1. Introduction
The 2021 Michigan-Ontario Ozone Source Experiment (MOOSE) was a joint Canadian and United States multi-institution campaign aimed at studying ozone, meteorology, and air pollution in and around Michigan and Ontario. The study area focused on Southeast Michigan (SEMI), and Western Ontario including Detroit (USA) and surrounding industrial areas, Windsor (Canada), Port Huron (USA), and Sarnia (Canada). This campaign included daily forecasting, stationary ground measurements, several ground mobile laboratories and instrumented aircraft flights.
Mobile laboratories are a useful tool in urban and industrial environments [
1], as they allow for good spatial coverage of multiple species of interest. Monitoring networks, such as the one operated by Michigan EGLE, provide long-term trends but are limited by the number and location of sites. In contrast, a mobile laboratory can provide detailed street-by-street mapping of pollutants for a defined timespan [
2]. Mobile laboratories also excel at point source measurements, since they adapt easily to changing wind directions and can follow concentration enhancements upwind to their sources [
3]. Highly-equipped and rapid-response mobile laboratories can also provide ratios of different species for each source. Finally, mobile laboratories measure concentrations at ground level, where people live, which is particularly relevant for hazardous air pollutants, and for campaigns in dense urban areas.
Mobile laboratory point source sampling involves driving downwind of a facility, in a direction roughly perpendicular to the wind, to measure a “plume”. A plume is an enhancement over background of one or more chemical species. Measurements on surrounding roads, and in different wind directions are used to help separate the contributions of the facility in question from other potential sources in the area. Downwind methods are limited by the prevailing wind and road access around a site. This can be mitigated by a flexible sampling schedule where sites are visited when the wind direction is favorable [
4]. Downwind point source measurements can also prove challenging in dense source areas. Sampling strategies for such areas include driving loops around facilities to separate neighboring emitters [
3] and conducting repeated measurements in dense neighborhoods under different winds [
2].
A common alternative methodology for emission fingerprinting is stack testing, where a probe is placed in the exhaust stack of a site, or at the exhaust of a sub-component of the facility such as a tank-top vent [
5]. Stack testing does not suffer from ambiguity in source attribution, but such studies are time consuming, dependent upon site access negotiations and safety protocols, and may require offline sampling (collecting an air sample for subsequent analysis). They rely on a human operator to identify the sampling spots, and so may miss leaks at inaccessible or unusual locations. Such methods are related to “leak detection and repair” (LDAR) programs that are in place at some large industrial sites such as refineries.
In this study, the Aerodyne Mobile Laboratory (AML) [
1,
4,
6] conducted mobile and stationary measurements of trace gases and VOCs in the SEMI region for six weeks between 21 May 2021 and 30 June 2021, during the MOOSE campaign. The AML was stationed overnight in Dearborn, Michigan, USA at the Salina Elementary School monitoring station operated by the Michigan Department of Environment, Great Lakes and Energy (EGLE). The AML measurements focused on study goals of the CHEmical Source Signatures experiment (CHESS), which included measurement of ozone precursors at key EGLE monitoring stations; and characterization of downwind emission plumes from individual point sources, area sources, and industrial sectors—the topic of this manuscript. Here, we characterize emissions based on the relative molar ratios of species that are enhanced over background in each plume. This is called a “chemical signature” or a “chemical fingerprint”. Next, we examine emissions in a dense industrial area by conducting repeated drive loops under different wind conditions. Finally, we examine plumes from cross-border emissions using measurements from two mobile laboratories. Trends in emission fingerprints and the challenges of emissions characterization in dense industrial areas are discussed.
3. Results
3.1. Point Source Chemical Fingerprints
Several facility categories were visited during the MOOSE campaign. These include: automakers; steel manufacturers; solvent users; fine chemical manufacturers and chemical waste sites; power plants; biogenic sources such as landfills and waste water treatment plants; natural gas sources such as compressor stations and distribution network natural gas leaks; and oil and gas sources such as refineries and terminal fuel storage tanks.
87 distinct point sources were visited as a part of the MOOSE campaign (
Figure 3 and
Table S1). Certain large or high-priority sources were visited several times. Not all sources visited exhibited clear or attributable emissions, despite good winds and mobile lab positioning (see discussion), and additional observations would be required to rule out emissions at these sites.
