Comparison of stationary and personal air sampling with an air dispersion model for children’s ambient exposure to manganese

Abstract

Manganese (Mn) is ubiquitous in the environment and essential for normal growth and development, yet excessive exposure can lead to impairments in neurological function. This study modeled ambient Mn concentrations as an alternative to stationary and personal air sampling to assess exposure for children enrolled in the Communities Actively Researching Exposure Study in Marietta, OH. Ambient air Mn concentration values were modeled using US Environmental Protection Agency’s Air Dispersion Model AERMOD based on emissions from the ferromanganese refinery located in Marietta. Modeled Mn concentrations were compared with Mn concentrations from a nearby stationary air monitor. The Index of Agreement for modeled versus monitored data was 0.34 (48 h levels) and 0.79 (monthly levels). Fractional bias was 0.026 for 48 h levels and - 0.019 for monthly levels. The ratio of modeled ambient air Mn to measured ambient air Mn at the annual time scale was 0.94. Modeled values were also time matched to personal air samples for 19 children. The modeled values explained a greater degree of variability in personal exposures compared with time-weighted distance from the emission source. Based on these results modeled Mn concentrations provided a suitable approach for assessing airborne Mn exposure in this cohort.

INTRODUCTION

Manganese (Mn) is an essential dietary nutrient and in normal amounts supports healthy functioning and development. Excessive exposure to Mn is associated with central nervous system health effects.¹ Potential environmental sources of Mn include emissions from metallurgic processing and mining operations, contamination of soil and water through agricultural practices, toxic waste sites, and air emissions from the use of the gasoline additive methylcyclopentadienyl manganese tricarbonyl.^2,3 The predominant routes of Mn exposure are ingestion and inhalation. Levels of ingested Mn are regulated by homeostatic mechanisms in the body and few cases of toxicity by ingestion have been reported.^4,5 Several studies have highlighted concerns associated with inhalation of Mn including rapid uptake in the lungs and more direct access to the brain.^6–8

For those living in close proximity to manganese-ore and alloy production activities, there is concern that chronic exposure to ambient air Mn, even at low levels, may lead to significant impacts to motor, cognitive, or behavioral functions. Both occupational and environmental Mn exposure studies demonstrate an inverted U-shaped continuum of dysfunction where both low and high levels of exposure result in neurological deficits.^9–11 Several children’s studies have assessed the potential links between exposure to ambient air Mn and neurobehavioral and cognitive effects where the primary sources of exposure were ferroalloy mining and processing activities.^12–18 Multiple studies found significant association between the level of Mn in hair and measures of neurobehavioral function^{11,12,14,16,17} and motor/ sensory function.^13,15,18

What is lacking is a clear understanding of the environmental levels, the duration of exposures, and the pathways of exposure that pose the greatest risk to children. Frequently the reported measure for exposure has been proximity to the Mn source such as zone of residence, rural versus mining,¹² and distance from a ferromanganese alloy production plant.^16,18 Levels of Mn in ambient air were measured in four studies at a single location^13,14,17,19 and ranged from a single sampling event¹⁴ to three times per week over the course of 24 months.¹⁹ Personal air samples were collected in two studies: 48 h samples with the fine particulate matter (PM_2.5) fraction analyzed for Mn¹⁹ and 24 h samples with the PM₁₀ fraction analyzed for Mn, although only time spent at home was considered.¹⁵

Sampling for ambient levels and personal exposures may provide direct measures of Mn exposure, but are time-consuming and expensive. Additionally, sampling only captures levels of exposure associated with the location and environmental conditions during the time of sampling, a limitation when the overarching concern is cumulative exposure. Proximity measures used as surrogates for Mn exposure are inexpensive and easy to obtain, but are limited in their ability to capture other factors that impact ambient air Mn concentrations such as wind direction, precipitation and terrain. Air dispersion modeling may provide a viable alternative to ambient Mn exposure assessment.

