Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

ATMOSPHERIC ENVIRONMENT: X 6 (2020) 100070

Contents lists available at ScienceDirect

Atmospheric Environment: X
journal homepage: http://www.journals.elsevier.com/atmospheric-environment-x

Assessment of particulate toxic metals at an Environmental


Justice community
Olivia S. Ryder a, *, Jennifer L. DeWinter a, Steven G. Brown a, Keith Hoffman b, Betsy Frey b,
Ali Mirzakhalili b, 1
a
Sonoma Technology, Inc. (STI), 1450 N. McDowell Blvd., Suite 200, Petaluma, CA, 94954, USA
b
Department of Natural Resources and Environmental Control Delaware (DNREC), 100 W. Water Street, Suite 6A, Dover, DE, 19904, USA

A R T I C L E I N F O A B S T R A C T

Keywords: Exposure to particulate matter containing toxic metal species can have negative health implications. The com­
Toxic metals munity of Eden Park, Delaware, is surrounded by numerous industrial facilities, the Port of Wilmington, and a
Air quality highly trafficked freeway, all of which can be sources of toxic metals. The U.S. Environmental Protection
Positive matrix factorization
Agency’s EJ Screen ranks the community in the 88th percentile nationally for the Environmental Justice (EJ)
Community air monitoring
index of cancer risk and the 90th percentile for the EJ index of PM2.5 exposure. To understand the ambient
Environmental justice
Wilmington Delaware concentrations and sources of toxic metals and particulate matter, we collected hourly PM10 metals measure­
ments using an Xact 625i at a temporary monitoring site in the community. Additionally, PM2.5, wind speed and
direction, NOx, SO2, and black carbon were monitored. Conditional bivariate probability function maps and time
variation maps were constructed using the OpenAir statistical analysis package, and EPA Positive Matrix
Factorization was used to determine the sources of the measured metals. We found that transient concentration
events, where arsenic and lead concentrations were more than an order of magnitude larger than the study
average, occurred intermittently, but concentrations were generally below health benchmarks. Chromium and
nickel had common sources, including a metal plating facility and a chemical additive facility, and unidentified
sources were responsible for spikes in chromium levels up to 17 times above average. With Positive Matrix
Factorization, three sources were identified: soil/dust, concrete manufacturing dust, and brake wear. During
periods of moderate air quality (PM10 ¼ 55–154 μg/m3) concrete dust contributes 32% to PM10 concentrations,
while soil dust contributes 12%. Overall, the local industry and intermodal traffic emissions are large contrib­
utors to the ambient air pollution in the community, though transient high-concentration events from other
sources are also present.

1. Introduction Environmental Protection Agency (EPA) (Environmental Protec, 2018a,


2018b).
Metals can cause adverse impacts on the environment and human
health (Jaishankar et al., 2014; Tchounwou et al., 2012). Common � Arsenic is most commonly used as a wood preservative and in pes­
sources include industrial emissions, mobile source emissions, and ticides, and it is found in soil. Industrially, arsenic is used to make
soil/dust. Anthropogenic metal sources include combustion of heavy oil metal alloys for automobile batteries, semiconductors, and light
(e.g., vanadium and nickel), industrial facility emissions (e.g., arsenic emitting diodes; inhalation exposure has been linked to lung cancer
and lead), and mobile sources (e.g., iron and aluminum), among others (Tchounwou et al., 2012; U.S, 2007a).
(Becagli et al., 2012; Brown et al., 2007; Hays et al., 2011; Miller et al., � Nickel is used to make metal alloys for automobile batteries, semi­
2007). Of the metals and metalloids present in PM10, antimony, arsenic, conductors, and light-emitting diodes. Nickel may be released into
beryllium, cadmium, chromium, lead, manganese, mercury, nickel, se­ the air from industrial furnace stack emissions or from coal and oil
lenium, and uranium have specific risk thresholds determined by the combustion. It is also found at low levels in soil and plants. Prolonged

* Corresponding author.
E-mail address: oryder@sonomatech.com (O.S. Ryder).
1
Present address: Oregon Department of Environmental Quality

https://doi.org/10.1016/j.aeaoa.2020.100070
Received 20 September 2019; Received in revised form 24 January 2020; Accepted 23 February 2020
Available online 9 March 2020
2590-1621/© 2020 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license
(http://creativecommons.org/licenses/by-nc-nd/4.0/).
O.S. Ryder et al. Atmospheric Environment: X 6 (2020) 100070

exposure to high levels of nickel can produce adverse effects, at three different sites in the northeastern United States (Song et al.,
including reduced lung function, chronic bronchitis, and lung cancer 2001). The analysis revealed six common sources among the three sites,
(U.S, 2007b). including coal combustion characterized by sulfur and selenium pres­
� Chromium is found in soil, rocks, plants, and volcanic emissions (U.S, ence, sea salt characterized by sodium and sulfur, and oil combustion
2012a) and is emitted from industrial and other anthropogenic point characterized by nickel and vanadium. Jeong et al (2019) compared
sources such as welding, chrome plating, cement production, and PM2.5 metals at two near-road sites, performing PMF analysis to eluci­
coal combustion (Tchounwou et al., 2012; U.S, 2012a). Chronic date the contributions of exhaust and non-exhaust emissions. They
exposure has been linked to nasal septum atrophy and cancer (U.S, found that 9–19% of emissions measured were from exhaust, 2–6% were
2012a). from brake wear, and 3–4% were from re-suspension of road dust (Jeong
� Manganese is naturally found in food, rocks and soil, and is used in et al., 2019). Brown et al. (2007) identified eight sources of PM2.5 from
products such as fertilizer, fireworks, and paints. Industrially, man­ three years of data collected in Phoenix, Arizona, including soil/dust,
ganese is used in steel production to improve steel strength (U.S, mobile sources, and industrial sources (Brown et al., 2007).
2012b). Prolonged inhalation exposure to elevated levels of man­ The measurements described herein were taken at Eden Park in New
ganese results in cognitive effects and permanent nervous system Castle County, Delaware (Fig. 1). Eden Park, located in Wilmington,
damage (U.S, 2012b). Delaware, is the most populous city in the state with a population of
� Lead is released from production of lead compounds and alloys, or >70,000 people. The community is in the 88th percentile nationally for
from coal and oil burning. Health impacts of lead exposure have been the Environmental Justice (EJ) index of cancer risk based on the Na­
studied extensively, and the results of chronic exposure to lead range tional Air Toxics Assessment (NATA), the 90th percentile for the EJ
from cerebrovascular disease and cardiovascular effects to repro­ index of PM2.5 exposure, and the 86th percentile for the EJ index of air
ductive and neurological problems (U.S, 2007c). toxics respiratory hazard, and has a known issue with high concentra­
tions of PM10 (Environmental Protec, 2017). The EJ index was created
Soil and dust containing various concentrations of metals can by EPA to assess the risk associated with environmental and human
become entrained by wind, but also via agitation by car tires on road­ health conditions on minority and low-income communities with the
ways or unpaved lots (Seinfeld and Pandis, 2006; Pant and Harrison, aim of improving environmental protection and mitigating risk for all
2013). Vehicular emissions such as tire degradation and brake wear can populations. The EJ index, as described by EPA, “is a way of combining
also be sources of metals (Grigoratos and Martini, 2015). For example, demographic information with a single environmental indicator—such
Hays et al (2011) found, among others, barium, copper, and zinc asso­ as proximity to traffic—that can help identify communities that may
ciated with tire and brake wear, and nickel and chromium associated have a higher combination of environmental burdens and vulnerable
with components of engine metal alloys during a study near a North populations.” The community is surrounded by sources of toxic metals,
Carolina interstate highway (Hays et al., 2011). In a review, Grigoratos including the Port of Wilmington and industrial facilities that include
and Martini (2015) summarize numerous studies with a table of 20 trace metal recycling and concrete manufacturing. This neighborhood is also
metals detected in brake wear dust (Grigoratos and Martini, 2015). In adjacent to Interstate 495, which has an annual average daily traffic
addition to those mentioned in the Hays study, iron, titanium, and value of 116,416 vehicles (Terminal Ave. to South Wilmington limits)
manganese comprise large concentrations of brake dust emissions. (Delaware Department of Transportation, 2019). The aim of this study
While efforts over the past decades have reduced particulate exhaust was to assess the air quality in the community for toxic metals and other
emissions from cars and trucks, non-exhaust emissions have not been a pollutants, and to assess the contribution of both local and regional
focus of control efforts and are thus increasingly becoming an important sources to PM10 concentrations.
source of particulate emissions on roadways (Adamiec et al., 2016; Reid
et al., 2016). These emissions are of particular importance in commu­ 2. Materials and methods
nities located near major roads (Sunyer et al., 2015; Edwards et al.,
1994; Hitchins et al., 2000; Finkelstein et al., 2004; Kim et al., 2004; 2.1. Sampling site and metal measurements
Kunzli et al., 2000; Baldauf et al., 2008).
Another source that is of concern in certain communities is concrete Measurements took place between October 19, 2018, and December
dust from cement plants. Rovira et al (2011) found that particles origi­ 12, 2018, in the Eden Park community at the intersection of Terminal
nating from such dust contained elevated levels of copper, lead, and and Wilmington Avenues (the southeast corner of Eden Park) as shown
manganese (Rovira et al., 2011). Additionally, due to the composition of in Fig. 1. An Xact 625i Multi-Metals Monitoring System (Cooper Envi­
concrete, the dust has high calcium content, especially compared to soil. ronmental) was used to measure PM10 metals on an hourly basis. Higher
Calcium is prevalent in the environment and is also found naturally in time resolution allows for data that can be paired with meteorological
the form of calcium carbonate, which can be converted to calcium oxide conditions to derive a more detailed understanding of sources and var­
through heating or other mechanical processes used in making concrete iances in time (Furger et al., 2017). The Xact 625i pulls ambient air into
and drywall (Oss and Padovani, 2002). In addition to the metals cited by an inlet at 16.7 L/min onto filter tape. After an hour of collection, the
Rovira et al. as components of concrete dust, particulate emissions of tape is automatically analyzed via nondestructive x-ray fluorescence
calcium can be released into the atmosphere as a result of concrete (XRF) while the next sample is collected. Hourly concentration values
manufacturing. Calcium may thus act as a fingerprint identifier for for each metal were compared to their respective minimum limit of
concrete emissions (Lioy and Yu, 2009). detection (MDL), as provided by the instrument manufacturer. Species
Given the ubiquitous nature of metal-containing particles and the with measured concentrations below the MDL greater than 50% of the
potential toxic effects of prolonged human exposure to them, under­ time were omitted from subsequent analysis. The metals included in the
standing sources of metal-containing particles in the air is key. Trace analysis and their frequency of measurement above MDL are shown in
metals in particles tend to remain preserved during transport and may Columns 1 and 2 of Table 1. In addition to Xact sampling, aerosols were
act as fingerprints for a specific source (Phillips-Smith et al., 2017a). In also collected onto filters using a Thermo Partisol 2025 with a PM10 inlet
such cases, receptor modeling can be used to elucidate particle sources for a 24-h period every third day. The filters were analyzed by DNREC
(also referred to as factors). Positive matrix factorization (PMF) is a for metals via XRF with EPA Method IO 3.3.
receptor model that has been used in numerous previous studies for this
purpose (Brown et al, 2007, 2012; Hays et al., 2011; Jeong et al., 2019; 2.2. Gas and particle phase measurements
Phillips-Smith et al., 2017b; Kim et al., 2003; Song et al., 2001). For
example, Song et al. (2001) used PMF analysis on PM2.5 metals collected Hourly NO–NO2-NOx and trace level SO2 were measured using TAPI

