In the case of acetic acid collected on Tenax traps, our calibration data with added amounts of acetic acid from a diffusion tube vapor stream showed nonlinearity of peak response vs. amount of acetic acid a t low microgram levels, suggesting adsorption. Hence, it is recommended that calibration data be obtained on the same day that samples are run. Acknowledgments The authors wish to thank W. B. Prescott and F. N. Santacana for their helpful suggestions. Literature Cited W., Chang, R. C., Zlatkis, A., J. Chromatogr. Sei., 12, 175 (1974). (2) Mieure, J. P., Dietrich, M. W., ibid., 11,559 (1973). (3) Williams, F. W., Umstead, M. E., Anal. Chem., 40, 2232 (1968).
( 1 ) Bertsh,
(4) Dravnieks, A., Krotoszynski, B. K., Whitfield, J. O’Donnell, A., Burgwald, T., Enuiron. Sci. Technol., 5,1220 (1971). (5) Environmental Protection Agency, Standards of Performance for Stationary Sources, Fed. Regist., 36, No. 247, Thursday, December 23, 1971. (6) Craven, D. A., Anal. Chem.. 42.1679 (1970). (7) Van Wijk, R., “Advances in Chromatography,” A. Zlatkis, Ed., p 122, U. Houston, 1970. (8) Zlatkis, A., Lichtenstein, H. A., Tishbee, A., Chromatographia, 6.67 . 11973). ~~(9) Pellizzari, E. D., Bunch, J. E., Carpenter, B. H., Enuiron. Sci. Technol., 9,552 (1975). (10) Pellizzari, E. D., Carpenter, B. H., Bunch, J. E., ibid., 9, 556 (1975). (11) O’Keeffe, A. E., Ortman, G. C., Anal. Chem., 38,760 (1966). (12) Altshuller, A. P., Cohen, I. R., ibid., 32,802 (1960). (13) Fuller, E. N., Schettler, P. D., Giddings, J. C., Ind. Eng. Chem., 58, (5),19 (1966). ~
- I
Received f o r review February 14, 1975. Accepted July 17,1975.
Sources and Elemental Composition of Aerosol in Pasadena, Calif., by Energy-Dispersive X-ray Fluorescence Robert H. Hammerle” and William R. Piersbn Ford Motor Co., Research Staff, P.O. Box 2053, Dearborn, Mich. 48121
In the fall of 1972 we performed a 29-day study to identify sources of atmospheric particulate matter and assess their contributions to the aerosol in Pasadena, Calif. Sizefractionated and unfractionated filter samples were collected automatically and were analyzed by an automatic X-ray fluorescence spectrometer. Nine elements proved consistently measurable with reasonable accuracy: Ca, Ti, V, Mn, Fe, Ni, Zn, Br, and Pb. All show orders-of-magnitude fluctuations, with a tendency to follow diurnal patterns related to meteorological factors. Most of the elements can be classified as small-particle elements (Ni, Zn, Br, Pb) or largeparticle elements (Ca, Ti, Mn, Fe), with V falling in an intermediate size range. Statistical analysis of interelement correlations and size distributions indicates that gasoline engine exhaust is the source of the Br and Pb; soil from the basin is the main source of the Ti, Mn, and Fe; much of the Ca probably is from cement dust contamination of the soil, and the rest of the Ca is indigenous to the soil. Proportionality and high correlation coefficients (No. 8) characterize elements from the same sources and correlation coefficients No. 4 characterize unrelated elements. Using appropriate gravimetric factors, we estimate that the soil contributed 8 pg/m3 and the primary exhaust particulate from gasoline engines contributed 5 pg/m3, on the average, to the aerosol mass at the Pasadena site during the 29-day period.
There is considerable interest in identifying the origins of the atmospheric aerosol. With this as a major objective, a coordinated effort by many investigators was initiated in 1969 in the Los Angeles basin. The results of that study [reported in a series of papers in J . Colloid Interface Sei., 39, 136-304 (1972)] stimulated planning in 1970 for the large-scale California Aerosol Characterization Experiment (ACHEX) ( I , 2 ) , the observational phase of which was conducted in Pasadena and other sites in August through late October 1972. 1058
Environmental Science & Technology
During planning of ACHEX, a major difficulty was recognized to be the lack of a means of obtaining chemical information on the aerosol with good time resolution. We believed that X-ray fluorescence on-line with a Si(Li) semiconductor detector could yield elemental analyses for a number of elements concurrently in the aerosol on a reasonably fast time scale. An apparatus embodying this idea was designed and built at the University of California Lawrence Berkeley Laboratory by Jaklevic et al. (3, 4 ) under the auspicies of the U.S. Environmental Protection Agency, with the intent that its first utilization would be in the Pasadena ACHEX, prior to final modifications by LBL and subsequent transfer to EPA. The present paper reports the results obtained in Pasadena with this system. It represents the first extensive series of measurements, with prompt analysis and readout in the field, of the elemental composition of particulate matter with an automatic X-ray fluorescence spectrometer. It demonstrates not only the value of the superior time resolution but also the advantage offered by the rapid response in permitting decisions to be made during the course of the experiment. Technical difficulties delayed commencement of measurements until the end of October 1972. By this time, the main part of the program involving most of the other ACHEX participants had been truncated owing to unseasonably smog-free weather, and therefore we are able to relate our results to other properties of the basin atmosphere to a less extent than had been anticipated. Whitby et al. ( 5 ) report that the size distribution of aerosols often has two modes separated by a particle-deficient region at approximately 1-2 pm. This bimodality may be the result of different formation processes for the two size modes. Any attempt to identify aerosol sources by means of elemental composition should therefore include measurements of composition as a function of particle size. This was done in the present study by measuring the composition of the undifferentiated aerosol and the portion below 1.5 pm concurrently.