Stationary Air Sampling

Air dispersion models are widely used to estimate exposure concentrations for criteria air pollutants, and in some cases for hazardous air pollutants (HAPs). In 2005, The United States Environmental Protection Agency (US EPA) promulgated the AMS/EPA Regulatory Model (AERMOD)²⁰ as the air dispersion model for predicting fate and transport of air emissions from point sources for use in regulatory requirements, environmental and health standards, and facility design criteria. AERMOD is a steady-state plume model that incorporates meteorological data and terrain data as well as point source emissions data. Evaluations of the performance of AERMOD relative to other air dispersion estimation techniques indicted that AERMOD provides location-specific exposure risk estimates with higher accuracy, simulates pollutant dispersion at or under 50 km accurately, and tends to generate lower concentration estimates for point sources at the point of maximum air concentration.^21–23

The Communities Actively Researching Exposure Study (CARES) in Marietta, OH provided an opportunity to gain a greater understanding of the link between concentrations of ambient air Mn and children’s exposure.²⁴ Marietta is home to the longest operating ferromanganese refinery in the nation, Eramet Marietta Incorporated (EMI). CARES is one of the first studies to characterize children’s personal exposure to airborne Mn. Although AERMOD has demonstrated acceptable performance, the CARES study offered an opportunity to compare modeled ambient air Mn concentrations as an alternative to both stationary and personal air sampling, as well as surrogate exposure measures such as time-weighted distance (TWD) from the primary emission source.

MATERIALS AND METHODS

Study Population

Participants enrolled in CARES were children aged 7 to 9 years, recruited between October 2008 to March 2013 using a volunteer sampling strategy of letters through schools and local radio and newspaper advertisement.¹¹ For a subset of children in 2009 and 2010, personal air samples were collected for a 48-h period.¹⁹ All participants participating in personal air sampling reported living in a non-smoking residence. Data for the current analysis were obtained from participants enrolled and sampled in 2009. The University of Cincinnati Institutional Review Board approved this study. All parents signed an informed consent and the children signed an informed assent.

Personal Air Sampling

Personal air samples were collected for a 48 ± 2-h period using a two-stage personal modular impactor capable of sampling PM_2.5, PM_2.5–10, orboth.¹⁹ Participants wore backpacks equipped with an air pump, battery packs, and a Tygon tube connecting the air sampling pump to the personal modular impactor sampler for 48 h. Pumps were calibrated to 3 LPM (liters per minute) before each sampling period. Airborne particles were collected on a 37 mm preweighed Teflon membrane filter with a 2 μm pore size. Ten percent of the samples were laboratory blanks and another 10% were field blanks to protect the validity of the sample results. Families were instructed to keep the air sampler in or near the child’s breathing zone as much as possible for the entire sampling period; for example, during sleep the equipment could be laid on a nightstand. Parents maintained a daily log during the sampling period that recorded activities and location of the child, including details such as time spent at home, outdoors, in vehicles, at school, and outside the Marietta area. Location, but not activity level, was used in the statistical analyses. Following the sampling period, a technician visited the home and conducted a brief questionnaire to determine the child’s potential exposure to particle sources during the sampling period. Questions included the child’s exposure to cigarette smoke, use of heating and cooking devices, candle use, and time the sampling equipment was not with the child or outside of the Marietta area. Personal air samples collected from 22 April 2009 through 24 May 2009 were used in this subanalysis.

A stationary PM_2.5 air sampler measured the levels of ambient air Mn in Marietta. Given the purpose of this modeling was to provide exposure estimates for use in health studies, PM was considered appropriate. Particles of this size are capable of reaching the small airways and are associated with increased health effects. Air samples were collected using a Harvard-type PM_2.5 impactor with a high volume pump calibrated to 10 LPM before each sampling period.¹⁹ Particles < 2.5 μm were collected on the same Teflon filters used in the personal sampling. Ten percent of the samples were laboratory blanks and another 10% were field blanks to protect the validity of the sample results. The sampler was positioned on the rooftop of the Rickey Science Center at Marietta College, ~ 8 km from EMI (Figure 1). The building is centrally located within the highest population density of the study area and thus represents an exposure scenario relevant to a number of study participants.¹⁹ The height of the building plus the height of the sampler combined was 17 m from the base elevation. Samples were collected for 48 h on the average three times per week from October 2008 to September 2010. Stationary air samples collected from 1 January 2009 through 31 December 2009 were included in the current analysis.

Figure 1.

Contour Map for AERMOD (AMS/Environmental Protection Agency Regulatory Model) annual ambient Mn concentrations (ng/m³) for study area. Note: It is assumed that modeled Mn is in the fine particulate matter (PM_2.5) fraction.

