Accurate emission inventories and their temporal trends must be incorporated into pollutant inventories to allow for reliable modeling of the country's past, present, and future air quality. Measured carbon monoxide (CO) and nitrogen oxide (NOx) concentrations from two urban areas show that the CO/NOx vehicular emission ratio has decreased at an average rate of 7–9% per year from 1987 to 1999. This amounts to a factor of nearly 3 over the 12 years. The current U.S. Environmental Protection Agency tabulations of estimated pollutant emission trends indicate a rate of decrease smaller by a factor of 2–3. The trend in maximum ambient CO levels in U.S. cities suggests a 5.2 ± 0.8% per year average annual decrease in CO vehicular emissions, which implies a 2–3% annual increase in NOx emissions from vehicles. Thus over the decade of the 1990s, annual U.S. CO emissions from vehicles have decreased from ∼65 to ∼38 Tg, representing approximate decreases of 6 and 3% in the annual global fuel-use CO emissions and in total global anthropogenic CO emissions, respectively. It is expected that the volatile organic compound (VOC)/NOx vehicular exhaust emission ratio has decreased similarly, implying that the character of atmospheric photochemistry in U.S. urban areas has changed significantly over the decade.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 In urban areas, photochemical processes produce elevated ambient ozone levels from precursors emitted by anthropogenic sources. These precursors include the fuels, volatile organic compounds (VOCs) and CO, and the catalysts, oxides of nitrogen (NO + NO2 = NOx), required for the photochemical reactions. The efficiency with which atmospheric photochemistry produces ozone is a sensitive function of the VOC to NOx ratio [Seinfeld et al., 1991; Schere and Hidy, 2000, and references therein]. The United States has expended large resources to reduce emissions of the photochemical precursors, and it is argued that these efforts have been effective [e.g., U.S. Environmental Protection Agency (EPA), 2000a]. However, corroboration and quantification of emission reductions by measurements of ambient precursor concentrations are difficult because (1) spatial and temporal heterogeneity of the sources combined with changing meteorological conditions cause large variability in the ambient levels and (2) accurate measurements over many years are required to reveal trends resulting from small annual changes.
 During the last three decades, implementation of increasingly sophisticated control technology has greatly reduced the emissions per kilometer driven from automobiles [EPA, 2000a], which are the dominant urban source of VOCs, CO, and NOx. The effect on total vehicular emissions is smaller because increasing vehicle distance traveled has partially offset the resulting reductions and because control efforts are primarily focused on automobiles and light-duty trucks with less emphasis on controlling the emissions of heavy-duty diesel-powered trucks (Sawyer et al., 2000). However, the EPA estimates that over the 1988–1998 decade the total on-road vehicle emission ratio of CO/NOx has decreased by 30% owing to a 29% decrease in CO emissions and a 1.4% increase in NOx emissions [EPA, 2000a]. Our goal in this paper is to quantitatively test this estimate through ambient measurements.
 A careful approach is required to effect this test. First, we use measurements from two sites that are at altitudes well up in the boundary layer so that the sampled air is well mixed with regard to the many very local sources and represents the emissions from a relatively large region. Second, we select a diurnal time window, the morning rush hour, when the ambient measurements are dominated by vehicle emissions. Third, we collect data at the same time of day, in the same season, and at the same site over years, and focus on temporal trends at specific sites so that the results are relatively immune to diurnal, seasonal, and spatial differences in emissions. Finally, we focus on ratios of ambient concentrations rather than the concentrations themselves so that we are relatively immune to changing meteorological conditions, changing traffic density, or driving patterns, etc. Using this approach, the time scales of the measurements presented here are long enough to discern statistically significant trends in vehicle emission ratios.
 From the following analysis we conclude that the ratio of CO to NOx in U.S. vehicular emissions has decreased significantly over the past decade. A substantial decrease in CO emissions combined with a smaller increase in NOx emissions is responsible for the decrease in the ratio. These changes in automotive emissions are not accurately captured in the EPA emission inventory. Finally, we point out the implications of this change in emissions for local-, regional- and hemispheric-scale atmospheric photochemistry.
