Indoor-Outdoor Relationships and Infiltration Behavior of Elemental Components of Outdoor PM_2.5 for Boston-Area Homes

Abstract

In order to investigate the relationship between indoor and outdoor elemental concentrations and to characterize the infiltration behavior of elemental PM_2.5 constituents, we conducted an analysis of indoor and outdoor PM_2.5 elemental data collected during a comprehensive particle characterization study of nine nonsmoking homes in Boston, MA. Using data from nighttime periods when little or no particle-generating activity occurred, analyses focused on six elements that were consistently detected in both indoor and outdoor samples and that spanned a range of particle sizes: sulfur, nickel, zinc, iron, potassium, and silicon. Results showed that outdoor levels of all the elements were highly correlated with their corresponding indoor levels. Correlations remained high for different air exchange rate conditions, building characteristics, and seasons, suggesting that variability in ambient elemental infiltration into residences may not be a large source of variability affecting personal-ambient correlations for these elements. Elemental infiltration factors showed strong relationships with air exchange rate and season and were suggestive of an effect of particle size, which was likely obscured by remaining indoor source impacts. Analyses of this small dataset provided an indication that several elements—in particular nickel—could potentially serve as accurate tracers for infiltration of total PM_2.5 mass- and size-resolved particles into residential buildings. Similar to previously reported findings for sulfur, these elemental tracers showed the poorest performance for smaller and larger particle sizes.

INTRODUCTION

Numerous epidemiological studies have reported significant associations between ambient PM_2.5 concentrations and adverse health effects (e.g., Pope et al. 2002; Stieb et al. 2002; Samet et al. 2000; HEI 2003). In addition, several recent epidemiological studies have shown associations between ambient elemental PM_2.5 constituents and adverse health effects (Laden et al. 2000; Burnett et al. 2000; Mar et al. 2000). In order to properly interpret these findings, exposure studies should be applied to assess and understand the validity of the use of ambient particle levels as surrogates of personal exposure for not only PM_2.5 but also for specific PM_2.5 constituents.

Since people spend the majority of their time indoors (> 85%) where a significant portion of their total personal exposures to ambient particles can occur (Klepeis et al. 2001), ambient particle infiltration into indoor environments is one of the more important factors affecting the relationship between ambient concentrations and corresponding personal exposure to ambient particles in epidemiological studies. With the exception of sulfate (Sarnat et al. 2002), however, there are few data available to describe the infiltration behavior of PM_2.5 elemental species into residential spaces. The few studies that have reported the elemental composition of both outdoor and indoor particles (i.e., Hidy et al. 2000; Jones et al. 2000; Janssen et al. 1999; Özkaynak et al. 1996; Clayton et al. 1993; Koutrakis et al. 1992; Lewis 1991) identified large indoor source contributions for many elements, thus limiting their use for determination of elemental infiltration factors.

In order to examine the relationship between indoor and outdoor elemental concentrations and to characterize the infiltration behavior of elemental PM_2.5 constituents, we analyzed residential indoor and outdoor PM_2.5 elemental data from a comprehensive particle characterization study conducted in Boston, MA. Previous analyses of these study data quantified ambient infiltration factors for PM_2.5 and particles within specific size fractions (Long et al. 2001) and have been used to evaluate the utility of sulfur as a tracer for PM_2.5 infiltration into residences (Sarnat et al. 2002). Including only nighttime periods with few if any human activities, this analysis focused on six elements (sulfur, nickel, zinc, iron, potassium, and silicon) that were consistently detected in indoor and outdoor samples and that span a range of particle sizes.

Improving our understanding of ambient infiltration of elemental PM species is not only important for assessing exposure and potential health effects of these specific PM constituents but also because elemental species may have utility as tracers of PM of ambient origin (Sarnat et al. 2002, 2000; Ebelt et al. 2000; Oglesby et al. 2000). Since it is not currently possible to directly measure indoor levels of PM of ambient origin, indoor tracers of ambient PM can be valuable tools for apportioning the contributions of ambient sources to total indoor PM levels. Previous analyses of these data have demonstrated that sulfur is a strong tracer of ambient PM_2.5 and particles with diameters between 0.06 and 0.5 μm in Boston-area residences (Sarnat et al. 2002). Given that sulfur served as a less-precise tracer for very small (< 0.06 μm) and large particle sizes (> 0.7 μm), there is interest in determining whether other elements that are more typically found within these size ranges can serve as more accurate tracers. In addition, although most homes will have few if any indoor sulfur sources, homes with kerosene heaters and smokers have been demonstrated to have indoor sulfur contributions (Koutrakis et al. 1992; Leaderer et al. 1999), necessitating the use of alternative tracers of ambient PM in some homes. This study thus expanded previous analyses for sulfur to additional elemental species and evaluated whether other elements have utility as potential tracers of PM_2.5 infiltration into residences.

METHODS

Study Design

As described previously (Abt et al. 2000a; Long et al. 2000, 2001; Sarnat et al. 2002), indoor and outdoor particle mass concentration, elemental composition, and size distribution data were collected as part of a comprehensive particle characterization study in Boston-area homes during 1998. Long et al. (2000) contains the most complete description of the study design, sampling methods, and data quality assurance procedures. Briefly, nine nonsmoking Boston-area homes were sampled for one or two week-long periods during spring–summer and fall–winter 1998. Five of the nine study homes were sampled during each of two seasons. All homes were located within 30 miles of downtown Boston in suburban neighborhoods.

