High-resolution mapping of motor vehicle carbon dioxide emissions

Authors


Abstract

A fuel-based inventory for vehicle emissions is presented for carbon dioxide (CO2) and mapped at various spatial resolutions (10 km, 4 km, 1 km, and 500 m) using fuel sales and traffic count data. The mapping is done separately for gasoline-powered vehicles and heavy-duty diesel trucks. Emission estimates from this study are compared with the Emissions Database for Global Atmospheric Research (EDGAR) and VULCAN. All three inventories agree at the national level within 5%. EDGAR uses road density as a surrogate to apportion vehicle emissions, which leads to 20–80% overestimates of on-road CO2 emissions in the largest U.S. cities. High-resolution emission maps are presented for Los Angeles, New York City, San Francisco-San Jose, Houston, and Dallas-Fort Worth. Sharp emission gradients that exist near major highways are not apparent when emissions are mapped at 10 km resolution. High CO2 emission fluxes over highways become apparent at grid resolutions of 1 km and finer. Temporal variations in vehicle emissions are characterized using extensive day- and time-specific traffic count data and are described over diurnal, day of week, and seasonal time scales. Clear differences are observed when comparing light- and heavy-duty vehicle traffic patterns and comparing urban and rural areas. Decadal emission trends were analyzed from 2000 to 2007 when traffic volumes were increasing and a more recent period (2007–2010) when traffic volumes declined due to recession. We found large nonuniform changes in on-road CO2 emissions over a period of ~5 years, highlighting the importance of timely updates to motor vehicle emission inventories.

1 Introduction

The accumulation of carbon dioxide (CO2) in the atmosphere, due mainly to fossil fuel combustion, is the largest positive radiative forcing that contributes to global climate change. Under likely future emission scenarios [Moss et al., 2010], average global temperatures are expected to increase by more than 1.5°C by 2100 relative to the nineteenth century, unless effective mitigation measures are implemented [Intergovernmental Panel on Climate Change, 2013]. A range of approaches have been suggested to reduce greenhouse gas emissions to levels that stabilize atmospheric CO2 concentrations and help to minimize adverse impacts of climate change [Pacala and Socolow, 2004; Williams et al., 2012]. The transportation sector plays a centrally important role as a source of CO2 emissions and as a focus of climate change mitigation efforts. In the U.S., ~30% of anthropogenic greenhouse gas emissions (in CO2 equivalents) are due to transportation [Environmental Protection Agency, 2013]. In addition to CO2, motor vehicles emit combustion by-products that lead to air quality problems and exert short-lived effects on climate [Fuglestvedt et al., 2008]. Relevant pollutants include carbon monoxide (CO), nitrogen oxides (NOx), volatile organic compounds (VOCs), and black carbon (BC). For motor vehicles, emissions of these pollutants can be related to CO2 emissions and fuel consumption using fuel-based emission factors [Ban-Weiss et al., 2008; Bishop and Stedman, 2008; Dallmann et al., 2012]. Reducing emissions of certain air pollutants, such as CO, VOCs, and BC, present opportunities to mitigate climate change and improve human health [Smith et al., 2009]. Emissions of NOx can lead to both cooling and warming, and therefore, there is a need to consider air quality management and climate change mitigation together.

Cities are estimated to account for ~70% of energy-related emissions of CO2 globally and will be foci of efforts to mitigate and adapt to climate change [Rosenzweig et al., 2010]. To evaluate the effectiveness of these efforts, reliable emission inventories and ambient measurements of greenhouse gases are needed. Available emission inventories include the Emissions Database for Global Atmospheric Research (EDGAR, version 4.2) and the Vulcan Project (VULCAN). EDGAR maps emissions of CO2 and other air pollutants globally at a spatial resolution of 0.1° × 0.1° (European Commission–Joint Research Center, 2011, http://edgar.jrc.ec.europa.eu, accessed July 2013). VULCAN maps emissions of CO2 in the U.S. at 10 km × 10 km resolution [Gurney et al., 2009]. Both of these data sets are comprehensive in nature and map anthropogenic emissions from all major sources including industrial, electricity generation, on-road, nonroad, and residential/commercial sectors. Emission estimates for all sectors are needed in order to reconcile emissions with atmospheric observations. Inventories with higher spatial resolution are needed to guide local efforts to mitigate greenhouse gas emissions [Parshall et al., 2010] and to assess human exposure to traffic-related air pollution. National- and state-level emissions of motor vehicle CO2 can be readily estimated in the U.S. using fuel sales data. However, additional uncertainties arise in downscaling annual emission totals to finer spatial and temporal scales [Bellucci et al., 2012]. High-resolution CO2 emission inventories are few in number; current examples include an inventory developed for Indianapolis at the building and street scale [Gurney et al., 2012], and an inventory of motor vehicle emissions for all of Massachusetts on a 1 km × 1 km grid [Gately et al., 2013].

Spatially and temporally resolved emission maps will be useful in efforts to separate anthropogenic and biogenic contributions to observed CO2 surface fluxes. The biosphere exerts major influences on the global carbon cycle, which vary by season and time of day [Pataki et al., 2003; Newman et al., 2013]. Constraining emission inventories with observations is important for improving reliability of inventory estimates. Measurement approaches to sampling atmospheric burdens of CO2 include by satellite [Morino et al., 2011; Kort et al., 2012], aircraft [Mays et al., 2009; Brioude et al., 2012], and ground stations [Wunch et al., 2009; McKain et al., 2012]. Inversion methods have been developed that constrain bottom-up emission estimates and surface fluxes with top-down measurements using chemical transport models [Streets et al., 2013]. Recent applications include aircraft observations to map CO2 emissions on a 4 km × 4 km grid in Houston [Brioude et al., 2012] and Los Angeles [Brioude et al., 2013] and to determine minimum siting requirements for an urban CO2 monitoring network [Kort et al., 2013].

