Fuel-based fine particulate and black carbon emission factors from a railyard area in Atlanta

Abstract

Railyards have the potential to influence local fine particulate matter (aerodynamic diameter ≤2.5 μm; PM_2.5) concentrations through emissions from diesel locomotives and supporting activities. This is of concern in urban regions where railyards are in proximity to residential areas. Northwest of Atlanta, Georgia, Inman and Tilford railyards are located beside residential neighborhoods, industries, and schools. The PM_2.5 concentrations near the railyards is the highest measured amongst the state-run monitoring sites (Georgia Environmental Protection Division, 2012; http://www.georgiaair.org/amp/report.php). The authors estimated fuel-based black carbon (BC) and PM_2.5 emission factors for these railyards in order to help determine the impact of railyard activities on PM_2.5 concentrations, and for assessing the potential benefits of replacing current locomotive engines with cleaner technologies. High-time-resolution measurements of BC, PM_2.5, CO₂, and wind speed and direction were made at two locations, north and south of the railyards. Emissions factors (i.e., the mass of BC or PM_2.5 per gallon of fuel burned) were estimated by using the downwind/upwind difference in concentrations, wavelet analysis, and an event-based approach. By the authors’ estimates, diesel-electric engines used in the railyards have average emission factors of 2.8 ± 0.2 g of BC and 6.0 ± 0.5 g of PM_2.5 per gallon of diesel fuel burned. A broader mix of railyard supporting activities appear to lead to average emission factors of 0.7 ± 0.03 g of BC and 1.5 ± 0.1 g of PM_2.5 per gallon of diesel fuel burned. Railyard emissions appear to lead to average enhancements of approximately 1.7 ± 0.1 µg/m³ of PM_2.5 and approximately 0.8 ± 0.01 µg/m³ of BC in neighboring areas on an annual average basis. Uncertainty not quantified in these results could arise mainly from variability in downwind/upwind differences, differences in emissions of the diverse zones within the railyards, and the influence of on-road mobile source emissions.

Implications:

In-use fuel-based black carbon and fine particulate emission factors for railyard activities were quantified by novel approaches using near-source high-time-resolution monitoring of ambient concentrations at two sites. Results can reduce the uncertainty in railyard emission inventories and the approach can be replicated and extended to assess trends and evaluate emission reduction alternatives.

Supplemental Materials: Supplemental materials are available for this paper. Go to the publisher's online edition of the Journal of the Air & Waste Management Association for information on dates of photographic surveillance of Inman yard, the algorithm for wavelet analysis, histograms and time series plots of Downwind/Upwind data, and boxplots of BC emission factors.

Introduction

Railyard emissions are thought to originate largely from diesel-electric locomotives called “switchers” that are used to gather cars and assemble them into trains. Switchers are potentially high emitters because they are typically older model locomotives and have low-power duty cycles (U.S. Environmental Protection Agency [EPA], 2011a). Emissions from switchers include primary fine particulate matter (aerodynamic diameter ≤2.5 μm; PM_2.5), elemental and organic carbon (EC/OC), nitrogen oxides (NO_x), sulfur dioxide (SO₂), hydrocarbons, carbon monoxide (CO), and carbon dioxide (CO₂). Diesel emissions have suspected negative effects on human health (World Health Organization [WHO], 2012). Black carbon (BC) from diesel and other fossil fuels absorb solar radiation, affecting visibility (Prasad and Bella, 2010) and climate (Roberts and Jones, 2004). Railyards have been identified as local sources of particulates (Kam et al., 2004, 2011), EC/OC (Sawant et al., 2007; Cahill et al., 2011), NO_x (Starcrest Consulting Group, 2004; Cahill et al., 2011), CO₂, SO₂, metals, and polycyclic aromatic hydrocarbons (PAHs) (Cahill et al., 2011).