The measurements here consist of concentrations and spatial maps of a vast suite of trace gas, VOC and combustion products. The quantification of emission magnitudes is beyond the scope of this manuscript, though such efforts are underway via inverse modeling efforts by Michigan EGLE [
24].
In
Table 1, we summarize point source fingerprints for select facilities of different types. The facility ID is listed along with a general category, and a list of noteworthy species. Fingerprint ratios are listed, with the denominator of the molar ratio defined either as the total of C6–C9 aromatics, or as CH
4, depending on the source. In the following paragraphs, each listed source is discussed.
3.1.1. MA130: Industrial Coatings
Facility MA130 develops and manufactures coatings for a variety of applications including automotive, pipeline and electrical insulation. An example downwind transect showing chemical ratios is shown in
Figure 4. The facility was visited twice, on 23 May 2021 and 4 June 2021
(SI Section S4.1, Figures S13 and S14). There was significant variability in the chemical signatures of this site, even within the same day, particularly in the acetone ratio relative to the sum of aromatics. One possible explanation is that the acetone originates from a different sub-source at the site (e.g., a different room’s stack emissions) than the aromatics. This is consistent with observations of a slightly poorer correlation to the sum (R
2 = 0.82,
Figure 4) versus any other single aromatic (R
2 0.91–0.99,
Figure 4).
3.1.2. MA237: Industrial Cleaning
Facility MA237 is an industrial cleaning facility that cleans bulk containers or totes with solvents. It was visited three times, with successful measurements on 15 June 2021 and 25 June 2021
(SI Section S4.2, Figures S15 and S16). Interestingly, at this site, the chemical signature varied significantly between the two visits: C
6H
7+ enhancements were negligible on 15 June but present on 25 June. Acetone was present on 15 June but absent on 25 June. We observed minor but well-correlated natural gas emissions during these plumes but do not attribute them definitively to the site due to their spatial location.
3.1.3. SA96: Adhesives Manufacturer
Site SA96, an adhesives manufacturer, is dominated by toluene emissions, with minor correlated phenol (C
6H
7O
+ via Vocus-PTR-ToF). SA96 manufactures adhesives, packaging, and construction materials. Some of their products include carton-sealing tapes, house flashing, lumber wrap, mailers, shrink film, and specialty fabrics. Raw materials include polyethylene resin, paper, and adhesives, among others. SA96 was visited on 29 May 2021 and 10 June 2021, with an additional observation on 2 June 2021 en-route to other sites
(SI Section S4.3, Figures S17–S22). In 2020, the EPA reported air releases of toluene (982,858 lbs.) from this facility as part of the Toxics Release Inventory Program.
3.1.4. WA236: Chemical Waste
Site WA236 is a chemical waste company with on-site storage. This facility is near several other sources including WA248, a facility treating waste oils and wastewater, and two automaker facilities: WA137, an assembly plant, and WA27, an engine plant
Figure S23). The chemical waste site WA236 dominates emissions of aromatics and other VOCs in this area. On several occasions, the AML tracked the plume far into residential neighborhoods
(SI Section S4.4, Figures S24 and S25).
The clearest separation of the automaker assembly plant WA137 and chemical waste facility WA236 occurred on 26 May 2021 with wind from the SE (
Figure 5). In this figure, a mixed VOC plume (@ symbol) is observed along with a broader, primarily acetone plume (* symbol) further to the north. Sharp and brief spikes of aromatics and CO show the impact of local traffic on these busy boulevards. We attribute the southern-most plume (@ symbol, enhancements of many VOCs) to the WA236 chemical waste facility. The northern-most plume (* symbol, primarily acetone on this day) originates from the automaker assembly plant WA137 or nearby.
Nearly all reported species by the Vocus increase during the plume from this site including C
4H
9O+ (methyl ethyl ketone + butanal), C
3H
5O+ (acrolein) and C
6H
7O+ (phenol). GC-ToF measurements of the WA236 chemical waste site show notable enhancements of halocarbons, primarily dichloromethane (CH
2Cl
2), aromatics and acetonitrile (CH
3CN) (
Figure S26). PCBTF, a paint solvent, is also elevated. Positive Matrix Factorization (PMF) [
25,
26,
27,
28] was used on the full Vocus mass spectral dataset to separate the chemical waste signature (WA263) from other nearby sources in a plume propagating through the neighborhood
(SI Section S4.4, Figures S27–S30). This analysis yielded key ions present in the WA236 plume (
Table S8). These ions and their potential chemical assignments are listed in the
SI, and include acetone or propanal (
m/
z 59.049), and select aldehydes or ketones (
m/
z 73.065).