Analysis of Sampling Filters

Teflon filters were analyzed by a commercial laboratory (Research Triangle Institute, Research Triangle Park, NC, USA).¹⁹ Samples were analyzed for Mn in the PM_2.5 fraction using a Thermo X Series II inductively coupled plasma mass spectrometer and for particle mass gravimetrically. Particle mass results were reported in μg/m³ and Mn in ng/m³. Limits of detection were 4.1 μg and 2.5 ng per sample, respectively, for PM_2.5 mass and Mn. 2.5

Ambient Air Modeling

Annual Mn emissions data, reported in pounds per year for EMI in 2009 were obtained from The US EPA’s Emissions Inventory System (EIS).²⁵ Emissions for five stack release points and one fugitive release were included (Table 1). The Ohio Environmental Protection Agency’s Emissions Inventory System²⁶ was accessed to obtain facility operating information and indicated that EMI operated 24 h per day 7 days per week at a consistent throughput across all seasons and a total operating time for each of the five stack emissions of 8181 h per year. However, assuming the processes (e.g., furnaces) are typically shutdown a few days per year for maintenance, the 8181h per year was entered as 341 days in the model. Only stack emissions were entered in the resulting model. Information reported for fugitive emissions was insufficient for appropriate model inputs. Fugitive emissions can be important for exposure modeling assessments, especially for locations in close vicinity of the sources; however, as fugitive emissions are not released through a confined controlled airstream (i.e. stack with baghouse or scrubber),²⁷ these emissions are expected to be generally of a larger size fraction aerodynamically than stack emissions and may have less buoyancy.²⁸ These coarser fugitive emissions would likely have less impact on the sampling in this study, which used a PM_2.5 stationary sampler located 8 km from EMI, and personal samplers located an average of 10.6 km from EMI.

Table 1.

2009 US EPA EIS reported emissions for Eramet Marietta

Release point ID	Release type	Stack height (ft)	Stack diameter (ft)	Fugitive height (ft)	Exit gas Row rate (ACFM)	Exist gas temp (F0)	Total emissions (lbs)
68501712	Stack	90	1.6		8280	105	9827
68503212	Stack	110	4		30,000	90	7758
68502412	Stack	90	5		29,000	90	7758
68500812	Stack	110	4		34,000	90	7758
68502012	Stack	90	5		11,000	90	7758
85855212	Fugitive			110			108,612

Abbreviations: EIS, Emission Inventory System; EPA, Environmental Protection Agency.

Surface meteorological data for 2009 and Automated Surface Observing System 1-min data from the nearest airport to the study area, Parkersburg, WV, and upper air data files for Wilmington, OH were downloaded from National Oceanic and Atmospheric Agency Climate Data Center websites.^29–31 The Automated Surface Observing System 1-min data were processed through AERMINUTE to calculate hourly average wind speed and directions.³² Local land-use and terrain data required by AERMOD were obtained from the Multi-Resolution Land Characteristics Consortium website.³³ National Land Cover Data for 1992 was preprocessed with AERSURFACE to obtain values for surface albedo and Bowen ratio based on land-use characteristics and surface roughness.³⁴ Outputs from AERSURFACE and AERMINUTE, along with local meteorological data, were inputs to AERMET, the meteorological preprocessor for AERMOD.³⁵ Resolution of the terrain data was 1/3 s National Elevation Data and was analyzed with AERMAP³⁶ to determine ground level elevations for the sampling domain including elevations for the stationary monitor and residences of study participants as well as local schools (n = 12) the participants attended (Figure 1). The final AERMOD model was executed with the dispersion option as the regulatory default, the dispersion coefficient as rural, the output type as concentration, the pollutant type as Mn, the terrain height options as elevated, and a flagpole receptor height only for the stationary air monitor. All other receptors were modeled at ground level elevation. The option to account for building downwash was excluded owing to the lack of information on building dimensions associated with each stack. However, given that the height of each stack was >90ft, there should be little to no effect on modeling results. The final AERMOD model and all preprocessing, except AERMINUTE, were executed with the integrated interface AERMOD View Version 7.6 from Lakes Environmental Software.³⁷ Output from AERMOD included 24 h and monthly ambient Mn μg/m³ concentration levels modeled for the stationary air monitor and each school and study participant residence.

Proximity Measure

A TWD measure was used within subanalyses for the CARES study.^18,19 It is calculated as a weighted measure of the distance of the child’s residence and school from the ferromanganese refinery using estimates of time spent at home and school. For comparison with personal air Mn levels, the amount of time at home and school were derived from daily activity logs.