The contribution of particulate matter from railyards to U.S. emissions, as estimated in the National Emissions Inventory (NEI), is small compared with on-road mobile sources or power plants (EPA, 2012). Switcher locomotives have been estimated to emit less than 0.1% of the total PM₁₀ (PM with an aerodynamic diameter ≤10 μm) and PM_2.5 in the United States (EPA, 2008a). Yet, emissions from railyards located close to residential areas are of new interest because of recent regulations (EPA, 2008b), intensity of operations in limited areas, and the fast growing economic activity of switchyards and intermodal terminals (Laurits R. Christensen Associates, 2009).

In Atlanta, PM_2.5 concentrations have been decreasing over the past 10 yr (Electric Power Research Institute [EPRI], 2012; Georgia Environmental Protection Division [Georgia EPD], 2012), but near Inman and Tilford railyards, the Fire Station 8 site (FS) has consistently showed the highest annual average PM_2.5 concentration reported at any of the Georgia state-run monitoring locations (Figure 1). Georgia EPD (2009) applied the American Meteorological Society/Environmental Protection Agency Regulatory Model (AERMOD) using emission estimates based on NEI methodology, and found that railyards contribute approximately 1.9 µg/m³ to the concentration of PM_2.5 at FS.

Figure 1. PM_2.5 annual arithmetic means at Atlanta urban sites (Georgia EPD, 2012).

Railyard emissions are viewed as highly uncertain (Simon et al., 2008). Recently, a 27-state committee called ERTAC Rail developed top-down nationwide railyard, line-haul, and shortline/regional emission inventories for the years 2007/2008 using confidential information from the railroad companies (Bergin et al., 2012). This inventory was used to update the 2011 EPA - NEI. Previous NEIs used the conventional approach quantify railyard emissions. Inventories were calculated multiplying state-level yearly average fuel consumption data by nationwide fleet average fuel-based yard emission factors. States currently estimate railyard emissions using methods based on the same approach (Sierra Research, 2004). Sources of uncertainty are estimated fuel use, distribution of consumption data to each rail category (i.e., switcher vs. Class I locomotives), allocation of emissions to county level using local activity data (National Cooperative Freight Research Program [NCFRP], 2010), and yard emission factors that don't necessarily represent the variability in engine technologies, specific yard operating conditions, and the yard fleet mix (Simon et al., 2008). Furthermore, the yard emission factors may not adequately account for yard-associated emissions (i.e., emissions from testing and maintenance of locomotives and drayage trucks) (Fritz and Cataldi, 1991) Disaggregated fuel consumption data required to address the fuel related sources of uncertainty are unavailable mainly because companies view fuel consumption as proprietary information (NCFRP, 2010).

Railyard emission estimates are developed mainly using emission factors for switchers that are an average of engine emissions over a cycle of stationary sequential operation at low and normal idle, and at eight other discrete power levels, called notches, weighted by numerical factors that reflect the time the engine is operated at each notch (CFR-40-92.101-133, 2011). These emissions factors have high reproducibility but may not represent real-world emissions from particular operating conditions (St. Denis et al., 1994; Cocker et al., 2004) and they may not have a quantitative indication of uncertainty. Previous work has been directed to obtain real-word emission factors from small samples of diesel-electric switcher locomotives measuring directly from the stack, varying fuel or type of engine (Fritz and Cataldi, 1991; Honc et al., 2006; Sawant et al., 2007), but little work has been aimed at quantifying their uncertainties, or to estimate emission factors that account for actual activities going on in and around railyards.

The objective of this work is to advance the understanding of railyard emissions by estimating PM_2.5 and BC fuel-based emission factors to reduce uncertainty in emission inventories. The emission factors will account for the particular operating conditions of the railyards using near-source high-time-resolution monitoring. This information may be used to improve air quality modeling results, aid in the development of effective air quality management strategies, and, as part of a joint government industry project (Congestion Mitigation and Air Quality Improvement Program, Georgia Environmental Protection Division [CMAQ-EPD], 2009), to assess the improvement in local air quality as cleaner technologies replace old switcher engines used at Inman and Tilford railyards.

