al. (1995) measured trace element concentrations in aerosol for the purpose of intercomparing di!erent sampling and analysis techniques. * Corresponding author. E-mail address: [email protected] (S. Huang). 1 Present address: Physics Department, New Mexico Institute of Mining and Technology, 801 Leroy Place, Socorro, NM 87801, USA. Studies at Mace Head also have been conducted as part of the Atmosphere/Ocean Chemistry Experiment (AEROCE). Daily aerosol samples were collected for that program from 1989 to 1994, and compared with three other AEROCE sites, Barbados, Bermuda, and Izan8 a, Mace Head has the lowest dust concentrations (with maxima an order of magnitude lower than at the other sites) but the highest sea-salt concentrations (Arimoto et al., 1995). Recently, Huang et al. (1997) derived the dry-deposition component of aerosol at Mace Head from "eld blanks. Here we focus on the sources of aerosol and their variations at Mace Head, using a variant of factor analysis, positive matrix factorization (PMF), for the analyses. Brie#y, as is the case for other factor-analysis techniques, PMF is based on the correlations among variables (elements in our case), which are assumed to be caused by certain underlying structures (sources). By solving the 1352-2310/01/$ - see front matter ( 2001 Elsevier Science Ltd. All rights reserved. PII: S 1 3 5 2 - 2 3 1 0 ( 0 0 ) 0 0 3 6 8 - X 1422 S. Huang et al. / Atmospheric Environment 35 (2001) 1421}1437 correlation matrix, PMF generates factor loadings (the arithmetic mean concentrations of the elements) and factor scores (the contributions of each factor to the sample) (Paatero and Tapper, 1993, 1994; Juntto and Paatero, 1994; Anttila et al., 1995; Paatero, 1996). PMF resolves sources for aerosol semi-quantitatively, and its resolving power for some elements was shown to be better than conventional factor analysis techniques (Huang et al., 1999a). The results from PMF are complemented by studies based on a new graphical analysis technique developed by Rahn (1999). Furthermore, meteorological maps and air-mass back-trajectories are used in our studies to identify the atmospheric transport pathways to Mace Head and to aid in interpreting temporal trends. The speci"c questions we address in this paper are (1) What are the main sources for aerosols at Mace Head? (2) What is the composition of aerosol associated with each source? (3) How do these sources vary with season? (4) How does the composition of aerosol at Mace Head compare with Bermuda, which receives pollution mostly from North America? 2. Materials and methods 2.1. Study site, sample collection and analysis Mace Head is located on the west coast of Ireland at 53.433N, 9.733W. The atmosphere there is not strongly polluted because the dominant wind direction is from the west, with a long fetch over the North Atlantic Ocean. While aerosol at Mace Head has been viewed as representing background aerosol, continental in#uences from Europe have been observed (Jennings et al., 1991; Merrill, 1994). A second sampling site in the North Atlantic, Bermuda, is at 32.243N, 64.873W, about 1000 km from the US east coast. In comparison to Mace Head, aerosol at Bermuda is well studied (e.g., Arimoto et al., 1992; Duce et al., 1976; Chen and Duce, 1983; Huang et al., 1996, 1999b). Brie#y, Bermuda's aerosol is strongly a!ected by large-scale changes in circulation that makes seasonality more dramatic than at Mace Head. In warm seasons, the Bermuda-Azores high-pressure system dominates conditions at Bermuda and causes substantial amounts of Saharan dust to be brought to the island. In cold seasons, air often #ows to Bermuda from North America, bringing materials, mostly pollutants, from North America to Bermuda. The atmospheres of Mace Head and Bermuda form an interesting contrast, because (1) both are in marine environments and have limited local pollution; (2) both are in#uenced by nearby continents, i.e., North America and Europe, respectively; and (3) both face the problem of pollution from distant continents being masked by nearer emissions. Therefore, by studying aerosol at Mace Head and Bermuda, we can obtain information on sources of aerosol over the two continents. While the atmospheres at Bermuda and Mace Head are similar in many aspects, some di!erences do exist. For example, compared to the intense transport of Saharan dust in warm seasons at Bermuda, signi"cant quantities of Saharan dust reach Mace Head only on rare occasions (Reville et al., 1990). On the other hand, largely because of Mace Head's higher latitude, conditions tend to be stormier than Bermuda, and consequently some marine in#uences are stronger. Another di!erence is that much of the air sampled at Mace Head come from the open ocean, with few pollution sources other than ship emissions, while upwind of Bermuda is North America, a strong source of pollution. Sample collection and analytical procedures used for the AEROCE studies have been described by Arimoto et al. (1995). Brie#y, daily high-volume aerosol samples were collected on Whatmant 41 "lters (Whatman Inc., Maidstone, UK) by using custom-made, computer-based sampling systems. Wind speed, wind direction, condensation nuclei (CN) counts, relative humidity, and the occurrence of rain were used to control the sampling and to minimize contamination from local sources. Due to frequent mid-latitude storms, some samples were collected over more than one day, and some samples were wet and thus discarded. Trace-element concentrations were determined by using instrumental neutron-activation analysis (INAA); total sulfate, NO~, NH~, and methane sulfonate 3 4 (MSA) for matching samples was measured at the University of Miami. 2.2. Data analysis To maximize the source-identi"cation power of factor analysis, we selected elements with strong source information and relatively few missing values (Huang et al., 1999a). In general, one should exclude those elements from the analyses with many values missing because of concentrations near detection limits, but this must be balanced by including as many elements as possible to increase the degrees of freedom for tuning the model. The fact that Mace Head has high concentrations of marine aerosol increases the di$culties of measuring elements from other sources. As a result, many data for these elements were not reported in the Mace Head data set due to detection-limit problems. The suite of trace elements we used in the PMF was: Na, Mg, Al, Cl, Ca, Sc, V, Cr, Fe, Co, Zn, Se, Br, Sb, I, Cs, Th, and sulfate (for simplicity, we refer to SO2~ as an 4 element). The percentage of below-detection-limit (BDL) values ranged from 10 to 66% (median 30%) for this set, and following the procedures of Huang et al. (1999a), mean concentrations of the elements were substituted for the missing values. Since few anomalous single-element S. Huang et al. / Atmospheric Environment 35 (2001) 1421}1437 factors (signs of including `weaka elements that degrade the PMF results) were generated, the factors extracted from this data set were considered reasonable. Because PMF, like other factor-analysis programs, is subjective