This study addresses the need for detailed, high-resolution maps of emissions from motor vehicles. We present a fuel-based inventory for vehicle emissions (FIVE), advancing an approach that has typically been used to support air quality analyses at regional/air basin scales [Singer and Harley, 1996; Pokharel et al., 2002; Dallmann and Harley, 2010; McDonald et al., 2012], and report emissions at much higher spatial resolution (1 km and finer scales) than most prior studies. High-resolution emission maps (HI-FIVE) are shown for various highly populated regions of the U.S. (California, Texas, and New York City metropolitan area in this study). We use publicly available data including fuel sales, road-level traffic counts, and time-resolved weigh-in-motion traffic count data, to demonstrate an emission mapping methodology that can be applied nationwide.

Another important feature of our approach is the separate mapping of on-road gasoline and diesel engine activity and emissions. This separation of emissions by engine type is essential to support inventory development for other air pollutants, for which the gasoline and diesel emission profiles are quite different. Furthermore, gasoline and diesel fuel consumption differ in their spatial and temporal patterns, and in their long-term growth rates over time. Future emission trajectories will also differ as emissions from the two engine types are regulated differently [Dallmann et al., 2011]. High-resolution emission mapping will be useful for assessing human exposure to traffic-related air pollution, since current air quality monitoring networks and computer modeling efforts do not typically capture proximity-related differences in human exposure to traffic-related air pollution [Kaur et al., 2007; Apte et al., 2011].

The objectives of this study are to (1) develop and compare maps of on-road CO2 emissions at resolutions of 10 km, 4 km, 1 km, and 500 m; (2) evaluate FIVE against published EDGAR and VULCAN inventories; and (3) provide diurnal, day-of-week, and seasonal time resolutions, commensurate with high-resolution spatial mapping of vehicle emissions. On-road emissions and activity are described separately for gasoline and diesel-powered vehicles operating in urban and rural areas. We also assess the effects of spatial resolution on the magnitudes and variability in emission fluxes. The finest resolution (500 m) used in this study approaches length scales needed for characterizing near-roadway exposure to traffic-related air pollution [Karner et al., 2010]. Our results are internally consistent, meaning that emissions are the same when aggregating from small to larger scales, and emissions are consistent with annually reported taxable fuel sales at the state and national levels. Finally, we assess changes in the magnitude and spatial patterns of vehicle emissions during the period 2000–2010 and consider the implications of these changes as they relate to needs for emission inventory updates.

2 Methods

2.1 State-Level Fuel Data

The standard approach to estimating motor vehicle emissions is to use estimates of vehicle travel subdivided by different vehicle categories and to multiply with CO2 emission factors normalized to distance traveled that are specific to road and vehicle type [Gurney et al., 2009; Gately et al., 2013]. Other air pollutants are estimated in a similar manner using emission factors derived from laboratory emission tests, with adjustment factors to account for variations in vehicle speed, engine load, ambient temperature, fuel properties, etc. In this way, a bottom-up emission estimate can be calculated, with uncertainties that depend on the reliability of both vehicle travel data and emission factors [Mendoza et al., 2013]. In this study, we use an alternative approach that starts with taxable fuel sales reports for each state (Table MF-2) [Federal Highway Administration (FHWA), 2011]. Fuel use can be allocated to finer spatial scales using relative rather than absolute estimates of vehicle travel. This fuel-based approach has been used previously to estimate vehicle emissions at air basin, state, and national scales [Singer and Harley, 1996; Pokharel et al., 2002; McDonald et al., 2012].

Sales of gasoline and diesel fuel intended for use in on-road engines are subject to highway fuel taxes. Dallmann and Harley [2010] estimate uncertainties in reported on-road fuel use at ±3–5% at the national level. The uncertainties are due to issues that arise in excluding tax-free fuel consumed by off-road engines and due to adjustments needed to reconcile aggregate fuel sales with amounts of fuel produced by refineries. In the U.S., gasoline is consumed mostly by on-road light-duty vehicles and diesel mostly by medium- and heavy-duty trucks. The U.S. light-duty vehicle fleet differs from that in Europe, where diesel engines are more widely used in passenger vehicles. In this study, on-road use of gasoline is apportioned to fine spatial scales for three regions of the U.S. (California, Texas, and New York City and surrounding areas), with an emphasis on mapping passenger vehicle emissions in urban centers. Trucks drive to a greater extent than passenger vehicles on highways outside of cities. Additional CO2 emissions from on-road diesel engines are mapped for California and Texas including both urban and rural areas.

A source of uncertainty in using state-level fuel sales data relates to whether the point of sale coincides with where fuel is consumed. In large states such as California and Texas, the contributions to total traffic from vehicles that cross state lines is expected to account for a relatively small fraction. If unaccounted for, long-haul trucking can result in a significant portion of diesel fuel that is sold in one state being consumed in other neighboring states [Lutsey, 2009]. However, state-level diesel sales reports are adjusted to reflect where fuel was used rather than where it was purchased, using reports filed by long-haul truck operators (International Fuel Tax Association, Inc., 2013, http://www.iftach.org, accessed January 2014). The adjustments are made based on truck distances traveled in each state. Fuel taxes paid by interstate truckers are thereby proportionally redistributed from states where fuel was purchased to states where truck travel took place.

Following McDonald et al. [2013], uncertainties in gasoline fuel sales reports are estimated by comparing each state's share of national gasoline sales (Table MF-2) and total vehicle travel (Table VM-2) [FHWA, 2011]. Uncertainty is calculated by subtracting the state share of national gasoline sales from shares of total vehicle travel and then normalizing by the state's share of gasoline sales. Some of the difference results from the variable composition of light-duty vehicle fleets, i.e., light trucks versus automobiles that exist across states, and the rest from interstate traffic. Mendoza et al. [2013] suggest that vehicle fleet composition is not a significant source of uncertainty when estimating on-road CO2 emissions in VULCAN. A positive difference suggests that the average fuel economy of a state's vehicle fleet is higher than the national average (i.e., more travel for a given amount of fuel burned), and/or there is a net import of fuel due to out-of-state fuel purchases being consumed within state borders. A negative difference suggests the opposite, and lower than average fuel economy and/or net export of fuel purchases. Using this approach, uncertainties in gasoline fuel sales and CO2 emissions are estimated as follows: ±5% for California, ±13% for Texas, and ±6% for New York/New Jersey/Connecticut. Uncertainties in diesel fuel volumes are computed similarly by comparing state shares of diesel fuel sales with truck travel (Tables VM-2 and VM-4) [FHWA, 2011]. Similar uncertainties of ±10% result for both California and Texas and may be due to incomplete accounting of interstate truck travel.