Experimental Methods

Monitoring sites

The study was carried out in Atlanta, Georgia, at locations near Inman and Tilford railyards (Figure 2). CSX's Tilford Yard is a hump terminal that handles approximately 80 trains per week and operates 10 switcher locomotives (Georgia EPD, 2009). Inman Yard is a large Norfolk Southern intermodal facility with 14 switcher locomotives (Georgia EPD, 2009). The yards are adjacent to each other, northwest of downtown Atlanta, inside the perimeter freeway I285 (Figure 2). Other pollution sources in the area include Howells Yard (a small intermodal yard with 15 tracks), Georgia Power Company's McDonough-Atkinson Plant, Ennis Paint, and a Metropolitan Atlanta Rapid Transit Authority (MARTA) garage facility. The McDonough-Atkinson Plant was being converted from coal to natural gas during this study.

Figure 2. Schematic of the location of the study. The two monitoring sites are at the Dixie (DX) and Fire Station 8 (FS) locations.

Two monitoring sites were used: Fire Station 8 (FS) (coordinates: 33.80176°N, −84.43559°W) and Dixie (DX) (coordinates: 33.79080N, −84.44026°W), north and south of the railyards (Figure 2). Sites are 1.3 km apart. The FS site is part of the Assessment of Spatial Aerosol Composition Network (ASACA) (Butler et al., 2003) and is located at approximately 300 m of the arrival section of Tilford Yard and 30 m of Marietta Boulevard NW (17,000 annual average daily traffic [AADT] approximately), which runs between the FS site and the railyards. Other roads with less traffic (>2000 AADT), such as Marietta Road, Bolton Road, and Perry Boulevard, surround and run through the railyards. DX is contiguous to the intermodal terminal at Inman Yard, approximately 80 m from the tracks. The MARTA garage is located southwest of DX.

Air pollutant measurements

BC (multiangle absorption photometer [MAAP]; model 5012; Thermo Scientific, Franklin, MA), PM_2.5 (1400ab tapered element oscillating microbalance [TEOM]; R & P Thermo Scientific, Franklin, MA; operated at 50°C), wind speed and wind direction (Young 03002-L wind sentry set; Young-Campbell Scientific, Logan, UT) were measured from December 2010 to December 2011 at both sites. CO₂ (NDIR 41i analyzer; Thermo Scientific, Franklin, MA) was measured from April to December 2011. Coarse particles were removed from the TEOM and the MAAP sample lines by model 2000-30EH 16.7 liters per minute (LPM) 2.5-μm cutoff cyclones (URG, Chapel Hill, NC). Three meters of 1/8 inch outer diameter (OD) Teflon tubing was used to draw 1 LPM to the CO₂ monitors. Samples were taken at a height of approximately 3 m. One-minute averages of all variables were logged as a text file to a field computer and later loaded to a database. CO₂ analyzers were calibrated with a CO₂ certified standard Nexair gas mixture. Railyard operations were recorded from the DX site using a camera (Hero Gopro 960; Woodman Labs, Inc., Half Moon Bay, CA) to take photos every minute on 42 days between September 15, 2011, and November 14, 2011. A table with the specific dates is available as supplemental material.

The pairs (one for each site) of CO₂ analyzers, TEOMs, and MAAPs were run for 2 weeks side by side at the Georgia Tech campus before deployment. One-minute concentrations measured with CO₂ analyzers and MAAPs were within 5%. Thirty-minute PM_2.5 concentrations reported by the TEOM instruments were within 5%. During monitoring at the railyards, zero and span checks of the CO₂ analyzers and flow checks for the TEOMs and MAAPs were carried out on weekly basis and monthly basis, respectively.