Study homes were typical of homes in New England, a region in the United States with four distinct seasons including cold winters and warm summers. With the exception of one home where a central air-conditioning system was used, home occupants typically opened windows and doors to promote air circulation during the warmer spring and summer months. Windows and doors were predominantly kept closed for the winter months as well as the majority of the fall sampling period. Previously reported mean nighttime air exchange rates of 2.0 and 0.8 h⁻¹ for spring–summer and fall–winter sampling periods, respectively, provide support for these qualitative observations regarding seasonal differences in ventilation characteristics and home tightness (Sarnat et al. 2002). During the winter months, five of the nine homes were heated with oil, while the remaining four had natural gas heating systems. None of the homes used kerosene heaters, woodstoves, or fireplaces—all well-recognized sources of elemental PM constituents (Koutrakis et al. 1992; Özkaynak et al. 1996; Leaderer et al. 1999)—during the sampling periods.

This article describes a subset of the sampling data for nighttime periods where PM_2.5 filters were analyzed for elemental composition. A total of 49 pairs of indoor and outdoor 12 h nighttime (i.e., 8 pm to 8 am) filters representing six of the nine study homes were selected for elemental analysis based on the minimal occurrence of recorded particle-generating activities in time-activity logs. Only filters collected during nighttime sampling periods were selected since the objective of the analysis was to investigate infiltration of ambient PM. With few human activities, it was assumed that there would be limited indoor source contributions of elemental species. For four of the six homes, filters were selected for both the spring–summer and fall–winter sampling periods. In addition to elemental composition, other data collected from each home included continuous size distribution measurements, both 12 h and continuous PM_2.5 mass concentrations, air exchange rates, and time–activity information.

Data Collection Methods

Twelve-hour PM_2.5 samples were collected on Teflon filters using Harvard Impactor (HI) samplers. Using well-established sampling protocols, 12 h time-integrated samples were collected at a flow rate of 10 lpm on preweighed 37 mm Teflon filters (Teflo^™, Gelman Sciences, Ann Arbor, MI, USA). As described in Long et al. (2000), field blanks comprised 10% of the total number of collected samples, and blank-corrected PM_2.5 mass concentrations were determined following gravimetric analysis.

Elemental analysis was conducted on a subset of the HI filters by X-ray fluorescence spectroscopy (XRF) at the Desert Research Institute (DRI; Reno, NV, USA) according to DRI standard operating protocols, which include rigorous quality control and assurance (Chow and Watson 1998; Watson et al. 1998; DRI 1998). For each element and sample, concentration uncertainties equal to one standard deviation of error estimates based on analytical precision were reported by DRI. Similar to previous studies (Thomas et al. 1993), three times the uncertainty was considered to be the uncertainty limit for each sample and element. Blank correction was necessary for only three elements (magnesium, aluminum, and phosphorous) where field blank concentrations exceeded the reported uncertainty limits. Similar to previous studies reporting aerosol elemental data (Thomas et al. 1993; Hidy et al. 2000), only concentrations that exceeded three times the sample-specific uncertainty limits after blank correction were reported. One indoor zinc measurement was removed from the dataset since it was determined to be a statistical outlier. Due to particle size effects for Mg and Na, DRI recommended that concentrations of these elements should be considered as qualitative measures only.

Real-time size distribution measurements were made using two particle-sizing instruments, the scanning mobility particle sizer (SMPS; Model 3934, TSI, Inc., St. Paul, MN, USA) and the aerodynamic particle sizer (APS; Model 3310A, TSI, Inc., St. Paul, MN, USA). These instruments provided particle count concentrations within discrete size bins between 0.02 to 0.5 μm (SMPS) and 0.7 to 10 μm (APS). As described elsewhere (Abt et al. 2000a; Long et al. 2000), these instruments alternately sampled both indoor and outdoor air from ports in a specially designed stainless steel sampling manifold. Size distributions were obtained over 5 min sampling periods, where indoor measurements were made at the 0, 5, 10, 20, 25, 30, 40, 45, and 50 min intervals of each hour and outdoor measurements were made at the 15, 35, and 55 min intervals. Fine particle SMPS and APS data were converted to volume concentrations (μm³/cm³) for 12 discrete particle size ranges (for the SMPS, 0.02–0.03, 0.03–0.04, 0.04–0.06, 0.06–0.08, 0.08–0.1, 0.1–0.2, 0.2–0.03, 0.3–0.4, and 0.4–0.5 μm; for the APS, 0.7–1.0, 1.0–2.0, and 2.0–3.0 μm). Median indoor and outdoor volume concentrations were computed for 12 h sampling intervals matched to the corresponding indoor and outdoor HI samples.

Other data collected during the comprehensive sampling activities include continuous air exchanges rates and detailed time–activity information. Air exchange rates (AERs) were measured in each home every 5 min using continuous injection of sulfur hexafluoride (SF₆) tracer gas and measurement via a photo-acoustic monitor (Abt et al. 2000a; Long et al. 2000). Time–activity information was recorded by the home occupants in 20 min intervals using a daily time–activity diary.