Friedlander (6) gives a good description of the basic theory for the use of elemental composition to identify sources. The approach is to solve the material-balance equations for each element measured. For example, the Ca in the aerosol can be written as Ca(wg/m3) = CcviCi i
(1)
where Ca content (fraction, by weight) in the particulate matter originating from the ith source c.I = - concentration (pg/m3) of particulate matter sampled that can be attributed to the ith source subject to the constraint Total (wg/m3) = ZCi i
(2)
The summations over i are meant to represent summations over all sources, such as soil, sea salt, gasoline engines, oil refining operations, and so forth. The assumption is made that for certain trace species the concentration in the aerosol emitted from a given source is constant in time, both a t the source and during transport through the atmosphere. This assumption can be tested from the elemental composition of the many aerosol samples collected sequentially a t one location. In the present study, elemental composition of size-fractionated and unfractionated aerosol was measured a t one Pasadena site for a series of 2-hr periods spanning 29 days. Sources are identified and variations in the elemental composition and size distribution of the particulate matter coming from the sources are described statistically. Contributions of the identified sources to the atmospheric particulate loading are deduced. A preliminary report on the present study has been given elsewhere (7). Model Our results were analyzed in terms of a “single-source model.” Let us imagine that, of the many sources of particulate matter, there is only one source which emits elements X and Y . Let us imagine further that the elemental composition of the particulate matter from this latter source remains constant in time, both as it is emitted and as it moves through the atmosphere. Then it follows that the concentrations, X and Y , in the gross particulate matter downwind of the source are always proportional to one another. If a large number of samples are taken a t a downwind site and analyzed for X and Y , the best least-squares line Y = A BX should have the following properties: (1) The intercept, A, should be zero; (2) the slope, El, should be equal to the Y / X ratio characteristic of the source; (3) the variation of Y about the least-squares line should be approximately equal to the error in the measurement of Y , and similarly for X; (4) the correlation coefficient should approach unity. The third condition follows from the assumption that the elemental composition of the aerosol is constant, because errors in the measurement of Y and X are then the only sources of variation about the line. The best leastsquares line is generally not the regression line, since both Y and X are subject to measurement errors, and also since, as explained by Kendall and Stuart (8) (pp 375, 418), regression is not the correct mode of treatment when a functional relationship between the variables (here, X and Y ) is expected. Let us assume that the concentrations of elements X and Y in the aerosol particles are apportioned among the vari-
+
ous particle-size ranges in a manner that remains constant in time and is the same as that found at the source. In that case, the least-squares lines
Y ( r < r’) = Ay
+ ByY(tota1)
(3)
X ( r < r’) = Ax
+ BxX(tota1)
(4)
and
should each satisfy criteria similar to those listed above for the interelement data. There may be many atmo8pheric aerosol systems to which this simple model is inapplicable, most likely owing to multiple sources for the same element, transfer of material between the gas and particulate phases, coagulation of aerosol particles, or variation in the elemental composition and the size distribution a t the source. Experimental Sampling Procedure. The sampling equipment was set up in the W. M. Keck Laboratory, California Institute of Technology. Air was drawn from a point 6.7 m above the roof (18.5 m above street level) through a 21-m vertical sampling pipe (7-cm i.d.) described by Whitby et al. (9). The air stream was analyzed continuously by withdrawing parts of it to a number of instruments. One part of the stream was drawn through a membrane filter (0.8 wm mean pore flow diameter) to collect the aerosol for elemental analysis. This filter is called the totalmass filter. Another part of the stream was drawn through a similar filter preceded by the first four stages of an Andersen Cascade Impactor. The filter in this case is called the backup filter. The function of the impactor stages was to remove particles larger than -1.5 wm aerodynamic diameter (p1I2d). The backup filter then collected aerosol particles smaller than 1.5 pm