2. Measurements and Data Interpretation
 We measured ambient ozone precursor concentrations at two sites located within urban areas but isolated from the direct influence of very local sources; hence they serve to characterize the average ambient urban concentrations. The Nashville, Tennessee, site was at the top of the 110-m high Polk Building in the urban center. Kleinman et al.  previously used measurements from this site to characterize the Nashville emissions. The Boulder, Colorado, site was on the western edge of the Boulder-Denver urban area on a mesa ∼150 m higher in elevation than the urban area that lies to the south and to the east. Parrish et al.  and Goldan et al.  previously characterized Boulder-Denver emissions from measurements at this site.
 CO and the total oxidized nitrogen compounds (NOy) were measured during short, intensive measurement periods. The CO concentrations were measured by standard monitoring instruments (TECO Model 48, nondispersive infrared absorption technique) that were carefully calibrated and zeroed (see Parrish et al.  for further details). In the NOy instruments, all of the NOy species were reduced to NO during sample passage through a catalytic convertor (Au-catalyzed reduction by added CO in Boulder and reduction over MoO2 in Nashville). The NO was then measured by its chemiluminescence reaction with ozone added as a reagent. Williams et al.  fully described and extensively intercompared NOy instruments with the two catalytic techniques.
 After emission to the atmosphere, NOx is oxidized on a time scale of hours to days to nitric acid, peroxyacyl nitrates, and other organic and inorganic nitrates. NOy comprises these product species plus unreacted NOx. Therefore, under the conditions considered here, the measured NOy represents the total of the originally emitted NOx. All of the Nashville measurements were made in the summer (22 July to 31 August 1994, 19 June to 28 July 1995, and 15 June to 15 July 1999), while the Boulder measurements were made in the spring (21 March to 7 April 1989), winter (3–15 February 1991), and fall (16 October to 4 November 1996 and 12–24 November 1998).
 We consider only data from the morning rush hour period (0600–0830 and 0600–1000 LT for Nashville and Boulder, respectively). Measurements from this period are most representative of solely fresh vehicular emissions, show the highest ambient concentrations due to high traffic volume and slow vertical mixing, and are affected minimally by any removal processes. Figure 1 shows the time series of data from one example day in Nashville. The concentrations of the species are highly correlated (r2 = 0.91, where r is the correlation coefficient) and are distinctly elevated during the morning high traffic period. This indicates that on-road vehicles are the dominant source of these species at this location and time of day.
 To determine the CO to NOx emission ratio from ambient measurements, the relative lifetimes of CO and NOy must be considered. CO is a long-lived species (lifetime of ≈1–3 months depending upon season), so regional background levels of CO make a significant contribution to the ambient levels in emission regions. In contrast, the lifetime of NOy is on the order of a few days, so background concentrations are proportionately much smaller. To extract the emission ratio, we derive the linear relationship between the measured CO and NOy. The slope gives the CO to NOx emission ratio, and the intercept approximates the regional background level of CO.
Figure 2 presents the data from the three Nashville and the four Boulder studies. Table 1 summarizes the parameters of the linear regressions. The slopes and intercepts were derived from weighted linear regressions that allow for uncertainties in both variables [Neri et al., 1989]. The weighting for each point was 1/σ2, where σ is the estimated uncertainty in each measurement; 10 ppbv for CO and 1 ppbv for NOy at both sites. In each data set the large r2 value indicates that CO and NOy are well fit by the derived linear relationship. All confidence limits given in Table 1 and elsewhere are estimated 95% confidence limits. The confidence limits on the slopes are propagated from the uncertainty of the CO calibration (±10%), the uncertainty of the NOy calibration (±10%), and the uncertainty in the slope from the linear regression. The slope derived from the linear regression is dependent upon the relative weighting of the CO and NOy values. A relatively high weight on the CO data attributes the total deviation of the data from a linear relationship to deviations in NOy only and vice versa. The total relative range in the slope that results from all possible weightings is r−2–1. Therefore the uncertainty in the slope is set equal to (r−2–1)/2.