Data analysis

We applied the carbon balance method (Singer and Harley, 1996) to calculate fuel-based emission factors, relating the amount of pollutant emitted to the amount of fuel burned ( eq 1):

1

where is the emission factor in units of grams of pollutant emitted per gallon of fuel burned, is the ratio of the mass of pollutant to mass of carbon from CO₂, and is the ratio of the mass of other carbonaceous species, such as unburned hydrocarbons or CO, to the mass of carbon form CO₂. Three methods were used to calculate , including what we refer to as the “delta,” the wavelet, and the regression approaches discussed below. It was assumed that CO₂ dominates the carbon balance for the railyard diesel sources, with carbonaceous species besides CO₂ (e.g., hydrocarbons and CO) playing a minor role in the carbon budget (Yanowitz et al., 2000). Consequently, is assumed to be significantly less than 1 and is neglected in our calculations. is the carbon content per gallon of diesel fuel specified by the Code of Federal Regulations (CFR-40-600.113–78) as 2778 g C/gal. Uncertainties in the properties of the fuel were neglected. All uncertainties reported were calculated as the 95% confidence interval of the mean.

All the approaches to calculate the ratio were based on averages from concentration data occurring when wind with velocities greater or equal to 0.5 m/sec and directions between 320° and 360° and between 0° and 90° at DX and between 170° and 280° at FS were measured. These wind sectors comprise approximately the complete area of the railyards.

The “delta” approach was based on the downwind–upwind difference in pollutant concentrations. The ratio obtained by this method () is in units of mass of pollutant emitted per mass of C ( eq 2):

2

where and are the mean pollutant (BC or PM_2.5) concentration and mean CO₂ concentration respectively in μg/m³, the subscripts DW and UW indicate when the average is from the downwind or upwind site, respectively. The factor of 12/44 is the atomic mass of carbon over the molecular mass of CO₂. The delta approach is thought to represent emissions from a broad mix of railyard sources.

A second method used wavelet analysis (Daubechies, 1992) to separate the concentration signals into high- and low-frequency components (Figure 3). The ratio Q _w calculated by this approach is in units of mass of pollutant emitted per mass of C ( eq 3).

Figure 3. (a) CO₂ concentration at DX from 12:00 a.m. on September 5, 2011, to 11:59 p.m. on September 6, 2011. The CO₂ concentration signal was separated into spikes and background components by wavelet analysis. (b) Spikes in CO₂ concentration.

3

where and are the mean pollutant (BC or PM_2.5) concentration and mean CO₂ concentration, respectively, in μg/m³. The factor of 12/44 is the atomic mass of carbon over the molecular mass of CO₂. It was assumed that the high-frequency components extracted by the wavelet-based algorithm are predominantly near-field emissions from a variety of railyard sources (e.g., drayage trucks, cranes, welding facilities, or switcher locomotives) and from diesel trucks and gasoline vehicles in the surroundings. Low-frequency contributions are assumed to be associated with nonrailyard activities and represent the background concentrations in the vicinity. A MATLAB (MathWorks, Natick, MA) algorithm was used for this analysis and it is available as supplementary material. Wavelet analysis has been applied previously by Klems et al. (2011) to a similar problem in order to determine the contribution of motor vehicles near a roadway intersection to the ambient ultrafine particle mass by correlating high-frequency contributions with fast changes in ultrafine particle chemical composition.

The regression approach, the third technique employed, focused on events of high BC concentration. Events were identified by selecting groups of 5–20 consecutive-minute data points when the maximum BC concentration of the set was greater than the mean plus 3 times the standard deviation of the BC concentrations occurring in the same hour at the same site and when a linear relationship with a correlation coefficient greater than or equal to 0.90 at a 0.95 confidence level between CO₂ and BC concentrations was obtained (Figure 4). Events were selected from data occurring for wind speeds and wind directions with the restrictions described for all the approaches. The ratio, Q _r, was calculated as the mean of the slopes of the BC to CO₂ regressions. The ratio of concentrations was converted to a ratio of mass of BC to mass of C from CO₂ by dividing it by the atomic mass of carbon over the molecular mass of CO₂. The minimum concentration measured during the event was taken as the baseline. This approach is likely to represent near-field brief emission events from a subset of railyard sources (e.g., a passing switcher or line-haul engine). A comparable approach was formulated by Dallmann et al. (2011) to measure BC emission factors from diesel exhaust emissions of trucks used to move containers with in a railyard and by Hansen and Rosen (1990) to measure BC emission factors from automobiles.