aerodynamic diameter. Information about the size distribution of the elements in the gross aerosol can then be obtained by comparing elemental analyses of the total-mass filter and the simultaneously collected backup filter. We verified by means of electron-microscope photographs that the cutoff was 1.5 wm. The total-mass and backup filters were changed simultaneously every 2 hr by automatic sample changers. These sampling stations were designed and built ( 4 ) a t the Lawrence Berkeley Laboratory, University of California, for the U S . Environmental Protection Agency. They can be set to change the sample at any desired interval; the 2 hr generally used in our study represents a compromise between time resolution and sensitivity. A third part of the stream was drawn through a /3 thickness gauge (Frieseke & Hoepfner Type FH 62A Dust Monitor) which determined the total particulate mass concentration by measuring the P-ray attenuation in the particulate mass collected on a filter tape. The /3 gauge has been evaluated and described by Husar (IO);stated accuracy is f25%. Other parts of the stream were withdrawn for measurements of such items as various gases, light scattering, and condensation-nuclei counts. Carbon monoxide was measured every 10 or 20 min with a Beckman 6800 chromatograph. Samples were collected continuously (except for short interruptions for maintenance and repairs, and longer interruptions for subsidiary experiments) from October 30 through November 27,1972. During this 29-day period, 289 total-mass-filter and 223 backup-filter samples were collected. Of the latter, 190 were collected simultaneously Volume 9, Number 12, November 1975
1059
with total-mass-filter samples. The 0 gauge was available only November 16 through November 27; during this period, 114 total-mass-filter samples were collected. Elemental Analysis. The elemental composition of the particulate matter was measured on-site with the automatic X-ray fluorescence spectrometer. The instrument is described in detail elsewhere (3, 4 ) . A molybdenum secondary-target X-ray tube was used for the excitation of the sample. A 20-min fluorescence X-ray spectrum of each total-mass filter and each backup filter was accumulated with a Si(Li) semiconductor detector and associated electronics (system resolution 180 eV FWHM at 5.9 keV) and a 1024-channel pulse-height analyzer. The spectrum was automatically analyzed a t the end of the accumulation period by a minicomputer system which used a spectrum-stripping program and the stored spectra of 16 elements (K, Ca, Ti, V, Cr, Mn, Fe, Co, Ni, Cu, Zn, As, Se, Br, Sr, and Pb). The results, in ng/cm2 of filter, were printed by a Teletype minutes after the spectrum was accumulated. Meanwhile, the next filter moved automatically into the spectrometer, and the process was repeated. Elements lighter than K are relatively difficult to detect because of their low fluorescence yields and especially the strong absorption of the fluorescence X rays by matter (including the particles themselves). The absorption problem moreover renders the measurement of these light elements highly unreliable. There can also be interference from heavy-element L, M, etc. X rays; for example, the K lines of S suffer interference from one of the M lines of Pb. Because of these limitations, a number of light elements which would have been of great interest (e.g., Na, Al, Si, S, C1) are not included in our measurements. Elements in the 40 < Z < 60 region are not reported in this study because of the unsuitability of the molybdenum K X rays for their fluoresence. Capability for excitation of these elements, and also for those slightly below Z 20, has been added to the instrument but not in time for the Pasadena study.
Table I. Mean Concentrations, RMS Fluctuation o f the Concentrations, and Inter- and lntrafilter Measurement Errors,a.b for Elements o n Total-Mass Filters Errors
Element
Pb Br
Fe Ti
Mn Ca Ni V
Zn
Mean, nglm3
R.M.S. fluctuation, ng/m3
lnterfilter
2140.0 718.0 268.0 33.2 9.3 239.0 10.8 8.6 75.7
1851.0 649.0 272.0 39.0 15.1 256.0 11.3 10.9 79.1
7.0 7.2 7.3 28.0C 28.0C 52.0 16.0C 50.0C 7.5
Intrafilter ng/mjb
%a
>6 >3 >5 10 4 > 18 3 5
>3
a Inter-filter error stand r d deviation of measurements o f a given element o n replicate filters.% lntrafilter error =standard deviation among repeated measurements o f a given element o n t h e same filter, with t h e filter left in place i n t h e spectrometer for t h e repeated measurements. c T h e inter-filter error for this element is n o t a constant percentage o f t h e individual determinations; t h e error itself is essentially constant and is approximately equal t o t h e intra-filter error. T h e percent of t h e mean IS given.
