United States
      Environmental Protection
EPA 420-R-96-002
 November, 1996
EPA  Impact  of the  Oxyfuel
      Program on Ambient CO

         United States
         Environmental Protection
EPA 420-R-96-002
 November, 1996
\vEPA  Impact  of the  Oxyfuel
         Program on Ambient CO

Impact of the Oxyfuel Program on Ambient CO Levels
      J. Richard Cook, Phil Enns, and Michael S. Sklar

      U.S. Environmental Protection Agency, National Vehicle and Fuel Emissions

      Laboratory, Ann Arbor, Michigan


A regression model judged to best represent mean quarterly ambient CO data indicated that sites

which implemented the winter oxyfuels program in 1992 experienced a significant reduction in

mean quarterly ambient CO that was sustained in subsequent years. This reduction occurred over

and above the long-term trend of a decrease in ambient CO and was not observed in areas that

did not implement the program.


Regression models were fit to quarterly mean ambient CO data from monitor sites which first

implemented a winter oxygenated fuel program in the winter of 1992/93 and still had the

program in place during the last quarter of 1994, or which had not implemented such a program

through 1994.  The regressions revealed a statistically significant trend of lower CO levels over

time, most likely due  to the effects of fleet turnover. According to the model which best

represents the data based on statistical criteria, CO levels were observed to fall by 2.1 to 3.4%


per year in oxyfuel and non-oxyfuel areas combined, based on the 95% confidence interval. The

analysis also revealed that ambient CO trends for the 1985/1986 winter season through the last

quarter of 1994 are consistent with the hypothesis that ambient CO levels have been reduced

significantly by implementation of the winter oxygenated fuels program. Ambient CO levels in

areas which implemented an oxygenated fuels program in 1992 fell by 3.1 to 13.6 percent

beyond what would have been expected from the long-term trend based on the 95% confidence

interval. This reduction was estimated from data covering 6 months over the winter season,

whereas the oxyfuel season is typically 4 months in duration for most areas.  Thus, a reduction

adjusted for the actual duration of the oxyfuel season, assuming in particular no spillover into the

nonoxyfuel season, would be about 5  to 20%. Individual cities typically do have some spillover,

and some have an oxyfuel season shorter or longer than 4 months.


A winter oxygenated fuels program is required by Section 21 l(m) of the Clean Air Act

Amendments of 1990 in certain areas  of the country that exceed the National Ambient Air

Quality Standards (NAAQS) for carbon monoxide (CO). In most areas, the program requires 2.7

weight percent oxygen on average to be present in the gasoline sold in the program areas, which

is equivalent to 15 volume percent methyl tertiary butyl ether (MTBE) or 7.8 volume percent

ethanol in gasoline. These two compounds are the most common oxygenates used to comply with

the oxygenated fuels program requirements. The oxygenates are intended to make the

combustion process more complete by providing more oxygen for combustion, thereby reducing


the rate of carbon monoxide production. As a result of this requirement, a large number of areas

implemented a winter oxygenated fuels program beginning in November 1992.

       Numerous dynamometer studies have demonstrated that the addition of oxygenates to

gasoline reduces CO emissions substantially in both older and newer technology vehicles under a

standardized driving cycle and at temperatures above SO0?.1"4 In these studies, CO emission

reductions have ranged from 2 to 10% per weight percent oxygen.5'6 Reductions have also been

found at temperatures between  20F and 55 F, but data are lacking at temperatures below

20F.7 Several tunnel and remote sensing studies have confirmed that emission reductions occur

under the specific real world conditions observed in those studies.8"11 Recent studies have

attempted to determine statistically whether the CO emission reductions observed under specific

laboratory, tunnel, or remote sensing conditions are translated into significant reductions in

ambient levels of CO in areas that have implemented a winter oxygenated fuels program.  In some

cities with winter oxygenated gasoline programs, a reduction in ambient CO concentrations of

approximately 10% has been observed and attributed to the use of the oxygenate.12"14 In other

cases, however, a reduction either could not be quantified or was not statistically significant.15"21

       The analyses in this study were designed to address whether ambient CO levels are

consistent with a hypothesis that ambient CO levels have been reduced significantly by

implementation of the winter oxygenated fuels program. This study initially grew out of an

investigation into a small upturn in nationwide mean second maximum CO values and quarterly

average CO levels, especially in oxyfuel areas, observed for the last quarter of 1993/first quarter

of 1994 relative to 1992/93. The 1993/94 increase in second-maximum CO levels was not found

to be a statistically significant departure from the long-term downward trend seen from the

1985/86 winter season through the last quarter of 1994.
The data used in the following analyses consisted of mean ambient quarterly CO data from 310

sites in approximately 150 Metropolitan Statistical Areas over the years 1980 through 1994

(Appendix). Quarterly means values were chosen for analysis in part because they were the most

readily accessible to the authors, and in part because quarterly averages would be less influenced

by short-term meteorological variations ignored in the analysis.  Data used in the analyses can be

obtained from the authors. One-hundred seventeen (117) sites were from areas where ambient


monitoring CO data were available, that had implemented the oxyfuel program in 1992/93, and
            - '                                I;

still had the program in place during the fourth quarter of 1994. One-hundred and ninety-three