Figure 4. Event associated with a locomotive at the DX site on September 17, 2011. At 2:14 p.m. a train passes by the monitoring site. An event is detected shortly after. The subplot shows the lineal regression of the event detected.

Results

Concentrations of BC and PM_2.5

Differences in annual average PM_2.5 concentrations between Georgia EPD Fire Station 8 and other urban sites have become smaller in recent years (Figure 1), due in large part to a combination of factors set in place by the 2008 economic downturn, higher-than-average annual rainfall in 2009 (National Oceanic and Atmospheric Administration [NOAA], 2012), and air quality policies. In 2011, annual average PM_2.5 and BC concentrations at DX and FS were comparable (Table 1). Annual average PM_2.5 concentrations are below the current National Ambient Air Quality Standard (NAAQS; 15 μg/m³), but above the proposed level (12 μg/m³) (EPA, 2011b).

Table 1. Concentrations of PM_2.5 and BC for FS, DX, and other Atlanta urban sites in 2011

Site	Method	PM2.5 (μg/m^3)^†	Method	BC (μg/m^3)^†
FS	TEOM	12.3 (7.1)	MAAP	1.5 (1.4)
DX	TEOM	13.1 (8.0)	MAAP	1.3 (1.2)
Fire Station 8	FRM^a	13.3 (5.8)
South DeKalb	FRM^a	12.4 (6.0)	TOR^b	1.3 (0.9)
Gwinnett Tech	FRM^a	12.5 (6.7)
E. Rivers School	FRM^a	11.4 (5.0)
Notes: ^aFederal Reference Method. ^bThermal/optical reflectance. ^†Mean (standard deviation).

Wind speed and direction and pollutants

During the study, the predominant wind direction was west-southwest at both the DX and FS sites (Figure 5). Average wind speeds of 1.5 m/sec at DX and 1.2 m/sec at FS were measured. The highest speeds were recorded when the wind came from the southeast and southwest quadrants at FS and from the northeast and southeast quadrants at DX. Structures and trees located southwest of DX and northeast of FS could have hindered wind circulation to some extent.

Figure 5. Wind Roses for (a) the DX site and (b) the FS site.

We plotted normalized pollutant concentrations to gain insight on the location of the sources that impact DX and FS (Figure 6). Pollutant concentration roses were constructed by normalizing the concentrations subtracting the mean and dividing by the standard deviation and adding one. Normalized pollutant roses show local concentrations of BC, PM_2.5, and CO₂ approximately 1.5 times greater than average coming from the direction where the railyards are located, that is, the northeast quadrant at DX and southeast quadrant at FS, as their main feature (Figure 6). There is a source of BC, PM_2.5, and CO₂ north of FS. FS could as well be impacted to some degree by BC, CO₂, and PM_2.5 emissions coming from activities on Marietta Boulevard. The roses suggest that BC is a better tracer for yard activities than PM_2.5. At both sites, directions of higher-than-average BC concentrations closely follow the layout of the railyard. PM_2.5 and CO₂ concentration roses at DX show sources south and west-southwest, respectively, but no significant BC is associated with those directions.

Figure 6. Normalized pollutant concentration roses for (a) BC, (b) PM_2.5, and (c) CO₂ at the DX and FS monitoring sites. Downwind sectors are marked in gray.

Somewhat higher concentrations of BC were measured at FS (Table 1). FS downwind conditions were measured 44.5% of the time, whereas DX was downwind 32.5 % of the time during the months of this study. Also, wind speed was slightly lower (1.7 m/sec on average) when FS was downwind than when DX was downwind (1.9 m/sec on average). Greater time downwind with lower wind speeds is one reason for the slightly greater BC concentrations at FS. It was much harder to detect PM_2.5 and CO₂ enhancements from the railyards due to greater background levels and variability for these contaminants, as well as the variety of their sources.