Data Analysis

Data for the various elemental species, and their relationship with total PM_2.5 mass concentration and volume concentrations within discrete size ranges, were characterized using descriptive statistics, graphical displays, and mixed model regression analysis. All elemental concentration data are reported as ng/m³. For all statistical analyses, significance is defined for the 0.05 level. Version 8 of the Statistical Analysis System (SAS Institute, Cary, NC, USA) was used for all data analyses.

Descriptive statistics were performed to describe the elemental composition of indoor and outdoor PM_2.5. Nonparametric Wilcoxon paired signed-rank tests were used to compare indoor–outdoor differences in elemental and particle concentrations, while both elemental and indoor–outdoor correlations were described using Spearman's correlation coefficients (R_S). The strength of the association between indoor and outdoor concentrations was investigated using both indoor–outdoor correlations and mixed-model regression analyses. Mixed-models were used to account for the repeated measures within each home. In the mixed-model regression analyses, indoor concentrations were modeled as dependent variables; outdoor concentrations were modeled as independent, fixed variables; and home was modeled as an independent, random effect. Because regression intercepts represent indoor concentrations when outdoor concentrations equal zero, they can be interpreted to indicate indoor source contributions.

Statistical analyses were conducted on subsets of the dataset corresponding to either high or low AER conditions. Since all but one of the study homes were naturally ventilated, higher AERs typically occurred in the spring–summer sampling period as homeowners opened windows and doors to increase ventilation during warmer weather (Long et al. 2000). As a result, effects on indoor–outdoor relationships attributed to air exchange rate are likely related to season as well. However, analyses were not conducted using data stratified by season since it is believed that season is serving as a surrogate for AER (although it is also true that if ambient source emissions vary by season, seasonal impacts on I/O ratios could result if size distributions differ or if outdoor concentrations fall to levels near or below the detection limit). Sampling periods were classified as having high AER conditions when the 12 h mean AERs exceeded 0.86 h⁻¹ or the overall median AER for all study homes. Conversely, homes with mean AERs less than the overall study median of 0.86 h⁻¹ were classified as having low AERs.

Mixed models and graphical displays were used to assess the utility of elemental indoor–outdoor (I/O) ratios as estimates of infiltration of PM_2.5 and particles of different size ranges into the study homes. Mixed models were performed by regressing either I/O ratios for PM_2.5 or the specific size ranges on elemental I/O ratios, with the intercept forced through the origin. A slope of one was used as an indicator of an unbiased (i.e., accurate) relationship between the two sets of I/O ratios and hence as evidence that an element could serve as a reliable tracer for the infiltration of total PM_2.5 or a particle size range into the study homes (Sarnat et al. 2002). Lower slopes indicated poorer agreement between the two sets of I/O ratios; for example, a slope of 0.5 indicated that on average the elemental I/O ratios were 50% greater than those for the other particle measure (i.e., either total PM_2.5 or a particular size range).

In addition, the mean deviation was used as an additional measure of the predictive ability of elemental I/O ratios to estimate ambient PM infiltration. Mean deviations, which serve as a measure of relative agreement, were calculated as the mean of the absolute mean deviation:

[1]

where [I/O]_{element, ij} is the I/O ratio for a particular element for home i on day j, and [I/O]_{PM2.5 (or PVsize), ij} is the I/O ratio of a corresponding particle measure for home i on day j.

RESULTS AND DISCUSSION

Indoor and Outdoor Elemental Composition

Table 1 provides summary statistics for all indoor and outdoor samples detected above the sample-specific uncertainty limits. Mean outdoor and indoor PM_2.5 concentrations were 13.7 and 10.8 μg/m³ for the 49 samples, respectively. Few of the 40 elements analyzed were consistently detected in both indoor and outdoor samples (i.e., Al, Ca, Fe, K, Ni, Si, S, and Zn). Notably, vanadium was not detected above its uncertainty limit in any indoor or outdoor samples despite its well-documented presence in fuel oil combustion emissions (Laden et al. 2000) and the high number of Boston-area homes that use home heating oil. Elemental concentrations were generally on the order of a few to tens of ng/m³ both indoors and outdoors, with only sulfur concentrations exceeding 1 μg/m³. Nonparametric Wilcoxon paired signed-rank tests showed that outdoor PM_2.5 and elemental concentrations were significantly higher than their corresponding indoor concentrations for PM_2.5, S, Ni, K, Zn, and Fe (p < 0.01 for all but K, where p = 0.02), and higher but insignificant for Al and Si (p = 0.7 and 0.2, respectively). Only for Ca were concentrations significantly greater indoors (p = 0.02).

TABLE 1 Summary of indoor–outdoor PM_2.5 compositional data for 12 h nighttime samples