to a degree, we evaluated the factors, or source signatures, by (1) extracting and evaluating 3- to 10-factor solutions in a stepwise manner, (2) examining the enrichments of the crustal and marine factors relative to average crustal material and bulk seawater, and (3) comparing the compositions of the factors extracted to known pollution sources. A total of 650 samples from Mace Head collected from 1989 to 1994 and 1168 Bermuda samples from 1988 to 1994 were used for the PMF studies and comparisons. The analytical uncertainties were used as weights for individual data points for PMF. The resulting rotational matrices are rather close to zero, suggesting factor rotation is not necessary. To study the variations of sources in di!erent seasons, the samples were further subdivided into four groups: March}May (spring), June}August (summer), September}November (fall), and December}February (winter). This set of seasons was used so that the results could be compared to those from Bermuda (Huang et al., 1999b). The full sample set and the seasonal subgroups were used for separating PMF analyses. 3. Results and discussion 3.1. Determining the number of underlying sources The number of underlying sources in the data sets was determined by examining the crustal and marine factors for the solutions with various numbers of factors extracted and comparing the composition of mineral aerosol and sea-salt factors against reference crustal material or bulk seawater (Huang et al., 1999a). Brie#y, we "rst calculated the crustal enrichment factors (EFs) for the crustal factors by dividing their X/Al ratios by the corresponding X/Al ratio in the crustal reference material, using the Mason (1992) compilation as a reference. Similarly, the marine EFs for the marine factors were calculated by dividing the X/Na ratio in the factors by the X/Na ratio in bulk seawater (Goldberg, 1992). The results were considered realistic if the EFs were close to unity, i.e., if the composition of the factors extracted resembled the reference material. We caution that certain trace elements may be naturally enriched in atmospheric sea salt by the bubble-bursting process (Weisel et al., 1984), and this could make the bulk seawater reference less than ideal for marine aerosol. The EFs for the crustal and marine factors for the 3- to 10-factor solutions generated by PMF and the enrichments calculated from the sample averages (Table 1) show that PMF is able to reconstruct the composition of 1423 the crustal and marine factors almost quantitatively. The substantial reduction of the crustal and marine enrichments from the sample data compared with the crustal and marine factors, sometimes reaching four-orders-ofmagnitude, demonstrates the power of PMF for extracting sources from aerosol data. In general, many elements have crustal and/or marine EFs close to unity, further suggesting that PMF reproduced the crustal and marine sources rather accurately. The discrepancies for some elements are discussed in the source composition section below. The optimal number of factors can be determined by comparing the elemental compositions of the crustal and marine factors for solutions with varying numbers of factors (Table 1). Lacking other criteria, we assume that the solution that produces crustal and marine EFs closest to unity is the best for the other sources as well, and that solution should be used as the basis for further interpretation. In addition, the uncertainties assigned to the elements in the crustal and marine factors need to be taken into account. In our case, when too few factors (e.g., 3 or 4) are extracted, the EFs of those minor elements (in terms of origins) are high, indicating that the crustal (and/or marine) factor is not `purea enough. When more than seven or eight factors are extracted, however, the crustal and marine compositions start to become less certain (e.g., V in the crustal factor and SO2~ 4 and Br in the marine factor). For the 6- to 8-factor solutions, the EFs of most elements are generally closer to unity relative to the other solutions without excluding many elements. In other words, the 6- to 8-factor solutions produce relatively clean crustal and marine factors whose compositions resemble their reference source materials. As a result, the discussion that follows is based on the 7-factor solution, which explains 89% of the variance in the data. 3.2. Source interpretation 3.2.1. General interpretation The seven factors are represented as variance explained by each factor, which is the square of the conventional factor loadings (Fig. 1), and they are discussed in order of appearance. The "rst factor is most loaded with typical crustal elements such as Al, Sc, Fe, and Th; it obviously represents the crustal source (with possible contributions from coal #yash). The crustal factor explains 17% of the variance in the data. The second factor is most loaded with marine elements such as Na, Mg, Cl, and Br, and clearly represents a marine source, i.e., atmospheric sea salt, which is expected because of the strong in#uence of the North Atlantic. This "rst marine, or sea salt, factor explains 16% of the annual variance of the system. The third factor is most loaded with V, Zn, Sb, and small amounts of Fe, Co, Se, and Cs. Noncrustal V in the 1424 S. Huang et al. / Atmospheric Environment 35 (2001) 1421}1437 Table 1 Enrichments of the crustal and marine factors to crustal reference material and bulk sea-water respectively Number of factors extracted Element 3 4 5 EF of sample average 6 7 8 )0.71 0.53 1 )42 )239 0.92 0.90 )0.065 )0.064 1.1 0.84 2.2 )51 )66 )1.5 )101 1.9 1.8 (a) Crustal factors Na 3.1 Mg 1.2 Al 1 SO2~ 120 4 Cl 1086 Ca 1.2 Sc 0.85 V 0.93 Cr 1.0 Fe 1.1 Co 0.89 Zn 1.6 Se 139 Br 70 Sb )1.4 I 541 Cs 2.1 Th 1.9 0.94 0.68 1 )26 295 1.1 0.83 0.91 0.8 1.1 0.86 1.8 37 )20 )2.4 471 2.1 1.9 )0.30 0.53 1 )29 313 0.85 0.84 0.88 )0.098 1.1 0.82 1.7 )40 )15 )1.9 )84 2.0 1.8 1.7 0.62 1 )29 540 1.1 0.84 0.92 )0.060 1.1 0.79 1.7 47 )20 )1.7 388 2.0 1.8 )0.28 0.53 1 115 )219 0.92 0.82 1.0 )0.060 1.0 0.78 1.5 74 )57 )1.6 )90 1.9 1.8 (b) Marine factors Na 1 Mg 0.81 Al 69 SO2~ 1.3 4 Cl 0.93 Ca 0.89 Sc 209 V 28 Cr 677 Fe 40 Co )0.43 Zn )1.1 Se 2570 Br 0.64 Sb )0.55 I 46 Cs 4.3 Th 725 1 0.77 148 0.57 0.93 0.77 139 )3.0 )35 )6.0 )0.21 4.4 192 0.66 1.8 41 1.7 98 1 0.76 133 0.55 1.0 0.69 90 )2.9 )29 )4.4 )0.49 2.9 )31 0.70 2.2 20 1.3 )99 1 0.74 164 0.53 0.93 0.60 62 )2.1 )35 )3.3 )0.24 )0.75 )14 0.59 1.1 )2.2 0.5 )68 1 0.76 231 0.42 1.0 0.59 75 )2.2 )38 )5.3 )0.59 )1.0 )15 0.89 )1.1 )1.9 )0.30 )88 atmosphere is most often associated with the combustion of heavy fuel oil (e.g., Rahn and Lowenthal, 1984). Zn comes from many pollution sources, including incinerators and smelters. Sb also mostly comes from a variety of industrial sources, but especially nonferrous smelters (Nriagu, 1979). Given this combination of elements from di!erent sources, this third factor apparently is a mixed pollution signal, most likely from Europe. This factor explains 14% of the variance. The fourth factor is loaded with SO2~ and Se, the 4 two sulfur-group elements that have strong natural sources (e.g., marine and volcanic eruption) and strong 9 1 0.77 )7.5 )0.0059 0.91 0.45 63 )2.0 )45 )5.2 )0.63 6.3 218 0.20 )0.66 )1.5 )1.0 )58 10 )0.30 )0.56 0.62 0.39 1 1 )37 206 344 )321 0.85 )0.055 0.88 0.89 )0.077 )0.053 )0.065 )0.069 1.1 1.1 0.82 0.81 2.1 2.0 122 )44 )46 80 )1.4 )1.2 )98 239 1.9 0.2 1.8 1.2 261 43 1 6134 97892 10 0.6 15 12 1.0 2.0 59 6550 14135 999 9692 6.1 2.4 1 1 0.80 0.79 )12 )1.4 )0.045 0.1 0.87 0.93 0.82 0.51 )84 )0.57 )9.0 )0.036 )309 )4.6 )13 )0.17 )0.21 )0.0048 )2.5 6.2 2781 )3.1 )0.0035 0.05 )5.4 )0.014 )5.9 )0.46 )0.41 0.024 446 )345 1 0.94 11554 2.6 1.0 1.3 18769 1394 34443 7392 176 587 10343 0.77 947 115 86 24728 anthropogenic sources (e.g., coal combustion). These substances are likely modi"ed after emission and are thus considered secondary aerosol. This factor can represent a coal source, which would be expected for aerosol at Mace Head because many countries in Europe, especially Eastern Europe, burn coal for industry, power, and heating. It could also represent marine background aerosol since both SO2~ and Se have marine sources (e.g., 4 Mosher and Duce, 1983; Charlson et al., 1987; Arimoto et al., 1992; Ellis et al., 1993). This factor, which we refer to as the sulfur-family or sulfur-group factor, explains 8% of the variance and is discussed in detail below. S. Huang et al. / Atmospheric Environment 35 (2001) 1421}1437 1425 Fig. 1. Sources for aerosol at Mace Head expressed as variances explained by the factors, 7-factor solution. The "fth factor is loaded with Cr, Co, and a small amount of Fe. There are two possible interpretations for this factor: (1) it is an artifact of impurities in Whatmant 41 "lters, or (2) it is a pollution signal. It is well known that the Cr blanks in Whatmant 41 "lters are high, and hence accurate determination of Cr concentrations is di$cult (Huang et al., 1997). Thus, a single-element factor (Cr) might have been generated as a result of the high and variable "lter blanks. This interpretation, however, does not explain the fact that Co and Fe, which are associated with ferrous-industries, covary with Cr. Therefore, an equally plausible interpretation is that this factor represents a pollution signal from ferrous or other metallic industries. We call this the `ferrous factora; it explains the same amount of the variance (8%) as does the sulfurfamily factor. The sixth factor is loaded with moderate amounts of primary marine elements (Na, Mg, Cl, and Ca) that are major components of sea spray, with elements that have been linked to marine biogenic emissions (Se and I), and with V, which is also important for marine biota (Weisel, 1981). Therefore, this factor may represent a 1426 S. Huang et al. / Atmospheric Environment 35 (2001) 1421}1437 second marine component, possibly mixed with continental pollution; it explains 11% of the variance. The seventh factor is loaded with iodine (I) alone. The sources for I in the atmosphere are diverse and not well quanti"ed (Cicerone, 1981). Likely sources include photochemical production from seawater, pyrogenic sources such as coal burning, and the dye industry. Exactly what source this factor represents is not clear, but its existence suggests that the variability of I is distinctively di!erent from the other elements. This factor explains 9% of the variance. 3.2.2. The relations among factors To study whether these factors are related to each other, we examined the correlations of the factor scores. No strong correlations were found: the largest R2 is 0.33 for the general pollution factor vs. sulfur-group factor, and the only statistically signi"cant correlations are the I factor vs. crustal, general pollution, and sulfur groups, and the sulfur-group factor vs. general pollution. Thus, the factors are essentially independent of each other. For the log-transformed factor scores discussed below, the largest R2 is 0.3 for the general pollution factor vs. sulfur-group factor, although more pairs are shown to have statistically signi"cant correlations than in the untransformed data. The relations among factor scores were further investigated by plotting them against each other on log scales (Fig. 2). The "rst marine factor (which presumably represents clean maritime air) has a weak negative correlation with the general pollution and ferrous factors. The second marine factor correlates only weakly with the I factor, con"rming that it represents a distinctive source signal. The I factor increases with the "rst marine factor, indicating marine in#uences of some sort. The crustal and ferrous sources increase with the general pollution signal, suggesting that all three are associated with continental emissions from Europe. The most interesting pattern is for the sulfur-group vs. the general pollution factor, which reveals two components of the sulfur-group factor (Fig. 2). The sulfur-group factor does not vary with the general pollution factor until a threshold (roughly the mean factor score) is reached, above which both factors increase together. Samples Fig. 2. Matrix plot of the factor scores on log scales. The weak correlations among the crustal, general pollution, and ferrous sources re#ect the e!ect of similar transport pathways of these factors. The `doublea correlation between the general pollution and sulfur-group factors is highlighted in a thick-bordered box. S. Huang et al. / Atmospheric Environment 35 (2001) 1421}1437 in the `pollutiona tail are probably derived from coal combustion. Meteorological maps and air-mass trajectories matched to the daily samples show that recirculating #ow due to the westerlies, passed over United Kingdom, Ireland and/or the Belgium, Netherlands, and Luxemburg (Benelux) region. Those samples toward the lowpollution end were transported from the open ocean, 1427 indicating that they had originated in clean marine air. About 25% of the samples have SO2~ and Se that are 4 clearly pollution derived, another 20% are associated with clean aged marine air, and the rest appear to be a mixture of marine and pollution sources. To investigate the sulfur-group factor further, we examined the sulfate plots in Fig. 3a by splitting the Fig. 3. Log}log scatter plot of X/Al vs. X/Sb (or X/Se) and X/Na, where X stands for any element: (a) Mg, SO2~, Cl, Ca, Br, Se, and I; (b) 4 Sc, V, Cr, Fe, Co, Zn, Cs, and Th. 1428 S. Huang et al. / Atmospheric Environment 35 (2001) 1421}1437 Fig. 3. (continued). samples into two groups, one with pollution factor scores '0.4 so that they were presumably pollution derived, and the other with scores )0.4 so that the samples were relatively cleaner. The value of 0.4 was estimated as the in#ection point in Fig. 2. For the Fig. 3-type of plot, we "rst assumed that Al, Na, and Sb represented crustal, marine, and pollution sources, respectively. Then we plotted X/Al vs. Sb/Al (or Na/Al) using log}log scales, where X is the element of interest. The crustal component of aerosol appears as a 03 zone in the X/Al vs. Sb/Al plot, the marine component appears as a 453 zone in the X/Al vs. Na/Al plot, and the pollution component appears as a 453 zone in the X/Al vs. Sb/Al plot. The crustal X/Al (or marine X/Na) ratios can be read from the crustal S. Huang et al. / Atmospheric Environment 35 (2001) 1421}1437 (or marine) tail from this type of plot (Rahn, 1999). Since the values of Sb/Al and Na/Al do not overlap, we combined both plots into a single graph (Fig. 3a and b). Two broad 453 zones, falling on top of one another, are observed in the SO2~/Al vs. Sb/Al part of SO2~ (a), Fig. 4 4 3a, indicating two sources. The samples that are associated with moderate to heavy pollution (pollution factor score '0.4) are in the lower half of the 453 zone (SO2~/Sb"10,000}80,000), and those that are asso4 ciated with less pollution (pollution factor score )0.4) are in the upper half (SO2~/Sb"80,000}700,000). 