2.2 Road-Specific Traffic Count Data

In the U.S., traffic count data collected from the Highway Performance Monitoring System (HPMS) are available at the roadway level for highways and other principal arterial roads. Each state is responsible for collecting its own traffic data and reporting to the Federal Highway Administration. For example, in California, highways are sampled comprehensively using portable detectors that are moved periodically. Partial day and 24 h counts are typically used to characterize traffic on high-volume, urban highways, whereas 7 day counts are done on rural, low-volume roadways (California Department of Transportation, http://traffic-counts.dot.ca.gov/, accessed November 2013). Random sampling methods are used to quantify vehicle travel on smaller roadways. The precision of these estimates is strongly influenced by site selection and sample size. The Federal Highway Administration provides guidelines to transportation agencies on how to meet required precision levels in traffic sampling [FHWA, 2013]. The precision requirements are more stringent for heavily trafficked roadways (principal arterial roads and larger) than for lower volume roadways. Statewide traffic estimates are required to be within ±10% for major roads and freeways in large urban areas. The precision requirements are even higher for rural interstates at ±5%. In this study, we use traffic count data collected by the California Department of Transportation to map emissions in California. Counts are reported for an annual averaged day with totals for all vehicles, totals for all trucks, and trucks by axle category (two axle/six tires and three or more axles). For other states, HPMS traffic counts for all vehicles and for trucks specifically are available from the Freight Analysis Framework (Federal Highway Administration, http://www.ops.fhwa.dot.gov/freight/freight_analysis/faf/, accessed November 2012).

2.3 Fuel Apportionment

Prior work has shown that emissions from on-road vehicles are correlated with population density, road density, and traffic counts [Saide et al., 2009; Shu and Lam, 2011; Brondfield et al., 2012; Gately et al., 2013]. Traffic counts and road density were used in this study to apportion emissions spatially. We estimated the fraction of statewide fuel that is consumed on highways and major urban arterial roadways, for which traffic count data are available. This was done separately for passenger vehicles and for trucks using data tables that report vehicle travel on different road types (Table VM-2) and by vehicle class (Table VM-4) at the state level [FHWA, 2011]. Not all vehicle travel can be accounted for considering only those roads for which traffic counts are explicitly available (i.e., traffic count × roadway length), and the difference is made up by travel on urban and rural arterial roads (Tables S1–S3). Vehicle travel is then used as a proxy for fuel use, where total vehicle travel and travel by trucks with three or more axles are used as the proxies for on-road gasoline and diesel fuel use, respectively. We estimate that for both California and Texas ~70% of gasoline (Table S1) and ~80% of diesel fuel usage (Table S3) can be accounted for based on driving on roads where traffic has been counted explicitly.

We choose counts of trucks with three or more axles to map diesel fuel use rather than counts for all trucks, because more than half of the two-axle/six-tire trucks are gasoline powered (Table S4), and these trucks tend to drive more within cities. Most diesel fuel is consumed by larger trucks, which have a significant fraction of their travel between cities and in rural areas. In California, route-specific truck counts are reported by axle category (California Department of Transportation, http://traffic-counts.dot.ca.gov/, accessed November 2012). In other states, the proportion of medium- and heavy-duty trucks is reported at the state level by road class (Table VM-4) [FHWA, 2011]; these fractions are applied to total truck counts on individual roadways. Using counts of trucks with three or more axles (rather than totals for all trucks) as the proxy for diesel fuel use results in ~10–15% more diesel fuel use being assigned to rural areas with a similar reduction for urban areas (Tables S2 and S3).

The dominant fractions of on-road fuel consumed on roadways with explicit traffic count data (i.e., ~70% of gasoline and ~80% of diesel) are allocated from a state-level to specific grid cells, using vehicle travel from traffic count data as a spatial surrogate (i.e., traffic count × roadway length). Again, separate counts for total vehicles and for trucks with three or more axles are used as proxies for gasoline and diesel fuel use, respectively. The differences between statewide fuel sales and fuel quantities accounted for as outlined above (accounting for ~20–30% of fuel use) are assigned to remaining portions of the roadway network (i.e., those roads without traffic count data), using road length as a proxy. This is done separately for urban and rural grid cells to ensure that rural emissions are not overestimated, as travel on rural roads is expected to be lower [Brondfield et al., 2012]. For example, in California ~30% of total vehicle travel occurs on roads without traffic counts, and of this subset ~90% is urban and ~10% is rural (Table S1). However, the length of all roadways in rural areas of California is ~2 times larger than in urban areas. Separate urban road length and rural road length are therefore used as spatial surrogates for apportioning emissions. The roadway network has been mapped nationally, and urban boundaries used throughout this analysis are as defined by the U.S. Census Bureau (http://www.census.gov/geo/maps-data/data/tiger-line.html, accessed July 2013). In this analysis, only urbanized areas of 50,000 or more people are classified as urban. All other areas are considered rural, including urban clusters with greater than 2500 people but less than 50,000 people.

For individual urban areas, vehicle travel is also reported by road class (Table HM-71) [FHWA, 2011]. For selected urban areas with populations of 500,000 or more, fuel used by gasoline engines was constrained by comparing total reported vehicle travel within that urban area relative to the corresponding state total. We are unable to constrain diesel fuel consumption within individual urban areas in the same manner, since separate urban tables are not available for truck travel. However, diesel emissions are still constrained by statewide taxable fuel sales (Table MF-2), and between urban and rural areas using state-level reports of truck travel by road class (Table VM-4) [FHWA, 2011]. The same emissions mapping approach outlined above was repeated at grid resolutions of 10 km, 4 km, 1 km, and 500 m. Fine-scale mapping of emissions (i.e., at 1 km and 500 m resolutions) was only done for urban areas.