	Outdoor conc. (ng/m^3)					Indoor conc. (ng/m^3)					I/O Ratios
Constituent	N > LOD	Mean	SD	Median	Maximum	N > LOD	Mean	SD	Median	Maximum	N	Mean	SD	% > 1.15	I/O correlations Spearman R
Total PM2.5	49	13717	7203	11843	41402	49	10820	6815	9300	30986	49	0.78	0.21	10	0.82
Al	36	17	13	13	65	29	17	11	14	49	24	1.19	0.84	42	0.39
As	0					4	5	2	5	7			NC		NC
Ba	1	133	NC	133	133	0							NC		NC
Br	26	4	2	3	8	23	3	1	3	5	16	0.81	0.23	6	0.78
Ca	35	23	12	19	51	46	27	12	25	63	35	1.46	0.75	51	0.48
Cl	0					6	122	114	85	316			NC		NC
Cu	19	5	2	4	11	18	6	6	5	31	12	1.27	1.29	25	0.12
Fe	48	41	25	33	117	49	29	18	24	80	48	0.81	0.43	10	0.78
K	46	42	20	39	84	48	37	23	34	169	45	1.01	0.47	31	0.73
Mg	10	25	12	22	46	14	24	12	20	45			NC		NC
Na	23	109	38	91	192	33	121	63	95	285	18	1.24	0.72	39	0.28
Ni	34	6	2	6	12	24	6	3	4	11	22	0.80	0.25	9	0.72
P	12	16	10	13	34	16	11	4	10	19			NC		NC
Pb	20	8	3	7	15	10	7	2	7	11	8	0.81	0.25	13	0.62
Pd	2	14	0.3	14	14	1	15	NC	15	15			NC		NC
Se	18	4	2	4	9	8	2	0.4	2	3	7	0.52	0.27	0	0.54
Si	49	64	41	54	225	49	59	29	53	173	49	1.09	0.61	35	0.62
Sr	1	2	NC	2	2	0							NC		NC
S	49	1510	1184	1072	6356	49	1212	1175	830	4955	49	0.74	0.23	4	0.92
Zn	46	14	8	12	38	47	10	5	10	21	45	0.83	0.31	18	0.79
Zr	0					3	3	0.4	3	3			NC		NC
The following elements were not detected above their uncertainty limits in any sample: Ti, Va, Cr, Mn, Co, Ga, Rb, Yt, Mo, Ag, Cd, In, Sn, Sb, La, Au, Hg, Tl, and Ur.
Only Al, P, and S were detected above their uncertainty limits in field blanks and thus required blank corrections.
One indoor zinc measurement was removed from the dataset because it was determined to be a statistical outlier.
Due to particle size effects for Mg and Na, concentrations of these elements should not be considered quantitative.
I/O ratios and correlations are presented when the number of matched I/O samples > 5.
NC = not calculated.
Correlations in bold indicate statistical significance (p < 0.05).

Correlations among the outdoor elemental concentrations are consistent with the reported major ambient source types for these elements in the literature (USEPA 1996), as two groupings of highly significant correlations were observed (Table 2a). Sulfur was significantly correlated with Ni (with a moderate Spearman's correlation coefficient of 0.38), and this is consistent with regional transport of fossil fuel combustion emissions as a source of both of these elements. Sulfur was only significantly correlated with one other element, Al, and the source of this correlation (0.43) is not known. With the exception of Al, which was only significantly correlated with Ca (0.47), the crustal elements (Ca, Si, K, and Fe) were all significantly correlated with each other, with Spearman's correlation coefficients ranging from 0.48 to 0.86. In addition, moderate significant correlations between 0.35 and 0.59 were observed between the three elements with potential industrial sources—Ni, Zn, and Fe. Interestingly, elemental correlations were slightly higher for indoor samples, both for all data (Table 2b) and the censored dataset (Table 2c), where indoor data were removed for I/O ratios exceeding 1.15 to minimize the impacts of indoor source contributions.

TABLE 2 PM_2.5 and elemental correlation coefficients for outdoor and indoor datasets

	S	Ni	Zn	Fe	K	Si	Ca	Al
Outdoors
S		0.38	−0.13	−0.10	−0.24	0.005	0.07	0.43
Ni	0.38		0.36	0.35	0.09	0.33	0.24	0.15
Zn	−0.13	0.36		0.59	0.68	0.48	0.31	−0.03
Fe	−0.10	0.35	0.59		0.66	0.86	0.70	0.17
K	−0.24	0.09	0.68	0.66		0.64	0.50	0.20
Si	0.005	0.33	0.48	0.86	0.64		0.82	0.31
Ca	0.07	0.24	0.31	0.70	0.50	0.82		0.47
Al	0.43	0.15	−0.03	0.17	0.20	0.31	0.47
Indoors
S		0.37	−0.06	0.09	−0.10	0.16	−0.11	0.43
Ni	0.37		0.26	0.34	0.20	0.17	0.19	0.33
Zn	−0.06	0.26		0.71	0.61	0.65	0.66	0.17
Fe	0.09	0.34	0.71		0.68	0.87	0.69	0.09
K	−0.10	0.20	0.61	0.68		0.71	0.70	0.13
Si	0.16	0.17	0.65	0.87	0.71		0.77	0.30
Ca	−0.11	0.19	0.66	0.69	0.70	0.77		0.27
Al	0.43	0.33	0.17	0.09	0.13	0.30	0.27
Indoors—excluding indoor data where I/O > 1.15
S		0.52	−0.03	0.09	−0.03	0.33	−0.11	0.53
Ni	0.52		0.31	0.29	0.08	0.08	0.29	0.38
Zn	−0.03	0.31		0.72	0.71	0.65	0.65	0.13
Fe	0.09	0.29	0.72		0.71	0.91	0.65	0.11
K	−0.03	0.08	0.71	0.71		0.72	0.77	0.18
Si	0.33	0.08	0.65	0.91	0.72		0.78	0.36
Ca	−0.11	0.29	0.65	0.65	0.77	0.78		0.34
Al	0.53	0.38	0.13	0.11	0.18	0.36	0.34
Correlations in bold indicate statistical significance (p < 0.05).