4 In the SO2~/Al vs. Na/Al part of SO2~ (a), Fig. 3a, 4 4 those samples associated with high pollution have high SO2~/Na ratios and di!er from those in the marine tail. 4 Those with low pollution factor scores are in the upperright end of the marine tail and approach a baseline, the seawater SO2~/Na ratio (0.25, Goldberg, 1992), indicat4 ing that they were from sea salt. The double-ratio plot of Se shows a pattern similar to that of SO2~, with the 4 Se/Sb ratios being 1}10 and 10}80 for more polluted and less polluted samples, respectively. The baseline Se/Na ratio is 3}4 orders of magnitude higher than the bulk seawater ratio, and probably represents a marine biogenic Se/Na ratio. To summarize, a substantial fraction of the sulfate is clearly pollution derived, but the background SO2~}Se is more than likely a mixture from the 4 clean marine and pollution sources. We also plotted SO2~/Al vs. Se/Al and Na/Al in SO2~ 4 4 (b), Fig. 3a. The two groups of samples overlap, further con"rming that the SO2~/Se ratio is the same for both 4 groups } this is why PMF produced only one factor for the two di!erent sources. The two-component nature of the sulfur-group factor exposes a weakness of factor analysis: it cannot distinguish sources with similar compositions. One way around this problem in the case of the sulfur-group factor is to examine its relationship with other factors or sources. Therefore, measurements of multiple elements or constituents from di!erent sources are essential for solving this problem. Another way to improve the PMF resolution to this sulfur-group factor is to include data for MSA, NO~, and NH` in PMF; the fraction that 3 4 correlates with MSA is considered marine derived while that correlated with NO~ and NH` is considered an3 4 thropogenic. While similar conclusions can be drawn about the pollution and marine contributions of SO2~ 4 and Se, we found that including MSA, NO~, and NH` 3 4 increased the residuals of most elements, especially SO2~, Cr and I. The higher residuals are probably due to 4 (1) the e!ects of di!erent sources and (2) the analysis of di!erent samples by di!erent analytical techniques. Examining the relationships among sources further demonstrates that PMF can separate signals in air masses coming from the same direction as long as the sources have distinct compositions (e.g., the crustal and general pollution signals). In other words, the correlations built 1429 into the system due to similar transport pathways do not a!ect the source-identi"cation power of PMF. 3.3. Graphical representation of major sources Table 1 shows that the compositions of the crustal and marine factors agree well with source references, especially for the major elements, but there are still some uncertainties with respect to the minor elements. To further check the accuracy of the crustal and marine compositions derived from PMF, we used `double ratioa plots (Fig. 3a and b). 3.3.1. Marine component Mg, SO2~, Cl, Ca, Br, Se, and I are clearly dominated 4 by a marine source because they all have well-de"ned marine zones (Fig. 3a, 453 zone on the right-hand side of the plots). The marine tails of Mg and Cl are especially narrow, indicating a pure marine source. Ca and Br also have narrow marine zones. The marine zones of SO2~, 4 Se, and I are somewhat noisy, indicating the existence of other sources and/or secondary processes such as photochemical conversion. For Sc, V, Fe, Co, Cs, and Th, the X/Al vs. Na/Al part of the plots is noisier than the X/Al vs. Sb/Al part, suggesting that marine sources contribute little to these elements. However, we do observe weak `marinea tails when the Na/Al ratios are high, i.e., strong marine and/or weak crustal sources (Fig. 3b). There are two possible and not mutually exclusive explanations for these `enrichmentsa: (1) They may represent analytical di$culties for Al and possibly X in the presence of massive loadings of marine aerosols because the marine end can have either high Na or low Al concentrations. (2) The `marinea tails are real, i.e., they represent the composition of giant, locally generated, marine aerosol particles. The X/Na ratios read from these tails (as well as the PMF results) are enriched by factors of 10}105, which are within the range of trace-element enrichments reported for atmospheric sea salt (Buat-MeH nard, 1983; Weisel et al., 1984). 3.3.2. Crustal component Among the elements shown in Fig. 3a, only Ca shows a crustal asymptote. Actually, two crustal tails can be seen: One is clearly established, with (Ca/Al) "2, and #3645 the other less so, with (Ca/Al) "0.2. These two crus#3645 tal ratios suggest that Ca has two crustal sources. It is likely that some Ca at Mace Head originated from Saharan dust, where Ca is depleted relative to most crusts (Arimoto et al., 1995), while the other fraction came from Ca-enriched local rock, soil or coal #y-ash. This element also has two crustal ratios in Bermuda aerosol (0.5}1 for Saharan dust, and 0.2}0.4 for North American sources) as it does in aerosol at Narragansett, Rhode Island (0.1}0.3 for local crustal aerosol and 1430 S. Huang et al. / Atmospheric Environment 35 (2001) 1421}1437 possibly some Saharan dust, and 0.4}0.8 for a crustal source from the midwestern U.S. (e.g., Huang et al., 1999a, b). The X/Al ratios of Sc, Fe, and Th are invariant with Sb/Al and Na/Al, and hence they come mainly from crustal or coal #y-ash sources (Fig. 3b). The crustal asymptotes of these elements are not as narrow as the marine tails for marine elements because (1) 453 tails are vertically compressed two-fold relative to horizontal tails (Rahn, 1999); (2) there are sometimes discrepancies between radionuclides that have short and long halflives with INAA, which are probably di$culties inherent to INAA; and (3) the measurements of low concentrations of Al are a!ected by high concentrations of sea salt. The crustal asymptotes of V, Cr, Co, Zn, and Cs contain many fewer data points, suggesting that the crustal sources for these elements are weak. 3.3.3. Pollution component In Fig. 3a, the X/Al vs. Sb/Al part (i.e., the left-hand side) of the plots shows that SO2~ and Se have signi"cant 4 pollution source(s), at least as indicated by Sb. Furthermore, SO2~ covaried more strongly with Se than with 4 Sb, suggesting that SO2~ and Se have similar source(s). 