Statistical Analysis

All statistical analyses were performed using SAS 9.4 (SAS Institute, Cary, NC, USA). Summary statistics including mean, standard deviation, median and range were calculated. Normality of variables was evaluated with the Shapiro–Wilk test and histograms. Natural log transformation (ln) was used to achieve normality as necessary. Agreement of modeled to measured levels of Mn was assessed with multiple statistics including Fractional Bias (F_B) and Index of Agreement (IOA). Regression analysis was used to evaluate the relationship between personal air Mn and modeled Mn levels.

Modeled versus Personal Air Mn

For participants with complete 2009 personal air sampling data (n = 19), a modeled level of ambient air Mn was calculated based on the dates of personal monitoring. A time-weighted 48 h ambient Mn concentration was determined for each study participant. Simple linear regression between personal air Mn and modeled ambient air Mn was used to assess the amount of variability in personal air Mn exposure attributable to ambient air Mn from outdoor sources.²¹ In addition, a regression between personal air Mn versus TWD was used to obtain the amount of variability in personal air Mn exposure that could be explained by proximity to EMI.

Modeled versus Measured Mn

Ambient air Mn concentrations modeled by AERMOD at the same location and height as the stationary air monitor were compared with measured Mn concentrations from stationary air monitor samples. Temporal scales of 48 h, monthly and yearly were included in the assessment. For comparison of Mn levels based on 48 h, modeled 24 h Mn concentrations were time matched to dates of 48 h stationary air samples and averaged to 48 h levels (n = 132). The calculated monthly Mn levels from the stationary air samples were compared with the modeled monthly Mn concentrations from AERMOD. Annual values were compared based on the average of the monthly Mn concentrations produced by AERMOD and the average of stationary air samples across all sample dates in 2009. Statistical approaches used to compare the modeled and measured ambient Mn exposure levels (across time scales) included plots of modeled versus measured Mn concentrations, the normalized mean square error, F_B and IOA.^23,38

Modeled Exposure Concentrations for Study Participants

For each participant enrolled in 2009 (n = 88), ambient air Mn exposure concentrations were calculated based on the modeled annual air Mn concentrations spatially allocated to geographic locations associated with schools and homes of participants and the amount of time participants spent at home and at school. Output from AERMOD included monthly ambient Mn concentration levels for each school and participant residence; annual ambient Mn concentrations were calculated by averaging the monthly concentrations. A time-weighted annual ambient Mn concentration was determined for each study participant with time at home and school assumed to be 70% and 30%, respectively, based on completed daily logs from CARES.¹⁸

RESULTS

Modeled versus Personal Air Mn

Descriptive statistics for participants with complete personal air sampling data from 2009 (n = 19) are detailed in Table 2. Personal air and 48 h modeled air Mn was 11.11 and 27.88 ng/m³, respectively. The relationship between personal air Mn and 48 h modeled Mn concentrations was examined (Table 3 and Figures 2 and 3). Initially, the 48 h modeled Mn concentration was not significantly related to personal air Mn levels. Examination of the regression plots, studentized residuals and Cook’s D indicated the presence of an outlier. The data point had a personal air Mn level of 42.35 ng/m³ and a 48 h modeled ambient Mn level of 1.156 ng/m³ and was in close proximity to the primary emission source (TWD = 4.6 km). When compared with modeled levels at a similar distance, the value for the 48-h time frame is low (1.56.8 and 135.6 versus 1.2 ng/m³), indicating that the model may have underestimated the level given the combination of close proximity and short time frame. The regression model was rerun without the outlier and the R² increased from 0.1030 to 0.4196 and the level of 48 h modeled ambient air Mn was found to be positively associated with the level of personal air Mn (P = 0.0037). A post hoc robust regression run on the same data resulted in the same parameter estimates for the regression of log personal air Mn with log 48 h modeled Mn.

Table 2.

Summary statistics for 2009 personal air Mn samples in Marietta, OH and time-matched AERMOD modeled ambient air Mn concentrations.