Although only samples from nighttime periods with few if any human activities were analyzed, indoor/outdoor (I/O) ratios shown in Table 1 still indicate the contributions of indoor sources for a number of elements, including Ca, Al, Si, and K. For each of these elements, the mean I/O ratio frequently exceeded one, with 51, 42, 35, and 31% of I/O ratios exceeding 1.15 (i.e., ≫1) for Ca, Al, Si, and K, respectively. We chose an I/O ratio of 1.15 (i.e., indoor concentrations were 15% greater than the corresponding outdoor concentrations) as an indicator of the contributions of indoor source events based on our review of the few cases where limited indoor activities were known to occur and we observed an impact on elemental I/O ratios. In particular, we observed on several occasions that brief resuspension events (e.g., walking, checking sampling equipment) were associated with elevated I/O ratios of the larger elements (Ca, Al, Si, and K).

The presence of indoor sources for these elements (i.e., Ca, Al, Si, and K), along with wintertime Zn, was also indicated by positive and statistically significant regression intercepts for indoor–outdoor mixed models (data not shown). Other studies have shown these elements to have major indoor sources that include resuspension of soil and crustal materials for Al and Si (Koutrakis et al. 1992; Janssen et al. 1999), and both cooking and resuspension for K and Ca (Özkaynak et al. 1996; Oglesby et al. 2000; Janssen et al. 1999). Despite the likely presence of indoor sources of these elements, indoor–outdoor correlations were moderate to high for these elements (R_S = 0.39, 0.48, 0.62, and 0.73 for Al, Ca, Si, and K, respectively), indicating that outdoor concentrations could explain much of the variability in indoor concentrations.

In contrast, I/O data indicate that S, Ni, Zn, and Fe were least impacted by indoor source contributions. Mean I/O ratios for these compounds ranged from 0.74 to 0.83, while I/O ratios exceeded 1.15 for only 4, 9, 18, and 10% of the S, Ni, Zn, and Fe samples, respectively (Table 1). Again, indoor–outdoor correl-ations were strongest for these compounds, with Spearman's correlation coefficients of 0.92, 0.72, 0.79, and 0.78 for S, Ni, Zn, and Fe, respectively.

Based on these analyses which indicate that indoor source contributions affected indoor elemental concentrations during some sampling periods, we focused subsequent analyses on elemental infiltration behavior and tracer performance on the six elements that were consistently detected in both indoor and outdoor samples, that were largely retained following removal of data where I/O ratios exceeded 1.15 (i.e., at least 20 paired indoor and outdoor samples), and that spanned a range of particle sizes: S, with a literature mass median diameter of 0.52 ± 0.26 μm (Milford and Davidson 1985); Ni, with a literature mass median diameter of 0.98 μm (Milford and Davidson 1985); Zn, with a literature mass median diameter of 1.13 μm (Milford and Davidson 1985) and “typical” size range of 0.65–20 μm (Alzona et al. 1979); Fe, with a literature mass median diameter of 3.42 μm (Milford and Davidson 1985) and “typical” size range of 3.6 to 20 μm (Alzona et al. 1979); K, with a literature mass median diameter of 3.76 μm (Milford and Davidson 1985); and Si, with a literature mass median diameter of 3.9 μm (Milford and Davidson 1985). Observations were excluded when I/O ratios exceeded 1.15, since data show that these elevated I/O ratios are indicative of indoor source contributions. For most elements, few observations were removed, ranging from 2 samples for both sulfur and nickel to 17 samples for silicon (or 35% of the 49 detects).

Characterization of Elemental Infiltration Behavior

Data were collected during periods of limited indoor PM source contributions; therefore, I/O ratios for S, Ni, Zn, Fe, K, and Si can be used as estimates of elemental infiltration factors (F_inf, also known as the effective penetration efficiency, or P_eff). This is consistent with the definition of F_inf as the equilibrium fraction of ambient particles that penetrate indoors and remain suspended (Wilson et al. 2000). Figure 1 shows that I/O ratios for these elements were highly variable; for example, I/O ratios ranged from 0.49 to > 1 for Ni, and 0.23 to > 1 for Fe. Average I/O ratios were higher for all elements for high AER conditions compared to low AER conditions. For S, Ni, Zn, and Fe, which appear to be the least impacted by potential indoor source contributions, median I/O ratios were 0.61, 0.65, 0.68, and 0.55, respectively, for low AER conditions and 0.77, 0.75, 0.77, and 0.78, respectively, for high AER conditions. I/O ratios equal to or higher than 1 were typically observed for homes with high AERs, when windows and doors were left open. In addition, a number of very low I/O ratios (i.e., < 0.4) were observed, and these values were predominantly observed in the home with a central air-conditioning unit and very low AERs.

FIG. 1 Distributions of elemental indoor/outdoor ratios stratified by low and high AER conditions.

Figure 2 further illustrates the strong relationship between AER and I/O ratios for several of the elements. For S, Ni, Zn, and Fe in particular, lower I/O ratios are typically associated with lower AERs. As AERs approach 1–2 h⁻¹, a steady upward trend in I/O ratios is exhibited for most elements including S, Ni, Zn, and Fe. For all of these elements, infiltration factors appear to plateau at values between 0.8 and 1 as AERs increase to values exceeding 1 or 2 h⁻¹.