The sum of the measurements in this area leads us to several conclusions. The chemical waste disposal facility WA236, to the south of the automakers, dominates emissions of aromatics and other VOCs, including odorous oxygenated VOCs, in this area. On several occasions, the mobile lab tracked the plume far into residential neighborhoods. The automaker assembly plant WA137 may also be emitting a mix of acetone and/or aromatic species. Only a few measurements were carried out around WA27, the engine plant, limiting our ability to discern emissions from this site. Several additional sources in this region contribute to a complex source environment, including WA248, the waste oils facility in the same block as the WA236 chemical waste facility, and an unknown VOC source north of WA27 with ppm-level measured aromatic concentrations.
3.1.5. MA141: Natural Gas Compressor Station
Site MA141 is a natural gas compressor station, visited twice on 23 May 2021 and 15 June 2021
(SI Section S4.5, Figure S31). In contrast to many of the other industrial sources described here, MA141 is in a rural area, isolated from other nearby sources, which simplifies measurements and attribution. As expected, the main observed emissions are methane and ethane, the components of natural gas, which are perfectly correlated (R
2 = 1.00). The ethane/methane ratio changes slightly between visits, with a ratio of 0.081 on 23 May and a ratio of 0.073 on 15 June, likely reflecting the makeup of the compressed gas itself. These ratios are slightly higher than expected based on Michigan average heating values of natural gas consumed by month of 1058 BTU (May 2021) and 1057 BTU (June 2021) [
29]. These heating values correspond to ethane/methane ratios of approximately 0.064 and 0.062, assuming no other components in the gas besides ethane and methane. However, the gas transiting through the MA141 compressor station may not be destined for Michigan consumers or may not reflect the state average. Other species roughly correlated with the natural gas plume are HCHO and NOx; enhancements in CO
2 are not clearly distinguishable above instrument noise, and CO is uncorrelated since it is dominated by sharp plumes from other sources such as traffic. For this reason, we report only the ratios of HCHO and NOx to CH
4 and only for those plumes with R
2 > 0.75. Combustion tracers are expected at compressor stations since the compressor engines themselves run on natural gas, with a certain amount of “slip” (unburned natural gas) escaping with the compressor exhaust [
20].
3.1.6. WA238 and WA240: Natural Gas Distribution Network Leaks
Natural gas plumes containing correlated ethane and methane with no other correlated tracers were common in the study area. Two spots in particular (WA238 and WA240) were observed repeatedly throughout the campaign with methane enhancements in the part-per-million level in an area which we refer to as the Dearborn Loop (see
Section 3.2 for a more detailed description of this area). Their ethane/methane ratios are summarized in the table below with additional discussion in
Section 3.2. The ethane/methane ratios of 0.06–0.09 are similar to those measured at the MA141 compressor station discussed previously and are consistent with the expected ethane/methane ratio in distribution grade natural gas.
3.1.7. WA0 and WA87: Steel Manufacturer and Automaker
Permission was obtained to drive on facility property for a major source area along the Dearborn Loop: the complex comprised of an automaker (Facility WA87) and steel manufacturer (Facility WA0). In the
SI (Section S4.6, Figures S32–S38), we describe 5 unique aromatic plume fingerprints, plus a 300 m section of road that showed as many as 4 overlapping plumes with different signatures. This facility duo is complex enough to warrant a dedicated study of its own.
3.1.8. WA22: Refinery
Finally, site access was secured at the refinery in Dearbon (Facility WA22). These results are presented in the
SI (SI Section S4.7, Figures S39–S47), and similar to the automaker/steel manufacturer WA87/WA0 described above, demonstrate that no single chemical fingerprint is appropriate for such large and complex facilities. In the following section, we describe an alternate sampling strategy used in the dense industrial area surrounding the refinery and automaker/steel manufacturer.