4 This similarity is expressed as the sulfur-group factor in PMF. In addition to SO2~ and Se, the elements V, Co, Zn, 4 and Cs show 453 zones in the left part of each plot, indicating strong pollution sources (Fig. 3b). Among them, Zn covaried tightly with Sb, suggesting similar pollution sources, probably from nonferrous industries. V, Co, and Cs covaried with Sb to lesser degrees, suggesting in#uences of other pollution sources. The independence of Cr relative to Sb and Na suggests a unique source for Cr, which is also consistent with the ferrous factor from PMF. In summary, at least "ve out of seven factors (i.e., the crustal, marine, general pollution, sulfurgroup, and ferrous factors) can be reproduced graphically, lending credibility to the PMF results. 3.4. Source composition Not only do the PMF and graphical approaches agree well with respect to sources, but the (X/Al) and #3645 (X/Na) ratios read from log}log scatter plots and .!3*/% generated from PMF also are in excellent agreement (Table 2). The di!erences between the compositions of the crustal and marine factors and their source references are worth examining. First, the ratio of Mg to Al derived from PMF is systematically lower than the average crustal ratio by a factor of two, which brings it close to the value in soil (Rahn, 1976 and associated references). Second, the light crustal element Sc is consistently slightly depleted (i.e., EFs(1) with respect to Mason's crustal reference (1992), while the `heavya crustal elements Cs and Th are slightly enriched, agreeing with previous studies on fractionation from soil (e.g., SchuK tz and Rahn, 1982). Third, the enrichments of V, Fe, and Co are within the range of crustal references, while that of Zn is slightly higher than unity (Rahn, 1976). Table 2 The crustal X/Al and marine X/Na ratios, derived from the graphics and PMF! (X/Al) Crust (X/Na) Marine Element Graphical PMF Crustal reference Graphical PMF Seawater reference Na Mg Al SO2~ 4 Cl Ca Sc V Cr Fe Co Zn Se Br Sb I Cs Th * * 1 * * 2,0.02 0.0002 0.002 0.001 0.80 0.00025 0.0015 * * * * 1.5E!04 1.6E!04 )0.098 0.13 1 0.98 )0.35 0.41 0.00022 0.0016 )7.5E!5 0.63 0.00024 0.0013 9.1E!05 )0.0017 )4.0E!6 )5.5E!4 7.1E!05 1.6E!04 0.35 0.26 1 0.0085 0.0016 0.45 0.00027 0.0017 0.0012 0.62 0.00031 0.00086 1.2E!06 3.1E!05 2.5E!06 6.2E!06 3.7E!05 8.9E!05 1 0.1 * 0.25 1.8 0.02}0.05 (6E!8 to 1E!6) (1E!5 to 1E!4) * (2E!4 to 5E!3) (5E!7 to 5E!6) * (2E!5 to 7E!5) 0.0055 * (1E!4 to 6E!4) (3E!7 to 1E!6) (IE!7 to 3E!7) 1 0.098 2.2E!04 0.094 1.74 0.022 6.9E!09 )4.1E!7 )2.5E!7 )5.0E!6 )6.9E!5 )9.2E!7 )1.9E!7 0.0055 )7.0E!8 )1.0E!5 )2.1E!8 )1.1E!8 1 0.13 9.5E!07 0.22 1.81 0.038 9.1E!11 1.9E!07 4.8E!09 9.5E!07 3.8E!08 9.5E!07 8.6E!09 0.0062 2.9E!08 5.7E!06 2.9E!08 9.5E!11 !The values in parenthesis were read from the `marinea tails. S. Huang et al. / Atmospheric Environment 35 (2001) 1421}1437 The enrichments of some marine and pollution elements in the crustal factor (e.g., SO2~, Cl, Se, and I) and 4 the crustal elements in the marine factor (e.g., Al and Sc) are caused by the incomplete separation of the sources. The depletion of some elements in the crustal and marine factors (e.g., Na in the crustal factor and Cr in the crustal and marine factors) is due to their strong loadings on other sources. Evidently, PMF sometimes extracts more than needed from weak sources to form a factor with a strong source. In other words, depletions in crustal or marine factors are more than likely due to improper tuning of PMF. Consequently, the values of these enriched or depleted elements need to be examined with great caution, and this is why the graphical technique is essential for con"rmation. Pollution elements are also depleted in the marine factor, probably for the same reason. Thus, PMF's resolution is limited, and the source compositions from PMF should be viewed as semiquantitative. So far, the composition of known sources (e.g., crustal aerosol and marine sea salt) cannot be preset in PMF (or other factor-analysis techniques) to give more accurate estimates of other sources of unknown composition. However, PMF does provide solutions to minimize the `impuritya of the factor, that is, the method produces only negligible amounts of `loadingsa (or mass) for the elements that do not belong. In addition, PMF assigns large uncertainties to these loadings. For example, even though the crustal EF of Se is 74 for the 7-factor solution, its mass in the crustal factor is only (1% of the total Se. As another example, Al in the marine factor accounts for only (2% of the total mass although it is enriched by '200 for the 7-factor solution. For Sb, which is mostly pollution derived, PMF assigned an uncertainty that is greater than 100% of its loading in the crustal factor and hence excluded it from the crustal factor. PMF and the graphical technique are complementary. For example, sometimes when crustal or marine ratios cannot be read with any certainty from the plots (e.g., Mg in the crustal factor), PMF will produce numbers for them. On the other hand, unlike the Ca plot (Fig. 3a) discussed above, PMF produces only one crustal Ca/Al ratio, which represents the mean of the two sources and is thus less informative than the plot. While marine X/Na ratios for Mg, Cl, and Br are accurately produced by both techniques, the correct marine SO2~/Na ratio can 4 be read only from the SO2~ plots in Fig. 3a, because the 4 marine SO2~/Na ratio derived from PMF is only half of 4 the seawater ratio. This discrepancy must have been caused by improperly tuning of PMF to form the sulfurgroup factor. Furthermore, in the sea-salt factor, iodine cannot be qunti"ed because PMF produced large uncertainties, while in the I plot (Fig. 3a), the I/Na ratio is about two orders of magnitude higher than the seawater ratio, probably due to enrichment during the formation of sea salt (Seto and Duce, 1981) or photochemical processes (Za"riou, 1974). 1431 The unexpected behavior of SO2~ and I in the sea-salt 4 factor can be explained as follows: (1) they are artifacts caused by the formation of other major factors as mentioned above; and (2) the graphs represent the sum of several sources of SO2~ and I, whereas PMF reveals 4 more-subtle characteristics. The fact that the sulfate plots clearly show the proper marine SO2~/Na ratio supports 4 the "rst explanation. The arithmetic mean concentrations of the elements in various factors produced from PMF are listed in Table 3. Arithmetic means rather than geometric means are used because they are more appropriate for our source-composition studies than geometric means (which are more appropriate for comparing populations). Furthermore, the arithmetic mean concentrations suit our studies of source strengths well because they are controlled more by high concentrations than low concentrations. The crustal source accounts for &90% of the concentrations of Al and Sc, 80% of Th, &70% of Fe, and &30% of Co (Table 3). The mixed pollution source accounts for &80%, 90% and 95% of V, Zn and Sb, respectively. The ferrous industry produced all of the Cr, 36% of the Co, and &10% of the Fe. (Note that some crustal-derived Cr may have been included in this factor.) Sea salt accounts for two-thirds of the Na, Br, Cl, Br, and