Variable	n	Mean	SD	Median	Range
Age (years)	19	8.6	0.9	8.8	7.0–9.8
Time-weighted distancea (km)	19	10.3	3.7	9.9	4.6–16.3
Personal air Mn (ng/m3)	19	11.1	13.3	6.5	1.7–52.0
48 h Modeled air Mn (ng/m3)	19	27.9	46.8	8.8	1.2–156.8
Log personal air Mn (ng/m3)	19	2.0	0.8	1.9	0.5–4.0
Log 48 h modeled air Mn (ng/m3)	19	2.3	1.4	2.2	0.1–5.1

Abbreviations: AERMOD, AMS/EPA Regulatory Model; PM_2.5, fine particulate matter. Note: Measured Mn is in the PM_2.5 fraction and comparisons arepartly based on the assumption that modeled Mn is also in the PM₂₅ fraction.

^aWeighted measure of the distance of the child’s residence and school from the ferromanganese refinery using estimates of time spent athome and school.

Table 3.

Coefficients and model fit statistics for simple linear regressions of 2009 log personal air Mn samples in Marietta, OH and time-matched AERMOD modeled ambient air Mn concentrations.

	n	Estimate	SE	t-value	p-value	R2
Intercept	19	1.540	0.382	4.03	0.0009	0.103
Log 48 h modeled Mn		0.199	0.142	1.40	0.1804
Intercepta	18	0.981	0.306	3.21	0.0055	0.4196
Log 48 h modeled Mna		0.377	0.111	3.40	0.0037
Intercept	19	3.136	0.527	5.95	<0.0001	0.2331
Time-weighted distanceb		−0.109	0.048	−2.27	0.036

Abbreviations: AERMOD, AMS/EPA Regulatory Model; PM_2.5, fine particulate matter. Note: Measured Mn is in the PM_2.5 fraction and comparisons are partly based on the assumption that modeled Mn is also in the PM_2.5 fraction.

^aAfter removal of outlier.

^bWeighted measure of the distance of the child’s residence and school from the ferromanganese refinery using estimates of time spent at home and school.

Figure 2.

Regression plot of log personal air Mn with log AERMOD (AMS/Environmental Protection Agency Regulatory Model) 48 h modeled air Mn (n= 19) in Marietta, OH (2009). Note: Measured Mn is in the fine particulate matter (PM_2.5) fraction and comparisons are partly based on the assumption that modeled Mn is also in the PM_2.5 fraction.

Figure 3.

Regression plot of log personal air Mn with log AERMOD (AMS/Environmental Protection Agency Regulatory Model) 48 h modeled air Mn (n= 18) in Marietta, OH (2009). Note: Measured Mn is in the fine particulate matter (PM_2.5) fraction and comparisons are partly based on the assumption that modeled Mn is also in the PM_2.5 fraction.

A regression between personal air Mn and TWD was used to assess the amount of variability in personal air Mn exposure that could be explained by proximity to EMI. The R² was 0.2331 and TWD was determined to be negatively associated with personal air Mn levels (P = 0.036) (Table 3 and Figure 4). No outliers were detected in the analysis between personal air Mn and TWD.

Figure 4.

Regression plot of log personal air Mn samples (n = 19) with time-weighted distance in Marietta, OH (2009). Time-weighted distance= weighted measure of the distance of the child’s residence and school from the ferromanganese refinery using estimates of time spent at home and school. Note: Measured Mn is in the fine particulate matter (PM_2.5) fraction.

Modeled versus Measured Mn

Sample statistics were produced for modeled and measured Mn concentrations for each time scale: 48 h (n = 132), monthly (n = 12), and annual (Table 4). All measures of agreement between measured and modeled Mn concentrations are based partly on the assumption that all modeled emissions are in the PM_2.5 fraction. The normalized mean square error indicates a substantial improvement in model performance for the monthly time scale compared with the 48 h time scale. The IOA indicates a doubling in the agreement between measured and modeled Mn concentrations from the 48 h to the monthly temporal scale (0.3437 versus 0.7894). The model bias, represented by F_B, is lower for the monthly time scale compared with 48 h. However, the direction of the bias changes; a positive bias or overestimation by the model for 48 h levels (F_B = −0.0256) to a negative or underestimation by the model for monthly levels (F_B =- 0.0194). The ratio of modeled ambient air Mn to measured ambient air Mn at the annual time scale was 0.94, indicating a high level of agreement; however, there was only a single time point at the annual scale.

Table 4.

Summary and evaluation statistics for comparison of stationary air monitor Mn concentrations in Marietta, OH to time-matched AERMOD modeled ambient air Mn concentrations.