Table 1. Parameters of Linear Regressions Between CO and NOy for the Seven Data Sets
 The intercepts from Nashville (111–194 ppbv CO) in Table 1 are generally similar to the lower levels seen in the eastern United States in summer [Chin et al., 1994; Novelli et al., 1998]. In Nashville, r2 drops below 0.95 only in 1995; this same year also has a higher CO intercept, corresponding to higher regional background CO levels. This larger scatter and higher background are consistent with the large, episodic influence of Canadian forest fires, which had a particularly strong impact on the CO levels in the Nashville area in 1995 [Wotawa and Trainer, 2000]. In Boulder the intercepts are also reasonably consistent with regional background levels.
Figure 3 shows the temporal trends of the CO to NOx vehicular emission ratios derived from the ambient data of Figure 2; two other experimental determinations of the emission ratio are included for comparison. Table 2 gives the parameters from linear fits to the temporal trends of the log-transformed data. The 1995 value from Figure 2a for Nashville (8.5 ± 1.3) compares well with the result for the same city and time (8.4 ± 2.3) calculated independently from emission factors derived from remotely sensed vehicle exhaust and from roadway tunnel measurements [Harley et al., 2001]. This comparison corroborates the use of the slope of the CO to NOy relationship as an accurate measure of the CO to NOx emission ratio from vehicles in an urban area. The Nashville 1995 data set has been discussed previously by Kleinman et al. , who reported a slope of 9.3, statistically in agreement with the value derived here.
Table 2. Temporal Evolution of CO to NOx Automotive Exhaust Emission Ratios
Rate of Change, % per year
Years of Data
B. MacRae of the Air Pollution Control Division, Colorado Department of Public Health and Environment (CDPHE) provided the CDPHE inventories to us.
 A similar correlation analysis of CO with NOx for the Los Angeles area has been presented for the summer 1987 Southern California Air Quality Study. The Los Angeles datum in Figure 3 and Table 1 comes from the inverse of the slope derived from the linear regression of ambient NOx versus ambient CO concentrations measured from 0700–0800 LT at eight sites in the California South Coast air basin [Harley et al., 1997]. This value agrees well with the CO to NOx ratios calculated from this same data set [Fujita et al., 1992]. This Los Angeles result also agrees well with the extrapolation back to 1987 of the temporal trend we derive for Nashville (Figure 3).
 The Boulder ratios in Figure 2 are significantly larger than the Los Angeles-Nashville trend throughout the period of measurements. This difference is consistent with the colder seasons of the Boulder data and with the special air pollution control measures instituted in the Boulder-Denver urban area. For gasoline fueled vehicles, CO emissions are significantly increased by lower ambient temperatures [Stump et al., 1989]. NOx emissions are also increased, but to a lesser extent, so the emission ratio is higher in colder weather. The emission inventories compiled by the Colorado Department of Public Health and Environment (CDPHE) reflect this effect. In the 1988 inventory the estimated CO to NOx emission ratio was 15% higher in winter than in summer, although in the 1995 inventory the ratio was the same in the two seasons to within 1.5%. The primary air quality concern in the Boulder-Denver urban area is violation of the ambient CO standard in winter, so control efforts have been focused on minimizing CO emissions in this season. These efforts include the nation's first oxygenated fuels program, initiated in 1988, and an increasingly strict inspection and maintenance (I/M) program. Oxygenated fuels reduce CO emissions by ∼10% and may simultaneously increase NOx emissions [National Science and Technology Council Committee on Environment and Natural Resources, 1997; Ragazzi and Nelson, 1999], and the I/M program is more stringent with regard to CO emissions than NOx emissions. The Boulder 1996 data were collected after the oxygenated fuel program had ended for the season; this may partially account for this point lying above the trend.