2
O.S. Ryder et al. Atmospheric Environment: X 6 (2020) 100070

Fig. 1. Satellite image of Eden Park and the surrounding area. Inset is a wind rose plot for Eden Park between October 19 and December 12, 2018.

T200U and Thermo 43i-TLE continuous gas analyzers. Black carbon


mΔθ;Δu jC�x
(BC) was monitored with a Magee Scientific Aethalometer 633, and CBPFΔθΔu ¼
PM2.5 was measured using a TAPI T640 continuous particle monitor. nΔθ;Δu
Wind speed and direction were measured using a Vaisala WXT-520 in­
Where mΔθ;Δu is the number of samples in wind sector Δθ that have a
strument. The data were reported at 60-min averages and wirelessly
wind speed interval of Δu and whose concentration (CÞ is greater than
transmitted to a data logger, where they were quality-assured weekly.
the threshold x. In the case of the plots shown here, we use the 90th
percentile. This is divided by nΔθ;Δu , the total number of samples within
2.3. Data analysis
the wind sector and speed interval.
Since both the Xact data and filter measurements have associated
2.3.1. Individual metals
uncertainties, in addition to linear fit analysis, Deming regression was
Analysis began with investigations of individual metals. Here, the
performed to compare the filter and Xact data. Xact measurements were
OpenAir R package was used to elucidate diurnal trends and generate
averaged to daily values, and the regression calculation was performed
pollution rose plots and conditional bivariate probability function
using the Deming package in R. The standard error input for the filter
(CBPF) plots. Diurnal profiles were generated using the time variation
measurements was the experimental error value. The standard error
function within OpenAir. This analysis produces average diurnal plots
input for the Xact measurements was calculated using the equation
for each day of the week, an average daily diurnal profile, and plots of
below. Since the MDL used was for hourly measurements rather than 24
the average concentration on a day-of-the-week and monthly basis.
h measurements, the estimation of standard error is a conservative upper
Pollution rose plots are a variation on wind rose plots, except the color
limit.
bar represents species concentration rather than wind speed. Some­
times, high concentrations of species which arise under low wind fre­ StandardErrorXact ¼ 0:1 � ½species�dailyaverage þ 0:5 � ½MDL�Hourly
quency conditions are difficult to elucidate with pollution rose plots
alone. CBPF plots can be used to provide an additional view of the data. 2.3.2. Positive Matrix Factorization
CBPF plots are heat maps showing the probability that the concentration EPA PMF version 5.0 was used to determine source factors of the Xact
of a species will be greater than some specified value for a given wind and BC measurements. This method has been described elsewhere, but
direction and speed. On these plots, warmer colors represent higher briefly, the model accepts a data matrix of measured concentrations and
concentrations, and the location on the plot indicates the direction and uncertainties for a given set of species and decomposes the data set into
wind speed under which the high concentration occur. In comparison to factors (Norris et al., 2014; Paatero and Tapper, 1994). These factors
conditional polar plots, CBPF plots provide additional insight into the represent potential sources that then need to be interpreted by the user
wind speed and direction dependency of a species’ concentration. As as to what each factor represents, given additional information about the
reported by Uria-Tellaetxe and Carslaw, CBPF is defined as: measurements (e.g., wind direction, nearby industrial sources, diurnal
variability, etc.). Each data point has a user-specified uncertainty value.
Here we used the sum of squares of the species-specific limit of detection

3
O.S. Ryder et al. Atmospheric Environment: X 6 (2020) 100070

Table 1
Table of species measured with the Xact, detection limits, and the percentage of measurements above the minimum detection limit; data are sorted by this percentage.
Gray text indicates the species were above MDL less than 50% of the time. Column 4 shows the median concentration and interquartile range from the 55-day sampling
period of this work as measured with the Xact. Columns 5 and 6 show the median concentrations from the Xact and collocated filter measurements for the 17 days with
filter measurements. Columns 7–9 are the r2, slope, and intercept of the linear fit between the filter and Xact measurements. Columns 9–10 are the slope and intercept
determined through Deming regression analysis.
Species MDL Measurements Xact Xact, 17 days with Filter Filter vs. Xact
(60 > MDL (%) (Median filter measurement measurement
Linear Linear Linear fit Deming Deming
min), [IQR]), μg/ (Median [IQR]), μg/ (Median [IQR]),
fit r2 fit slope intercept regression regression
μg/m3 m3 m3 μg/m3
slope intercept