Downwind–upwind differences and high-frequency components

Enhancements in PM_2.5, BC, and CO₂ concentrations come from the directions where the railyards are located. PM_2.5, BC, and CO₂ enhancements are statistically significant (two-sample t tests, with P < 1E-10 in the least satisfactory conditions with 99% confidence). Yet. PM_2.5, BC, and CO₂ downwind–upwind differences have large variability, showing standard deviations much larger than their means (Table 2). This variability will lead to uncertainty in the emission factors calculated by this method. Histograms of downwind–upwind differences and concentrations time series are presented in the supplemental materials.

Table 2. Downwind–upwind concentration differences for the DX and FS sites

	Downwind Site
Pollutant	DX	FS
PM2.5
Mean (μg/m^3)	1.5	1.9
Standard deviation (μg/m^3)	±8.6	±10.8
Uncertainty of the mean (μg/m^3)	±0.1	±0.1
Number of observations	25,105	31,123
BC
Mean (μg/m^3)	0.7	1.0
Standard deviation (μg/m^3)	±1.2	±1.7
Uncertainty of the mean (μg/m^3)	±0.01	±0.01
Number of observations	27,161	40,998
CO2
Mean (ppm)	6.3	5.9
Standard deviation (ppm)	±28.6	±33.2
Uncertainty of the mean (ppm)	±0.3	±0.3
Number of observations	18,110	33,865

At both sites, means of the high-frequency components of PM_2.5, BC, and CO₂ concentrations obtained by the wavelet approach are higher when the wind blows from the railyards than from any other direction. Wavelet analysis helps to rectify the noise and baseline drift of the instruments to a considerable degree, and reduces to some extent the interference of the signals from sources with extremely high frequencies (i.e., fast-moving gasoline vehicles and diesel trucks). This is apparent in the variability of the results of wavelet approach, which is less than the variability of the results of the delta approach (Table 3). Consequently, the uncertainty derived from this variability could be expected to be smaller in the wavelet approach than in the delta approach. Yet, as mentioned before, spikes could be predominantly near-field emissions from a variety of railyard sources but also from diesel trucks and gasoline vehicles. This contribution from nonrailyard sources could still confound the results.

Table 3. High-frequency components from wavelet analysis for the DX and FS sites

	Downwind Site
Pollutant	DX	FS
PM2.5
Mean (μg/m^3)	1.8	1.9
Standard deviation (μg/m^3)	±3.0	±4.0
Uncertainty of the mean (μg/m^3)	±0.02	±0.02
Number of observations	57,908	82,223
BC
Mean (μg/m^3)	0.7	0.9
Standard deviation (μg/m^3)	±1.2	±1.3
Uncertainty of the mean (μg/m^3)	±0.01	±0.01
Number of observations	53,805	73,134
CO2
Mean (ppm)	8.2	6.6
Standard deviation (ppm)	±19.1	±11.1
Uncertainty of the mean (ppm)	±0.1	±0.1
Number of observations	51,711	50,697

Greater enhancements in PM_2.5 and BC concentrations were found when FS was downwind (Table 2). The same result was observed by the wavelet approach. The means of PM_2.5 spikes and BC spikes were greater when wind blew from the railyards to FS than when it was blowing from the railyards to DX (Table 3). Results from this part of our analysis are comparable to those obtained by Campbell and Fujita (2009), at the Roseville railyard in California for 2008 whom measured a downwind–upwind delta of 0.73 ± 0.01 and 1.14 ± 0.01 µg/m³ of BC and 2.5 ± 0.6 and 2.4 ± 0.7 µg/m³ of PM_2.5 at two monitoring sites. Our results support the modeling study by Georgia EPD (2009), which estimated that the railyard emissions led to an additional 1.9 µg/m³ of PM_2.5.

Emission factors

Means of BC and PM_2.5 emission factors obtained by the delta and the wavelet approaches were similar between both sites (Table 4). For both approaches, FS reported higher emission factors than DX. Results obtained at FS could be confounded by emissions from traffic. There is also uncertainty related to the emissions of the different zones within the railyards. FS is located near the arrival section of Tilford Yard, where there is also a turntable and fuel storage and repair facilities. The DX site is close to tracks where a mix of locomotives cruise, accelerate, and idle. The intermodal terminal of Inman Yard where there is heavy-duty diesel truck traffic is also close by. Emission factors calculated by the delta approach when the wind is not blowing from the railyards are presented in the supplemental materials (Table S2). As shown, the small values derived (approximately an order of magnitude less than when using concentrations found from the downwind–upwind pairing) support our results.