(193) sites were from areas where ambient monitoring CO data were available and had never

implemented an oxyfuel program. These data were selected so that a direct comparison could be

made of changes hi ambient CO from areas which had implemented an oxyfuel program with

changes in ambient CO from areas which had not implemented such a program. Data from areas

which had implemented an oxyfuel program prior to 1992/93 were excluded, as were data from

areas which had implemented a program in 1992/93 but discontinued it. Data were analyzed on a

season-by-season basis (i.e. the fourth quarter of one year and the first quarter of the following

year) rather than a year-by-year basis, as is done in the U.S. EPA's National Air Quality and


Emissions Trends Report.22'23  For the 1994/95 winter season, only data for the fourth quarter of

1994 were available.

       Mean quarterly CO values are the mean of all one hour monitor values for that quarter.

The minimum detectable limit for individual monitor readings is 0.5 ppm.  Values below 0.5

ppm are recorded as 0.25 ppm. Readings are reported in 0.1 ppm increments as a convention.

Monitors are audited by EPA for accuracy (how close measurements are to known standards) on

an annual basis and precision (repeatability of measurements) on a weekly basis. Accuracy of the

monitors must be within 20%  for standards in the following ranges: 2-8 ppm,  15-20 ppm, 35-45

ppm, and 80-90 ppm.  Precision must be within 15% for an 8-10 ppm standard.

       Mean quarterly CO data were used rather than second maximum CO values because

mean values are likely to be less sensitive to day to day fluctuations such as meteorology since

they represent three months of measurements rather than a single measurement. Data for some

sites were missing for certain quarters in certain years.

       There were substantially more missing values for 1980 through the 1984/85 winter season

than for later seasons. Of the 9300 potential observations for 1980 through 1994, 1261 values are

missing. Only 93 sites  have the full 30 observations.  However, for data from 1985/86 through

1994, only 136 values  are missing out of 5890 total data points, and 219 sites have  the full 19

observations. Also, there was much more scatter in data for 1980 through 1984/85.  Furthermore,

since there is a potential non-linearity in the secular trend, using data from earlier years to


linearize the trend may increase the error in the later years. Thus, the analyses below focus on

data from  1985/86 through 1994, although results from the complete data set are included as well

for comparative purposes.

       It should also be noted that in the analyses described below, no effort was made to

estimate or interpolate missing points. Consequently, different sites and years receive slightly

different weight in the various regression calculations.

       Figure 1 displays box plots of the data for oxyfuel and non-oxyfuel areas by year. Higher

average levels of mean ambient CO for Oxyfuel program sites are clearly shown. The plots also

reveal positive skewness in the distributions, suggesting that at any time a few sites are likely to

produce unusually high readings.


These data were used to estimate a series of regression models. The following linear model

forms the basis for this study:

                   Ln(CO) = B0 + B,*PROG + B2*YEAR + B3*POST92

                             + B4*OXPROG + B5*PXYEAR +e                   (1)


                _   i 0 if non-oxyfuel program site
                   \ i  -c    _c   i
                      1  if oxyfuel program site

       YEAR   =     Calendar Year - 1986 (limited data set); or

                      Calendar Year - 1980 (complete data set)

       POST92  =   ( Otfpre-Quarter4,1992

        ^           { 1  if Quarter 4,  1992 or later
       OYPROr -  ( 0 if pre-Quarter 4, 1992 or non-oxyfuel program site

                      1 if Quarter 4, 1992 or later and oxyfuel program site
       DYVP A P      / if non-oxyfuel program site
       r AI IlAK  =   l-irr-.AT-.-r   r   i
                      YEAR if oxyfuel program site
       e         =    error term

       In the model above, calendar year 1986 represents the 1985/86 oxyfuel season, 1987

represents the 1986/87 oxyfuel season, and so on. The coefficients have the following


       B0 -  Intercept for non-program sites;

       B, -  Effect (on intercept) of any time invariant factors unrelated to the oxyfuel program

             that nevertheless systematically differ between program sites and non-program

             sites variables;

       B2 -  Long-term reduction trend in ambient CO levels for all sites (both oxyfuel

             program and non-program sites);

       B3 -  Shift in Ln(CO)-intercept beginning in Winter 1992/93 for all sites (both oxyfuel

             program and non-program sites);

        B4 -  Shift in the Ln(CO)-intercept for oxyfuel sites only, beginning in 1992/9324;


       B5 -  Difference in the slope of the regression lines between oxyfuel areas and non-

             oxyfuel areas.