-
Results and Discussion There were nine elements for which the average concentration in the 289 total-mass filters was larger than the interfilter error in the measurements: Ca, Ti, V, Mn, Fe, Ni, Zn, Br, and Pb. For each of the nine elements, Table I shows its average atmospheric concentration; the rms variability in the concentration; the interfilter error, that is, the standard deviation of measurements of a given element on replicate filters; and the intrafilter error, that is, the standard deviation among repeated measurements of a given element on the same filter (with the filter left in place in the spectrometer for the repeated measurements). Figure 1 shows the record of atmospheric concentrations of a few elements for the entire sampling period. Similar plots could be constructed for the other elements of the foregoing list. Interelement Correlations a n d Functional Relationships. The best-fit lines Y = A B X for possible interelement functional relationships were determined using the method of Bartlett ( I I ) , which consists of averaging the points in the first third of the plot of Y vs. X ,averaging the points in the last third of the plot, and drawing a straight line between the averages. This method ignores the data in the middle third of the plot. We refer to relationships deduced in this way as “best-fit’’ lines. The “best least-squares” lines (8, 12) were also determined. In this treatment, errors in both X and Y are taken into account (in contrast to the common least-squares pro-
+
1060
Environmental Science & Technology
io130
11/16
Iill3
I I I20
I112i
Flgure 1. Diurnal patterns of Pb, Br, Fe, Ni. and Zn for the entire sampling period. Generally the concentrations of all of the elements maximize at the same time, even though Pb and Br are the only ones shown here with a source in common. Note the close similarity between the Pb and Br patterns
cedure that takes into account only errors in Y ) .The errors in X and Y were assumed to be normally distributed and independent of the magnitudes of X and Y . The error ratio between Y and X was assumed to be that given by the interfilter error estimates listed in Table I, and the errors in X and Y were assumed to be uncorrelated or only slightly correlated ( r < 0.5). For most pairs of elements the best least-squares line agrees with the best-fit line. The slopes and intercepts for the best-fit lines with their rms errors are given in Table 11. The variation in Y about the best-fit lines is given in Table 111. The intrafilter error is not known for the elements Pb, Br, Fe, and Zn. The interfilter percent error in the determination of each of these elements is about 7%, probably due to differences in flow rates and variations in filter positioning in the spectrometer. These effects cancel in the ratio between two elements on the same filter; hence, the errors in the ratio are less than or equal to the interfilter errors.
3 m
-?
30 b N
3 3 om m e
m N 3
h
O
N
- -
b-
EZ
; ? " N O * N a* 3 w -m -m 'I-" ?c? - 0 coo 0 0 m N N
N A OLCj a 0 ab
m
b
-"
a3 om m 0 3
.-
2
- 0 9
: 2 ; ? - " ? ? aa N m 3rd Vm V b - e 53 3 -e -co -3 -* --1 'I? 0-"'I? ? ? 0c??c? m o b o N O eo 0 0 3 0 0 0 -
n
e
-
--
N
m
m
3
3
m
N
N
c L
m
N 3
n n
--
? -a -b " 7 'I" eo c oo 3 z m -0
Volume 9, Number 12, November 1975
1061
The interfilter errors ( i a ) are also shown in Table 111. The interfilter errors for Ca, V, Mn, Ti, and Ni are precisely what is expected from the statistical error in the determination of the peak area over a background signal. The intrafilter errors for these elements should also be equal to the statistical errors. The best interelement correlation seen in this study is that between Br and P b (Table 111; Figure 2). The intercept is zero (within f 2 a ) . The slope has very small error limits. These facts indicate that the Br and P b arise from a common source. The correlation of P b with CO (correlation coefficient 0.98) P b ( ~ g / m= ~ )0.888
+ 0.365 X CO (ppm)
(5)
confirms the expectation that all three species, Br, Pb, and CO, originate predominantly from automobiles and other gasoline engines. The variation of Br about the line in Figure 2 is twice the interfilter standard error. ,To determine whether this violates criterion # 3 of the single-source model, we have analyzed the Br/Pb data in other ways. The Br/Pb ratio for all total-mass-filter samples taken together shows a normal distribution but with a standard deviation (15%) that is broader than expected (10%). An analysis of variance in the Br/Pb ratio shows that the among-day variance (0.0095) is significantly larger than the within-day variance (0.0017). If the Br/Pb ratio varied due to random errors, the two variances would be equal. Hence, we conclude that there is a day-to-day variation in the Br/Pb ratio in the atmospheric particulate matter-Le., the model criterion #3 is not met in the case of Br vs. Pb. Loss of Br from the particulate phase is probably responsible for the variation in the Br/Pb ratio. In support of this, Figure 2 shows that the slope of the regression line is below the Br/Pb ratio (0.386) in tetraethyl fluid. The stoichiometry of the major compounds reported in automobile exhaust particulate matter (13-18) is such that the Br/Pb mass ratio should be approximately equal to the tetraethyl fluid value. Ninomiya et al. (19) find a Br/Pb ratio in