one-third of the Ca and about one-third of the marine elements were grouped into the second marine factor (Table 3). In the second marine factor, the source of these elements is probably related to sea-salt production, but in this case also associated with secondary marine aerosols. Apportioning the marine and pollution contributions of SO2~ and Se is less straight-forward because (1) sev4 eral factors are involved ("rst and second marine, and the sulfur-group factors); (2) the sulfur-group factor has two components; and (3) the estimate of sea-salt SO2~ from 4 PMF is o!set by a factor of two. Still, a "rst-order apportionment can be achieved by considering all the factors of SO2~ or Se involved (Table 4). The results 4 show that at Mace Head, the natural (marine) sources of SO2~ and Se are stronger than their anthropogenic sour4 ces. (When MSA, NO~, and NH~ were included in the 3 4 PMF analysis, the marine and pollution contributions were 67% and 32% for SO2~, and 55% and 41% for Se, 4 respectively.) We caution that our estimates are limited because they are based on acceptable sampling conditions of only &40% (Huang et al., 1997). For these samples, which generally represent marine #ow, roughly one-third are polluted and the other two-thirds clean. For the unsampled (60%) air, roughly one-third was due to out-ofsector winds, with the rest due to other sampling criteria, e.g., no wind, high CN, or conditions too wet to sample. The sector control and CN counting excluded the direct or easterly transport of heavy pollution from the United Kingdom or Europe, but the control of humidity and 1432 Table 3 The predicted and observed arithmetic mean concentrations of the elements! Element Crustal Marine Na Mg Al SO2~ 4 Cl Ca Sc V Cr Fe Co Zn Se Br Sb I Cs Th 3.1 (1.0) )0.0030 0.30 (2.5) )0.0043 6.7E!4 (13) 0.0022 (9.6) 0.29 (4.4) 0.060 (25) 5.4 (1.1) )0.010 0.07 (5.3) 0.0043 (53) 2.4E!8 (34))1.8E!8 )1.4E!6 6.0E!4 (1.7) )6.9E!7 )3.6E!6 )1.8E!5 0.0035 (8.2) )7.6E!8 7.7E!6 (4.0) )4.1E!6 0.0017 (2.2) )3.7E!7 9.9E!5 (4.6) 0.017 (1.1) 2.3E!4 (2.9) )1.1E!7 8.0E!5 (1.8) )4.6E!5 1.7E!4 (12) )3.3E!8 3.1E!6 (4.9) )2.9E!8 2.5E!7 (49) )0.0040 0.0046 (58) 0.033 (1.4) 0.044 (18) )0.013 0.014 (16) 7.3E!6 (1.1) 5.3E!5 (8.4) )3.1E!6 0.021 (1.9) 7.8E!06 (3.2) 4.4E!05 (14) 3.8E!06 (36) )7.3E!5 )1.9E!7 )2.3E!5 2.3E!06 (4.8) 5.1E!06 (3.2) !Percentage error given in parenthesis. General pollution Sulfur group Ferrous Second marine I Factor )0.012 0.012 (44) )7.0E!5 1.9 (1.4) )0.0071 0.0052 (63) 1.1E!7 (19) )1.3E!6 )2.4E!6 2.4E!04 (39) )4.2E!8 1.6E!5 (52) 8.9E!5 (4.5) 0.0010 (9.2) 1.6E!6 (14) )1.6E!5 5.5E!7 (14) 1.5E!7 (63) )0.0036 0.0048 (77) 0.0012 (8.4) 0.0098 (16) 0.0061 (66) 0.0042 (27) )6.3E!9 )6.4E!7 6.9E!04 (2.0) 0.0025 (7.3) 8.9E!6 (3.4) 3.7E!5 (28) 7.1E!7 (92) )2.6E!5 2.5E!6 (19) 2.1E!5 (42) 4.1E!7 (26) 2.5E!7 (46) 1.3 (1.7) 0.13 (4.5) )4.4E!4 0.084 (15) 2.0 (1.9) 0.046 (7.8) )1.2E!8 9.2E!5 (3.8) )3.0E!6 )3.9E!5 )3.6E!8 )4.0E!6 1.0E!4 (3.4) )4.8E!5 )1.8E!7 3.8E!4 (7.6) 1.3E!7 (50) 2.6E!7 (30) )0.0054 0.015 (44) )4.1E!5 )0.017 )0.027 0.045 (9.0) 3.9E!7 (5.2) )1.2E!6 )2.0E!6 0.0014 (7.5) 3.2E!7 (30) 1.1E!4 (8.5) 3.2E!5 (8.6) 0.0027 (4.5) )3.1E!7 0.0020 (2.4) 9.8E!7 (8.6) 4.1E!7 (22) Total predicted Total observed 4.4 (0.9) 0.47 (2.9) 0.037 (1.4) 2.4 (1.5) 7.4 (1.0) 0.19 (4.3) 7.8E!6 (1.1) 7.5E!4 (1.6) 6.9E!4 (2.0) 2.8E!2 (1.9) 2.5E!5 (2.1) 1.9E!3 (2.2) 3.3E!4 (2.3) 2.1E!2 (1.2) 8.4E!5 (1.8) 0.0025 (2.4) 7.5E!6 (3.4) 6.4E!6 (4.3) 4.2 (3.6) 0.51 (4.6) 0.047 (8.7) 2.4 (3.4) 7.3 (3.9) 0.21 (2.7) 7.3E!6 (8.3) 1.1E!3 (9.0) 6.9E!4 (7.5) 3.0E!2 (7.4) 2.8E!5 (7.2) 2.4E!3 (8.3) 3.8E!4 (3.4) 2.0E!2 (4.6) 1.1E!4 (8.2) 2.8E!03 (5.4) 1.0E!5 (5.3) 1.0E!5 (5.5) S. Huang et al. / Atmospheric Environment 35 (2001) 1421}1437 Concentration in factors, lg m~3 S. Huang et al. / Atmospheric Environment 35 (2001) 1421}1437 Table 4 The marine and pollution contributions of SO2~ and Se to the 4 mean concentrations Source Contribution Marine SO2~ 4 First marine factor Second marine factor Sulfur-group factor (marine fraction) Total 12%! 4%! 0.54"]79%! 59% Pollution SO2~ 4 General pollution factor Sulfur-group factor (pollution fraction) Total 3%! 0.46"]79%! 39% Marine Se Second marine factor Sulfur-group factor (marine fraction) I factor Total 32%! 0.54"]27%! 0.5#]10%! 52% Pollution Se General pollution factor Sulfur-group factor (pollution fraction) I factor Total 30%! 0.46"]27%! 0.5#]10%! 47% !From Table 3. "Fraction of the scores of the sulfur-group factor with the general-pollution factor score )0.4. #Assumed fraction. rain also eliminated some samples from the marine sector (Merrill, 1994). This `double exclusiona is likely to cancel out some of the preferential sampling e!ect in our data set, and therefore, our "gures for the pollution/marine contributions of SO2~ and Se should be generally valid. 4 Although the sources and their compositions remain the 1433 same regardless of wind sector, the crustal and pollution fractions of the aerosols are likely to be underrepresented in our data set. To obtain information on how well each element was predicted by PMF, we compared the total predicted concentrations with the observed concentrations for various elements (Table 3). Most elements and SO2~ are 4 predicted within 10% of the observed values; Co, Zn, and Se within 10}20%, and Al, Sb, and Cs within 20}30%. V and Th are underpredicted by 33 and 35%, respectively, suggesting that additional factors may be needed to explain V and Th. 3.5. Transport pattern and source variation 3.5.1. General transport pattern As stated previously, air#ow to Mace Head can generally be classi"ed as clean marine or polluted continental. Air-mass trajectories, calculated by R. Platner and J. Merrill based on the method described by Merrill (1989), show that during 20% of a typical year, the air reaching Mace Head has passed over nearby lands, including Ireland, the United Kingdom, and the Benelux region (and their pollution sources), while another 20% of the air#ow passed over more-distant parts of Europe. To study the general #ow patterns chemically, we examined the variations of the signals of major sources at Mace Head over time (Fig. 4a and b). For this analysis, we "rst counted the total number of samples in each month and then the number of samples whose scores were in the upper 25-percentile range for each factor in the month. Dividing the latter by the former, we obtained the frequency of the presence of the upper 25-percentile scores. The crustal, general pollution, and ferrous factors have the highest variability among factors, generally showing winter minima and late spring}early summer maxima. High concentrations of the crustal and pollution Fig. 4. Frequencies of the upper 25-percentile factor scores vs. time for major sources at Mace Head: (a) The crustal, general pollution, and ferrous factors; (b) Sea salt and sulfur factors. 