The density and carbon weight fractions of gasoline and diesel fuel reported by Kirchstetter et al. [1999] were used to convert fuel sales volumes to equivalent mass rates of CO2 emissions. By-products of incomplete combustion such as CO and unburned hydrocarbons that are coemitted with CO2 were ignored because they account for minor fractions of total fuel use. McDonald et al. [2013] report that fleet average light-duty vehicle emission factors for CO were 30–40 g/kg fuel burned in 2002, and had decreased to ~20 g/kg by 2010. The emissions of nonmethane hydrocarbons were in turn an order of magnitude lower than for CO. The fraction of total fuel carbon emitted as CO in diesel exhaust is also minor [Dallmann et al., 2012].

2.4 Weigh-in-Motion Data

Traffic count data from weigh-in-motion (WIM) detectors are used to specify variations in CO2 emissions over time. Binary data were obtained from the California Department of Transportation and converted to a readable format using commercially available software (http://www.dot.ca.gov/hq/traffops/trucks/datawim/, accessed January 2014). WIM stations exist to enforce size and weight limits on trucks, and stations count traffic and also weigh vehicles while they are in motion. Traffic counts are collected continuously on a year-round basis, with each vehicle classified based on the number of axles. Raw data are archived at high temporal resolution. Using WIM data, separate temporal activity profiles were developed for passenger vehicles and heavy-duty diesel trucks on multiple time scales: decadal, seasonal, day-of-week, and diurnal. We acquired and analyzed the complete WIM data set for the state of California, including about 70 counting locations, for each year from 2000 to 2010. Prior studies have also made use of WIM data, typically for one specific year using data from a subset of the available sites [Marr and Harley, 2002; Gurney et al., 2009].

Representative temporal profiles were developed to describe variations in vehicle traffic for urban and rural areas on finer time scales. Only the most complete WIM traffic count data from 2010 were used for this analysis. To classify WIM stations as urban or rural, we looked for a peak in passenger vehicle activity on weekday mornings. Morning traffic peaks associated with commuting are found in and near cities, but not in rural areas (see Figure S1). Separate diurnal traffic profiles were developed for Monday-Thursday, Friday, Saturday, and Sunday following Marr and Harley [2002], with separate profiles resulting based on urban and rural WIM station data. Day-of-week variations in daily traffic totals are reported for passenger vehicles and for trucks. Seasonal variations in traffic are described using factors calculated for each month and averaged across sites with at least nine complete months of data.

The WIM data set is well suited for analysis of changes in emissions over time, since the locations of counting stations have remained fixed for long periods. In earlier years, data are intermittent and available ~60 days per year. By 2010, data availability is close to 100% of days on a year-round basis at most sites. Traffic counts for each year were normalized to corresponding counts for 2007. This year was chosen as a reference point because traffic and fuel use in California reached a peak then and subsequently declined. Data were included for each site and year if at least 60 days of complete measurements were available. Stations were classified based on observed growth rates between 2000 and 2007 as either high (top 10%), average (10th–90th percentile), or low. This analysis was done separately for passenger vehicles and heavy-duty trucks.

3 Results and Discussion

3.1 On-Road CO2 Emissions and Comparisons With VULCAN and EDGAR

Vehicle emissions (on-road gasoline + diesel) within major urban centers show up prominently on the maps for both California and Texas (see Figure 1). Overall, emissions from motor vehicles in urban areas are higher than in more sparsely populated areas, by about an order of magnitude. Emissions are greatest near urban cores (>1000 tC km−2 yr−1) and decrease as one moves to the periphery (300 tC km−2 yr−1). In both California and Texas, on-road emissions of CO2 are concentrated in a few metropolitan areas. Los Angeles and San Francisco-San Jose account for ~50% of on-road CO2 emissions in California, and Dallas-Fort Worth and Houston account for ~30% of on-road emissions in Texas. Emissions due to vehicle travel on highways outside of the major urban areas are also apparent in Figure 1, with larger relative contributions to emissions on rural highways coming from diesel trucks.

Figure 1.

On-road emissions of CO2 for California (left) and Texas (right) for (a and b) FIVE, (c and d) VULCAN, and (e and f) EDGAR. All maps are on a 10 km grid for the year 2002, except for Figures 1e and 1f which are mapped at 0.1° resolution. The same color scale applies to all panels. See Table 1 for details on marked urban areas.

Table 1. Comparison of Annual On-Road CO2 Emission Estimates From FIVE (This Study), VULCAN, and EDGAR for 2002a
DomainGasoline Engines (106 tC)Diesel Engines (106 tC)On-Road Totalb (106 tC)On-Road VULCANc (106 tC)On-Road EDGARd (106 tC)
  1. a

    Uncertainty bounds give 95% confidence intervals.

  2. b

    On-road total = gasoline + diesel. Presented values are rounded and so may not sum exactly.

  3. c

    The VULCAN inventory, version 2.2, can be found at http://vulcan.project.asu.edu.

  4. d

    The EDGAR inventory, version 4.2, can be found at http://edgar.jrc.ec.europa.eu.

  5. e

    CO2 emissions calculated from national taxable gasoline and diesel fuel sales.

  6. f

    Uncertainty at the urban level is calculated as the propagation of errors in state-level fuel sales reports and spatial apportionment using traffic count data (see text).

  7. g

    A map of urban and rural grid cells for the VULCAN and EDGAR inventories can be found in the supporting information. Grid cells were classified as urban if their centroids were within urban boundaries as defined by the U.S. Census Bureau. Individual metropolitan areas listed above are shown as boxes in Figure 1.