       Coefficient B, will reflect the fact that it was the sites with conditions leading to higher

CO concentrations that were chosen by Congress to have an oxyfuel program. With the natural

log transformation used in this model, the last four regression coefficients are usefully interpreted

as exponential rates of decay in ambient CO over time (B2 and B5), incremented or decremented

by the intercept shifts associated with introduction of the oxyfuels program (B3 and B4). The

transform also can serve to stabilize the variance over time. The need for doing this is somewhat

evident in Figure 1. Figure 2 demonstrates that the transformation did, in fact, stabilize the


       Least squares regression analyses were performed on the logarithms of the mean ambient

CO data from individual sites, using the SAS PROC GLM procedure. Long-term trend (YEAR)

presumably captures reductions in ambient CO resulting from mobile source control programs.

The one-time shift in the intercept for all sites (POST92) captures any CO shift for 1992/93 not

associated with the oxyfuel program. Possible effects of the oxyfuel program include two

interaction terms - a one-time shift in the intercept (OXPROG) of the oxyfuel program site trend

lines, and the change in slope (PXYEAR) of these lines.

       The purpose of this analysis is to address the primary hypothesis that the oxyfuel program

has reduced CO emissions. Analyses of ambient CO data cannot prove the hypothesis, but can


determine if the data are consistent with it. In particular, after adjusting for trend (YEAR and

PXYEAR) a determination can be made of whether the direction and significance of any shift in

average CO in the oxyfuel program sites (OXPROG) is distinct from a general shift across all

sites (POST92) following implementation. A significant negative value of B4 and a

nonsignificant value of B3 supports the primary hypothesis that the oxyfuel program has reduced

CO emissions.

       The other variables potentially account for much of the variation in the data. If, as

expected, the replacement of older vehicles with newer, cleaner vehicles more than offsets the

yearly increase in vehicle-miles traveled (VMT) in both oxyfuel and non-oxyfuel areas, a

negative value for B2 will be observed. A difference in this trend between sites in oxyfuel areas

and nonoxyfuel areas would produce a significant value for B5, although the direction of this

value was not anticipated in this research. Similarly, any general shift in the level of ambient CO

coinciding with oxyfuel program implementation (B3), possibly due to meteorology,  is not

predictable a priori but may be present.


Table 1 presents regression coefficient estimates for two models using the limited data set of

1986 and later data  one is represented by equation (1) (Model A) and one is a reduced model

(Model B). Analagous regressions using the complete data set, which includes the data from

1980 through 1984/85, are included for comparative purposes (Models C and D). The R2 values


for the Models in Table 1 are relatively low - 0.190, 0.190,0.186, and 0.184 for Models A, B, C,

and D, respectively. These values are low because the models group many sites into two

categories and do not account for the effect of each site on mean ambient CO.25 The coefficient

estimates for all models support claims about the collective efficacy of mobile source programs.

The underlying trend in mean ambient CO, represented by the variable YEAR, is significant in

all four models. The coefficient for this variable (B2) ranges from -0.025 to -0.030, which

suggests an average annual decrease in ambient CO of 2.5% to 3.0%. The long-term decline in

ambient CO levels in both oxyfuel and non-oxyfuel areas are likely due to a combination of fleet

turnover towards cars built to meet more stringent standards, improved durability of vehicle

emission controls, vehicle inspection and maintenance programs in some areas, and more

stringent controls on stationary sources that more than offset the effects of economic growth and

increases in VMT. The step change associated with implementation of the oxyfuel program is

apparent in the coefficient for the OXPROG term seen in all four models, where the estimated

average effect across program sites is 11.4% in Model A, 8.3% in Model B,  14.0% in Model C,

and 5.0%  in Model D. This coefficient is statistically significant in all models except Model D.

       As mentioned earlier, because of missing values, data scatter, and the potential non-

linearity of the overall trend, use of the complete data set is likely to introduce more error into the

results.  Thus, Models A and B, based on the limited data set,  are judged to be superior to

Models C and D.  Between Models A and B, B appears most satisfactory as a representation of

anticipated benefits from the oxyfuel program, based on statistical criteria. As indicated above,

this model includes all the terms in equation 1 except PXYEAR, the interaction term


representing the difference in the time trend between oxyfuel and nonoxyfuel areas. Where

PXYEAR was included in Model A, it was the only marginal t statistic not significant at the 5%

level. Since this interaction term was non-significant in Model A it was removed from the model.

Although POST92 was also non-significant, it had to be retained in the model because OXPROG

represents the interaction between POST92 and PROGRAM. Eliminating POST92 would bias

OXPROG upward through confounding of coefficients. The significance of the OXPROG term

in Model B, in contrast with the non-significance of POST92, .indicates that there was a

statistically significant step decrease in ambient CO for areas which implemented the oxyfuel

program beginning in 1992/93, but there was not a statistically significant, corresponding

decrease in nonoxyfuel areas. No mobile source program that could significantly affect ambient

CO other than the oxyfuel program was implemented nationwide at this time.  Many of the

oxyfuel areas have vehicle emission inspection and maintenance programs, but in all or virtually

all cases these began prior to 1985/86, and most have not made significant changes since

(California did make changes in 1990).