3. Comparison With Emission Inventories and Ambient Monitoring Data
Figure 4 compares the CO to NOx emission ratios in vehicle exhaust derived from emission inventories with those derived from the observations. The observed decrease in this ratio over time is much greater than that estimated by the EPA in current tabulations of national U.S. emissions [EPA, 2000a]. The observed ratio decreased by 38% over the five years spanned by the Nashville measurements and by 53% over the nine years spanned by the Boulder measurements. The combined Nashville-Los Angeles trend and Boulder trend suggests that the CO to NOx emission ratio from vehicles in the United States has decreased by 7–9% per year, corresponding to a total decrease from 1987 to 1999 of a factor of 2.4–3.1. These decreases are a factor of 2–3 larger than the annual average decrease of 3.5% (3.4% annual decrease from the fit in Figure 3) for 1988–1998 included in the national emission inventory [EPA, 2000a]. Comparison of the ambient and inventory ratios shows that the EPA inventory seriously underestimates the CO to NOx ratio in vehicle exhaust before 1990, but overestimates that ratio by nearly a factor of 2 by the end of the decade.
 The EPA national emission inventory does provide a basis on which to judge if Nashville and Boulder, the two urban areas where the present measurements were carried out, are nationally representative. For 1996 (the most recent inventory detailed to the county level) this inventory gives summertime weekday vehicular CO/NOx ratios for Nashville (Davidson County) of 11.9 and Boulder (Boulder County) of 10.2. These values are similar to the national annual average inventory ratio (11.1). This comparison suggests that the emissions of the vehicle fleet in Boulder and Nashville are reasonably representative of the average national vehicle fleet.
 The cause of the inaccuracy of the EPA emission inventories deserves thorough investigation. On-road mobile exhaust emissions in inventories are derived using emission factors from the Mobile Source Emissions Factor (MOBILE) model [National Research Council, 2000]. These emission factors give vehicle emissions in grams per mile developed from information on the emissions characteristics of the vehicle fleet. Combining emission factors with estimates of vehicle miles traveled yields emissions estimates. The particular versions of the model used in the national inventory were MOBILE 5A for 1970–1994 and MOBILE 5B for later years [EPA, 1998]. CDPHE (B. MacRae, Air Pollution Control Division, CDPHE, personal communication, 2000), also using MOBILE 5A, has developed an independent emission inventory for the Boulder-Denver metropolitan area. Estimates from the CDPHE inventory shown in Figure 4 are in excellent agreement with the Boulder measurements. For 1988, ratios of 23.1 and 20.1 are given for the winter and summer, respectively, and the rates of decrease through 2001 correspond to 7.1 and 5.9% per year for the respective seasons. It would be useful to compare the techniques used in developing these two inventories (both based on the MOBILE model) to resolve the differences between them and perhaps to begin to understand the large discrepancies between the EPA inventory and the measurements.
 Decreasing CO and/or increasing NOx vehicular emissions must cause the observed decrease in the CO to NOx emission ratio; ambient measurements of CO can help to quantify the contribution of the former. The EPA reports ambient CO measurements [EPA, 2000b] from over 350 predominately urban and suburban monitoring sites nationwide. These data therefore represent some of the highest observed ambient levels of CO in populated areas, and vehicular emissions are primarily responsible for them. Figure 5 shows that these levels have indeed decreased over the last decade. The average annual national decrease is 5.2 ± 0.8% with similar values found for Nashville and Boulder, again indicating that the vehicle fleets in the two urban areas studied here are nationally representative. The observed annual decrease in CO ambient levels suggests that decreasing CO emissions account for most of the observed decrease of 7– 9% in the CO to NOx emission ratio. However, NOx emissions from vehicles must have increased by 2–3% annually to fully account for the observed decrease in the emission ratio.
 The EPA also reports ambient NO2 measurements [EPA, 2000b] from 225 monitoring sites. The reported trend over the last decade is not an increase but a decrease of ∼1% per year. However, this trend cannot be directly compared with the trend in vehicular NOx emissions for several reasons. First, in 1990, on-road vehicles accounted for only ∼30% of the NOx emissions compared with ∼60% of the CO emissions. Thus ambient levels reflect vehicle emissions more clearly for CO than for NOx. Second, for NO2 the arithmetic mean annual concentration is tabulated, while for CO the second highest maximum 8-hour average reported in each year is tabulated. The latter statistic captures the peak concentrations caused by vehicle traffic maxima. Third, the routine monitoring method for NO2 is subject to interferences, and the instrumentation has inadequate sensitivity to accurately quantify mean NO2 levels; Demerjian  discusses these issues. Consequently, we cannot use the ambient data to check our inferred rate of increase of vehicle NOx emissions.