U.S.e321 ± 1195 ± 5416 ± 12398406
California36.6 ± 1.88.0 ± 0.844.6 ± 2.040.156.1
Urbanf, g28.7 ± 3.24.5 ± 0.633.2 ± 3.330.450.4
Los Angeles (LA)13.5 ± 1.52.0 ± 0.315.4 ± 1.513.827.8
San Francisco/San Jose (SF-SJ)5.9 ± 0.70.7 ± 0.16.6 ± 0.76.410.7
San Diego (SD)2.83 ± 0.320.29 ± 0.043.12 ± 0.323.14.2
Texas27.0 ± 3.58.6 ± 0.935.6 ± 3.631.828.4
Urbanf, g17.4 ± 2.83.4 ± 0.520.7 ± 2.919.220.9
Dallas/Fort Worth (DAL-FW)5.2 ± 0.91.1 ± 0.26.3 ± 0.96.07.9
Houston (HOU)4.0 ± 0.70.7 ± 0.14.7 ± 0.74.55.7

Gasoline CO2 emissions dominate over diesel especially in urban areas, accounting for 80–90% of the total (Table 1). In rural areas, diesel is relatively more important (30–40% of total emissions). Inventories that assume diesel trucks account for a constant fraction of total vehicular traffic at all locations will erroneously overassign CO2 emissions to urban areas. Larger errors will likely result in the spatial assignment of other pollutants such as NOx and black carbon (BC), for which the diesel contribution to total emissions is greater than for CO2 [Ban-Weiss et al., 2008; McDonald et al., 2012]. This is particularly true in Texas, where ~60% of diesel truck CO2 emissions occur outside of urban areas, in contrast to passenger vehicles for which ~60% of emissions are within urban areas (Table 1).

Comparisons of on-road emissions estimates from FIVE (this study), VULCAN, and EDGAR are shown in Table 1 and Figure 1. We focus on three spatial scales: national, state, and urban. Note that in addition to including on-road motor vehicle emissions, VULCAN and EDGAR map many other sources of emissions that are not considered in this study, including industrial, electricity generation, nonroad, and residential/commercial sectors. Also, results are not shown for the New York City metropolitan area because diesel emissions were not mapped in this study. Differences among emission estimates increase as the domain of interest becomes smaller. Nationally for the U.S., all three inventories of on-road CO2 emissions agree to within 5% (Table 1). At the state level for California and Texas, VULCAN estimates are ~10% lower than this study. On-road CO2 emissions in EDGAR are 30% higher for California and 20% lower in Texas compared to the present study. The differences in emission estimates between EDGAR and FIVE increase when one focuses in on specific urban areas. EDGAR indicates on-road CO2 emissions in California cities that are higher than this study by 40–80% and also overestimates emissions in Texas cities by as much as 20–30%. VULCAN estimates are in closer agreement with FIVE for all of the cities that we evaluated; differences are within the uncertainty of our estimates.

As the national inventories are closely aligned, differences in the ways that motor vehicle emissions are disaggregated from national totals down to state and urban scales must account for most of the discrepancies. EDGAR is a global inventory that disaggregates national data, and uses road density as a spatial proxy, which may cause overestimation of emissions in population centers. VULCAN is for the U.S. only and first estimates vehicle emissions at a county level, projects the emissions onto a road atlas, and then aggregates to 10 km grid cells. VULCAN estimates are expected to agree more closely with the present study since similar traffic data sets are used. Similar findings have been reported by Gately et al. [2013], who compared their vehicle estimates of CO2 emissions for Massachusetts with those from other inventories and found that VULCAN agreed to within 5%, whereas EDGAR overestimated by 23% on average. EDGAR also assigns most of the on-road vehicle emissions to be within cities and shows only minimal amounts in outlying areas (Figures 1e and 1f). VULCAN has an urban to rural emission distribution that is similar to FIVE, but differences can be seen in specific cities. For example, in Dallas and Fort Worth, VULCAN shows two hot spots, while FIVE has emissions that are more evenly distributed over the broader region. For the Los Angeles area, our approach predicts higher emissions from traffic in inland communities than VULCAN.

3.2 Mapping Emissions at Higher Resolutions

To illustrate how maps of on-road emissions of CO2 are affected by spatial resolution, we evaluate emissions at resolutions of 10 km, 4 km, 1 km, and 500 m for the Los Angeles area (Figures 2 and 3). The traffic count data and roadway network, used to spatially apportion fuel usage in this study, are available at a roadway segment level. Therefore, the spatial resolution of the underlying traffic data is still much finer than the highest resolution emission maps (500 m) shown in this study. We chose to model vehicle emissions at these spatial resolutions because they represent a range of length scales commonly used in regional and local air quality models [Kleinman et al., 2004; Kim et al., 2009; Brioude et al., 2011; Joe et al., 2013], satellite retrievals of tropospheric air pollutants [Russell et al., 2010; Brauer et al., 2012; Kort et al., 2012], and assessment of near-roadway air pollution [Karner et al., 2010].

Figure 2.

Effect of spatial resolution on on-road emission fluxes of CO2 in Los Angeles at (a) 10 km, (b) 4 km, (c) 1 km, and (d) 500 m. Emissions are shown for the year 2010.

Figure 3.

Distribution of emissions from lowest to highest for urbanized portions of Los Angeles (see Figure 2). For example, emission fluxes exceed ~7000 tC km−2 yr−1 for ~10% of the urbanized land area in Los Angeles. Above this dividing line, grid cells contain major highway segments. The dashed lines at 1 km and 500 m resolutions show results of an emission factor sensitivity analysis (see text). Emissions are shown for the year 2010.

Increased spatial resolution is expected to result in sharper gradients, since emissions on heavily traveled highways are more accurately mapped and concentrated in smaller areas. For reference, the 10 km resolution emission map (Figure 2a) uses the same grid system as VULCAN. There are clear differences between the coarse- and finer-resolution emission maps (see Figure 2). The highway network is readily apparent based on the much higher on-road CO2 emission fluxes compared to surrounding areas, but only in the higher-resolution maps. The highway network is clearly distinguishable at 1 km, and emissions are brought into even sharper focus at 500 m resolution. With the 10 and 4 km grids, highway emissions are more widely distributed, and as a result, strong horizontal emission gradients tend to disappear. Traffic on major arterial roadways is also visible in the 500 m resolution emission map. Finer details are especially apparent for the Los Angeles area because traffic count data are available for many of the major surface streets.