3.2. VOC Concentrations in an Industrial Area
Dearborn and River Rouge are two cities in Wayne County, Michigan that border Detroit to the west. This area, including the south-western-most part of Detroit, is home to numerous industrial facilities, including automakers, steel manufacturers, a refinery, power generation, terminal stations, rail yards, and more. Residential and shopping areas are in these cities as well. The area is bisected by the Rouge River, a tributary of the Detroit River, used for shipping. The state-run Dearborn monitoring station used as the home base for the mobile laboratory is in this area, and as such, a wealth of measurement data was collected at and around the Dearborn site.
The density of sources in Dearborn and surrounding areas prompted a different sampling strategy than the other point sources targeted during MOOSE. In Dearborn, we defined a standard route that looped through and around the dense source area. This “Dearborn Loop” was repeated multiple times throughout the campaign, at different times of day, and during different dominant wind directions. Such a sampling strategy can allow for triangulation of observed emissions under different wind conditions to a given point source.
Dominant wind directions, as measured at the Dearborn Site were from the SW, SSE, NNW and E directions. The mobile wind measured during these loops shows similar trends, though with less clear distinctions between primary directions, which we attribute to the challenges of measuring wind while driving, and to real wind variability within street canyons. Wind rose plots are shown in the
SI, (Section S4, Figures S48 and S49). The wind data suggest defining three general wind flows: Southwesterly (135–270); Northwesterly (270–360); and Easterly (0–135).
Figure S50 shows a breakdown of the driven loops by wind direction.
Given the spatial and chemical complexity of sources in this area, we focus on a few key indicators: (1) the sum of C6–C9 aromatics, expected from fuel storage, refinery operations and storage, paint, coatings and solvent use, and combustion; (2) ethane, expected from natural gas leaks, combustion sources, and select refinery sources; and (3) carbon monoxide, expected from traffic, generators, and other industrial combustion sources.
Figure 6 shows Dearborn Loop average concentrations for the sum of aromatics under southwesterly winds. The Dearborn Loop route is about 8.2 square kilometers in size and the bin size is 0.001 decimal degrees. Additional Figures in the
SI show analogous maps for other wind directions
(SI Section S4.1.1, Figures S51–S53). Under Southerly winds, we see aromatic hotspots downwind (east) of the automaker (Facility WA87) and steel plant (Facility WA0), which are nestled together in the same area. Enhancements are also present downwind of the petroleum terminal (eastern-most section of the refinery outline) and sections of the Rouge River. Aromatic enhancements are also observed on the elevated highway that transects the loop, with the exact location of the enhancements varying depending on wind direction. These highway enhancements could include contributions from highway traffic itself or to Rouge River or refinery sources. Refinery impact is also observed under Northerly flow. Distinct enhancements in aromatics are observed near a petroleum terminal.
Ethane hot spots across all wind directions
(SI, Section S4.1.2, Figures S54 and S56) show persistent natural gas leaks at several points of the route. One leak in particular (WA238) was routinely measured under an overpass, where natural gas may have been accumulating. Model estimates of this leak have been performed by Olaguer [
30], and this leak and other natural gas distribution leaks were sampled by Batterman et al. [
31] Under Southerly flow, a persistent ethane (and CH
4) signature was present downwind of a natural gas power generation plant (Facility WA13), but the absence of well-correlated combustion tracers, and the proximity of the transect to the source, suggests a ground-level leak of unburned natural gas.
Finally, CO emissions
(SI Section S4.1.3, Figures S57–S59) show persistent enhancements downwind of the automaker and steel complex in both Southwesterly and ENE winds. Enhancements along roadways, particularly the section of elevated highway crossing the River Rouge are evident.
3.3. Cross-Border Emissions
In previous sections, we examined emissions from facilities in the SEMI study area. Here, we show that facilities on the Canadian side of the international border impact the Michigan airshed. The AML spent time sampling in and around Port Huron, Michigan, which is across the St. Clair River from Sarnia, Ontario. Sarnia is home to a dense cluster of refineries and petrochemical facilities [
32], and one of the goals of these measurements was to investigate trans-border transport of emissions. Measurements in Port Huron were also coordinated with the Ontario mobile laboratory run by MECP: the Trace Atmospheric Gas Analyzer or TAGA [
32]. Both laboratories drove riverside routes at the same time. The Canadian riverside was not accessible for the entire route. No HCHO was available on the TAGA. The SO
2 analyzer was not on board the AML during this drive.