automotive exhaust particulate of 0.45. Loss of Br from particulate matter has been documented by many investigators. Robbins and Snitz ( 2 1 ) reported that the Br/Pb ratio in isolated automobile exhaust particulate decays with a half-life of approximately 15 min; Ter Haar and Bayard (18) reported that most of the Br in isolated automobile exhaust particulate is lost within 1 hr. Actually, in view of our observed variation in Br/Pb ratio, these decay times seem surprisingly short. The largest Br/Pb daily averages (0.39) occurred on October 30 and November 11, the two windiest days in the sampling period, presumably because the more rapid advection results in a younger aerosol reaching the site. The lowest Br/Pb daily averages, 0.29, were observed on November 4 and 27, which were days with low wind speeds and which were also the smoggiest days in the sampling period. The day with lowest wind speed, November 21, had a Br/Pb daily average of 0.32, significantly higher than the minimum, and no smog. The Br/Pb relationship is evidently not straightforward; conditions associated with smog as well as low wind appear to be associated with low Br/Pb ratios. The Br/Pb ratios in the 2-hr samples are not correlated with P b concentration. As discussed later, the highest P b concentration for most days occurs during rush hours. Since fresh automobile exhaust particulate should have the maximum Br/Pb ratio, we might have expected Br/Pb to be positively correlated with Pb. However, the wind speed is also generally lowest during the rush hours, and this will 1062
Environmental Science & Technology
tend to decrease the Br/Pb ratio, perhaps leading to the observed lack of correlation. The Ti and Fe concentrations are approximately proportional to each other, with a zero intercept (within f 2 a ) and a well-defined slope. Apparently, therefore, most of the Ti and Fe originate from a common source. The slope of the best-fit line is within the range (Ti/Fe = 0.12 to 0.24) expected ( 2 4 ) for soil dust. Other sources might be involved, however, since the variation in Ti about the best-fit line is approximately twice the standard measurement error for Ti. An analysis of variance for the Ti/Fe ratio indicates a significantly larger variance among days (0.885) than within days (0.294). Most of the daily averaged ratios are close to the grand mean, but the daily average for November 11 is more than twice the grand mean. November 11 was a windy, rainy day, and the concentrations of all elements including Fe and Ti on the filters were very low as a consequence. Because of the resulting lack of accuracy, it is not certain that the high daily average on November 11 is significant. If November 11 is dropped from the analysis of variance, the among-day variance (0.355) is only 1.4 times the within-day variance (0.252) indicating a single source for Ti and Fe-i.e., soil dust. Manganese is also expected in soil dust ( 2 4 ) . However, the concentrations of Mn and Fe are not proportional to one another in our experiments, though they are well correlated. The nonzero intercept suggests that Fe, to a greater extent than Ti or Mn, is present in sources besides soil. An analysis of variance for the Mn/Fe ratio shows a few anomalous days when the daily-averaged ratio is substantially different from the grand mean, but generally on those days there was very little particulate matter in the filters and the accuracy of the measurements is poor. We find that Ti and Mn are not proportional to one another; the intercept is 2.4 times the standard error. The Ti/Mn slope is smaller than the expected 3.6 for soil dust. If one were to force the best-fit line to go through the origin, the slope would be 3.6. In general, then, the high correlation coefficients among the three elements, Ti, Mn, and Fe, and the Ti/Fe and Mn/Fe ratios, support the expectation that all three elements are derived principally from the soil. However, a clear-cut proportionality (zero intercept) exists only between Ti and Fe. The lack of proportionality in the other pairings could be caused by variability in the concentrations of Fe, Ti, and Mn in soil dust, or by other sources of Fe. Miller et al. ( 2 4 ) showed that the elemental composition of soil varies with location and that the elemental composition of the dust from soil is different from the elemental composition of the parent soil. However, the error limits for some of their analyses are not reported. Miller et al. ( 2 4 ) and Friedlander ( 6 ) conclude that the Fe in the Pasadena aerosol July 3, 1969, comes from metallurgical processes. The iron and steel industry in Los Angeles County has been estimated to emit 3.4 tons/day of gross particulate matter, out of a total primary man-made emission of 130 tons/day (25), and hence the possibility of metallurgical operations as a secondary source of Fe cannot be discounted. Our correlations and Ti/Fe and Mn/Fe ratios, however, indicate that soil was the major source of Ti, Mn, and Fe during November 1972. [Strong correlations between Fe, Mn, and other elements prominent in soil, including Al, were used by John et al. ( 2 6 ) to identify the sources of these elements as soil dust in the San Francisco Bay area.] Calcium is proportional to Fe. The best-fit line satisfies all four criteria of the one-source model. Calcium is almost
900
-
0.4
dy-. ).e. .