Generally, traffic count data on local roads are difficult to obtain and are not sampled or archived in a comprehensive manner as compared to the highway network. Yet nonhighway vehicle activity constitutes a large fraction (about half) nationwide [FHWA, 2011]. To improve characterization of traffic patterns on local roads especially within cities, the integration of new data sources would be useful. Such sources include mobile phones which can record, store, and report data from global positioning systems while operating in vehicles that are in motion [Herrera et al., 2010]. Such data are currently used by major internet and vehicle navigation service providers to show real-time traffic information. Traffic sensing on arterial roads is also increasing due to the proliferation of well-instrumented city streets and intersections. These sensors include traditional inroad traffic sensors as well as other technologies such as radio frequency identification readers used for automated collection of tolls [Landt, 2005]. The trend toward increasing vehicle activity information and open data platforms is clear, and new data sources are likely to increase empirical support for high-resolution emissions mapping in coming years.

The question of spatial resolution is relevant when integrating satellite observations of air pollutants with the development of bottom-up emission inventories, especially pollutants for which motor vehicles are a dominant source of emissions, such as in certain urban areas for NOx [McDonald et al., 2012] and CO2 (Figure S2). Satellite columns of NO2 have been retrieved with global coverage at spatial resolutions of ~13 km × 24 km using the Ozone Monitoring Instrument [Levelt et al., 2006]. The spatial resolution shown in Figure 2a aligns approximately with current capabilities of low-Earth orbiting satellites to map NOx emissions from space. Spatial averaging techniques have been utilized to resolve NOx even more finely down to 5–10 km [Russell et al., 2010]. In general, current satellite observations cannot resolve sharp pollutant emission gradients near roadways (Figures 2c and 2d) but may be appropriate for constraining emissions at urban and larger scales. For satellite retrievals of CO2, the Greenhouse Gases Observing Satellite is capable of resolutions of ~10 km at nadir [Kort et al., 2012]. The Orbiting Carbon Observatory (OCO-2), a geostationary satellite to be launched in 2014, will have a footprint of ~3 km2 at nadir [Boesch et al., 2011] and is also capable of achieving high temporal resolution (i.e., hourly data) over individual locations [Streets et al., 2013].

To address further the question of what can be gained from increasing spatial resolution of emission maps within urban areas, the distribution of CO2 emission fluxes (tC km−2) in the Los Angeles area are ordered from lowest to highest for various spatial resolutions (Figure 3). There is a dividing line at emission levels of ~7000 tC km−2 yr−1 that separates freeways from other smaller roadway types. Figure 3 shows that on-road CO2 emission fluxes over the most heavily traveled grid cells that include major highway segments, increase by a factor of 3 when spatial resolution increases from 10 km to 1 km. When the resolution is further increased from 1 km to 500 m, an additional increase of ~60% is seen. Interestingly, little increase is seen in emission fluxes when increasing the resolution from 10 km to 4 km. Highly resolved emission maps are important for understanding transportation microenvironments. Karner et al. [2010] reviewed near-roadway air quality studies and found that pollutant concentrations typically fall to background levels at downwind distances of 115–570 m. A recent air quality modeling study in and around the Port of Oakland found twofold increases in predicted concentrations of elemental carbon in locations closest to diesel truck emissions, when spatial resolution in the model was changed from 1 km to 250 m [Joe et al., 2013].

A substantial fraction (about half) of vehicle activity occurs nationwide on roads characterized by high driving speeds (interstates/freeways + rural principal roads). It is well known that most conventional internal combustion engines have lower rates of fuel consumption when driving on highways compared to stop-and-go city driving. To understand potential effects of fuel economy differences on CO2 emission maps, we performed a sensitivity analysis allowing CO2 emission factors to vary by road type. The shape of the curve relating CO2 emissions (grams emitted per unit distance traveled) and vehicle speed is parabolic, with a minimum emission rate at vehicle speeds of 70–80 km h−1 [Barth and Boriboonsomsin, 2008]. Emission factors for CO2 increase by a factor of 2 when average vehicle speeds drop from 50 to 25 km h−1 but vary relatively little at higher speeds. Therefore, the largest changes in CO2 emission factors are anticipated on congested urban arterial roadways where average vehicle speeds are commonly below 50 km h−1 especially at peak traffic times. Fuel economy penalties associated with stop-and-go driving and traffic congestion may result in redistribution of CO2 emissions away from highways relative to estimates in the present study. This can be seen in Figure 3, which shows the results of the sensitivity analysis when emission factors include speed adjustments for different road types. Emissions are first reestimated at a state level by road type, taking into account a higher CO2 emission factor (g CO2 km−1) for urban arterial roads. State-level emissions are then gridded using the same spatial surrogates outlined in section 'Fuel Apportionment'. A detailed calculation is provided in the supporting information (Table S5). These two approaches to CO2 emission apportionment give similar results. Note that hybrid vehicles are expected to flatten the vehicle speed-CO2 emissions relationship at lower speeds (i.e., similar fuel economy under stop-and-go driving conditions) [Fontaras et al., 2008], making the results potentially even more similar if there is a high penetration of hybrid vehicles in the urban fleet.

Figure 4 presents examples of high-resolution CO2 emission maps (HI-FIVE) for other urban areas and demonstrates the extensibility of the general approach used here. The New York City example (Figure 4a) illustrates an application of the emissions mapping approach across a densely populated multistate region. New York City also provides a point of contrast to Los Angeles in terms of population density. The highway network in New York is not as extensive, and on-highway emissions only become clearly visible in suburban areas on Long Island, and in neighboring states of New Jersey and Connecticut. The San Francisco Bay area (Figure 4b) features contrasts between a dense urban center (San Francisco) and other nearby lower density cities (e.g., San Jose). In San Jose, a network of highways akin to Los Angeles exists. Emission maps for Houston and Dallas-Fort Worth are also shown in Figure 4, to provide examples of even lower density sprawling cities, with resulting emission fluxes that tend to be lower on average than other cities considered here. Consistent mapping of vehicle emissions at high spatial resolution could facilitate more detailed study of relationships between urban form and transportation emissions, by allowing for subregional and cross-city comparisons [Marshall, 2008; Gately et al., 2013].