Table 4. Emission Factors for the DX and FS sites

	Downwind Site		Other
Emission Factor	DX	FS	Works
EFBC (g BC/gal fuel)
Delta approach	0.6 ± 0.04	0.9 ± 0.05
Wavelet approach	0.5 ± 0.01	0.7 ± 0.01
Regression approach	3.1 ± 0.2	2.4 ± 0.2	3.8^b
(g PM2.5/gal fuel)
Delta approach	1.3 ± 0.1	1.8 ± 0.1
Wavelet approach	1.2 ± 0.02	1.6 ± 0.03
Regression approach^a	7.2 ± 0.8	4.8 ± 0.6	4.5^c, 4.7^d
Notes: ^aPM2.5 emission factor was not calculated directly by the regression approach but estimated from the ratio of BC to PM2.5 and the BC emission factor from the regression approach. ^b Sawant et al. (2007). ^cExpected fleet average PM10 emission factor for 2011 (EPA. 2009). ^d Fritz and Cataldi (1991).

BC emission factors from the regression approach are higher than those obtained from the delta and wavelets approaches (Table 4), which is anticipated because the BC events, identified when BC levels rise by 3 standard deviations or more above the mean value during the hour of the event, are likely due to activities with high BC emissions (i.e., switchers or line-haul engines). Results of the regression approach are comparable to elemental carbon emission factors of 3.8 g of BC per gallon of diesel fuel measured directly from the stacks of switcher locomotives (Sawant et al., 2007). The DX site was equipped to photograph railyard activity to link with pollutant data and investigate the possibility of the recorded events originating from sources other than the railyards. Photos indicate locomotives, either idling or passing by, shortly (1–3 min) before an event was registered. During the event shown (Figure 4), the wind was blowing north-northeast, from the railyards to DX, with speeds that varied between 1 and 2.5 m/sec. The minimum concentration measured during the corresponding hour was taken as baseline. Overlapping signals of concentrations of BC and CO₂ were registered on the downwind monitoring site, whereas the upwind site showed steady concentrations. Photographs also showed that when no locomotives were present and the wind was blowing from the direction of the railyards, BC and CO₂ concentrations were poorly correlated. The scenario depicted in Figure 4 is an example of the many events used to determine the emissions factors by the regression approach.

Events of high BC concentrations detected at DX were generated inside the railyards and were less likely to be influenced by other sources. At FS, there is the possibility that some of events were influenced by traffic on Marietta Boulevard. The regression approach yields a smaller average emission factor for FS (Table 4). Some events with higher BC concentrations were detected at DX, but on average BC concentrations during events show an increase of about 3 µg/m³ of BC at both sites and their respective standard deviations were comparable, as high as 6 µg/m³ and as low as 1 µg/m³ above baseline (Figure 7). Differences between FS and DX regression approach results (Table 4) likely derive from the higher variability in CO₂ concentration at FS. Incremental CO₂ concentrations at FS used in the regression approach show an average and standard deviation approximately 2 and 1.4 ppm greater than at DX (Figure 7), leading to lower emissions factors. Given that BC is found to be a good tracer of railyard activity, and that emission factors calculated by the regression approach show little dependency on the hour of the day or the day of the week (Figures S5–S8), we infer that most of the events detected at FS were generated inside the railyards.

Figure 7. Events of high BC and corresponding CO₂ concentrations at (a) DX and (b) FS. The minimum concentration measured during each event was taken as baseline. Events were centered at the time when the maximum BC concentration was measured (t). Average concentrations 5 min before and 5 min after are shown along with standard deviations (σ) and uncertainties of the mean (σ _x ).

BC emission factors calculated by the regression approach show similar frequency distributions at the two sites (Figure 8), with 423 and 399 events detected at FS and DX, respectively. Several events, likely coming from high-emitting locomotive engines, produced BC emission factors 1 order of magnitude higher than the PM₁₀ emissions standards published by EPA (2009).