 A continuing increase in vehicle NOx emissions is supported by some evidence. Emissions from heavy-duty diesel-powered trucks are especially rich in NOx but contain little CO [EPA, 2000a]; much less emphasis has been placed on controlling these emissions. Yanowitz et al.  show that the on-road diesel NOx emission per fuel volume burned has been stable, while diesel fuel consumption has been increasing [Sawyer et al., 2000]. This implies that diesel NOx emissions have been going up through the 1990s. The sales of light trucks (which include pickups, vans, and sports utility vehicles) have increased through the 1990s. Compared with cars, these vehicles have lower fuel economy and less stringent tailpipe emission standards for CO, hydrocarbons, and NOx. The first-order effect of this change is that more CO and NOx are now emitted per distance traveled, but the ambient ratio is not as affected since both CO and NOx are higher than they would be in the absence of more light trucks.
4. Conclusions and Implications
 We have shown that the ratio of CO to NOx emitted by U.S. vehicles has decreased rapidly, by 7–9 % per year over the last 12 years; this rate of decrease is underestimated by EPA emission inventories. The trend in ambient CO measurements shows that the decrease in the ratio is primarily due to a decrease in CO emissions (5.2 ± 0.8% per year). We infer that NOx emissions have increased (2–3% per year) to fully account for the changing ratio. Here we discuss the implications of these changes for tropospheric photochemistry on local, regional, and hemispheric scales.
 VOC and CO automotive emissions are controlled by the same technology with approximately the same efficiency, which implies that the VOC to NOx ratio in urban atmospheres has decreased in parallel with the CO to NOx ratio, also much differently than estimated in current emission inventories. Observational evidence for similar efficiencies of control for CO and VOC vehicle emissions comes from two sources. First, comparisons of vehicle fleets with very different levels of emission control show that reductions in CO and VOCs are highly correlated. Lonneman et al.  measured CO and VOCs in the Lincoln Tunnel in 1970 and 1982. CO and VOCs decreased by factors of 4.2 ± 1.4 and 3.9 ± 1.1, respectively, in the time between the two studies, a decrease that reflects the increasing utilization of catalyst-equipped vehicles. In the early 1990s, Zhang et al.  measured CO and VOCs from vehicle exhaust by remote sensing in 22 urban areas around the world. The vehicle fleet profile varied widely (for example, they included Los Angeles and Bangkok, Thailand), but the fleet mean percentages of CO and VOCs were highly correlated (r2 = 0.81, excluding fleets with significant fractions of two-stroke vehicles). A second line of evidence is more recent tunnel studies. Measurements taken in the Murfreesboro Pike Tunnel in Nashville [Harley et al., 2001] showed that the VOC/CO ratio in vehicle exhaust was virtually identical in the summers of 1995 and 1999, 0.29 ± 0.04 and 0.28 ± 0.06 ppbC/ppbv, respectively. Measurements in the Caldecott Tunnel in the San Francisco Bay area showed that VOC and CO emissions decreased by similar amounts between 1994 and 1997, 43 ± 8% and 31 ± 5%. These two studies indicate that the vehicular VOC/CO emission ratio remained nearly constant over the decade of the 1990s. The tunnel and remote sensing studies do not take account of nontailpipe sources of vehicular VOC emissions. However, their contribution is relatively minor, 7–35%, on a mass basis [Pierson et al., 1999]. Thus we conclude that the VOC to NOx ratio has declined in U.S. urban areas at a rate similar to the decline in the CO to NOx ratio.