Figure 4.

High-resolution maps of CO2 emissions (HI-FIVE) from gasoline-powered vehicles for (a) New York City, (b) San Francisco-San Jose, (c) Houston, and (d) Dallas-Fort Worth. Only emissions in urbanized areas are shown; maps are for the year 2010. Horizontal resolution is 500 m in all cases. Note also that the 0–40 km distance scale is the same for all four areas.

3.3 Variability in Emissions on Short Time Scales

Both spatial and temporal resolutions are needed when emissions estimates are used as input to atmospheric models, and in turn for comparison with atmospheric observations. In this study, weigh-in-motion traffic count data were analyzed to characterize variations in vehicular CO2 emissions on various time scales: diurnal, day-of-week, and seasonal. Different temporal profiles are developed for light- and heavy-duty vehicle traffic, as shown in Figure 5. The diurnal activity patterns on weekdays clearly differ between the two vehicle types. Weekday passenger vehicle traffic exhibits two distinct (morning and evening) commuter-related peaks. In contrast, truck traffic shows a single midday peak. These findings are consistent with prior work [Marr and Harley, 2002]. Because passenger vehicles dominate on-road CO2 emissions in urban areas (Table 1), the diurnal pattern of urban on-road emissions is similar to that for light-duty vehicles (Figure 6). For other pollutants where heavy-duty diesel emission factors are significantly higher than for gasoline vehicles, like for BC [Dallmann et al., 2013], the diurnal profile of emissions is expected to follow more closely the activity pattern for heavy-duty trucks. This highlights the need to apply separate temporal profiles to characterize traffic and associated CO2 emissions for light- and heavy-duty vehicles.

Figure 5.

(a and b) Diurnal, (c and d) day of week, and (e and f) seasonal variations in counts of passenger vehicles shown in Figures 5a, 5c, and 5e and heavy-duty diesel trucks shown in Figures 5b, 5d, and 5f. Diurnal profiles are for weekdays (Monday-Thursday); for profiles on other days, see the supporting information. Each marker represents observations at an individual weigh-in-motion traffic count location. Colored bands represent 95% confidence intervals for the means across all urban (red) and rural (green) sites.

Figure 6.

Diurnal (shaded) and day of week (dashed lines with labels) patterns of on-road CO2 emissions for (a) California and (b) Texas in 2010. Red and green shading denote emissions in urban and rural areas, respectively. Total (urban + rural) emissions are shown in gray. Ratios of day-specific emissions to the weekly average are labeled for each day. Uncertainty estimates indicate 95% confidence levels for diurnal (dark bands) and day of week (error bars) emissions. Confidence intervals are calculated using uncertainties in weigh-in-motion traffic count data shown in Figure 5. Refer to the supporting information (Figure S1) to see which areas were classified as urban and rural.

While large decreases in weekend truck traffic and their emissions are well documented [Marr and Harley, 2002; Harley et al., 2005], day-of-week variations for passenger vehicle traffic are also of interest (Figure 5). Urban emissions of CO2 are found to increase by ~10% through the workweek between Monday and Friday, due to changes in passenger vehicle activity, and then decrease by ~20% and ~30% relative to Friday peak levels on Saturdays and Sundays, respectively (Figure 6). Day-of-week variations in traffic can lead to noticeable changes in on-road CO2 emissions, given that vehicle emissions are the largest anthropogenic source of CO2 in many California and Texas cities (Figure S2). Given a design goal of urban CO2 monitoring networks to detect changes in emission fluxes that differ by 10% or more from the average [Kort et al., 2013], then comparing CO2 emissions on Fridays to other (especially weekend) days could serve as a repeatable real-world test case for detection capabilities of emerging CO2 monitoring networks, at least in urban areas where motor vehicle emissions tend to dominate.

There are also clear contrasts between urban and rural vehicle activity patterns. The differences are most apparent for passenger vehicles (Figure 5); heavy-duty truck traffic shows similar activity patterns throughout the urban and rural areas considered here (Figure S1). Passenger vehicle traffic in rural areas follows a similar diurnal profile to heavy-duty trucks and does not have commuter peaks as seen in urban areas on weekdays. As a result, when emissions are aggregated to the statewide level, emissions are highest on weekdays in the late afternoon/early evening hours rather than in the mornings (Figure 6), which is consistent with VULCAN for the contiguous U.S. [Nassar et al., 2013].

Diurnal and day-of-week traffic profiles from this study are similar to results reported by Marr and Harley [2002] using weigh-in-motion traffic count data from the mid-1990s for California. This suggests that diurnal and weekly patterns in vehicle activity have not changed much over longer time scales. We find smaller weekend decreases in heavy-duty truck traffic compared to Marr and Harley [2002], who reported that weekend truck traffic decreases by ~80% and ~60% compared to weekday averages in urban and rural areas, respectively. Analysis of a larger and more complete WIM data set in this study show decreases in weekend truck traffic of ~70% in urban and ~50% in rural areas. The seasonal activity patterns shown in Figures 5e and 5f are new, and all of the temporal profiles reported here are based on analysis of a more extensive database of traffic counts. Seasonal emission cycles are important to resolve, as they can be used to assign contributions to atmospheric observations from different emission sources [van der A et al., 2008]. CO2 fluxes due to biosphere-atmosphere exchange and natural gas combustion also exhibit seasonality [Pataki et al., 2003]. Noteworthy increases of 35–40% are observed in heavy-duty truck traffic between January and August. The peak in August, especially in rural areas, may be linked in part to harvesting of crops. Passenger vehicle traffic in rural areas also exhibits strong seasonality, with year-round variation of ~40%. The peak occurs during summertime, presumably due to increased recreational and vacation-related driving. In contrast, seasonal variations in passenger vehicle traffic are limited to ~10% of mean levels in urban areas.