FIG. 2 Relationships between elemental indoor/outdoor ratios and AERs. Filled circles represent spring–summer samples, while open circles represent fall–winter samples.

The strong relationship between AER and elemental infiltration behavior for these elements is consistent with PM penetration and deposition theory and previous study findings. Previous studies have shown that ambient infiltration of PM_2.5 and particles within discrete size ranges are strongly related to home ventilation characteristics, due to both the direct impact of high AERs as well as their indirect impacts on both penetration and deposition behavior (Abt et al. 2000b; Long et al. 2001; Vette et al. 2001). Long et al. (2001) demonstrated that penetration efficiencies for PM_2.5 are reduced in homes that are kept tight with low AERs. In contrast, for homes where windows and doors were left open and AERs were high (≫1 h⁻¹), penetration efficiencies for PM_2.5 and particles with sizes between 0.02 and 1 μm were near unity. Deposition theory also supports increased ambient infiltration for high AERs, since deposition losses are likely minimized by very brief indoor residence times at high AERs. I/O ratios for most elements show less variability for high AER conditions compared to low AER conditions, which is consistent with the idea that penetration efficiencies plateau and that depositional losses are minimized at high AERs.

Figure 2 does not show a consistent relationship between AER and I/O ratios for either K or Si, nor does it show an effect of particle size for these larger crustal elements. This was unexpected because both penetration and deposition theory support reduced ambient infiltration of larger particles such as K or Si for low AER conditions when homes are tight, since their size would limit their penetration through building cracks and crevices and increase their depositional losses (Hinds 1982). Prior studies that measured indoor–outdoor particle size distributions have also reported reduced penetration efficiencies and increased deposition rates for larger particles under low AER conditions (Long et al. 2001; Vette et al. 2001). Given that these elements are known to have significant indoor sources (Koutrakis et al. 1992; Oglesby et al. 2000; Moschandreas et al. 1979; Janssen et al. 1999), it is likely that the relationship with AER has been obscured by indoor source contributions for these elements since I/O ratios less than 1.15 (and even those less than 1) can be biased high due to indoor source contributions. Alternatively, studies have observed bimodal size distributions for the crustal elements like Si and K (Horvath et al. 1996), such that it is possible that a portion of these elements was present in the accumulation region and exhibited similar infiltration behavior to that of Ni and S.

Regression slopes from indoor–outdoor mixed models appear to reflect the size dependence of ambient particle infiltration (Table 3). For example, higher regression slopes of 0.83 and 0.94 were observed for S and Ni, respectively, whereas a substantially lower slope of 0.56 was observed for Fe. Although the number of samples used in the different regressions was variable, slopes for the indoor–outdoor mixed models were highly significant for all elements for unstratified data and for models where data were stratified by AER. Slopes for the larger crustal elements were in general smaller than those for S and Ni, again reflecting the possible role of particle size on ambient infiltration. Slopes for all elements but silicon were higher for high AER conditions compared to low AER conditions, again suggesting increased elemental penetration efficiencies for open, airy homes.

TABLE 3 Summary of indoor–outdoor mixed-model regression and correlation analyses

		All data ^b mixed-model coefficients				Low AER mixed-model coefficients				High AER mixed-model coefficients
Constituent	N	Slope	Intercept	Spearman R	N	Slope	Intercept	Spearman R	N	Slope	Intercept	Spearman R
Sulfur	47	0.83 ^a	−0.10	0.91 ^a	25	0.67	0.09	0.92	22	0.87	−0.09	0.95
Nickel	20	0.94 ^a	−0.001	0.89 ^a	8	0.81	−0.0008	0.83	12	1.13	−0.002	0.92
Zinc	37	0.49 ^a	0.004	0.87 ^a	20	0.45	0.004	0.79	17	0.68	0.001	0.85
Iron	43	0.56 ^a	0.006	0.83 ^a	22	0.44	0.008	0.79	21	0.65	0.005	0.90
Potassium	31	0.64 ^a	0.005	0.87 ^a	14	0.59	0.005	0.91	17	0.69	0.005	0.92
Silicon	32	0.69 ^a	0.006	0.88 ^a	17	0.70	0.003	0.84	15	0.67	0.01	0.94

Table 3 also shows that intercepts from the indoor–outdoor mixed models were small and nonsignificant, suggesting limited indoor source impacts for these six elements. Regression intercepts were also nonsignificant for analyses stratified by both high and low AER conditions. Since several studies have shown that indoor source impacts are most pronounced at low AERs when indoor residence times are higher (Long et al. 2000; Abt et al. 2000a), these findings indicate that indoor particles were primarily of ambient origin following removal of I/O ratios exceeding 1.15.

Although particle infiltration behavior (as reflected in indoor–outdoor ratios and mixed model regression slopes) was variable with AER and between different elements, indoor–outdoor correlations were uniformly strong for each of the six elements, with significant Spearman's correlation coefficients ranging from 0.79 to 0.95. These indoor–outdoor correlations were higher than those observed for the dataset including I/O ratios over 1.15, and thus suggest that these data may have been influenced by indoor source contributions. Correlations remained high for low AER conditions, although they were slightly higher for all elements for high as compared to low AER conditions. Other studies have not observed such uniformly high correlation coefficients for these elements (i.e., Hidy et al. 2000; Janssen et al. 1999), and one likely explanation involves the impacts of indoor source contributions in these studies. Although it is also true that these studies sampled in buildings with characteristics that may differ from those included in this study (i.e., the air-conditioned Birmingham, AL homes in Hidy et al. 2000), the current study included one newer home with a central heating and air-conditioning system. Despite the very low AERs associated with the reliance on this system during both seasons (nighttime means of 0.16 and 0.31 h⁻¹ for the summer and winter sampling periods, respectively), indoor–outdoor correlations for this home were high and significant as well, with values ranging from 0.66 (Si) to 0.97 (S) for sample sizes of 5 to 10.