Ca 3.00 � 100 0.586 1.241 [1.426] 1.215 [0.922] 0.86 1.70 0.67 0.81 0.07
10¡4 [1.528]
Fe 1.70 � 100 0.405 0.690 [0.461] 0.796 [0.454] 0.75 1.16 0.18 0.74 0.04
10 4 [0.821]
Zn 6.70 � 100 0.021 0.025 [0.009] 0.020 [0.013] 0.85 1.01 0.00 1.06 0.00
10 5 [0.029]
S 3.16 � 100 0.380 0.420 [0.249] 0.498 [0.211] 0.96 1.29 0.19 1.03 0.07
10 3 [0.345]
K 1.17 � 100 0.107 0.155 [0.079] 0.196 [0.095] 0.68 1.12 0.05 0.84 0.01
10 3 [0.152]
Ti 1.60 � 99 0.026 0.049 [0.032] 0.044 [0.025] 0.77 1.72 0.02 1.06 0.00
10 4 [0.054]
As 6.30 � 96 0.001 0.001 [ 0.001] 0.001 [0.001] 0.87 1.09 0.00 1.73 0.00
10 5 [0.001]
Ba 3.90 � 94 0.015 0.023 [0.023] 0.0313 [0.028] 0.86 0.76 0.00 0.68 0.00
10 4 [0.029]
Br 1.00 � 94 0.002 0.002 [0.002] 0.002 [0.002] 0.96 1.27 0.00 1.20 0.00
10 4 [0.002]
Al 0.1 94 0.748 0.966 [0.794] 0.275 [0.183] 0.84 4.97 0.17 4.79 0.14
[0.851]
Cl 1.73 � 94 0.133 0.157 [0.364] 0.145 [0.417] 0.90 1.03 0.00 0.95 0.02
10 3 [0.343]
Mn 1.40 � 89 0.007 0.014 [0.015] 0.020 [0.015] 0.66 0.92 0.00 0.69 0.00
10 4 [0.018]
Ni 1.00 � 87 0.001 0.002 [0.001] 0.001 [0.001] 0.70 1.35 0.00 2.08 0.00
10 4 [0.002]
Cr 1.20 � 85 0.002 0.004 [0.005] 0.005 [0.002] 0.46 0.87 0.00 1.11 0.00
10 4 [0.005]
Cu 7.90 � 82 0.006 0.010 [0.008] 0.026 [0.025] 0.47 0.22 0.00 0.28 0.00
10 5 [0.011]
Si 0.178 78 0.358 1.228 [1.611] 1.219 [0.844] 0.84 1.93 0.91 0.86 0.09
[1.402]
Sb 5.20 � 55 0.006 0.007 [0.001] 0.001 [0.006] 0.06 0.22 0.10 0.02 0.01
10 3 [0.008]
Pb 1.30 � 53 0 [0.002] 0.002 [0.001] 0.003 [0.003] 0.90 0.68 0.00 0.68 0.00
10 4
Se 8.10 � 48 7.39 � 10-5 0 [0] 0 [0] 0.78 0.49 0.00 0.51 0.00
10 5 [0]
V 1.20 � 43 0 [0.001] 0 [0.001] 0.001 [0.001] 0.47 1.12 0.00 0.89 0.00
10 4
Ag 1.90 � 41 0.001 0.002 [0.001] 0.001 [0.002] 0.29 0.18 0.00 0.20 0.00
10 3 [0.003]
Sn 4.10 � 38 0.002 0.003 [0.001] 0.001 [0.002] 0.12 0.11 0.00 0.12 0.00
10 3 [0.006]
In 3.10 � 16 0 [0.002] 0.001 [0] 0 [0.002] 0.28 0.09 0.00 0.08 0.00
10 3
Cd 2.50 � 10 0 [0] 0 [0] 0 [0.001] 0.04 0.04 0.00 0.03 0.00
10 3
Co 1.40 � 3 0 [0] 0 [0] 0.002 [0.002] 0.02 0.01 0.00 0.00 0.00
10 4
P 5.20 � 2 0 [0] 0 [0] 0 [0] NA NA NA NA NA
10 3
Hg 1.20 � 1 0 [0] 0 [0] 0 [0] 0.01 0.04 0.00 0.74 0.04
10 4
Bi 1.30 � 1 0 [0] 0 [0] NA NA NA NA NA NA
10 4
Tl 1.20 � 0 0 [0] 0 [0] NA NA NA NA NA NA
10 4

(LODi ) and species- and sample-specific analytical uncertainty output by hourly metals data set were set to the minimum detection limit for the
the Xact (ai;j ): metal divided by 2 (Norris et al., 2014; Brown et al., 2015). The corre­
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sponding uncertainty value was set to 5/6 * MDL, as is typically done
U ¼ LOD2i þ a2i;j during PMF analysis. While assessing PMF solutions, individual species
were removed to determine the sensitivity of the solutions. Any species
Solutions ranging from three to five factors were considered. For the whose concentration was above the MDL less than half of the time was
purposes of the PMF calculation, data reported as 0 concentration in the

4
O.S. Ryder et al. Atmospheric Environment: X 6 (2020) 100070

not included. Additionally, species with a signal-to-noise ratio of 0, as a result (Environmental Protec, 2018a).
calculated in EPA PMF, were also removed. Of the remaining species, it
was found that inclusion of chlorine resulted in a factor that was 3. Results and discussion
dominated by chlorine with few other elemental contributions. The
same effect was observed for both sulfur and bromine. Considering the 3.1. Xact and filter sample comparison
types of sources present in the region, factors dominated by these species
did not explain the variance of the metals. These species were therefore Xact hourly data were averaged to 24 h and compared to the collo­
not used in PMF, leaving 15 metal species and BC. The concentrations cated filter samples analyzed by XRF to check the Xact measurements.
and uncertainties of crustal species of aluminum, silicon, calcium, iron, The average concentration of species measured by the Xact over the
and titanium were adjusted to account for the oxide compounds ac­ whole study (55 days) and over the days when corresponding filter
cording to known oxide ratios (Simon et al., 2011; Chow et al., 2015). measurements were taken (17 days), as well as the averages from the
Overall, 1148 samples were run with a random seed number to start the filter measurements, can be found in Table 1. Linear fit and Deming
analysis. Gaseous and meteorological data were not included in the regression statistics between Xact and filter data are also shown in
model input, but they were used to assess the PMF factors. Table 1. For all species measured above MDL more than 50% of the time,
Boot strapping (BS), displacement (Disp) and bootstrapping the average metal concentrations measured by Xact and XRF were not
enhanced by displacement (BS-Disp) were run to assess the validity of significantly different from each other within the overlapping inter­
the results (Paatero et al., 2014; Brown et al., 2015). In BS, EPA PMF quartile ranges (IQR). All species, with the exception of antimony
randomly selects non-overlapping blocks of consecutive samples, (whose signal resembled noise), exhibited modest agreement. Calcium,
creating a new input file with the same dimensions as the original, zinc, sulfur, arsenic, barium, bromine, chlorine, and lead had r2 values of
running PMF on the re-sampled data set, and mapping these new BS 0.85 or above, indicating good agreement between the two methods.
factors to the original factors. In DISP, each data point in the profile Tremper et al (2018) compared results obtained with a CES XACT 625
matrix is perturbed to determine an uncertainty estimate of rotational with those measured using ICP-MS (Tremper et al., 2018). They stated
ambiguity. BS-DISP combines these processes, by displacing the BS that differences between these two methods resulting in slopes that
resampled data. Taken together, all the approximations of rotationally deviate from unity could result from sampling temperature differences
accessible spaces for randomly located solutions represent both the or variable blank levels in filter paper. These may also be applicable in
random uncertainty and the rotational uncertainty for the modeled so­ this study. Additional error may have arisen from storage of 24-hr filter
lution to the complete data set (Brown et al., 2015). samples for 1 month prior to chemical analysis, versus collection and
immediate analysis for 1-hr Xact samples in this work. It is also possible
2.3.3. Black carbon that deviations in slope from unity may be attributable to differences in
The Magee Scientific AE33 Aethalometer was used to measure PM2.5 instrument sensitivities. The analysis that follows utilizes the Xact data.
black carbon. The AE33 collects ambient aerosol onto a filter tape; every
5 min, the absorbance of the material deposited on the tape is measured 3.2. Wind and PM trends
at seven wavelengths, ranging from 370 to 960 nm. The BC concentra­
tion is from the 880 nm channel. BC absorbs uniformly across all The inset in Fig. 1 shows a wind rose plot illustrating the distribution
wavelengths at which absorbance is measured, but brown carbon, which of winds at the Eden Park sampling site for the duration of the study,
is produced by combustion of wood or other biomass, absorbs primarily October 18, 2018–December 12, 2018. During this time, winds from the
in the lower wavelengths of the measured range (Park et al., 2006; Jeong west dominated: winds originated from between 230� and 320� more
et al., 2004). A commonly used method is to assume that BC is only from than 60% of the time. Over the course of the sampling period, hourly
wood burning and fossil fuel combustion, so that the difference in PM2.5 varied between 0.6 μg/m3 and 42.2 μg/m3, with a mean value of
absorbance at the 470 nm and 960 nm wavelengths can be used to 10.7 � 7.4 μg/m3 (1σ). PM2.5 concentrations peaked around 7:00 a.m.
calculate the quantity of BC from wood burning (BCwb) and fossil fuels and were lowest at approximately 6:00 p.m. Monday through Friday,
(BCff). For each hour of BC measurement, we used this method to with little diurnal trend seen on the weekend. For the same time period,
calculate BCwb and BCff. More information on these calculations can be hourly PM10 varied between 1.1 μg/m3 and 224.2 μg/m3, with a mean
found in the accompanying supporting information. value of 28.5 � 26.7 μg/m3 (1σ). On average, PM10 concentrations
reached a maximum between 7:00 a.m. and 11:00 a.m. and were highest
2.3.4. Back trajectories from Monday through Thursday, with lower concentrations on Friday
HYSPLIT back trajectory analysis was performed using the NOAA and no strong diurnal trend on the weekend. On an hourly basis, the sum
HYSPLIT trajectory model to determine the origin of air masses where of all species measured by the Xact accounted for 21.1 � 6.3% (1σ) of
spikes in arsenic and lead were observed (Stein et al., 2015). The model total PM10 mass on average, but varied over the course of the study to as
was using NAM 12 km meteorological data at 50 m, 100 m, and 250 m high as 80% and as low as 0.6%. Of the metal contribution to PM10, up to
elevation in back trajectory mode for 6 h preceding the spike in arsenic 5% was from metals both defined as EPA air toxics and measured above
or lead. their MDL over 50% of the time.
Figs. S1a and S1b show time series and diurnal profiles of NOx, BC,
2.4. Comparison to health benchmarks and SO2 concentrations. Over the sampling period, the median SO2
concentrations were 0.60 ppb [IQR ¼ 0.50 ppb], NOx concentrations
The U.S. EPA has developed dose-assessment response benchmarks were 20.3 ppb (IQR ¼ 21.7 ppb] and BC concentrations were 0.66 μg/m3
for toxic metals. The report lists the long-term chronic inhalation [IQR ¼ 0.95 μg/m3]. SO2 concentrations showed a diurnal profile with
thresholds for non-carcinogenic and carcinogenic effects (Environ­ multiple sharp peaks during the weekdays, typically occurring between
mental Protec, 2014a). The values for non-carcinogenic effects are the 8:00 a.m. and 12:00 p.m. Monday through Thursday. On Friday through
levels above which sensitive populations may start experiencing adverse Sunday, SO2 concentrations showed a less pronounced morning peak.
health problems other than cancer, such as breathing difficulties. The However, the median weekday vs. weekend concentrations of SO2
cancer data are unit risk estimate (URE) values in units of 1/(μg/m3). remained comparable (median weekday SO2 ¼ 0.60 ppb [IQR ¼ 0.50
Unit Risk Estimates can be transformed to 1-in-a-million cancer bench­ ppb]; median weekend SO2 ¼ 0.60 [IQR ¼ 0.40 ppb]). On average, NOx
marks by inverting them and dividing by a million, indicating that if a exhibited a strong diurnal profile, sharply peaking between 6:00 and
population of 1 million people is exposed to a certain concentration over 7:00 a.m. daily Monday through Thursday. Friday exhibited this same
a 70-year lifetime, one or more people are expected to develop cancer as profile, but the peak was less pronounced, and weekend trends did not