1434 S. Huang et al. / Atmospheric Environment 35 (2001) 1421}1437 elements were observed most frequently in May; smaller peaks appeared in July and September (Fig. 4a). The day-to-day meteorological maps published by the German Weather Services show that the #ow patterns for the days with high crustal and pollution components frequently were from the west, with recirculating #ow passing over nearby land, and occasionally over Scandinavia. The #ow patterns associated with high crustal aerosol and pollution in July were similar to those in May but subject to more local in#uences. In September, the #ow associated with strong pollution transport was at times westerly, passing over nearby land, and at other times easterly, coming from the continent. Outbreaks of pollution transport were observed in all seasons, even in the winter, when the westerlies were the strongest. In agreement with data presented in Arimoto et al. (1995), the sea-salt factor shows winter}spring maxima and summer minima, which is typical of marine environments (Fig. 4b). The second marine factor varies little over time, but the upper-25-percentile concentrations of Se occur in April, May (peak), and June (not shown), and therefore Se's seasonal cycles could be explained by a combination of the spring bloom of marine phytoplankton, periods of intense pollution transport, superimposed on the winter}spring maximum of sea-salt production. The sulfur-group factor does not show a strong seasonal trend, probably because of the combined e!ect of its two components (Fig. 4b). While the marine component of the factor weakly tracks the primary marine source, the high values (i.e., the coal component) follow the general mixed pollution signal. The I factor has fewer values in winter than in other seasons, possibly due to slower photochemical conversion at that time of year. 3.5.2. Source variation in diwerent seasons To study how sources of aerosol at Mace Head vary over the course of a year, we grouped the data by season. In most seasons, aerosols at Mace Head show great similarities, but those in fall are very di!erent: the crustal and pollution sources are much less distinguishable from each other than in other seasons. When extracting more than four factors, the crustal elements with short-lived radionuclides (e.g., Al, V) began to separate from the other crustal elements determined by the long-irradiation procedure (e.g., Sc, Fe, Th), so that two `crustala factors are formed, representing no source information but rather suggesting systematic analytical di$culties. Therefore, only four factors (crustal plus pollution, marine, ferrous, and the sulfur-group) were extracted for the fall season, with 75% variance explained. Meteorological maps show a mixture of #ow patterns in fall, including less direct and more variable #ow from Europe and sometimes the Mediterranean region and Africa in association with a nearby high-pressure system. Because of the slow movement of air in the high-pressure system, aerosols from di!erent sources become homogenized, and hence fewer sources can be identi"ed. One di!erence between the annual and the seasonally grouped data is that a residual oil signal appeared in summer and winter. In winter, the samples with air#ow from the clean oceanic direction tend to have higher and more variable amounts of oil signal than general pollution signal. On the other hand, most of the samples have lower but quasi-constant oil relative to general pollution signal, whose origins seemed to be from nearby lands. High ratios of oil vs. general pollution also appeared when the aerosol was transported from Scandinavia. These observations suggest that maritime air can bring aerosols with strong residual oil signals. Furthermore, winter is the season with intense westerly transport. Since the Northeastern US burns much residual oil in winter, some of the pollution aerosol generated may be transported clear across the North Atlantic Ocean. In summer, the oil signal generally increases with the general pollution signal, suggesting sources from nearby land. Another possible source for the oil factor is emissions from ships in the North Atlantic. The disappearance of the oil factor from the all-season data set further suggests that this oil signal is neither strong nor well de"ned. Another weak and variable source is the I source, which appeared only in spring, casting some doubt on its validity. The concentrations of most elements in most factors are rather consistent over the seasons, suggesting that the results from PMF are stable unless there are systematic errors in the technique (e.g., low SO2~ in the "rst marine 4 factor). The contribution of SO2~ from the sulfur-group 4 factor was rather consistent in summer, fall, and winter, and lower in spring. The second marine source of Se was especially important in spring and winter, probably because of cycles in biological productivity and intense air}sea exchange by breaking waves. 3.6. Comparison of aerosols at Mace Head and Bermuda While analyzed independently, similar factors were found for the two sites (crustal, marine, general pollution, ferrous, sulfur-group, oil, second marine, and I). It reveals that even di!erent continents can give rise to remarkably similar natural and anthropogenic sources (Fig. 5). For the two major natural sources, the crustal source is weaker and varies less at Mace Head than at Bermuda in summer and fall, while both the sea-salt and the second marine sources were stronger at Mace Head. Transport of pollution is weakest in winter at Mace Head because of strong westerly #ow during that season while at Bermuda semiannual cycles in pollution transport are evident with strong spring and weaker fall maxima (Huang et al., 1999b). The sulfur-group factor at Mace Head clearly represents a secondary aerosol signal with relatively strong marine and weak pollution components, whereas at S. Huang et al. / Atmospheric Environment 35 (2001) 1421}1437 1435 Fig. 5. Sources for aerosol at Bermuda expressed as variances explained by the factors, 8-factor solution. Bermuda it was not well formed in two ways. First, it lacked the long marine tail in the factor-score plot vs. the general pollution factor: With less clean oceanic air at Bermuda, the marine end of the sulfur-group factor was truncated by almost one order of magnitude. Second, SO2~ and Se tend to join the other pollution elements at 4 Bermuda and form a mixed pollution signal, especially in winter. Even when they split from the general pollution factor, there were always some other pollution elements accompanying them (e.g., Sb in spring, Zn and Sb in summer, and V in fall). The pollution component of the sulfur-group factor for Bermuda is further revealed as it covaried positively with the general pollution factor: hence, with weaker marine biogenic sources upwind, this secondary aerosol signal was rather weak and more associated with pollution. The oil source is comparatively weak at Mace Head compared with Bermuda, occurring in summer and winter at Mace Head but present all year round at Bermuda. The strong oil signal at Bermuda has been attributed to transport from the Northeastern U.S. (Chen and Duce, 1983). 