Utility of Elemental Tracers of Ambient PM

For a PM component to serve as an accurate tracer of ambient fine particles in indoor microenvironments, four criteria must be met: (1) a tracer must be present in sufficient quantities in both indoor and outdoor air; (2) a tracer must be chemically conserved from the outdoor to the indoor environment; (3) a tracer must have insignificant indoor source contributions, such that its presence in indoor environments is primarily due to ambient infiltration; and (4) a tracer must have similar physical behavior (i.e., penetration, deposition, chemistry) to that of other constituents of ambient PM_2.5 (Sarnat et al. 2002). As described earlier, S, Ni, Zn, Fe, K, and Si were not only detected consistently in both indoor and outdoor samples, but they are also known to be conservative (Koutrakis et al. 1992). These as well as previous analyses have indicated that S, Ni, Zn, and Fe are likely to have been least impacted by indoor source contributions in the study homes (Koutrakis et al. 1992; Jones et al. 2000; Özkaynak et al. 1996; Moschandreas et al. 1979), so these four elements best meet the third criterion.

Sarnat et al. (2002) previously developed a methodology for evaluating how well sulfur represents the behavior of PM_2.5 and specific fine particle size fractions. This methodology, which uses mixed models and graphical displays to assess the ability of the I/O ratios for sulfur to estimate corresponding I/O ratios for PM_2.5 and the various particle sizes, was also used in this analysis for Ni, Zn, and Fe. Figure 3 shows both the graphical displays and mixed-model output for regressions of PM_2.5 I/O ratios on elemental I/O ratios. As described previously by Sarnat et al. (2002), regression analyses showed that I/O sulfur ratios were strongly associated with the corresponding I/O PM_2.5 ratios, with an overall slope of 1.02 and mean deviation of 14%.

FIG. 3 Indoor/outdoor PM_2.5 ratios versus elemental I/O ratios. Filled circles represent samples collected during high AER conditions, while open circles represent samples collected during low AER conditions.

Figure 3 also shows that strong associations were observed between I/O PM_2.5 ratios and I/O ratios for Ni, Zn (high AER conditions only), and Fe, indicating that these elements also have utility as predictors of ambient PM_2.5 infiltration. All mixed-model slopes were highly significant (p < 0.0001 for all but the low air exchange model for Zn, where p = 0.007), despite variability in the number of samples. Ni, in particular, showed strong associations with PM_2.5 I/O ratios that were comparable with those involving S, regardless of AER conditions. For Ni, similar regression slopes of 0.95 and 0.94 were obtained for high and low AER conditions, neither of which differed significantly from one.

Figure 3 shows that tracer performance of Zn and Fe was dependent on AER conditions, with both elements overpredicting PM_2.5 I/O ratios at low AER conditions. Both Zn and Fe I/O ratios provided highly accurate predictions of I/O PM_2.5 ratios at high AER conditions, with regression slopes of 0.98 and 1.01, respectively. Mean deviations for high AER conditions were both 19% for Zn and Fe, only slightly higher than the values of 12 and 15% for Ni and S, respectively. Although the predictive ability for Ni was only slightly reduced for low AER conditions, reductions were more substantial for Zn and Fe. A slope of 0.46 was obtained for Zn, possibly reflecting significant indoor source contributions (i.e., cooking) in a few homes that strongly influenced the regression results. For Fe, a lower slope of 0.88 (but not significantly different from one) and a higher mean deviation of 38% were obtained, again possibly reflecting the enhanced impacts of small indoor sources (i.e., resuspension) under low AER conditions. It is possible that indoor source contributions for Fe may have obscured any effect of its large particle size (i.e., mass median diameter of 3.42 μm), since the reduced penetration of Fe in tighter homes compared to PM_2.5 would have been expected to result in a slope exceeding one.

Similar to Sarnat et al. (2002), I/O ratios for Ni, Zn, and Fe were also compared to I/O ratios for 12 particle size intervals to examine how well elemental infiltration factors served as tracers of particles of different sizes. Figure 4 summarizes the mean deviations for these comparisons and provides an indication that Ni may be a better tracer of ultrafine (< 0.1 μm) particles than S. Ni was shown to predict the infiltration for particles with diameters between 0.04 and 0.5 μm with the greatest accuracy, and for other particle sizes mean deviations for Ni were similar to those of S (Figure 4). For particles between 0.04 and 0.5 μm, mixed model regressions between 0.88 and 0.95 were observed for the nickel I/O regressions (Figure 5). Similar to previous findings for S (Sarnat et al. 2002), Figure 5 shows that Ni I/O ratios overpredicted I/O ratios for very small (< 0.04 μm) and very large particles (> 0.7 μm). This finding is consistent with particle penetration and deposition theory, because both very small ultrafine particles and coarse-mode particles would be expected to have greater penetration and deposition losses than Ni, which is associated with accumulation-mode particles. AER was found to have little impact on the strength of the associations between I/O nickel ratios and the I/O ratios for the particle sizes between approximately 0.04 and 0.5 μm (data not shown).