5
O.S. Ryder et al. Atmospheric Environment: X 6 (2020) 100070

have a well-defined peak in concentration. This is consistent with high the Eden Park concentration (0.0021 � 0.0037 μg/m3 [1σ]) falls below
commuter vehicle traffic during the mornings on weekdays (median the benchmark. Manganese has a reported non-cancer inhalation
weekday NOx ¼ 22.3 ppb [IQR ¼ 22.8 ppb]) and less traffic on the threshold of 0.3 μg/m3, which is well above the average value measured
weekends, when fewer commuters are on the road (average weekend at Eden Park (0.0162 � 0.0267 μg/m3 [1σ]).
NOx ¼ 14.95 ppb [IQR ¼ 15.75 ppb]). BC tracked NOx trends closely,
even during the weekends when a strong NOx diurnal pattern was not 3.4. Individual species analysis
present (average weekday BC ¼ 0.748 μg/m3 [IQR ¼ 1.103 μg/m3];
average weekend BC ¼ 0.483 μg/m3 [IQR ¼ 0.596 μg/m3]). Since BC is Initially, the individual metal species were analyzed separately and
also associated with vehicle emissions, the correlation with NOx is trends were compared to elucidate how metals were co-varying with
expected. each other as well as with NOx, SO2, PM10, and BC. Fig. 2 shows a time
series of the 15 species used subsequently in the PMF analysis as well as
3.3. Comparison to health benchmarks NOx and PM10. This section discusses trends in BC and in toxic species:
(1) chromium and nickel, (2) manganese, and (3) arsenic and lead. We
Table 2 lists the toxic metals for which the EPA has developed a dose- also discuss high-concentration species barium and copper, which trend
assessment response benchmark (Environmental Protec, 2014a) and together, and calcium. Other species on the EPA air toxics list are not
also lists the average concentrations and standard deviations from this discussed here, since their concentrations were below the minimum
work (55-day average) for comparison. It should be noted that the detection limit more than 50% of the time. Pollution roses, CBPF plots,
chronic inhalation benchmarks are for exposure over a 70-year lifetime, and diurnal profiles for metals, calculated using the OpenAir R package
while the concentrations reported from this work are for a much shorter (Carslaw and Ropkins, 2012), can be found in Figures S2 through S12.
time scale. Health risk screening is typically employed to compare
concentrations to benchmarks in order to prioritize further analyses. 3.4.1. Black carbon
Despite the discrepancy in time scale, the concentration comparisons Over the course of the study, median BC concentrations from wood
here are performed as a screening to identify which compounds may be burning (BCwb) were 0.12 μg/m3 (IQR ¼ 0.15 μg/m3), and BC concen­
of highest importance in the community for future focus. trations from fossil fuel combustion (BCff) were 0.52 μg/m3 (IQR ¼ 0.81
The average arsenic concentration is above the cancer threshold (1 μg/m3). This equates to BC contributions of 20 � 14% from biomass
� 10 6 benchmark ¼ 0.00023 μg/m3) indicating an approximate cancer burning and 80 � 14% from fossil fuel. Diurnal plots of BCwb and BCff
risk of 7.3-in-a-million. Arsenic is far below the non-cancer threshold of along with NOx are shown in Fig. S1b. The data show that BCff has a
0.015 μg/m3. strong morning peak that coincides with NOx, while BCwb exhibits
Chromium concentrations at Eden Park have a mean of 0.0049 � higher concentrations during the evening hours.
0.0084 μg/m3 (1σ), which falls below the Cr(III) non-cancer standard of
0.1 μg/m3. The chronic exposure cancer values are for Cr(VI), the more 3.4.2. Chromium and nickel
toxic valence state form, and not Cr(III) (Environmental Protec, 2018a). Chromium concentrations during the study had a median value of
Though total chromium is measured at Eden Park, Cr(VI) can be esti­ 0.002 μg/m3 [IQR ¼ 0.005 μg/m3]. Concentrations were highest
mated as 0.0125 of total chromium, based on EPA guidance (McCarthy, Monday to Thursday and decreased slightly on Friday and Saturday,
2018; Environmental Protec, 2018c). With this method, Cr(VI) at Eden with very low levels seen on Sunday. As was seen in other species, there
Park is estimated to be 6.13 � 10 6 μg/m3, which falls below the cancer were periodic pulses in chromium concentration ranging from 1 to 2 h,
threshold (1 � 10 6 benchmark ¼ 8.3 � 10 5 μg/m3). We note that the when chromium concentrations spiked up to 17 times the mean. These
national Cr(VI) to total chromium ratio of 0.0125 may not be appro­ spikes do not consistently correspond with spikes in other species such
priate if the chromium is locally produced by sources such as chromium as lead and arsenic, though chromium is moderately well correlated
plating, but the ratio of Cr(VI) to total chromium would have to be above with nickel (r2 ¼ 0.65). Chromium does not track with BC (r2 ¼ 0.07) or
0.15 to exceed the 1 � 10 6 benchmark. NOx (r2 ¼ 0.07), indicating that the main source of chromium is not
For lead, Eden Park mean concentrations were 0.0019 � 0.0053 μg/ likely related to traffic patterns. This is confirmed through CBPF anal­
m3 (1σ), which is below the non-cancer threshold and National Ambient ysis, which shows the greatest probability of high chromium concen­
Air Quality Standard of 0.15 μg/m3 (Environmental Protec, 2016). For trations when winds are above 4 m/s from the west-northwest (Fig. 3). A
nickel compounds, which have a non-cancer threshold of 0.09 μg/m3, nickel, cadmium, and chromium plating facility is 0.95 miles (1.5 km)
northwest of Eden Park and may contribute to the chromium and nickel
Table 2 concentrations (Fig. 1). Concentrations are also high under low wind
Comparison of chronic inhalation benchmarks for non-cancerous effects (Col­ speeds from the southeast, corresponding to the locations of I-495; fa­
umn 2), the unit risk assessment benchmarks for cancerous effects (Column 3), cilities for soil treatment and recycling, where metals and non-metals are
and the mean and 1σ values from this work (Column 4). separated; an oil pipeline storage facility; a chemical additive facility
Species Chronic Inhalationa Study Period supplying minerals for colorants, batteries, and specialty applications;
Mean � 1σ and a salt and packaged ice melt manufacturer. Some of these facilities
Non-Cancer Hazard Cancer 1-in-a-million This Work (μg/
can handle chromium in various forms and may be the source of chro­
Index (μg/m3) Benchmark (μg/m3) m3) mium from the southeast.
Similar to chromium, nickel CBPF plot analysis indicates that high
As 0.015 0.00023 0.0017 �
0.0026 concentrations predominantly occur with winds from the west-
Crb 0.1 8.3 � 10 5
0.0049 � northwest, with the highest concentrations being more likely when
0.0084 winds are above 5 m/s (Fig. 3). As was the case with chromium, nickel
Pb 0.15c – 0.0019 � appears to have a second contributory source to the southeast. The Toxic
0.0053
Ni 0.09 – 0.0021 �
Release Inventory (Environmental Protec, 2018d) lists Prince Minerals,
0.0037 a manufacturer of specialty chemicals and industrial additives to the
Mn 0.3 – 0.0162 � southeast, as a known contributor of manganese and nickel to the at­
0.0267 mosphere. Collectively, there appear to be multiple, intermittent sources
a
Data from Dose-Response Assessment Tables, EPA. of chromium and nickel, though as indicated by the moderate correla­
b
Chronic inhalation values specific to Cr(VI). tion, the emissions of chromium and nickel may not always occur
c
NAAQS standard. coincidentally (Fig. 4A). In addition, part of the correlation may be due