3.4 Decadal Emission Trends

Due to the high data and time demands required to update bottom-up emission inventories, a common practice in air quality planning is to scale baseline emission inventories to represent conditions in other years. The scaling of emissions attempts to reflect effects of both increasing population and vehicle activity, as well as the effects of advances in vehicle and emission control technologies. An underlying implicit assumption often made when scaling baseline inventories is that the spatial distribution of emissions remains the same over time, and that any increases in traffic, for example, occur uniformly at all locations throughout the domain of interest. The EDGAR on-road inventory provides an example of this approach (Figure S5). The spatial distribution of on-road emissions remains constant over a 5 year period (from 2002 to 2007). We use weigh-in-motion traffic count data to consider whether nonuniform changes in vehicle activity have occurred since 2000. If population growth leads to new housing being built in suburban areas rather than as urban in-fill, evolution in spatial maps of emissions should be expected. Two time periods considered in this analysis are 2000 to 2007 and a later period from 2007 to 2010 that was affected by a major economic downturn. We choose to evaluate these two time periods because the recession provides a useful test case for detecting changes using weigh-in-motion traffic count data.

In maps showing the rate of change in on-road CO2 emission by location (Figure 7), differences among sites are evident during the period from 2000 to 2007 as described above. See the supporting information (Figure S6) for trends in weigh-in-motion traffic count data used in Figure 7. Since 2007, effects of the recession on traffic emissions were unevenly distributed. Prior to the recession (Figure 7a), high-growth areas (>4% yr−1) are observed in inland areas of San Bernardino and Riverside counties, located east of Los Angeles. California's Central Valley also saw higher than average growth in on-road emissions (>2% yr−1). Little change in traffic was observed near the coast in Los Angeles, the San Francisco Bay area, or San Diego. A 2% yr−1 detection limit applies to this analysis, because when compounded over the period from 2000 to 2007, this matches the uncertainty in traffic counts from weigh-in-motion detectors. Using results from Li et al. [2010], we estimate ~15% uncertainty when traffic counts from WIM are compared with those obtained from video counting. After 2007 (Figure 7b), many locations saw decreases in emissions in excess of 2% yr−1. In general, locations with a high diesel truck traffic fraction exhibited the largest decreases in on-road CO2 emissions after 2007, and these locations tended to be outside of major metropolitan areas.

Figure 7.

Annual growth rates in on-road CO2 emissions in California for (a) 2000–2007 and (b) 2007–2010. Boundaries are shown for the five largest air basins in California: South Coast (SC), San Francisco Bay Area (SF), San Diego (SD), San Joaquin Valley (SJV), and Sacramento Valley (SV). Gray pixels are 10 km × 10 km grid cells with emission fluxes of >100 tC km−2 yr−1.

As mentioned above, the EDGAR inventory shows little change in the spatial pattern of on-road emissions over a 5 year time period from 2002 to 2007. However, changes in on-road emissions of CO2 at the ~70 weigh-in-motion stations show increases as high as 50%, with many locations increasing by more than 25% (Figure S5). In the supporting information, we show that our approach to mapping emissions using year-specific taxable fuel sales and traffic count data identifies areas with higher than average growth in on-road emissions (Figure S7). Annual changes in emissions are not spatially homogenous, and this highlights the importance of periodically updating emission inventories as new information becomes available.

4 Conclusions

In this study, fuel sales reports and traffic count data were used to create a fuel-based inventory for vehicle emissions (FIVE) of CO2 at various spatial resolutions, for major urban centers in the U.S. Passenger vehicles account for 80–90% of on-road CO2 emissions in cities, whereas heavy-duty diesel trucks are relatively more important in rural areas and account for 30–40% of the on-road total. Results from FIVE were compared with other emission inventories, VULCAN and EDGAR. All three inventories agree within 5% at the U.S. national level. EDGAR appears to overestimate on-road CO2 emissions in the largest cities in California and Texas by as much as 20–80%, while VULCAN estimates are in agreement with FIVE. We also show that spatial resolution has important effects on the mapping of motor vehicle emissions. At grid resolutions of 10 and 4 km, strong emission gradients that are known to exist near highways are not apparent. The highway network becomes clearly distinguishable at grid resolutions of 1 km. Increasing the resolution from 1 km to 500 m leads to further increases in CO2 emission fluxes by ~60% for grid cells that contain segments of heavily trafficked highways.

Over shorter time scales (diurnal, day-of-week, and seasonal cycles), there are large contrasts in on-road vehicle emission patterns in urban and rural areas, and between light- and heavy-duty vehicles. In urban settings, daily on-road emissions of CO2 are found to increase by 10% through the workweek, with a maximum on Fridays, followed by decreases from the Friday peak of 20–30% on weekends. This weekly cycle in traffic-related CO2 emissions could serve as a useful test case for evaluating the ability of urban CO2 monitoring networks to detect future emission changes. We also find significant seasonal variability in both passenger vehicle and heavy-duty truck traffic in rural areas. Year-to-year changes in vehicle activity were found to be nonuniform across California between 2000 and 2007. High-growth areas where on-road emissions increased by >4% yr−1 were concentrated in fast-growing suburbs to the east of Los Angeles. Between 2007 and 2010, decreases in vehicle emissions were seen over many parts of California. Changes of up to 50% in on-road emissions were found over a period of ~5 years, highlighting the need for timely updates to motor vehicle emission inventories.

Acknowledgments

Initial support for this research, including CO2 emission mapping and analysis of weigh-in-motion traffic count data, was provided by the California Air Resources Board. Traffic count data used to generate the high-resolution emission maps are available from the California Department of Transportation (http://traffic-counts.dot.ca.gov/) and Federal Highway Administration (http://www.ops.fhwa.dot.gov/freight/freight_analysis/faf/). The authors also thank the California Department of Transportation for providing weigh-in-motion traffic count data (http://www.dot.ca.gov/hq/traffops/trucks/datawim/). Additional support from the QUALCOMM foundation was provided through the QUEST Scholars Research program at UC Berkeley. The results reported herein are those of the authors, and our findings and conclusions may not reflect the opinions or policies of the research sponsors.