       Models C and D are included here to show that including the data from earlier years have

a significant impact on the results of the analysis.  Whereas the slope change variable (PXYEAR)

was not significant in Model A, it is slightly greater than zero and significant in the analogous

Model C.  This would seem to indicate that the average decrease in ambient CO over time for

areas which implemented the oxyfuel program is slightly lower than for nonoxyfuel areas. When

this term is dropped in  Model D, the coefficient for OXPROG (B4) is still negative, but

marginally insignificant. Statistically, this could be explained by collinearity of OXPROG with


PXYEAR in Model C.  Because the PXYEAR term is significant when data from earlier years

are included, Model C is judged to be superior to Model D.

        Figure 3 depicts the regression results using Model B. In Figure 3, the individual points

represent mean quarterly ambient CO for all oxyfuel areas or nonoxyfuel areas in a given season.

Specifically, these points are the mean of mean CO values for each quarter at each site in that

class (For oxyfuel areas, this is the mean of 236 total possible data points or fewer where there

were missing values at given sites in a given season. For nonoxyfuel areas, this is the mean of

386 total possible data points or fewer where there were missing values at  given sites in a given

season). The regression lines for oxyfuel and non-oxyfuel areas depict the average y-intercept

values across all sites.

       In summary, the results of Model B are consistent with a long term CO reduction trend of

2.8%. They are also consistent with an oxyfuel program effect of about an  8% one-time step

decrease in ambient CO over and above the long-term trend beginning in the winter of 1992/93,

for areas which implemented an oxyfuel program at that time. The 95% confidence interval

associated with the long- term trend is 2.1  to 3.4 %  and for the step change coinciding with

oxyfuel program implementation it  is 3.1% to 13.6%. It should be mentioned that the estimate of

the step change is based on data from the fourth quarter of one year and the first quarter of the

next, covering a total of 6 months. However, in most oxyfuel areas, the oxyfuel season is four

months in duration. Assuming no step decrease in ambient CO during the non-oxyfuel season

(i.e., no spillover of oxygenated gasoline use into the non-oxyfuel season),  a 5 tc 20% step


decrease in ambient CO during the months the program is in effect would be expected, based on

the 95% confidence interval. However, there are data from Denver indicating that the spillover

effect is significant; thus, an assumption of no spillover at all is too strong. Also, some cities

have oxyfuel programs of less than or more than 4 months.

       One potential concern with this analysis is the possibility of correlation among errors.

Errors will not be independent if seasonal deviations from the trend line are correlated among

sites. Since the air quality data from both oxyfuel and non-oxyfuel sites are distributed across a

wide geographic area, it is unlikely that, given the sample size, such correlation would change the

statistical conclusions.

       In response to reviewer comments, several modifications of these analyses which are not

presented in this paper were performed. First, we observed that mean ambient CO values for the

fourth quarter of one year tended to be higher than values for the first quarter of the following

year. This is likely due to more inversion and less wind movement in the fourth quarter. When

we added a term to the model to account for this, it did not significantly affect our results.

Models were fit in which the effect of each site was included as well. Trend lines were fit to the

logged time-series data for each site which were parallel to one another. The site factor captured,

for example, differences among sites in traffic density and dispersion conditions. However,

including a site factor in the analysis possibly confounded the program site-program

implementation interaction, which is the central concern of this analysis. The potential

confounding could not be avoided since every site belongs to either the non-program or program


set, but not both.  To avoid this possible confounding, analyses with the site factor were also

done for oxyfuel and non-oxyfuel sites separately. In addition, data were used to fit models in

which a site-year interaction term was included. This effectively allowed the trend line slope to

vary among sites. Finally, the models were fitted to a data set comprised of data only from those

sites for which no missing values existed. In this way, all included sites had equal weight and

gaps in the time series for individual sites were removed. The results from these additional

analyses, not detailed here, were consistent with the hypothesis of an oxyfuel program benefit

and can be obtained from the authors.


The analyses discussed in this memo have a number of limitations. Most importantly, they do not

explicitly control for meteorological factors such as temperature, wind speed, and mixing height;

however, using mean ambient CO values rather than second maximum CO values for each

season should  make the analyses  less sensitive to short-term meteorological fluctuation, since a

second maximum value can reflect one severe meteorological episode within the season. Also,

since the data for each oxyfuel season include means for two 3-month quarters, mean ambient

CO levels will be based partly on levels outside the  oxyfuel season. Moreover, these analyses

include data from areas across the nation, and the results cannot be used to address whether

benefits can be found in one  specific region with a specific type of meteorology, such as the

Northeast. In addition, the analyses do not account for differences in types of oxygenate used or

levels of oxygen in the fuel among oxyfuel areas. It should also be noted that the step decrease


associated with oxyfuel program implementation cannot by this analysis alone be attributed

solely to specific fuel properties such as oxygen content. Furthermore, the models assume that

the decrease in ambient CO over time is log-linear. Two lines were fit to the data, one for years

before 1992/93, and one for 1992/93 and later years. In the preferred Model B, these lines were

forced to have the same slope; that is, we assumed that the rate of decrease over time in mean

ambient CO remained the same before and after implementation of the oxyfuel program. Model