6
O.S. Ryder et al. Atmospheric Environment: X 6 (2020) 100070

Fig. 2. Time series of select species over the course of the sampling period at Eden Park. Units are μg/m3 unless otherwise stated on the y-axis. Red dashed lines
indicate the median value measured across the study period and are shown for context. (For interpretation of the references to color in this figure legend, the reader is
referred to the Web version of this article.)

to measurement biases. A comparison of nickel’s Xact measurements to and are centralized directly at the sampling location (Fig. 3). While the
filter measurements was reasonable (r2 ¼ 0.70), but chromium showed arsenic signal tends to remain low, there are intermittent spikes of very
more scatter between the collocated measurements (r2 ¼ 0.46). Due to high arsenic concentrations. There are twelve such instances of high
the intermittent concentration spikes and combination of sources in the concentrations, and these events fall on Tuesdays, Wednesdays, Fridays,
area, the signal may not be sufficiently strong to extract a specific Saturdays, or Sundays. The highest spikes are 0.044 μg/m3, occurring on
chromium/nickel factor in PMF. Sunday, November 11, at 4:00 p.m.; 0.034 μg/m3 on Saturday, October
20, at 6:00 p.m.; 0.030 μg/m3 on Wednesday, November 21, at 4:00 p.
3.4.3. Manganese m.; and 0.022 μg/m3 on Friday, November 23, at 10:00 p.m. They range
Over the sampling period, the median manganese concentration was between 5 and 25 times larger than the mean arsenic concentration
0.007 μg/m3 [IQR ¼ 0.018 μg/m3]. The manganese time series (Fig. 2) (0.002 μg/m3) during the study. Since these spikes typically occur after
shows time periods (lasting between 1 and 12 h) of higher manganese business hours, it is unlikely they are associated with an industrial
concentration spikes, which happen sporadically on weekdays. The source. HYSPLIT back trajectory analysis performed for the 6 h prior to
largest peak, occurring at 4:00 p.m. on November 6, 2018, is 13.7 times the peak of each arsenic spike showed no consistent origin of winds at
larger than the mean (0.016 μg/m3). CBPF plots show that higher con­ 50 m, 100 m, or 250 m elevation. To further investigate the spikes in
centrations of manganese occur predominantly when wind speeds from arsenic, pollution roses were created for each of the spikes, including 3 h
the northwest are more than 3 m/s. There is also a weaker source present before and after the peak as a buffer. However, the pollution roses did
under lower wind speeds (<2 m/s) southeast of Eden Park, which could not show consistent trends in wind speed or direction that would help to
possibly relate to the specialty chemical manufacturer mentioned above elucidate the source of the sudden increases in arsenic.
for nickel (Fig. 3). The CBPF plot for lead looks similar to that for arsenic (Fig. 3), where
the highest concentrations are more likely to be found directly at the site
3.4.4. Arsenic and lead under low wind conditions (<1 m/s). However, there also appear to be
Diurnal analysis of arsenic shows that this metal has very low con­ additional locations where lead may be found. For example, higher
centrations (median ¼ 0.001 μg/m3, IQR ¼ 0.001 μg/m3) and no concentrations of lead are found when winds are from the northwest and
consistent diurnal profile from day to day. CBPF plots indicate that high northeast. Again, like arsenic, lead has no diurnal structure and has very
concentrations are most probable under very low wind speeds (<1 m/s) low concentrations (median ¼ 0 μg/m3, IQR ¼ 0.002 μg/m3) throughout

7
O.S. Ryder et al. Atmospheric Environment: X 6 (2020) 100070

Fig. 3. Conditional bivariate probability function plots for species over the course of the study. The radial axis indicates wind speed in m/s, and the color bar
indicates the probability that a species will be greater than the 90th percentile. (For interpretation of the references to color in this figure legend, the reader is
referred to the Web version of this article.)

Fig. 4. Gray markers show chromium vs. nickel (left) and arsenic vs. black carbon from wood burning (right). Gray dashed lines are linear fits to data.

the study, except for sporadic times when the concentrations spike be­ of wood treated with preservatives. To assess whether soil was respon­
tween 4 and 39 times above the mean value of 0.002 μg/m3. Here, lead sible for the spikes, the arsenic signal was compared with the signals of
spikes appear sporadically throughout the week, and HYSPLIT back potassium and BC. Since potassium has multiple environmental sources,
trajectory analysis showed no consistent origin for the air masses leading the signal was apportioned to crustal potassium and non-crustal potas­
up to the spikes. Due to their similar CBPF maps and intermittent spiking sium using a literature-derived mass reconstruction equation, where
behavior, the lead and arsenic time series were compared to each other non-crustal potassium can be estimated as
as a determinant of how closely they may be related. 33% of the lead
Knon ¼ 1:2 � ðKtotal 0:6Fetotal Þ
spikes are unassociated with spikes in arsenic, 33% correspond with an crustal

increase in arsenic but below the 99th percentile value determined for (Simon et al., 2011; Chow et al., 2015). Crustal potassium was then
arsenic, and 33% of the lead spikes coincide with arsenic spikes. It is calculated as the difference between the total potassium signal and the
clear that both species are typically low except for aberrant events of non-crustal potassium. Scatter plot analysis showed that the spikes in
unknown origin. arsenic were not correlated with high concentrations of crustal potas­
Two additional potential sources of arsenic are soil and local burning sium, indicating that soil is unlikely to be responsible for the increased

8
O.S. Ryder et al. Atmospheric Environment: X 6 (2020) 100070

levels of arsenic. 3.4.5. Barium and copper


To assess whether the arsenic spikes resulted from local burning Barium and copper exhibit similar patterns to one another in their
activities, the biomass burning contribution to BC was compared to concentration patterns and diurnal variability, correlating with an r2 of
arsenic. This analysis showed the spikes were not correlated with high 0.77. The median concentration during the study was 0.015 μg/m3 (IQR
concentrations of wood burning BC (Fig. 4B). Since the spikes are spo­ ¼ 0.029 μg/m3) for barium and 0.006 μg/m3 (IQR ¼ 0.011 μg/m3) for
radic and do not correlate with either soil or wood burning, the two copper. There are sporadic periods of high concentrations of both spe­
likely local sources, the origin of the arsenic spikes remains unknown. cies lasting between 2 and 10 h throughout the week, including Sundays.
CBPF plots for both species show concentrations above the 90th
percentile when the winds are low at <1 m/s. They both exhibit diurnal

Fig. 5. PMF profiles for the 3-factor solution where bars are the concentration of each species, and the markers are the percentage of species and a time series trace
for each factor and PM10. Aluminum, silicon, calcium, iron, and titanium include adjustments for oxides.