Figure 8. Frequency distributions of emission factors obtained from BC events at the FS and DX sites.

Results of the application of the regression approach to estimate PM_2.5 emission factors were less satisfactory and are not presented. This was expected, given the noise in TEOM data on time scales less than 30 min. However, PM_2.5 emission factors could be estimated using the ratio of BC to PM_2.5 obtained from wavelet and delta approaches (0.43 ± 0.02 g BC/g PM_2.5 at DX and 0.5 ± 0.02 g BC/g PM_2.5 at FS). Using these ratios, emission factors of 7.2 ± 0.6 g PM_2.5/gal fuel at DX and 4.8 ± 0.6 g PM_2.5/gal fuel at FS are obtained.

Total BC and PM_2.5 emissions can be estimated based on the fuel use at the railyards and the fuel-based emission factors calculated in this study. Line haul and switching activity at Tilford and Inman railyards consumed 1.6 and 2.5 million of gallons of diesel fuel, respectively, during 2011. This was calculated using the method described (Georgia EPD, 2009), which is based on scaling state-level yearly average fuel consumption dividing the gross ton-miles transported in the yard by system-wide fuel combustion efficiency. Gross ton-miles data have been provided in the past for each railyard by Norfolk Southern and CSX Transportation (Georgia EPD, 2009). System-wide fuel combustion efficiency for 2011 was obtained from data contained in the Class I Railroad Surface Transportation Board R-1 Annual Report from each company (Norfolk Southern, 2011; CSX Transportation, 2011). This estimation does not include the fuel consumed in other activities occurring in the yard. Approximately 11.7 tons of BC and 26 tons of PM_2.5 per year were emitted from the railyards in 2011.

Conclusion

In-use emission factors were quantified for diesel-electric engines and supporting activities at the Inman-Tilford railyard complex in Atlanta, Georgia, using near-source high-time-resolution monitoring of ambient concentrations at two monitoring sites.

Three approaches were used to estimate the emission factors. The delta approach was based on the downwind–upwind difference in concentrations, the wavelet approach analyzed spikes of BC, PM_2.5, and CO₂ concentrations, and the regression approach utilized events of correlated BC and CO₂ concentrations. The delta and the wavelet approaches are thought to represent emissions of a broad mix of railyard sources, whereas the regression approach is likely to represent emissions from switchers and line-haul engines passing by monitoring sites. The average estimated emission factors from the delta and wavelet approaches are 0.6 ± 0.03 g of BC and 1.3 ± 0.1 g of PM_2.5 per gallon of diesel fuel burned at DX and 0.8 ± 0.03 g of BC and 1.7 ± 0.1 g of PM_2.5 per gallon of diesel fuel burned at FS. Emission factors estimated by the delta and wavelet approaches were statistically similar. The regression approach yielded an average emission factor of 2.8 ± 0.2 g of BC and 6.0 ± 0.5 g of PM_2.5 per gallon of fuel.

Railyard emissions led to average enhancements of approximately 1.7 ± 0.1 µg/m³ of PM_2.5 and approximately 0.85 ± 0.01 µg/m³ of BC on an annual basis. Events of high BC concentrations, likely generated by switchers and line-haul engines in the railyards, showed an average increase of about 3 µg/m³ of BC and about 6 ppm of CO₂ above baseline.

Uncertainties not quantified in these results arise in part from variability in downwind–upwind differences, differences in emissions of the diverse zones within the railyards, and influence of on-road mobile sources other than the ones of interest.

Acknowledgment

This work was funded by the Congestion Mitigation and Air Quality Improvement (CMAQ) Program through Georgia Department of Transportation and Georgia Department of Natural Resources. Funding also was provided by Georgia Power. The authors also thank Dixie Driveline & Spring Co. for allowing access to their property to set up and operate the monitoring sites and Universidad de La Salle, LASPAU, and COLCIENCIAS for providing a fellowship to B. Galvis.