Time scale	Approach	n	Mean (ng/m3)	SD (ng/m3)	Median (ng/m3)	Range (ng/m3)	NMSE	FB	IOA
48 h	Measured	132	16.31	18.81	9.54	0.04–102.99	2.169	0.0256	0.3437
	Modeled	132	16.74	26.74	4.89	0.13–171.84
Monthly	Measured	12	15.62	6.97	17.7	7.50–28.31	0.1694	−0.0194	0.7894
	Modeled	12	15.32	8.63	13.43	5.54–33.48
Annual	Measured	1	16.31
	Modeled	1	15.32

Abbreviations: F_B, fractional bias; IOA, index of agreement; NMSE, normalized mean square error; PM_2.5, fine particulate matter. Note: Measured Mn is in thePM_2.5 fraction and comparisons are partly based on the assumption that modeled Mn is also in the PM_2.5 fraction.

Plots of the modeled to measured Mn concentrations for both 48 h and monthly time scales provide a visual tool to evaluate model performance (Figure 5). Reference lines of 2:1 and 1:2 bracket the factor of 2 acceptance criterion of model performance recommended by EPA.³⁹ For the 48 h time scale, 42% of observations fall within the range and for the monthly time scale, 100% of observations are within the recommended range for acceptable model performance. Comparison at the annual time scale is limited as only data from 2009 were used. A plot of monthly means for measured and modeled Mn concentrations indicates higher modeled Mn concentrations in June and July and lower modeled values in October and November relative to measured levels in the same months (Figure 6).

Figure 5.

Scatterplot of AERMOD (AMS/Environmental Protection Agency Regulatory Model) modeled ambient air Mn (ng/m³) versus measured ambient air Mn (ng/m³) in Marietta, OH (2009). (a) AERMOD 48 h modeled ambient air Mn (ng/m³) versus 48 h measured ambient air Mn (ng/m³). (b) AERMOD monthly modeled ambient air Mn (ng/m³) versus monthly measured ambient air Mn (ng/m³). Reference lines of 2:1 and 1:2 bracket the factor of 2 acceptance criterion (over- or underprediction) of model performance recommended by Environmental Protection Agency (EPA); line representing 1:1 agreement between measured and modeled also shown. Note: Measured Mn is in the fine particulate matter (PM_2.5) fraction and comparisons are partly based on the assumption that modeled Mn is also in the PM_2.5 fraction.

Figure 6.

Plot of AERMOD (AMS/Environmental Protection Agency Regulatory Model) monthly average modeled ambient air Mn (ng/m³) and monthly measured ambient air Mn (ng/m³) in Marietta, OH (2009). Note: Measured Mn is in the fine particulate matter (PM_2.5) fraction and comparisons are partly based on the assump tion that modeled Mn is also in the PM_2.5 fraction.

Modeled Exposure Concentrations for Study Participants Modeled ambient air Mn concentrations were obtained for the schools (n = 12) and residences (n = 88) of CARES participants enrolled in 2009. From the AERMOD output, a contour map of annual ambient air Mn concentrations, including the locations of the stationary air monitor, EMI, and the residences and schools of study participants, was generated (Figure 1). Meteorological data from 2009 showed that wind typically blew to the North and East. From the contour map, it is clear that ambient air Mn concentrations varied across a gradient of distance from EMI, with concentrations generally decreasing with increased distance. Residences at similar distances from the plant varied in their Mn exposure potentials because of other geographic and meteorological conditions. Including participant’s time at school via TWD did not greatly modify the mean distance from EMI (10.9–10.6 km) but did change the mean exposure more substantially (27.0–33.9 ng/m³) (Table 5). Modeled Mn levels for residences and schools were combined to obtain a time-weighted annual ambient air Mn concentration for CARES participants (Table 5). The annual ambient Mn exposure levels ranged from 5.8 to 159.9 ng/m³. The cumulative distribution of modeled annual ambient air Mn concentrations demonstrated that 26% of the values are above 50 ng/m³, the reference concentration for chronic exposure to inhaled manganese set by EPA.⁴⁰

Table 5.

Summary statistics of exposure measures for study participants (n = 88).