9
O.S. Ryder et al. Atmospheric Environment: X 6 (2020) 100070

variability, co-varying with NOx (Ba/Cu r2 ¼ 0.71; Cu/NOx r2 ¼ 0.58), Table 3


and have a morning and evening peak. Their concentrations are highest Ratios of aluminum, iron, potassium, and calcium to calcium calculated from
Monday through Thursday, are lower on Friday, and are lowest on the elemental concentrations reported for two types of soil less than 6 miles (9.7 km)
weekend. These trends suggest that copper and barium concentrations from Eden Park (Rows 1 and 2). Rows 3 and 4 show calculated elemental ratios
relate to traffic patterns and vehicle activity. These species, along with for Factors 1 and 2 from the PMF solution.
antimony and zinc, can be from vehicle brake wear, which would be Data Al Fe K Ca
consistent with the high concentrations under low wind speed, diurnal Gleneig Soil 12 16 1 1
profiles, and trends with NOx observed here (Hays et al., 2011; Dillner Neshaminy Soil 17 18 1 1
et al., 2005; Thorpe and Harrison, 2008; Lough et al., 2005; Hjortenk­ Factor 1 14 12 0.2 1
Factor 2 0.5 0.2 0 1
rans et al., 2007).

3.4.6. Calcium sampled and tested in the laboratory via either Atomic Absorption or
The median calcium concentration was 0.586 μg/m3 (IQR ¼ 1.528 Inductively Coupled Plasma Mass Spectrometry for elemental compo­
μg/m3) during the study. The greatest probability of high concentrations sition. While the ratios from Factors 1 and 2 do not match perfectly to
(Fig. 3) was when winds were above 5 m/s from the west-northwest. On either soil type, Factor 1 ratios for each element are of the same order of
weekdays, median calcium concentrations were 0.883 μg/m3 (IQR ¼ magnitude as both soil types. For Factor 2, in comparison, potassium
2.09 μg/m3); however, on the weekends, they were 4.5 times lower. The ratio values are one order of magnitude higher and calcium ratio values
average diurnal profile (see Fig. S5) indicates that calcium tracks PM10 are two orders of magnitude higher. This suggests that Factor 2 is not
(r2 ¼ 0.78) but not NOx (r2 ¼ 0.035) or BC (r2 ¼ 0.049). The reduced crustal and may instead derive from local concrete manufacturing. The
calcium weekend signal and correlation with PM10 might be explained if most common type of concrete is portland cement, which typically
the source of the calcium is nearby cement facilities that run on a ranges between 60% and 70% CaO species (Oss and Padovani, 2002).
Monday to Friday schedule only. The four main mineral forms of CaO are SiO2⋅3CaO, SiO2⋅2CaO,
Al2O3⋅3CaO, and 4CaO⋅Al2O3⋅Fe2O3, which is consistent with contri­
3.5. Positive Matrix Factorization analysis butions of silicon, aluminum, and iron to Factor 2.
Further analysis was conducted by developing pollution roses and
In the PMF analysis, solutions ranging from 3 to 5 factors were CBPF plots for each factor, shown in Fig. 6. The CBPF plots show that
considered. A 5-factor solution appeared to over-fit the data, with two Factors 1 and 2 are highest when winds are from the west and northwest.
highly correlated crustal factors (slope of 0.98 and an r2 value of 0.87). The highest probability of concentrations at or above the 90th percentile
The 4-factor solutions produced two factors related to crustal species for both Factors 1 and 2 occurs at wind speeds above 5 m/s, indicating
and two factors that correlate with BC. As in the 5-factor solution, the that both factors arise from sources farther away from Eden Park, and
two crustal factors co-varied, with a slope of 0.87 and an r2 value of 0.87. that these factors impact the site only when winds are sufficient to
This indicates that the 4-factor solution was likely also overfitting the transport PM10 particles. The pollution rose plots (Fig. 6d and e,
data. A 3-factor solution offered the most interpretable solution and was respectively) also confirm the directionality of the source emissions and
reproducible, with crustal (Factor 1), calcium (Factor 2) and brake wear show that the greatest concentrations for Factor 1 occur 20% of the time
(Factor 3) factors. from the west and 24% of the time from the west-northwest. Similarly,
For BS, 400 runs were performed with a block size of 22 and mini­ 24% of Factor 2 comes from the west.
mum correlation r2 value of 0.8. BS runs of the 3-factor solution showed The combined results indicate that Factor 1 is likely a local soil/dust
that 95% of the bootstrapped Factor 1 and 100% of the bootstrapped source. To the west and northwest of Eden Park is a large open dirt lot on
Factors 2 and 3 mapped to the corresponding base factor. 0.5% of Factor South Market Street used by construction vehicles, unpaved parking
1 was unmapped. The BS results showed all species falling within the areas used to house salvage vehicles, and parkland. All of these are
interquartile range, zero cases with swaps, and a change in the sum-of- possible sources of soil that could contribute to Factor 1. In the same
squares object function (%dQ) of less than 0.1%. For BS-Disp, the active direction is Diamond Materials, a large facility that specializes in con­
species were set as iron, copper, and silicon, and the results showed zero crete and asphalt recycling and production, and Contractors Materials
cases with swaps. LLC, another large facility listed as a hot mix asphalt plant. The
The details of the resultant factors are described below. PMF factor elemental composition of Factor 2, in combination with the location of
profiles and time series are shown in Fig. 5. The amount of each indi­ the concrete facilities, indicates that Factor 2 likely derives from these
vidual metal species and black carbon is shown in bars, while the per­ concrete/asphalt dust sources. The diurnal profiles of both factors peak
centage of each total species measured that contributes to each factor’s between 10:00 a.m. and noon Monday through Thursday, with lower
total mass is shown by red markers. Summing the percentage values of a concentrations on Fridays and very low concentrations on the weekend
species over all three factors equals 100%. (Figs. S13–S14). They also both co-vary with PM10. For both factors,
Factor 1 is dominated by aluminum, chromium, iron, manganese, these trends are consistent with trucks or large vehicles driving over soil
nickel, and titanium. While silicon is also present, the largest percent of in open lots and entraining the soil into the air, or with trucks moving
silicon is in Factor 2 rather than this factor. Factor 2 also contains strong concrete or concrete components and creating dust during business
contributions from aluminum, chromium, manganese, nickel, and tita­ hours.
nium. Additionally, Factor 2 features high calcium and potassium The contribution of soil/dust and concrete to PM10 was assessed
components and a lower contribution from iron. Upon initial compari­ during air quality periods defined as “Good” (PM10 ¼ 0–54 μg/m3),
son, these two factors appear similar, as they both contain crustal ele­ “Moderate” (PM10 ¼ 55–154 μg/m3), and “Unhealthy for Sensitive
ments. The two factors have modest independence, with a slope of 0.56 Groups” (PM10 ¼ 155–254 μg/m3) according to the EPA’s Air Quality
and an r2 of 0.57. In order to distinguish the two potential sources, in­ Index (AQI) guidelines (Environmental Protec, 2014b); see Table 4).
formation on the composition of local soil was needed. When the air quality was defined as “Good,” median concrete dust and
A recent laboratory analytical report of soil composition in the state soil dust contributions to PM10 were 8% (IQR ¼ 17%; n ¼ 1003) and 7%
of Delaware contains data from two different soil types, Gleneig and (IQR ¼ 7%; n ¼ 1003), respectively. Under “Moderate” conditions,
Neshaminy, from within 6 miles (9.7 km) north-northwest of Eden Park concrete contributed a larger portion to PM10, with a median value of
(Galloway and Johnson, 2012). These data were used to generate ratios 33% (IQR ¼ 21%; n ¼ 130), compared to a median contribution from
of the elemental contents of the soil in the area for comparison to the soil of 12% (IQR ¼ 8%; n ¼ 130). Similarly, during periods when air
ratio of elements in PM measured in this work (Table 3). The soil was

10
O.S. Ryder et al. Atmospheric Environment: X 6 (2020) 100070

Fig. 6. PMF solution Factors 1–3 plotted as conditional probability function maps (Panels A–C), pollution roses (Panels D–F), and diurnal average. The bottom panel
shows the average diurnal variability for each factor.