Figure 7 below shows a river-side drive of the AML and MECP TAGA from south to north, with dominant winds from the refineries to the East. Though there are numerous individual facilities on the Canadian side, three refineries dominate the area. We number them clusters 1, 2 and 3 from south to north. Just north of Cluster 2 are two additional petrochemical sites: a rubber manufacturer with reported 1,3-butadiene emissions, and a styrene manufacturer. South and inland of Cluster 1 is another petrochemical plant producing ethylene. The map in
Figure 7 clearly shows enhancements of aromatics on both sides of the border surrounding Clusters 2 and 3, whereas concentrations are lower near the Cluster 1 facilities to the south.
We also observe three distinct formaldehyde plumes on the US side, with enhancements above background in the 4–5 ppb range. Hydrocarbon and aromatic tracers (C8 aromatics shown) are also enhanced, though not perfectly correlated with the two northern-most plumes associated with clusters 2 and 3. Only modest levels of hydrocarbons and aromatics are observed downwind of the southernmost cluster 1.
Figure S61 shows HCHO data as a function of time during this transect along with additional tracers.
The observed enhancements in C8 aromatics downwind of the Cluster 2 area agrees with available emission inventory reports [
33]. The Cluster 1 refinery reports 9 tonnes of xylene annually; a refinery in Cluster 2 reports 18 tonnes annually, and for Cluster 3, a refinery/chemical plant reports about 9 tonnes of xylenes annually. However, located in the Cluster 2 area close to the refinery is a separate styrene production facility that releases 14 tonnes of ethylbenzene annually (for comparison the other sites release only 7, 2, and 2 tonnes of ethylbenzene annually). The Cluster 2 area is very much the dominant C8 aromatic release area. Thus, the combined emissions of total C8 aromatics for the three areas are Cluster 1: 16 tonnes, Cluster 2: 34 tonnes, Cluster 3: 11 tonnes.
Next, we investigate the three broad HCHO plumes observed. HCHO emissions may originate from combustion processes, and refinery operations include numerous such processes. Examining the combustion tracers CO and CO
2 (
Figure S61), and ignoring narrow spikes due to traffic on the drive route, we observe broad increases moderately correlated with HCHO downwind of Cluster 3 and the northern part of Cluster 2, but not with Cluster 1.
A second possible explanation for enhanced HCHO is rapid atmospheric oxidation of reactive alkenes. Formaldehyde plumes have been observed downwind of refineries in Houston, Texas, for example, and attributed to reactive hydrocarbon emissions from refineries [
34]. No 1-s data aboard the AML is reported for reactive alkenes, but two 10-min samples measured via GC-EI-ToF showed elevated propene at concentrations ranging from of 0.58–1.9 ppb in Port Huron on this day (background concentrations in Port Huron on this day were about 0.25 ppb, at a time when fast instrumentation showed no VOC plume). The mobile GC measurement of 1.9 ppb included 3-s of exhaust, and so may be contaminated by traffic. However, stationary samples were taken riverside in Port Huron later this day, with wind from the South, yielding propene at 0.58 ppb. These stationary measurements are likely to have sampled Canadian petrochemical Clusters 1 and 2 shown in
Figure 7 based on wind direction. A full breakdown of the GC measurements is shown in the
SI.
4. Discussion
As part of this study, we observed characteristic emission signatures from different types of industry. Automakers in the study area are characterized largely by VOCs from paint. Automakers also have engine plants, where exhaust measurements are expected, but these emissions were not clearly distinguishable from surrounding traffic, due to the location of such facilities in dense trafficked areas. Industrial/chemical sites (e.g., coatings, solvent use) are characterized by solvent emissions related to their respective processes. Compressor stations are characterized by natural gas emissions and combustion exhaust species. Landfills are dominated by methane and biogenic VOCs without correlated ethane, with some combustion tracers possible depending on combustion equipment on site. Other ubiquitous emission sources not highlighted here include gas stations, and on-road exhaust measurements.