B also makes an additional assumption that the slope is the same for both oxyfuel and

nonoxyfuel areas. Finally, size of the estimated benefit of oxygenated fuel was sensitive to the

terms and data included in the model. Thus, although the model judged to be superior based on

statistical criteria is suggestive of an 8% program benefit and all models indicated a benefit,

because of inherent limitations of the analysis, the quantitative results must be interpreted with


                              Comparison to Other Studies

Ambient Air Quality Studies.  A 3.1 to 13.6% step decrease (up to 5 to 20% if adjusted on the

assumption of no spillover) in ambient CO is consistent with the estimates from several studies

of the ambient air quality effects of oxyfuels which attempted to control for meteorological

effects. For instance, CO reductions at monitoring sites in California during winter were

measured at 10-35% lower for the  1992/93, 1993/94, and 1994/95 oxyfuel seasons compared to

earlier years.  Dolislager of the California Air Resources Board estimated that a 5-10% reduction

in CO was attributable to the oxyfuel program, using NOX and CO as independent indicators of


the effects of meteorology on air quality.13'14 It should be pointed out, however, that California

uses 2 weight percent oxygen in the fuel compared to 2.7 weight percent in most other oxyfuel

areas. Assuming that the relationship between oxygen content and CO benefit is linear,

Dolislager's results imply that a 2.7 weight % oxygenate program in California would result in a

6.8% to 13.5%  CO benefit. A CDC study estimated CO benefits in 11 western states using 1.5-

2.7 weight percent oxygen during the period 1986 through 1992.12 The CDC researchers found

that areas using oxygenates in these states had approximately 10% greater reductions in ambient

CO than those that did not. Reductions of this magnitude were still found when efforts were

made to account for temperature and wind speed.

       However, analyses at the University of Colorado, which did not control for meteorology,

generally found ambient CO reductions associated with the oxyfuel program were not

statistically significant.18"20 These studies  used a regression technique known as structural time

series analysis. The authors analyzed data from the Denver CO area; Albuquerque, NM; Phoenix,

AZ; Las Vegas, NV; and Reno, NV. Only data from Phoenix showed a statistically significant

effect of the oxyfuel program. However, depending on the sites, reductions of 7-18% were

required to observe statistical significance. In a number  of other studies of ambient air quality

effects of oxyfuels, a reduction either could not be quantified or was not statistically


Emissions Studies. CO exhaust  emissions from vehicles operating at temperatures of 50F and

higher are reduced by oxyfuels by about 2 to 10% per weight % oxygen, which represents a


reduction of 5 to 27% for a blend containing 2.7% oxygenate by weight.5 A recent tunnel study in

California, with a 2% oxygenate program, indicated a 21% reduction in CO emissions during the

oxyfuel period relative to the non-oxyfuel season.11 Studies by Bishop and Stedman using remote

sensing found 6-16% decreases in CO emissions during the oxyfuel season.8'9 Conversely,

another remote sensing study in Raleigh, North Carolina did not find a significant reduction in

CO that could be attributed to the oxyfuel program, but this study was confounded by seasonal

temperature differences.26

Model Predictions.  EPA's Complex Model for CO predicts a reduction in CO emissions with a

2.7 weight percent oxygen blend of about 9.2% for normal emitting 1990 model year vehicles

and 5.0% for high-emitting vehicles.6  EPA's MOBILESa emission factor model predicts an

average reduction in CO emissions for the on-road fleet of about 30% for a 2.7% oxygenate by

weight gasoline.5 EPA's Air Quality Trends  Report indicates that on-road mobile sources account

for about 60% of the nation's total annual CO emissions and 70% of total winter CO emissions.22

Thus a 30% emission reduction in winter for motor vehicles would translate into about a 21% air

quality benefit for that season, provided the  on-road mobile source contribution to urban CO

emissions is the same in winter as it is for the entire  nation.

                                Potential Future Work

A potential follow-up to this study could include analyses for December and January only. All

areas with an oxyfuel program are selling oxyfuel during these two months. In addition, many


CO exceedances occur during these two months. It might also be worthwhile to analyze data

from the areas which initially implemented an oxyfuel program but subsequently dropped it.

Another option would be to compare the ratios of mean quarterly ambient CO values in summer

and winter before and after implementation of an oxyfuel program. Also, analyses could be

restricted to the Northeast and North Central regions to assess the impact of the program in areas

where winter temperatures are colder. Moreover, several recent studies have attempted to adjust

air quality data for meteorological effects.l3-14'27"29 The applicability and reliability of these

methodologies to these data should be evaluated. Yet another option for potential future work

would be to perform a meta-analysis combining all sites, but allowing each to have its own

unique CO trend and downward CO shift beginning in 1992/93. Similarly, a nested regression

technique could be used, whereby only those sites whose trend deviated significantly from the

overall trend would have a unique trend. Finally, more emission studies at low temperatures.

could be done to determine if CO emission benefits found at temperatures above 50F for

oxygenated fuel are also found at lower temperatures.