quality was “Unhealthy for Sensitive Groups”, the median concrete


Table 4 contribution was 58% (IQR ¼ 14%; n ¼ 7) and the median soil contri­
Median contribution of soil dust, concrete dust, and brake wear to PM10 mass
bution was 19% (IQR ¼ 4%; n ¼ 7). Across all AQI categories, concrete
during time periods defined as “Good,” “Moderate,” and “Unhealthy for Sensi­
dust is the highest contributor, and is by far the largest portion of PM10
tive Groups,” based on the AQI index.
under moderate and unhealthy AQI, showing the importance of a local
AQI Air n Concrete Dust Soil Dust Brake/Tire
emission source on the community.
Quality (hours) Wear
Designation In contrast to Factors 1 and 2, Factor 3 presents a markedly different
median IQR median IQR median IQR signature, with comparatively high concentrations of arsenic, lead, zinc,
Good (PM10: 1003 8% 17% 7% 7% 4% 3% copper, barium, and BC. The presence of zinc and copper has been linked
0–54 μg/ to tire wear (Jeong et al., 2019; Hjortenkrans et al., 2007; Councell et al.,
m3)
2004; Harrison et al., 2012; Wahlin et al., 2006; Apeagyei et al., 2011;
Moderate 130 33% 21% 12% 8% 2% 3%
(PM10: Pant and Harrison, 2013). Zinc was identified as the most abundant
55–154 μg/ heavy metal present in tire wear due to the addition of ZnO and ZnS
m3) during tire’s production’s rubber vulcanization process (Adamiec et al.,
Unhealthy 7 58% 14% 19% 4% 0% 1% 2016; Ozaki et al., 2004). Additionally, Adamiec et al. state that over the
(PM10:
155–254
course of a tire’s lifetime, the tire is expected to lose 1.5 kg in weight to
μg/m3) the environment due to road abrasion (Adamiec et al., 2016). Another
source of secondary vehicle emissions is brake linings, which produce
particles when the action of braking induces friction on brake pads and
drums. Contributions from iron, copper, barium, lead, nickel, chromium
and zinc have been reported in differing amounts depending on the

11
O.S. Ryder et al. Atmospheric Environment: X 6 (2020) 100070

brake composition and function of the brakes (Hays et al., 2011; Ada­ Declaration of competing interest
miec et al., 2016; Dillner et al., 2005; Lough et al., 2005; Hjortenkrans
et al., 2007; Westerlund, 2001; Blau, 2001; Thorpe and Harrison, 2008; The authors declare that they have no known competing financial
Environmental Protec, 1998). These signatures all indicate a source interests or personal relationships that could have appeared to influence
linked to vehicular emissions, which is further supported by the high the work reported in this paper.
contribution of BC in this factor and by the correlation (r2 ¼ 0.65) with
the NOx diurnal profiles (Fig. S15). For these reasons, Factor 3 was CRediT authorship contribution statement
assigned to secondary vehicle emissions.
The pollution plot and CBPF plot for Factor 3 are consistent with the Olivia S. Ryder: Formal analysis, Data curation, Writing - original
initial assignment that these particles are predominantly from vehicular draft, Writing - review & editing. Jennifer L. DeWinter: Conceptuali­
emissions. The pollution rose plot and CBPF maps indicate that this zation, Methodology, Software, Supervision, Project administration.
factor has its highest contribution under lower wind conditions, gener­ Steven G. Brown: Conceptualization, Methodology, Writing - original
ally under 2 m/s, implying a local source near Eden Park. The CBPF plot draft, Writing - review & editing, Supervision, Project administration.
also indicates that the highest probability of contributions in the 90th Keith Hoffman: Conceptualization, Investigation, Supervision. Betsy
percentile occurs when winds are minimal or low and from the south­ Frey: Conceptualization, Investigation, Supervision, Funding acquisi­
east, the direction of I-495 (see Fig. 1). tion, Project administration. Ali Mirzakhalili: Conceptualization,
In summary, a 3-factor solution was the best fit to the Eden Park data Funding acquisition.
and allowed for determination of a local soil/dust factor, a concrete dust
factor, and a brake/tire wear factor. The concrete and soil factors were
Acknowledgments
differentiated from each other by the high calcium concentration in the
concrete factor —likely deriving from the high concentration of calcium
The authors would like to thank Dr. Michael McCarthy for helpful
in limestone and other components used for concrete manufactur­
discussions and guidance during data analysis and Levi Stanton for in­
ing—and by the ratios of elements in the soil factor, resembling the
strument setup. This work was funded in part by an EPA Multipurpose
ratios from two local soil samples. The brake/tire wear factor had strong
Grant to the State of Delaware (AA06346501) and by State of Delaware
contributions from BC and metals including zinc, copper, and lead, all
Indirect Funds.
previously measured as components present in degradation of tires and
vehicle brakes; these contributions are consistent with a relationship to
vehicles. Appendix A. Supplementary data
Table 4 shows the contributions of brake/tire wear to PM10 under
different air quality conditions during the study. It also shows the Supplementary data to this article can be found online at https://doi.
contribution of BCBB and BCFF to black carbon during these times. org/10.1016/j.aeaoa.2020.100070.
Compared to concrete dust and soil dust, brake/tire wear contributed
less to PM10 across all air quality conditions. Brake/tire wear’s References
maximum contribution was under “Good” conditions, with a median of
4% (IQR ¼ 3%). Under all air quality conditions, the contribution of Adamiec, E., Jarosz-Krzemi� nska, E., Wieszała, R., 2016. Heavy metals from non-exhaust
vehicle emissions in urban and motorway road dusts. Environ. Monit. Assess. 188.
fossil fuel to black carbon was greater than the contribution of biomass Apeagyei, E., Bank, M.S., Spengler, J.D., 2011. Distribution of heavy metals in road dust
burning, with a maximum median value of 87% (IQR ¼ 6%) under along an urban-rural gradient in Massachusetts. Atmos. Environ. 45 (13),
“Moderate” conditions. 2310–2323.
Baldauf, R., et al., 2008. Traffic and meteorological impacts on near-road air quality:
summary of methods and trends from the Raleigh near-road study. J. Air Waste
5. Conclusion Manag. Assoc. 58, 865–878.
Becagli, S., et al., 2012. Evidence for heavy fuel oil combustion aerosols from chemical
analyses at the island of Lampedusa: a possible large role of ships emissions in the
We used a combination of high-time-resolution instruments to
Mediterranean. Atmos. Chem. Phys. 12 (7), 3479–3492.
measure numerous species to understand concentrations and sources of Blau, P.J., 2001. Compositions, Functions, and Testing of Friction Brake Materials and
toxic metals and other pollutants in the EJ community of Eden Park. This Their Additives.
type of customized, focused study can help community members, public Brown, S.G., et al., 2007. Source apportionment of fine particulate matter in Phoenix,
Arizona, using positive matrix factorization. J. Air Waste Manag. Assoc. 57,
agencies and local stakeholders understand and develop mitigation 741–752.
strategies for EJ communities that do not have routine air pollution Brown, S.G., et al., 2012. Receptor modeling of near-roadway aerosol mass spectrometer
monitoring. During this study, all species concentrations fell below the data in Las Vegas, Nevada, with EPA PMF. Atmos. Chem. Phys. 12, 309–325.
Brown, S.G., et al., 2015. Methods for estimating uncertainty in PMF solutions: examples
one-in-a-million cancer and non-cancer risk thresholds with the excep­ with ambient air and water quality data and guidance on reporting PMF results. Sci.
tion of arsenic, which was slightly above the cancer threshold. Transient Total Environ. 518–519, 626–635.
high concentration spikes of arsenic, lead, chromium, and nickel Carslaw, D.C., Ropkins, K., 2012. Openair — an R package for air quality data analysis.
Environ. Model. Software 27–28, 52–61.
occurred, but not with sufficient regularity to identify their sources. Chow, J.C., et al., 2015. Mass reconstruction methods for PM2.5: a review. Air Qual.
Spikes in lead and arsenic did not occur with spikes in soil or biomass Atmos. Health 8 (3), 243–263.
burning indicators, suggesting that arsenic and lead were not from local Councell, T.B., et al., 2004. Tire-wear particles as a source of zinc to the environment.
Environ. Sci. Technol. 38 (15), 4206–4214.
soil or burning. With PMF, three factors were identified: soil/dust,
Delaware Department of Transportation, Vehicle Volume Summary (Traffic Counts),
concrete dust, and brake wear/tire emissions from vehicles. On average, 2019.
concrete dust contributes 10% (IQR ¼ 20%) and soil contributes 7% Dillner, A.M., et al., 2005. A quantitative method for clustering size distributions of
elements. Atmos. Environ. 39 (8), 1525–1537.
(IQR ¼ 8%) of PM10. As the PM10 concentration increases from the
Edwards, J., Walters, S., Griffiths, R.K., 1994. Hospital admissions for asthma in
“Good” to “Unhealthy for Sensitive Groups” AQI category, concrete preschool children: relationship to major roads in Birmingham, United Kingdom.
emissions account for a greater portion of PM10 compared to soil/dust. Arch. Environ. Health 49, 223–227.
Results from this work are being used by DNREC to work with local U.S. Environmental Protection Agency, Toxicological Review of Trivalent Chromium: in
Support of Summary Information on the Integrated Risk Information System (IRIS),
community members and industrial representatives on reducing emis­ 1998.
sions of concrete dust in the Eden Park community. U.S. Environmental Protection Agency, Dose-Response Assessment for Assessing Health
Risks Associated with Exposure to Hazardous Air Pollutants, 2014.
U.S. Environmental Protection Agency, Air Quality Index: a Guide to Air Quality and
Your Health, 2014.
U.S. Environmental Protection Agency, NAAQS Table, 2016.