Variable	n	Mean	SD	Median	Range
Distance of residence from plant (km)	88	10.9	3.6	10.8	3.4–18.4
Time-weighted distancea (km)	88	10.6	3.5	11.0	3.5–17.7
AERMOD modeled annual Mn residence (ng/m3)	88	27.0	23.9	16.8	5.8–141.9
AERMOD modeled annual Mn, residence and school (ng/m3)	88	33.9	30.3	19.3	5.8–159.9
AERMOD modeled annual Mn Schools (ng/m3)	12	31.1	54.1	16.3	5.6–201.7

Abbreviations: AERMOD, AMS/EPA Regulatory Model; PM_2.5, fine particulate matter. Note: It is assumed that modeled Mn is in the PM_2.5 fraction.

^aWeighted measure of the distance of the child’s residence and school from the ferromanganese refinery using estimates of time spent at home and school.

DISCUSSION

This study investigated modeled ambient air Mn concentrations produced by US EPA’s air dispersion model, AERMOD, as an approach to estimate ambient air Mn exposure concentrations for children living near a ferromanganese refinery in Marietta, OH. To our knowledge, this is the first model of children’s exposure to ambient air Mn that accounts for time spent at home and at school. The results of the study indicated that modeled monthly and yearly ambient air Mn concentrations were in close agreement with Mn measured in the PM_2.5 obtained from a stationary monitor 8 km from EMI (Table 4). In addition, time-matched 48 h modeled ambient air Mn concentrations were significantly associated with personal air Mn concentrations and explained more than 40% of the variability in personal Mn exposures (R² of 0.42 versus 0.23; Table 3). It is important to note that the modeled concentrations were based solely on outdoor measurements while the personal samples included time spent indoors, presumably with decreased exposure levels. This likely contributed to the lower mean personal samples in comparison with modeled values. The spatial distribution of annual ambient air Mn levels generally decreased with increased distance from EMI. Collectively, these results suggest that ambient air Mn concentrations obtained from AERMOD provide a suitable approach for assessing airborne Mn exposure for children residing in Marietta, OH.

Comparison of the modeled ambient air Mn concentrations with measured Mn concentrations from the stationary monitor showed an improvement in model performance from 48 h to monthly time periods (Table 4). This mirrors the findings of a study assessing the performance of AERMOD in estimating sulfur dioxide concentrations from multiple emission sources relative to levels measured at three stationary monitors at 1 h, 3 h, 8 h, monthly and annual intervals.²³ The authors concluded that concentrations produced from AERMOD at longer time scales were more reliable and could provide useful exposure estimates in epidemiological studies where chronic exposure is the primary focus.²³

EPA’s monitoring of air pollutants at 63 schools in 22 states included Warren Elementary School, the school in closest proximty to EMI (3.8 km) and attended by 15 CARES participants.⁴¹ Air samples were collected by the EPA at Warren Elementary School on 14 days from 17 August 2009 to 3 November 2009 with a PM₁₀ sampler, with a reported average level of ambient air Mn of 146 ng/m³.⁴¹ In the current study, the school had the highest modeled ambient air Mn concentration at 201.7 ng/m³ (Table 5), a comparable figure to the EPA measured level.

The current study findings are consistent with a previous study by Haynes et al.,¹⁹ which demonstrated a significant association between personal and stationary Mn concentrations in the CARES population, and a negative association with TWD from the primary emission source. The modeled approach, however, has the advantage of considering factors across space and time, which increases its usefulness in assessing cumulative exposures.

A pilot study of Marietta residents identified an association between biomarkers of Mn exposure and AERMOD estimated Mn air exposure levels for the study region for 2006.⁴² Emissions from EMI and eight other Mn emission sources within a 20 mile radius of Marietta were used as model inputs. EMI emissions accounted for 73% of the total on-site and fugitive emissions. AERMOD was run using a receptor grid and study participanťs residences were located on the grid after processing, in contrast to the current study where participant’s homes and schools were the model receptors. Within a 15 mile area surrounding Marietta, modeled annual mean ambient air Mn levels were reported as 130 ng/m³, ranging from 10 to 18,130 ng/m³.⁴²

In a study of adult residents of Marietta, Bowler et al.⁴³ found an association between airborne Mn exposure and neurological end points using Mn concentration outputs from AERMOD in the exposure assessment. Model inputs included Mn air emission data from 2001. Details regarding the number of Mn emission sources and the magnitude of the Mn emissions were not provided. The model was executed using a 10 m×10 m receptor grid with concentrations modeled to the center of each grid. Residences of study participants were assigned the modeled air Mn concentrations associated with the grid where their residence was located. Modeled estimates for ambient air Mn within 8 km of EMI ranged from 40 to 960 ng/m³ with a mean level of 180 ng/m³. This research group also estimated annual PM concentration in Marietta residences using AERMOD and long-term data (2003–2013) from five sampling stations.⁴⁴ Emissions data from EMI were not considered. The modeled PM mean was 45 ng/m³ with a range of 7–340 ng/m³.⁴⁴