e': . , .A& . , '
, K(Br)=0718. (Pb)=2.140 2 3 4 5 6 Pb. pg/m3
"''
0
Br=-OOi3+03415Pb r.0.983
,
l-
1
I
I
7
9
1
0
Flgure 2. Linear regression between Br and Pb, two elements from a common source The regression line predicted on the basis of the Br/Pb ratio in tetraethyl fluid (TEL) in gasoline is shown. The "best-fit'' line (see text) is described by the 6 X where equation Y = A A = -0.013 f 0.019 ( 2 4ng/m3 Y = Br (ng/m3) B = 0.3415 & 0.0001 X = Pb (ng/m3) Note that the correlation coefficient is high ( r = 0.983), the intercept is close to zero (A = -0.0131), and the slope (6= 0.3415) is close to that expected on the one-source model with gasoline engines as the one source. Comparison between the slope of the regression line and the expected (TEL) value indicates that the vehicle exhaust particles have generally lost about 12% of their bromine
+
proportional to Ti; all criteria are satisfied except that the intercept is not zero within 2a. Calcium is proportional to Mn. The Ca/Fe ratio (the slope of the best-fit line) is 0.96, somewhat larger than that expected (24)for soil dust from a rural site (0.3-0.6), but similar to that found (24)for soil dust in the Pasadena urban area [0.6-1.4; Miller e t al. (2411. The two highest C a p e values reported for soil in the Pasadena area by Miller et al. (24)were 0.95 and 1.4, which pertain to soil taken near a street and from a construction site, respectively. The Ca enhancement probably stems principally from cement dust. Friedlander (6) believes that cement dust is the principal source of Ca in the Pasadena aerosol. Cement dust is high in Ca, nearly 50% by weight, and nearly devoid of the other elements measured in our study. On the basis of our data and the cited soil compositions, we would estimate that about half of the Ca in our samples is indigenous to the soil and the other half is cement dust (construction and roadway abrasion) mixed into and entrained within the soil. Calcium is poorly correlated to all the other elements except Ti, Mn, and Fe. This tends to strengthen the foregoing assessment. Vanadium and Ni appear to be proportional to one another. However, particle-size information (see later) shows that V and Ni actually come from different sources. Nickel is not well correlated with any element other than V. The measurement error for V is so large that variations about best-fit lines for any of the eight other elements are less than 2.1 times the V measurement error, so that criterion # 3 of the model can be satisfied by all elements relative to V. Moreover, V is also well correlated with Ti, Br, and Pb. Miller et al. (24)attribute both V and Ni to fuel-oil fly ash. However, they did not measure the atmospheric concentration of Ni; they did measure V but did not attempt to compare the concentration against an emission inventory. In the one sample of fuel-oil fly ash studied, they measured the V/Ni ratio to be 3.5, which does not agree with our atmospheric V/Ni ratio, 0.81 f 0.02. We conclude that Ni is a tag for a source or sources other than vehicles or soil, but we would decline to name fuel-oil fly ash. The remaining element, Zn, is not proportional to any other element measured, except perhaps V. Figure 3 shows
. . :c. .
.
2.
''?.-.;" ....'-.& 2
II
I
I
8
I
-
'
,
0
1
e.**' "
2
I
I
3
I
I
4
.
I
I
5 6 Pb. w / m 3
7
8
I
9
1
I 0
Figure 3. Linear regression between Zn and Pb The "best-fit'' line (see text) Y = A f 6 X where Y = Zn (ng/m3) A = 0.033 f 0.008 ( 2 ~ng/m3 ) X = Pb (ng/m3) 6 = 0.020 f 0.006 has a poor correlation ( r = 0.399) and large offset from the origin (1 # 0) characteristic of elements probably originating from independent sources
4
the scatter diagram for Zn and P b as an example. The approximate proportionality between V and Zn may be only a consequence pf the large uncertainties in the V measurements. Zinc has rather high correlation coefficients with Fe and Ti. However, there is much more Zn in the aerosol than would be anticipated if the source of Zn were soil dust. This, together with the lack of proportionality between Zn and Fe, indicates that Zn does not come from any of the same sources as the other elements measured in this study. It should be pointed out that almost all of the elements are significantly correlated to one another. A ,correlation coefficient of 0.25 would be 99% certain of being significant. The average coefficient in this study is 0.56. Undoubtedly this simply reflects that the emission rates from the various sources may be correlated in some cases and that the meteorological conditions, such as inversion heights, modulate all atmospheric concentrations alike. A high correlation coefficient by itself is not a criterion for ascribing two elements to a source in common. Elemental Size Distributions. Whitby et al. ( 5 ) report that the size distribution of aerosols, including the aerosol in Pasadena, often has two modes separated by a saddle at approximately 1-2 wm. This bimodality implies (and evidence advanced by Whitby et al. seems to corroborate) two classes of formation processes and a lack of mass transfer between the two modes. This being the case, the chemical compositions of the two modes ought to differ and therefore should be looked a t separately. This was part of the rationale for measuring compositions demarcated at 1.5 wm. The other reason was to test the size distribution of each element statistically to see if it met the size-distribution criteria of the