The authors would like to acknowledge the contribution of Stuart Romanow in performing some

of the initial data analysis that formed the basis of this study.  We would also like to thank the

following individuals for comments on previous versions of this manuscript: Phil Lorang,

Venkatesh Rao, Warren Freas, Rhonda Thompson, Larry G. Anderson, Pamela Wolfe, Carl

Howard, Douglas R. Lawson, David Mannino, S. T. Rao, Jim Hyde, Ron Melnick, Barry

McNutt, and H. T. McAdams.


J. Richard Cook and Phil Enns are Environmental Scientists and Michael S. Sklar is a Group

Manager at the U. S. EPA's Office of Mobile Sources. All three authors can be contacted at the

following address:  U.S. EPA, National Vehicle and Fuel Emissions Laboratory, 2565 Plymouth

Road, Ann Arbor, Michigan, 48105.



1.     Auto/Oil Air Quality Improvement Research Program.  "Technical Bulletin No. 6:

       Emission results of oxygenated gasoline and changes in RVP," 1991.

2.     U. S. Environmental Protection Agency. "CO Reduction With the Use of Oxygenated

       Blends," Memorandum from Greg Janssen to Phil Lorang, July 31,1991.

3.     Mayotte, S. C; Lindhjem, C. E.; Rao, V.; Sklar, M. S.  "Reformulated Gasoline Effects

       On Exhaust Emissions: Phase I: Initial Investigation of Oxygenate, Volatility, Distillation

       and Sulfur Effects,"  Society of Automotive Engineers (SAE) Paper 941973,1994a.

4.     Mayotte, S. C.; Lindhjem, C. E.; Rao, V., Sklar, M. S.  "Reformulated Gasoline Effects

       on Exhaust Emissions: Phase II: Continued Investigation of the Effects of Fuel Oxygen

       Content, Oxygenate Type, Sulfur, Olefins and Distillation Parameters," SAE Paper


5.     Howard, C. J.; Russell, A.; Atkinson, R.; Calvert, J. "Air Quality Benefits of the Winter

       Oxyfuel Program," Draft Technical Report Prepared For the Office of Science and

       Technology Policy, Executive Office of the President,  1996.


6.      Rao, V.,  "Development of an Exhaust Canbon Monoxide Emissions Model," S AE Paper


7.      Hood, J.; Farina, R. "Emissions From Light Duty Vehicles Operating On Oxygenated

       Fuels at Low Ambient Temperatures: A Review of Published Studies," SAE Paper


8.      Bishop, G. A.; Stedman, D. H.  "Oxygenated Fuels  A Remote Sensing Evaluation,"

       SAE Paper 891116,1993.

9.      Bishop, G. A.; Stedman, D. H. "On-Road Carbon Monoxide Emission Measurement

       Comparisons For the 1988-1989 Colorado Oxyfuels Program," Enviro. Sci. Technol,

       1990, 24, 843-847.

10.     PRC Environmental Management, Inc. "Final Report of the Performance Audit of

       Colorado's Oxygenated Fuels Program,"  December 1992.

11.     Kirchstetter, T. W.; Singer, B. C; Harley, R. A.; Kendall, G. R.; Chan, W.  "Impact of

       Oxygenated Gasoline Use on California Light-Duty Vehicle Emissions," Environmental

       Science Technology, 1996, 30, 661-670.


12.    Mannino, D. M.; Etzell, R. A. "Are Oxygenated Fuels Effective? An Evaluation of

       Ambient Carbon Monoxide Concentrations in Western States, 1986 to 1992,"  J. Air &

       Waste Manage. Assoc., 1996, 46, 20-24.

13.    Dolislager, L. J. "Did the Wintertime Oxygenated Fuels Program Reduce Carbon

       Monoxide Concentrations in California."  In  Proceedings of the 10th International

       Symposium on Alcohol Fuels, Colorado Springs, CO, November 7-10,1993.

14.    Dolislager, L. J. "The Effect of California's Wintertime Oxygenated Fuels Program On

       Ambient Carbon Monoxide Concentrations." Submitted to J. Air & Waste Manage.

       Assoc., 1996.

15.    Cornelius, W. L. "Effects of North Carolina's Oxygenated Fuel Program On Ambient

       Carbon Monoxide Concentration,"  Report to the North Carolina Department of

       Environment, Health, and Natural Resources Air Quality Section, 1995.

16.    Keislar, R. E.; Bowen, J. L.; Fujita, E. M.; Lawson, D. R. "Effect Of Oxygenated Fuels

       On Ambient Carbon Monoxide Concentrations in Provo, Utah,"  Report by Desert

       Research Institute, 1995.

17.    Vogt, D. An evaluation of the Effect of North Carolina's 1992-93 Oxygenated Fuel

       Program on Ambient Carbon Monoxide Levels in Urban Areas, State Center for Health


       and Environmental Statistics, North Carolina Department of Health and Environmental

       Resources, Special Report 80,1994.