12
O.S. Ryder et al. Atmospheric Environment: X 6 (2020) 100070

U.S. Environmental Protection Agency, EJSCREEN Environmental Justice Mapping and Paatero, P., Tapper, U., 1994. Positive matrix factorization: a non-negative factor model
Screening Tool: Technical Documentation, 2017. with optimal utilization of error estimates of data values. Environmetrics 5 (2),
U.S. Environmental Protection Agency, 2014 NATA: Assessment Results, 2018. 111–126.
U.S. Environmental Protection Agency, Dose-Response Assessment for Assessing Health Paatero, P., et al., 2014. Methods for estimating uncertainty in factor analystic solutions.
Risks Associated with Exposure to Hazardous Air Pollutants, 2018. Atmos. Meas. Tech. 7, 781–797.
U.S. Environmental Protection Agency, Technical Support Document: EPA’s 2014 Pant, P., Harrison, R.M., 2013. Estimation of the contribution of road traffic emissions to
National Air Toxics Assessment, 2018. particulate matter concentrations from field measurements: a review. Atmos.
U.S. Environmental Protection Agency, TRI Search, 2018. Environ. 77, 78–97.
Finkelstein, M.M., Jerrett, M., Sears, M.R., 2004. Traffic air pollution and mortality rate Park, K., et al., 2006. Comparison of continuous and filter-based carbon measurments at
advancement periods. Am. J. Epidemiol. 160, 173–177. the Fresno Supersite. J. Air Waste Manag. Assoc. 56 (4), 474–491. https://doi.org/
Furger, M., et al., 2017. Elemental composition of ambient aerosols measured with high 10.1080/10473289.2006.10464521.
temporal resolution using an online XRF spectrometer. Atmos. Meas. Tech. 10 (6), Phillips-Smith, C., et al., 2017a. Sources of particulate matter components in the
2061–2076. Athabasca oil sands region: investigation through a comparison of trace element
Galloway, R., Johnson, S.F., 2012. Statewide Soil Background Study: Report of Findings. measurement methodologies. Atmos. Chem. Phys. 17 (15), 9435–9449.
Grigoratos, T., Martini, G., 2015. Brake wear particle emissions: a review. Environ. Sci. Phillips-Smith, C., et al., 2017b. Sources of particulate matter components in the
Pollut. Control Ser. 22, 2491–2504. Athabasca oil sands region: investigation through a comparison of trace element
Harrison, R.M., et al., 2012. Estimation of the contributions of brake dust, tire wear, and measurement methodologies. Atmos. Chem. Phys. 17 (15), 9435–9449.
resuspension to nonexhaust traffic particles derived from atmospheric Reid, S., et al., 2016. Emissions modeling with MOVES and EMFAC to assess the potential
measurements. Environ. Sci. Technol. 46, 6523–6529. for a transportation project to create particulate matter hot spots. Transport. Res.
Hays, M.D., et al., 2011. Particle size distributions of metal and non-metal elements in an Rec.: J. Transport. Res. Board 2570, 12–20.
urban near-highway environment. Atmos. Environ. 45 (4), 925–934. Rovira, J., et al., 2011. Monitoring environmental pollutants in the vicinity of a cement
Hitchins, J., et al., 2000. Concentrations of submicrometre particles from vehicle plant: a temporal study. Arch. Environ. Contam. Toxicol. 60 (2), 372–384.
emissions near a major road. Atmos. Environ. 34 (1), 51–59. Seinfeld, J.H., Pandis, S.N., 2006. Atmospheric Chemistry and Physics: from Air
Hjortenkrans, D.S.T., Bergb€ ack, B.G., H€aggerud, A.V., 2007. Metal emissions from brake Pollution to Climate Change, second ed. John Wiley & Sons, Inc.
linings and tires: case studies of Stockholm, Sweden 1995/1998 and 2005. Environ. Simon, H., et al., 2011. Determining the spatial and seasonal variability in OM/OC ratios
Sci. Technol. 41 (15), 5224–5230. across the US using multiple regression. Atmos. Chem. Phys. 11, 2933–2949.
Jaishankar, M., et al., 2014. Toxicity, mechanism and health effects of some heavy Song, X.H., Polissar, A.V., Hopke, P.K., 2001. Sources of fine particle composition in the
metals. Interdiscipl. Toxicol. 7 (2), 60–72. northeastern U.S. Atmos. Environ. 35 (31), 5277–5286.
Jeong, C.H., et al., 2004. The comparison between thermal-optical transmittance Stein, A.F., et al., 2015. NOAA’s HYSPLIT atmospheric transport and dispersion
elemental carbon and Aethalometer black carbon measured at multiple monitoring modeling system. Bull. Am. Meteorol. Soc. 96, 2059–2077.
sites. Atmos. Environ. 38 (31), 5193–5204. Sunyer, J., et al., 2015. Association between traffic-related air pollution in schools and
Jeong, C.-H., et al., 2019. Temporal and spatial variability of traffic-related PM2.5 cognitive development in primary school children: a prospective cohort study. PLoS
sources: comparison of exhaust and non-exhaust emissions. Atmos. Environ. 198, Med. (3), 12.
55–69. Tchounwou, P.B., et al., 2012. Heavy metals toxicity and the environment. EXS 101,
Kim, E., Hopke, P.K., Edgerton, E.S., 2003. Source identification of Atlanta aerosol by 133–164.
positive matrix factorization. J. Air Waste Manag. Assoc. 53, 731–739. Thorpe, A., Harrison, R.M., 2008. Sources and properties of non-exhaust particulate
Kim, J.J., et al., 2004. Traffic-related air pollution near busy roads: the East Bay matter from road traffic: a review. Sci. Total Environ. 400 (1–3), 270–282.
Children’s respiratory health study. Am. J. Respir. Crit. Care Med. 170 (5), 520–526. Tremper, A.H., et al., 2018. Field and laboratory evaluation of a high time resolution x-
Kunzli, N., et al., 2000. Public-health impact of outdoor and traffic-related air pollution: ray fluorescence instrument for determining the elemental composition of ambient
a European assessment. Lancet 356, 795–801. aerosols. Atmos. Meas. Tech. 11 (6), 3541–3557.
Lioy, P.J., Yu, C.H., 2009. Contribution of Particle Emissions From a Cement-Related U.S. Department of Health and Human Services, 2007a. Toxicological Profile for Arsenic.
Facility to Outdoor Dust in Surrounding Community, p. 65. U.S. Department of Health and Human Services, 2007b. Toxicological Profile for Nickel.
Lough, G.C., et al., 2005. Emissions of metals associated with motor vehicle roadways. U.S. Department of Health and Human Services, 2007c. Toxicological Profile for Lead.
Environ. Sci. Technol. 39 (3), 826–836. U.S. Department of Health and Human Services, 2012a. Toxicological Profile for
McCarthy, M., 2018. NATA Remote Background Estimates Updated for 2017: Revised. Chromium.
Miller, A.L., et al., 2007. Role of lubrication oil in particulate emissions from a hydrogen- U.S. Department of Health and Human Services, 2012b. Toxicological Profile for
powered internal combustion engine. Environ. Sci. Technol. 41 (19), 6828–6835. Manganese.
Norris, G., et al., 2014. EPA Positive Matrix Factorization (PMF) 5.0 Fundamentals and Wahlin, P., Berkowicz, R., Palmgren, F., 2006. Characterisation of traffic-generated
User Guide. particulate matter in Copenhagen. Atmos. Environ. 40 (12), 2151–2159.
Oss, H.G., Padovani, A.C., 2002. Cement manufacture and the environment: Part I: Westerlund, K.-G., 2001. Metal Emissions from Stockholm Traffic - Wear of Brake
chemistry and technology. J. Ind. Ecol. 6 (1), 89–105. Linings.
Ozaki, H., Watanabe, I., Kuno, K., 2004. Investigation of the heavy metal sources in
relation to automobiles. Water, Air, Soil Pollut. 157 (1–4), 209–223.

13

You might also like