1436 S. Huang et al. / Atmospheric Environment 35 (2001) 1421}1437 In summary, for the most part, aerosols at Mace Head and Bermuda have similar sources. The greatest di!erence is that compared with Bermuda, mineral aerosol concentrations at Mace Head are much lower and vary much less with season. The secondary SO2~}Se source 4 and marine biogenic source of Se are stronger; and the oil source is weaker. Finally, compared with Bermuda the aerosol sources a!ecting Mace Head vary much less with season. 4. Conclusions By using a new factor analysis technique, PMF, we were able to study the sources and variations for aerosol at Mace Head and compare them nearly quantitatively with aerosol at Bermuda. This study shows that although Mace Head and Bermuda are considered remote oceanic sites, they are not isolated from pollution transported long distances, possibly even across the ocean. While the sources for aerosol at both sites are similar, Mace Head has weaker crustal and oil sources, and stronger marine sources, including secondary SO2~ and Se. Natural aero4 sols are most abundant in winter at Mace Head and in summer and winter at Bermuda. With data on multiple elements or constituents from di!erent sources, the twocomponent nature of the secondary SO2~}Se aerosol 4 was revealed. Futhermore, source variations of aerosol at both sites captured the change in global circulation, i.e., eastward transport is strong in cold season and weak in warm season. Acknowledgements We would like to thank D.L. Savoie and J.M. Prospero for the SO2~, NO~, NH~, and MSA data, J.T. Merrill 4 3 4 and R. Platner for the air-mass trajectories. This study is supported by NSF ATM ATM 94-14262 and ATM 9728983. References Anttila, P., Paatero, P., Tapper, U., Jarvinen, O., 1995. Source identi"cation of bulk wet deposition in Finland by positive matrix factorization. Atmospheric Environment 29, 1705}1718. Arimoto, R., Duce, R.A., Ray, B.J., Ellis, W.G., Cullen, J.D., Merrill, J.T., 1995. Trace elements in the atmosphere over the North Atlantic. Journal of Geophysical Research 100, 1199}1213. Arimoto, R., Duce, R.A., Savoie, D.L., Prospero, J.M., 1992. Trace elements in aerosol particles from Bermuda and Barbados: concentrations, sources and relationships to aerosol sulfate. Journal of Atmospheric Chemistry 14, 439}457. Buat-MeH nard, P., 1983. Particle geochemistry in the atmosphere and oceans. In: Liss, P.S., Slinn, W.G.N. (Eds.), Air}Sea Exchange of Gases and Particles. Reidel, Dordrecht, pp. 455}532. Charlson, R.J., Lovelock, J.E., Andreae, M.O., Warren, S.G., 1987. Oceanic phytoplankton, atmospheric sulphur, cloud albedo and climate. Nature 326, 655}661. Chen, L., Duce, R.A., 1983. The sources of sulfate, vanadium and mineral matter in aerosol particles over Bermuda. Atmospheric Environment 17, 2055}2064. Cicerone, R.J., 1981. Halogens in the atmosphere. Reviews of Geophysics and Space Physics 19, 123}139. Duce, R.A. et al, 1976. Trace metals in the marine atmosphere: sources and #uxes. In: Windom, H.L., Duce, R.A. (Eds.), Marine Pollutant Transfer. Lexington Books, Lexington, MA, pp. 77}119. Ellis, W.G., Arimoto, R., Savoie, D.L., Merrill, J.T., Duce, R.A., Prospero, J.M., 1993. Aerosol selenium at Bermuda and Barbados. Journal of Geophysical Research 98, 12673}12685. Franc7 ois, F., Maenhaut, W., Colin, J.-L., Losno, R., Schulz, M., Stahlschmidt, T., Spokes, L., Jickells, T., 1995. Intercomparison of elemental concentrations in total and size-fractionated aerosol samples collected during the Mace Head experiment, April 1991. Atmospheric Environment 29, 837}849. Goldberg, E.D., 1992. Seawater. In: Parker, S.P. et al. (Eds.), McGraw-Hill Encyclopedia of Science and Technology, Vol. 16, 7th edn. McGraw-Hill, New York, pp. 150}181. Huang, S., Arimoto, R., Rahn, K.A., 1996. Changes in atmospheric lead and other pollution elements at Bermuda. Journal of Geophysical Research 101, 21033}21040. Huang, S., Rahn, K., Arimoto, R., 1997. A graphical method for determining the dry-depositional component of aerosol samples and their "eld blanks. Atmospheric Environment 31, 3383}3394. Huang, S., Rahn, K.A., Arimoto, R., 1999a. Testing and optimizing two factor-analysis techniques on aerosol at Narragansett, Rhode Island. Atmospheric Environment 33, 2169}2185. Huang, S., Rahn, K.A., Arimoto, R., 1999b. Semiannual cycles of pollution at Bermuda. Journal of Geophysical Research 104, 30309}30318. Jennings, S.G., O'Dowd, C.D., O'Connor, T.C., McGovern, F.M., 1991. Physical characteristics of the ambient aerosol at Mace Head. Atmospheric Environment 25A, 557}562. Juntto, S., Paatero, P., 1994. Analysis of daily precipitation data by positive matrix factorization. Environmetrics 5, 127}144. Mason, B., 1992. Elements, geochemical distribution of. In: McGraw-Hill Encyclopedia of Science and Technology, Vol. 6. McGraw-Hill, New York, pp. 307}310. Merrill, J.T., 1989. Modeling long-range transport using trajectory techniques. In: Leinen, M., Sarthein, M. (Eds.), Paleoclimatology and Paleometeorology: Modern and Past Patterns of Global Atmospheric Transport. Kluwer Academic, Norwell, MA, pp. 207}225. Merrill, J.T., 1994. Isentropic air#ow probability analysis. Journal of Geophysical Research 99, 25881}25889. Mosher, B.W., Duce, R.A., 1983. Vapor phase and particulate selenium in the marine atmosphere. Journal of Geophysical Research 88, 6761}6768. S. Huang et al. / Atmospheric Environment 35 (2001) 1421}1437 Nriagu, J.O., 1979. Global inventory of natural and anthropogenic emissions of trace metals to the atmosphere. Nature 279, 409. Paatero, P., Tapper, U., 1993. Analysis of di!erent modes of factor analysis as least squares "t problems. Chemotherapy Intelligence Laboratory System 18, 183}194. Paatero, P., Tapper, U., 1994. Positive matrix factorization: a non-negative factor model with optimal utilization of error estimates of data values. Environmetrics 5, 111}126. Paatero, P., 1996. Least squares formulation of robust nonnegative factor analysis. Chemotherapy Intelligence Laboratory Systems 21, 1}13. Rahn, K.A., 1976. The chemical composition of the atmospheric aerosol. Technical Report, Graduate School of Oceanography, University of Rhode Island. Rahn, K.A., 1999. A graphical technique for determining major components in a mixed aerosol. I. Descriptive aspects. Atmospheric Environment 33, 1441}1455. 1437 Rahn, K.A., Lowenthal, D.H., 1984. Elemental tracers of distant regional pollution aerosols. Science 223, 132}139. Reville, W.J., Vernon, P.D., Cotter, M.P., 1990. Scanning electron microscopy of red dust falls in Ireland since November 1979. Ir. Nat. J. 23, 194}199. SchuK tz, L., Rahn, K., 1982. Trace-element concentrations in erodible soils. Atmospheric Environment 16, 171}176. Seto, F.Y.B., Duce, R.A., 1981. A laboratory study of iodine enrichment on atmospheric sea-salt particles produced by bubbles. Journal of Geophysical Research 77, 5339}5349. Weisel, C.P., 1981. The atmospheric #ux of elements from the ocean. Ph.D. Dissertation, University of Rhode Island. Weisel, C.P., Duce, R.A., Fasching, J.L., Heaton, R.W., 1984. Estimates of the transport of trace metals from the ocean to the atmosphere. Journal of Geophysical Research 89, 11607}11618. Za"riou, O.C., 1974. Photochemistry of halogens in the marine atmosphere. Journal of Geophysical Research 79, 2730}2732.