single-source model as outlined earlier. In measuring the elemental size distribution, we seek to ascertain the functional relationship between each element on the backup filters and the same element on the simultaneously taken total-mass filters. The intercepts, slopes, and variations about the best-fit line, as determined from our results, are shown in Table IV. In the case of Fe, the slope of the best-fit line (0.235) indicates that 23.5% of the Fe is in particles less than 1.5 wm in diameter. The error in the ratio between a given element on the backup filter and the same element on the total-mass filter is essentially the interfilter measurement error. These errors are also given in Table IV. The concentration of P b on the backup filter is proportional to that on the simultaneously taken total-mass filter. Similarly, the concentration of Br on the backup filter is proportional to the Br on the total-mass filter. The ratios Volume 9, Number
12, November 1975
1063
Table IV. Best-Fit Lines for Size Distribution Data, X (backup) = A + BX (total) (The standard errors are given for the slope a n d intercept. The standard interfilter errors are given in parentheses under the variation about the line) Intercept, Slope
ng/m3
Variation about line,
E le ment
Pb
0.7969 (0.00005) 0.7852 (0.00005) 0.235 (0.004) 0.774 (0.0005) 0.757 (0.0015) 0.301 (0.038) 0.17 (0.09) 0.56 (0.015) 0.06 (0.09)
-3.5 (25.4) -2.0 (8.4) 1.6 (4.2) -1.2 (1.9) 0.4 (0.3) 0.4 (0.8) 5.3 (2.5) 1.0 (0.6) 49 (48)
202 (120) 67 (41) 39 (5.4) 9.9 (4.4) 2.2 (1.4) 4.6 (2.1) 7.9 (4.8) 3.9 (3.4) 60 (37)
Br Fe Zn
Ni
Mn Ti V
Ca
ng/m3
Pb(backup)/Pb(total) and Br(backup)/Br(total) are equal, to within 1%. Size distributions of P b emitted from automobiles have been extensively measured (13-19, 27-37). Except for the Hirschler papers (13, 14) (where the collection method may have biased the result), there seems to be general agreement that (once allowance has been made for nonsuspendable particles) the mass median aerodynamic diameter of the airborne exhaust P b is submicron. Habibi et al. (26, 17, 33, 34) report P b mass median aerodynamic diameters larger than 1 pm but Mueller (38) points out that, if the particles larger than 9 pm are excluded to approximate the airborne fraction, Habibi’s residual distributions also have submicron mass median aerodynamic diameters. Huntzicker et al. (39) found approximately 80% of the airborne P b mass to be in particles smaller than 1.5 pm aerodynamic diameter at the Pasadena site in November 1972, in agreement with our values of Pb(backup)/Pb(total) and Br(backup)/Br(total). Atmospheric P b size distributions (22, 23, 39-43) reveal considerable disagreement as to what the mean size is and whether the size distribution varies with time and/or location. Gillette and Winchester (43) report that the P b size distribution is constant with time and location, a t least for sizes greater than 0.2 pm. Our results show that the Pb(backup)/Pb(total) ratio and the Br(backup)/Br(total) ratio were both constant in time during November 1972 a t ,the Pasadena site. Robbins and Snitz (21) have argued that the loss of Br from automotive exhaust particulate is rate-limited by the diffusion of Br from the interior of the particles to the particle surface. If they are correct, then Br in small particles should be lost much more rapidly than Br in large particles. We find that the Br/Pb ratio-Le., the slope of the best-fit line, on the backup filters is 0.3426 f 0.0003 which is almost identical to the Br/Pb ratio on the total filters (0.3415 f 0.00005). This means that we find no variation of Br/Pb ratio with particle size. Martens et al. (22) have also found the Br/Pb ratio to be independent of particle size. As they point out (22, 23), such results argue against the diffusion mechanism for Br loss. The question is still not settled, however (23,44). The concentration of Fe on the backup filter appears to be proportional to that on the total-mass filter. However, the variance of Fe(backup) about the best-fit line is 7.2 times that expected from measurement errors. An analysis 1064
Environmental Science & Technology
of variance for the ratio Fe(backup)/Fe(total) shows that the among-day variance is significantly larger than the within-day variance. The highest values for this ratio (Fe concentrated in smaller particles than usual) occurred on rainy days (November 14, 16, and 17). The lowest values occurred on dry windy days (October 30 and November 8); this suggests a connection between low ratio and high wind, as would be expected for soil dust. There is one day (November 11)that was both windy and rainy; for that day, the ratio was very close to the grand mean, so that rain might be a factor as well as wind but in the opposite direction. On rainy days the amount of Fe collected on the filter was very low. Systematic errors in the latter situation may confuse the picture, but in general one recognizes in these observations the role of wind in entraining soil dust, and also the role of rain in preventing entrainment and in removing particles by rainout and washout. The shift of the Fe size distribution in favor of smaller particles on rainy days is to be expected if, as is generally accepted (45), rainout and washout preferentially remove the larger airborne particles. Another explanation for the varying size distribution of the Fe is that the size distribution of the particles as “seen” by the sampling-tube