FIG. 4 Mean deviations between elemental I/O ratios and I/O ratios for discrete particle size intervals.

FIG. 5 I/O size-resolved data versus I/O nickel. Filled circles represent samples collected during high AER conditions, while open circles represent samples collected during low AER conditions. N = 20 for all plots. (Continued)

As shown in Figure 4, mean deviations for Zn and Fe were generally higher than those of S, indicating that for most particle sizes they served as less-accurate tracers. Given that Fe has been reported to have a mass median diameter of 3.42 μm (Milford and Davidson 1985), it was surprising that this element was not a better predictor than S or Ni for larger particle sizes. Fe I/O ratios were typically greater than I/O ratios for particle sizes greater than 0.7 μm, possibly reflecting the impacts of indoor Fe source contributions or the association of ambient Fe in these samples with accumulation-mode particles rather than coarse particles.

CONCLUSIONS AND IMPLICATIONS

Through the use of nighttime data collected during periods of limited human activity, this analysis observed consistently high indoor–outdoor correlations for a number of elemental species including S, Ni, Zn, Fe, K, and Si (Spearman R > 0.8). These correlations remained strong when data were stratified by high and low AER conditions, and by season (Spearman R > 0.7). Although the study included a small number of homes and samples, these findings support the validity of the use of residential outdoor data, and perhaps central site monitor data, as surrogates of residential indoor elemental concentrations in epidemiological investigations of PM_2.5 elemental constituents. Correlations remained high for different AER conditions, building characteristics, and seasons, suggesting that variability in ambient elemental infiltration into residences may not be a large source of variability affecting personal-ambient correlations for these elements. Spatial homogeneity of locally generated elements such as traffic-related pollutants or soil-derived elements could still be a source of bias for personal–ambient correlations (Oglesby et al. 2000), but these findings indicate that for elements dominated by regional transport (i.e., S and Ni), elemental concentrations from central site monitors may be acceptable surrogates of personal exposure for epidemiological studies. Notably, in one of relatively few epidemiological studies of PM_2.5 elemental constituents, Laden et al. (2000) observed significant associations with mortality for both sulfur and nickel.

Elemental infiltration factors showed strong relationships with AER and season, and were suggestive of an effect of particle size, which was likely obscured by remaining indoor source impacts. Notably, the results highlight the role of building tightness and indoor ventilation in impacting elemental infiltration factors, as infiltration factors were substantially reduced for low AER conditions. Additional data are needed to confirm whether these data are representative of other residential locations and housing types. However, based on these and previous findings (Janssen et al. 2002; Suh et al. 1994; Long et al. 2001), indoor air exchange rate should continue to be treated as a critical exposure factor that may potentially modify health effect estimates reported in PM epidemiological studies.

Although previous studies have demonstrated that ambient particle infiltration is strongly dependent on particle size (Abt et al. 2000; Long et al. 2001; Vette et al. 2001), a consistent relationship between particle size and elemental infiltration behavior was not observed, possibly due to indoor source contributions or because most elemental mass was found in similarly sized particles. Additional research is thus necessary to characterize better the variability of elemental infiltration factors, both as a function of ambient particle size distribution and building characteristics, especially in light of ongoing efforts to develop population exposure models for PM components, such as a refined version of the USEPA Stochastic Human Exposure and Dose Simulation (SHEDS-PM) model (USEPA 2002).

A striking finding of this study was the ability of Ni to serve as a strong tracer for total outdoor PM_2.5, as well as particles within discrete size ranges, in the study homes. Indeed, for this dataset, Ni served as a more accurate surrogate for some of the smaller particle size intervals (0.04–0.5 μm) than S. This is an important finding because there is a clear need for other tracers of ambient particles in addition to sulfur, since it is difficult for a single elemental tracer to represent accurately the entire PM_2.5 mixture due to heterogeneity with respect to source contributions, chemical composition, and size distribution. However, there remains uncertainty regarding the utility of Ni as a tracer in other regions and seasons because there may be differences in the predominant sources of Ni and their size distributions both by region and season (e.g., large Ni emissions from coal and oil combustion in the Northeast, with greater importance of natural gas combustion in other regions like Southern California). Also, because these data did not show elements with larger reported diameters (i.e., iron) to be more accurate tracers of larger particle sizes, additional research is also necessary to explore this finding, possibly through incorporation of elemental size distribution measurements in the study design, and to identify suitable tracers for coarse particles.

Acknowledgments

This article is based upon a field study conducted as part of Christopher Long's doctoral thesis in the Department of Environmental Health at the Harvard School of Public Health. The authors would like to thank Petros Koutrakis and Helen Suh for their generous mentoring and sage guidance. The authors would also like to express their sincere gratitude to all of the study participants as well as to George Allen, Jim Sullivan, Mark Davey, Denise Belliveau, and Jessica Sekula for their invaluable assistance during field and laboratory work. The field component of this study was funded by the Center for Indoor Air Research (CIAR) under contract #96-08A, with support also from the U.S. EPA STAR Graduate Fellowship Program.

Notes

^a Indicates significance at the 0.05 level (p < 0.05).

^b As described in the text, I/O ratios > 1.15 have been removed.