Measured chemical fingerprints can be compared to profiles tabulated in EPA’s SPECIATE database [
17]. This database reports emission profiles by weight, to the total weight of VOCs. Industrial solvent emissions, the SPECIATE category that would include the coatings (MA130), industrial cleaning (MA237) and solvent use (SA96) sites lists numerous source signatures or “profiles”. No direct matches are obvious, though the SPECIATE profiles vary dramatically in composition, just as with the range of ratios reported here. Another source category of interest is automotive paint (for example, profile 2546 in SECIATE). Molar ratios to summed aromatics are determined using the reported emissions profile in weight % of total VOCs, and individual species molar masses. The SPECIATE reference signature was dominated by toluene (C7 0.6 molar ratio to C7–C9 aromatics, no benzene reported), then the C8 aromatics (0.3 molar ratio). The C9 molar ratio is 0.1. Acetone had a molar ratio of 0.21 to the summed aromatics. This reference profile is within the measured ratios seen at the WA87/WA0 Automaker/Steel manufacturer for C8 and C9 aromatics, but exceeds the measured ratios for acetone and toluene. We note that other processes besides painting are occurring at this pair of sites, and that the SPECIATE profile predates (1989) the use of low-volatility solvents such as PCBTF.
The measurement of industrial point sources presented several challenges, notably related to source density, source complexity, source height, and the combination of wind direction and road access. Point sources in isolated areas, with surrounding road access and predictable source signatures, were the easiest to characterize. Examples of this type of source include landfills or compressor stations, which tend to be in more rural areas, and have significant emissions dominated by methane. Certain VOC point sources located outside of dense industrial areas also met these criteria, including the industrial cleaning facility MA237, the industrial coatings facility MA130, and the solvent use facility SA96.
The above sources also often were simple in their chemical and spatial emission characteristics (a single central emission point, with just a few chemical species). Other sources measured were much more complex and are better thought of as a collection of point sources. These include the refinery in Dearborn WA22, and the combined automaker/steel plant WA87/WA0. Cross-border refinery and petrochemical emissions from Canada also fall into this more complex category.
For complex sources and dense industrial areas such as Dearborn, neighboring point sources often present overlapping emissions in space. One sampling strategy that was developed was driving repeated loops through these dense areas under different wind conditions. This is discussed in greater detail in
Section 3.2. Many of the facilities along this route are large and complex, with no public road access within their fenceline, such as the WA22 refinery and the dual-complex automaker/steel maker (WA87/WA0) at the heart of the loop. Facilities such as these require a dedicated study to completely characterize. Though measurements such dense areas may not fully characterize individual sources, the data collected by the AML during MOOSE is useful in evaluating models by comparing real measured concentrations to gridded model outputs; such modeling efforts may then yield source contributions [
24].
Several facilities targeted for analysis had emissions expected from stacks at height. Examples of such facilities include power plants, refineries, and large chemical plants. In the case of power plants, the AML was often able to measure emissions of unburned fuel at ground-level (e.g., ethane and methane leaks at the WA13 power generation plant). Detecting combustion emissions from the stacks, which would require far-away transects in good winds, were most often difficult or impossible to distinguish above traffic and other nearby sources. A section of elevated highway along the Dearborn Loop provided an interesting opportunity to transect a refinery at height; however, refinery emissions proved difficult to distinguish from on-road traffic.
The fingerprints determined here were generally for fresh plumes (minute(s) old), measured near to their respective sources (e.g., hundreds of meters). Most species discussed (e.g., toluene; ethane) have photochemical lifetimes of days or months, and thus do not have time to undergo significant atmospheric processing before being measured. A notable possible exception are the emissions from Canadian refineries (
Section 3.3), measured 1–3 km downwind. In that case, we see evidence of propene, a reactive alkene, measured by GC-EI-ToF. We also observe distinct HCHO plumes, a species that can be both directly emitted but also produced as an intermediate in the oxidation of reactive compounds.
Certain facilities targeted for measurement exhibited no clear emissions even under favorable wind and road access conditions. Ruling out emissions from a given facility is much more difficult than a positive detection, particularly in dense source areas. Alternate sampling strategies would be required to rule out emissions from many of these sites, such as direct stack testing, or tracer-release measurements at the site.
Finally, the flexibility afforded by a mobile laboratory allowed scientists to find unexpected sources of emissions such as VOCs and trace them back to their source. In one such example, the WA236 chemical waste facility had emissions that dominated an area that included physically larger and more prominent automakers, and impacted an area spanning several residential and commercial blocks. This example, and others, demonstrate the power of mobile laboratories over stationary sampling in dense industrial areas.