18.    Wolfe, P.; Anderson, L. G.; Lanning, J. A.; Jones, R. H. "A Structural Time Series

       Assessment of the Effectiveness of the Oxygenated Fuel Program in Reducing Carbon

       Monoxide Concentrations in Five Western U.S. Cities," Air and Waste Management

       Association 87th Annual Meeting, Cincinnati, OH, Paper 94-WP91.04,1994.

19.    Anderson, L. G.; Wolfe, P; Barrell, R. A.; Lanning, J. A. "The Effects of Oxygenated

       Fuels On the Atmospheric Concentrations of Carbon Monoxide and Aldehydes in

       Colorado,"  In Alternative Fuels and the Environment, edited by F. S. Sterrett, Lewis

       Publishers, Boca Raton, FL, 1994.

20.    Wolfe, P.; Anderson, L. G.; Lanning, J. A.; Jones, R. "Techniques For Assessing the  .

       Effectiveness of Oxygenated Fuels,"  Paper to be Presented at the Air and Waste

       Management Association 89th Annual Meeting, Nashville, TN, Paper 96-WP89.06,1996.

21.    Heil, C. L. Assessment of the Anchorage Oxygenated Fuels Program on Ambient Carbon

       Monoxide Concentrations, Master of Environmental Quality Science Thesis, School of

       Engineering, University of Alaska, Anchorage, 1993.


22.    U. S. Environmental Protection Agency. National Air Quality and Emission Trends '

       Report, 1994, Report No. EPA-454/R-95-011,1995.

23.    If data were analyzed on a year-by-year basis, it would not be possible to compare

       ambient CO from one oxyfuel season to the next, since each season spans the latter part

       of one year and the first part of the next.

24.    Although an observed one-time, sustained step decrease in ambient CO beginning in

       1992/93 would most likely be due to the oxyfuel program, other changes could possibly

       contribute to such a decrease. According to the U.S. EPA's Complex Model for CO, fuel

       oxygen and sulfur content have  the heaviest influence on exhaust CO emissions, with

       very little influence from such parameters as RVP, distillation characteristics and olefms.

       However, there were no mandated changes in sulfur content or other fuel parameters in

       this time period, and the lower cost of higher sulfur crude oils would be expected to

       discourage significant sulfur reductions in 1992 and later years.


25.    Analyses which included a site factor were performed but are not included here. In these

       analyses, trend lines were fit to the logged time-series  data for each site which were

       parallel to one another. Adding this site factor increased R2 values to about 0.8 but did not

       affect the conclusions made in this paper.


26.    Rhudy, S. A.; Rodgers, M. O.; Vescio, N." Seasonal Measurements of Motor Vehicle

       Emissions by Remote Sensing: Raleigh, North Carolina Oxygenated Fuels Program,"

       Submitted to:  J. Air & Waste Manage. Assoc.

27.    Glen, W. G.; Zelenka, M. P.; and Graham, R. C.  "Relating Meteorological Variables and

       Trends in Motor Vehicle Emissions to Urban Carbon Monoxide Concentrations,"

       Atmospheric Environment, 1996, 30, 4225-4232.

28.    Rao, S. T.; Zalewski, E.; Zurbenko, I. G. "Determining Temporal and Spatial Variations

       in Ozone Air Quality," J. Air & Waste Manage. Assoc. 1995, 45, 57-61.

29.    Flaum, J. B.; Rao, S. T.; Zurbenko, I. G. "Moderating the Influence Of Meteorological

       Conditions On Ambient Ozone Concentrations," J. Air & Waste Manage. Assoc., 1996,

       46, 35-46.

 Table 1. Estimates and T-values (in Parentheses) for Variables Used in Regressions of Log

Mean Ambient CO data.







* YV-XXL-f/tVlX ^***


*Significant at the p<0.05 level

Figure 1. Box Plots of Oxyfuel and Nonoxyfuel areas by year (non-transformed data).

Figure 2. Box Plots of Oxyfuel and Nonoxyfuel areas by year (natural log transformed data).

Figure 3. Regression of Log Mean Ambient CO Air Quality Data Using Model B (described in



   79   80   81   82  ' 83
 84   85   86   87   88   89   90   91   92   93   94   95


SITE TYPE   *  * NONOXY   *-*-* OXY




* *








A a
k *


^ >

* *




  79   80   81   82   83   84   85   86   87   88   89   90   9'l   92  93   94   95   96

                     SITE TYPE  * * * NONOXY



 Metropolitan Statistical Areas (MSAs) or Counties and Number of Monitor Sites Included
                                   in Analyses
 MSA or County                                         Number of monitor sites
 Nonoxyfuel Areas:
 Birmingham, AL                                         5
 Tuscon, AZ                                             3
 Bakersfield, CA                                          2
 Mono County, CA                                        1
 Salinas, CA                                             1
 Vallejo-Fairfield-Napa, CA                                2
 Orange County, CA                                       3
 San Luis Obispo-Atascadero-Paso Robles, CA                1
 Santa Barbara-Santa Maria-Lompoc, CA                     2
 Santa Cruz-Watsonville, CA                               1
 Santa Rosa, CA                                          1
 Visalia-Tulare-Porterville, CA                              1
 Ventura, CA                                            2
 Greeley, CO                                             1
 Fort Lauderdale, FL                                       4
 Miami, FL                                              2
 Jacksonville, FL                                          4
 Tampa-St. Petersburg-Clearwater, FL                        7