inlet may be altered by the effect of the wind on the trajectories of the largest particles, in a manner analogous to the problem of isokinetic sampling from a moving stream. We found that 37% of the Fe measured by a sampling station on the roof did not reach a second sampling station a t the end of the sampling pipe (in contrast to Pb, which shows no significant losses in the sampling pipe). Presumably only the larger Fe-containing particles are lost, so the ratio Fe(backup)/Fe(total) on the roof should be 0.15. The slopes of the Ti and Mn lines are essentially the same as the slope for Fe. The slope of the Ca line is apparently smaller than that of Fe (i.e., the Ca tends, even more than does Fe, to reside in the larger particles). The concentration of Zn(backup) is proportional to Zn(tota1). Similarly, the concentration of Ni(backup) is proportional to that of Ni(tota1). It is especially interesting that the backuphotal ratios for all elements except V cluster about two values, 0.2 and 0.8 (Figure 4), signifying a segregation of particle composi-
-
3
I
0.4
LL
J 0.3
v)
I Flgure 4. Ratio between the mass on the backup filter and the mass on the total-mass filter, for each of the series of nine elements The errors in the ratio for Pb. Br, Ni, and Zn are too small to be shown on this scale, but the errors in the ratio (24are shown for V, Fe, Ti, Mn. and Ca
tion according to particle size. This is consistent with the concept of bimodal particle-size distributions ( 5 ) as arising from two categories of particle-generation mechanisme.g., condensation for the small-particle mode and abrasion for the large. In the small-particle class, with 80% of their mass residing in particles smaller than -1.5 wm, we observe Pb, Br, Zn, Ni; in the large-particle class, with 80% larger than 1.5 wm, we find Ca, Ti, Mn, Fe (Figure 4). All four small-particle elements have the same backup/total ratio within a few percent. Also, the ratio for each of these four is independent of time, which argues against any given one of them having multiple significant sources. Judging by the particle sizes, Ni and Zn are probably associated with combustion or other high-temperature processes. Vanadium is anomalous. Though the accuracy in its determination is poor, the slope of the best-fit line for the size distribution is well defined (Table IV). From the size distributions it is evident that V either comes from a source of medium-size particles-i.e., a different source from any of the other elements; or that it comes from two sources, one emitting small particles and the other emitting large ones. The latter alternative is favored by the idea of the bimodal size distribution. However, distribution of the V(backup)/ V(tota1) ratios is a normal distribution with a standard deviation only 50% of the mean (i.e., with % of the ratios falling between 0.8 and 0.3); there is thus no indication of two V sources. This evidence contrasts with observations of detailed size distributions and V/Al ratios in the San Francisco Bay area (46)showing the V arising from both smallparticle and large-particle sources identified as combustion and soil, respectively. Total P a r t i c u l a t e Mass. The average of the total particulate mass loadings for the 12-day period November 16 through 27 a t the site in Pasadena, as measured by the /3 thickness gauge described earlier, was 31.7 wg/m3. This average is consistent with values measured with the /3 gauge at this site in April 1972 (IO), but is substantially lower than the 100 wg/m3 found at the same site in August and September 1969 ( 4 7 ) . The low particulate loading in our study is probably related to the unusually light smog activity during the period. The alternative possibility of large systematic error in the P-gauge readings can probably be discounted. Beta-gauge measurements of aerosol mass have been shown to agree well with those obtained by gravimetric measurement of filters (IO). We have confirmed this to hold for one test sample obtained during November 1972. Random errors with the P gauge are -5% for a 10-hr average, or 25% for a 2-hr average. Errors due to atmospheric relative-humidity effects are said to be small (IO); moreover, the relative humidity for the latter part of November 1972 was quite constant. Each element measured on the total-mass filter correlates well with the P-gauge readings. With the exception of Ca and V, all of the correlation coefficients are approximately 0.75. Source Strengths. The strengths of the sources identified in this study are calculated from our experimental tagelement atmospheric mean concentrations, together with the source compositions in Table V given by Friedlander (6). The results are shown in Table VI and compared there to Friedlander’s estimates. We find that, a t the Pasadena sampling site du_ring November 1972, gasoline-engine exhaust particulate matter was on the average 5.35 pg/m3, and the average soil-dust concentration was 8.35 pglm3. During the latter part of November when the gross-mass measurements were available, gasoline-engine exhaust particulate matter was on the average 5.6 wg/m3 or 18%of the
Table V. Confpositions of Various Sources of Particulate Matter, in Percent (Taken from Friedlandera) Sea salt
Auto Fueloil Portland Soil dust exhaust fly ash cement
C com-
pounds
Na.
40.3b -2.5 1.4 8.2 20
30.6, 3.69
Mg AI Si
S
2.56 55.0 1.1 1.16
CI K
Ca Ti
6.8 1.5 1.5 0.4 0.006
V
Cr Mn Fe
0.11 3.2 0.002 0.004 0.008