The reported levels of modeled ambient air Mn using AERMOD for 2006,⁴² 2001,⁴³ and 2003–2013⁴⁴ data were higher compared with the levels reported in the current study. This difference is likely attributable to variations in model construction and declines in emissions from EMI over the time period in question. The EPA Toxic Release Inventory reports 2009 emissions from EMI as less than half that of 2006 and a third of 2001 emissions.⁴⁵ The EPA also reported that in 2009 EMI was operating below its average operating level for the past 5 years.⁴¹

There are limitations to using air dispersion models for exposure assessment. Uncertainty can be found in model inputs, including data for level of emissions, facility operating characteristics, terrain, and meteorology. All emissions data in EIS are self-reported and the annual emissions data are estimated.²⁶ Furthermore, the reported emissions in the EIS are for total manganese, which includes a wide range of particle sizes. However, the modeling results are compared only with the monitored results of Mn in the PM_2.5 fraction and this comparison is based partly on the assumption that modeled Mn emissions are also in the PM2.5 fraction. EIS also did not include sufficient data on fugitive emissions, which accounted for 73% of the total emissions in 2009, for inclusion in the model.²⁶ This limitation is most relevant to participants living or attending school in close proximity to EMI, resulting in the possible underestimation of their Mn exposures. However, a PM_2.5 sampler was used, reducing the impact of these generally coarser fugitive emissions on these measured values.

There is also sparse information available about the actual operating characteristics of the facility, such as if stack emissions are stopped for maintenance or if the facility is closed throughout the year. The overestimation of the model for the months of June and July may be explained by suspension or decrease of plant operations during the summer months. Another uncertainty is related to quantifying the confidence in model estimates. Currently, there is no viable approach to obtaining variability limits for modeled concentrations. The current state of practice primarily relies on evaluations of model performance such as the one presented in this study. Despite these uncertainties, air dispersion models offer advantages over stationary and personal sampling as well as proximity measures, each with limitations of their own. While the participants were well-dispersed geographically and likely representative of children in the area, the volunteer sampling strategy may limit generalizability.

Other potential sources of Mn exposure include soil and dust. Lucchini et al.¹⁵ found levels of soil Mn from the homes of children living in the mining areas were significantly higher compared with the soil from homes in reference communities and were associated with neurological outcomes of motor coordination and tremor intensity. In studies of children’s exposure to lead, soil and house dust have been identified as significant contributing factors for childhood lead exposure.^46–50 Dust has been identified by US EPA as a significant exposure pathway for children.⁵¹ Results from two studies in the mining district of Molango, Mexico emphasize the need to look at factors in the home, as well as the community, to fully understand Mn exposure.^52,53 Indoor and outdoor Mn concentrations were modeled based on samples collected at homes in the district.⁵² The multivariate model indicated that indoor Mn concentrations were directly impacted by outdoor Mn concentrations, open windows, and distance from the Mn smelter chimney. In the second study, a vehicular wake erosion model was developed to look at Mn exposure from road dust due to resuspension of particles.⁵³ The study quantified the increased risk to children when exposed to Mn based on vehicular traffic in an area where Mn has been deposited in soil and on surfaces.

The current study used a unique approach to estimate children’s airborne Mn exposure. Air dispersion models provided an improved estimate using exposure associated with both home and school geographical locations. The AERMOD model estimates of ambient air Mn accounted for Mn emissions, terrain, and weather within a spatial and temporal context, all factors that influence the magnitude of exposure to an air pollutant. To further elucidate the linkages from sources to environmental concentrations to biomarkers to effect, additional research is needed, including studies to evaluate the utility of the modeled time-weighted annual ambient air Mn levels in relation to Mn biomarkers. Additionally, assessment of the capability of modeled ambient air levels to identify important pathways of Mn exposure through linkages with other factors such as residential soil and dust and children’s activity patterns would also further the knowledge of exposure scenarios for children. Data from the current CARES participants provide an opportunity for research in both of these areas.