 A decline in VOC to NOx has significant implications for U.S. urban photochemistry. In nearly all U.S. urban areas the most reactive fraction of the ambient VOC mix is emitted by vehicles and constitutes the most important anthropogenic source to fuel the urban photochemistry that is responsible for elevated ozone levels. The character of atmospheric photochemistry in urban areas depends sensitively on the VOC to NOx ratio; declining ratios in vehicle exhaust must have changed this character significantly. The EPA air pollutant emission inventories will not be adequate for accurate modeling of urban air pollution and for its evolution over decadal time scales until they adequately capture the changing vehicle emission ratios.
 On regional to hemispheric scales, quantifying the CO budget is critical. Since hydroxyl radical (OH) is the primary oxidant for many trace species in the atmosphere, the budget of OH is of central importance to the understanding of tropospheric photochemistry. Reaction with CO is the major loss process for OH in the bulk of the troposphere; thus for otherwise constant photochemical conditions the OH concentration is approximately inversely proportional to the CO concentration. Consequently, decreasing U.S. CO emissions constitute a significant perturbation to tropospheric photochemistry. Hallock-Waters et al.  report that CO levels in the U.S. Mid-Atlantic region have been decreasing at ∼5 ppbv/year from 1989 to 1997. They attribute this trend to reductions in U.S. emissions and to the decreasing trend in global background CO levels that has been reported [Novelli et al., 1998; Khalil and Rasmussen, 1994].
 Reductions in U.S. emissions have contributed significantly to the decreasing trend in global background CO concentrations. To put the U.S. reductions in a global context, we can estimate the decrease in CO emissions over the decade of the 1990s. The Global Emissions Inventory Activity (GEIA) and Emission Database for Global Atmospheric Research (EDGAR) emission inventories [Olivier et al., 1999] estimate annual U.S. on-road vehicle emissions as 65 Tg CO in 1990. From the rate of decrease of 5.2% per year that we infer from ambient U.S. measurements we estimate that these emissions have decreased to 38 Tg CO in 2000. That reduction of 27 Tg CO emitted annually corresponds to a 6% reduction in annual global fuel-use CO emissions (480 Tg CO in 1990) or to a 3% reduction in total global anthropogenic emissions (970 Tg CO in 1990, which includes all land use and agricultural waste burning). CO emissions may be decreasing similarly in other countries currently implementing automobile catalytic convertors, e.g., western Europe.
Karlsdóttir and Isaksen  modeled global OH concentrations and the resulting changing methane lifetime for the 1980–1996 period based upon assumed emissions of CO, NOx, and VOCs. Their results indicate nearly constant global average CO levels in contrast to the decreasing trend found by Novelli et al. . For North America and western Europe, Karlsdóttir and Isaksen  assumed CO emission decreases of ≈15–20% from 1985 to 1996. Our results suggest larger decreases of ≈30% for that period in North America. Even larger decreases may have occurred in western Europe; Kuebler et al.  report that inventories and ambient measurements in Switzerland both indicate 55% decreases in CO emissions from 1985 to 1998. Inclusion of the larger CO emission decreases for North America, and similar or larger decreases for western Europe, would bring the model results into closer agreement with Novelli et al. . Larger decreases in CO in the model input would, in turn, result in a larger modeled increase in OH concentration and a larger decrease in methane lifetime, although the magnitude of these secondary effects is impossible to judge without repeating the model calculations. It is important that the global atmospheric modeling community recognize the full extent of the decreasing CO emissions from the United States and western Europe.
 In summary, our results show that although the EPA underestimated the CO emissions from on-road vehicles at the beginning of the decade of the 1990s, emission control efforts since that time have been more successful than generally realized. By the end of the decade, vehicle CO emissions (and likely VOC emissions as well) had decreased to below EPA estimates. The American public has invested substantial resources in vehicle control efforts; the success of these efforts should be more widely appreciated. At the same time we note evidence of increasing NOx emissions from vehicles also not accurately reflected in EPA inventories. These increases may indicate the need to more adequately control emissions from heavy-duty diesel-powered trucks.
 The NOAA Climate and Global Change and Health of the Atmosphere Programs funded this research. The Nashville data were collected as part of the Southern Oxidant Study. We are grateful to Carl Howard, Tom Ryerson, and John Holloway for constructive discussion and criticisms and to Stu McKeen for providing the U.S. emission inventory values.