 MSA or County                                         Number of monitor sites

 Orlando, FL                                             2
 West Palm Beach-Boca Raton, FL                           1
 Atlanta, GA                                             1
 Honolulu, HI                                            2
 Boise City, ID                                            1
 Chicago, EL                                              4
 St. Louis, MO-IL                                         6
 Peoria-Pekin, IL                                          1
 Springfield, IL                                            1
 Rockford, IL                                             1
 Fort Wayne, IN                                           1
 Gary, IN                                                 1
 Cedar Rapids, IA                                         1
 Des Moines, IA                                           3
 Wichita, KS                                              3
 Kansas City, MO-KS                                      5
 Owensboro, KY                                          1
 Lexington, KY                                            1
 Evansville-Henderson, IN-KY                               1
 Louisville, KY-IN                                         3
 Cincinnati, OH-KY-IN                                     3
McCracken County, KY                                    1
New Orleans, LA                                         2
Cumberland, MD-WV                                     1

 MSA or County                                         Number of monitor sites

 Springfield, MA                                          1
 Lowell, MA-NH                                          1
 Boston, MA-NH                                          2
 Grand Rapids-Muskegon-Holland, MI                       1
 Detroit, MI                                              5
 Rochester, MN                                           1
 Springfield, MO                                          1
 Missoula County, KY                                     1
 Omaha, NE-IA                                           2
 Lincoln, NE                                              1
 Douglas County, NV                                      1
 Nashua, NH                                              1
 Middlesex-Somerset, Hunterdon, NJ                         1
 Las Cruces, NM                                          2
 San Juan County, NM                                     1
 Santa Fe, NM                                            1
 Buffalo-Niagara Falls, NY                                  3
 Rochester, NY                                            2
 Albany-Schenectady-Troy, NY                              1
 Charlotte-Gastonia-Rock Hill, NC-SC                        3
 Columbus, OH                                            3
 Stuebenville-Weirton, OH-WV                              1
Toledo, OH                                              1
Dayton-Springfield, OH                                    2

MSA or County  	;	Number of monitor sites

Canton-Massilon, OH                                     1
Oklahoma City, OK                                       2
Tulsa, OK                                               2
Josephine County, OR                                     1
Eugene-Springfield, OR                                    1
Pittsburgh, PA                                            5
Reading, PA                                             1
Johnstown, PA                                           1
Harrisburg-Lebanon-Carlisle, PA                            1
Erie, PA                                                 1
Scranton-Wilkes-Barre-Hazelton, PA                         2
Allentown-Bethlehem-Easton, PA                           2
York, PA                                                1
Providence-Fall River-Warwick, RI-MA                      2
Charleston-North Charleston, SC                            1
Columbia, SC                                            1
Nasheville, TN                                           3
Johnson City-Kingsport-Bristol, TN-VA                      1
San Antonio, TX                                          2
Dallas, TX                                               2
Houston, TX                                             5
Beaumont-Port Arthur, TX                                 1
Fort Worth-Arlington, TX                                  2
Salt Lake City-Ogden, UT                                  4

MSA or County
Number of monitor sites
Burlington, VT
Norfolk-Virginia Beach-Newport News, VA-NC
Richmond-Petersburg, VA
Yakima, WA
Charleston, WV
Wheeling, WV-OH
Madison, WI
Milwaukee-Waukesha, WI
Racine, WI
San Juan-Bayamon, PR
Oxyfuel Areas
Oakland, CA
Chico-Paradise, CA
Sacramento, CA
Fresno, CA
Los Angeles-Long Beach, CA
San Francisco, CA
Riverside-San Bernardino, CA
San Diego, CA
Stockton-Lodi, CA
San Jose, CA
Modesto, CA
Bridgeport, CT
Stamford-Norwalk, CT


MSA or County                                         Number of monitor sites

Hartford, CT                                            2
New Haven-Meriden, CT                                  1
Wilmington-Newark, DE-MD                              2
Washington, DC-MD-VA-WV                             9
Baltimore, MD                                          4
Worcester, MA-CT                                       1
Minneapolis-St. Paul, MN-WI                              3
Bergen-Passaic, NJ                                       2
Philadelphia, PA-NJ                                      10
Newark, NJ                                              3
Jersey City, NJ                                          1
Monmouth-Ocean, NJ                                    2
New York, NY                                          4
Nassau-Suffolk, NY                                      1
Raleigh-Durham-Chapel Hill, NC                           1
Akron, OH                                              2
Medford-Ashland, OR                                    1
Portland-Vancouver, OR-WA                               3
El Paso, TX                                             4
Provo-Orem, UT                                         2
Seattle-Bellevue-Everett, WA                               5
Tacoma, WA                                            1
Spokane, WA                                            1