Environ. Sci. Technol. 2002, 36, 736-741
Spatial Applicability of Emission Factors for Modeling Mobile Emissions D. A. NIEMEIER* Department of Civil and Environmental Engineering, University of California, One Shields Avenue, Davis, California 95616
Driving cycles are used to create mobile emission factors. Emission factors are, in turn, used to estimate mobile source inventories, provide standards for new vehicle emissions testing, and facilitate comparisons of laboratory experiments. This study examines the spatial representativeness of the driving cycles underlying California’s CO, THC, and NOx emission rates that are applied when estimating regional mobile emissions inventories. Sixteen randomly selected vehicles were tested on a laboratory dynamometer using driving cycles representative of driving in different cities. A total of 214 tests, with repetitions, representing six driving cycles, were conducted on the 16 vehicles. We used a random effects analysis of variance to statistically examine the differences in the resulting emission rates. The study results suggest that California mobile source pollutant inventories prepared using emission rates based on the standard drive cycle may be off by as much as 30% for regions where traffic congestion and roadway networks differ significantly from those of Los Angeles.
Introduction Emission factors play a key role in regulating mobile source pollutant generation. They are the platform from which mobile source inventories are estimated (1, 2), they provide benchmarks for new vehicle emissions testing, and they are the prime way of normalizing for cross comparisons of pollutants in experimental research (e.g., refs 3-5). Most emission rates are measured under laboratory conditions using a dynamometer and a speed-time trace known as a driving cycle. Driving cycles are a convenient tool for comparing vehicle emission rates by age and technology and for gauging relative reductions in new car tailpipe emissions. However, their applicability in deriving emission rates for estimating regional mobile emissions inventories is less obvious. In regional air quality modeling, it is assumed that emission rates derived from a few laboratory driving cycle tests can be extrapolated to estimate emissions generated in any region, regardless of driving conditions and behavior. Here, we experimentally evaluate whether it is reasonable to assume that a few standard emission rates can be extrapolated to produce regional emissions inventories. To understand how emission rates are derived, it is necessary to examine the driving cycles used to produce them. Driving Cycles. The U.S. Federal Test Procedure (FTP), used in EPA’s MOBILE model, was created in the early 1970s * Corresponding author phone: (530) 752-8918; fax: (530) 7527872; e-mail:
[email protected]. 736
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 36, NO. 4, 2002
when six drivers from EPA’s West Coast Laboratory drove a 1969 Chevrolet over a single route in Los Angeles thought to be representative of the typical home to work journey at the time (6). Idle time, average speed, maximum speed, and number of stops per trip were computed for each of the six speed-time traces. After discarding one of the six traces, the trace with the actual time closest to the average was selected as the most representative speed-time trace. The selected trace contained 28 “hills” of nonzero speed activity separated by idle periods and was intended to replicate average rush hour driving behavior in Los Angeles. After slight modifications to accommodate the limitations of the belt-driven chassis dynamometers in use at the time, the final cycle, also known as the urban dynamometer driving schedule (UDDS) was finalized in the early 1970s. FTP is 7.46 mi in length, has an average speed of 19.6 mph and a maximum speed of 56.7 mph, and is 1372 s long, with 505 s of cold start and 867 s of running hot stabilized. The cycle has been the standard driving cycle for certification of lightduty vehicles beginning with the 1972 model year, despite the fact that vehicle and traffic patterns have changed significantly since 1969. In the early 1990s, the California Air Resources Board (CARB) created a second standard cycle, the Unified Cycle (UC), for estimating mobile source inventories in California. This cycle was created to mitigate deficiencies that had been identified with FTP, including an inadequate representation of (1) modal event frequencies and (2) hard acceleration/ decelerations (e.g., refs 7 and 8). UC was developed with chase-car data collected in the early 1990s in the greater metropolitan Los Angeles area (6, 9). The chase-car technique uses an instrumented vehicle (the chase vehicle) to follow a randomly selected vehicle (the target) in traffic. In addition to a range of variables (e.g., traffic conditions, roadway type, grade, etc.), the target vehicle’s speed is recorded using a laser rangefinder mounted on the chase vehicle. Literally hundreds of drivers across many types of routes can be followed while recording their driving behavior under a variety of operating conditions. The data are then used to build a driving cycle. Obviously, emission rates developed from UC are more likely to be representative of the mobile emissions in Los Angeles than those from FTP just because the cycle is based on more recent data. However, the UC emission rates are also likely to be more representative because they are based on a cycle drawn from data that reflects significantly more spatial coverage, which in turn incorporates a wide variety of operating conditions and roadway types. But how far can this notion of spatial representativeness be carried? Would emission rates from the Los Angeles-based UC be equally representative of emissions from urban driving in, for example, Sacramento? In other words, can emission rates derived from a standard driving cycle, or even a set of driving cycles, be used to reasonably characterize emissions from driving conditions and behavior across different cities? Both CARB and the EPA presume an affirmative answer to this question. The emission factors used in EPA’s MOBILE6 are based on 11 new facility-specific cycles representing travel under congested and uncongested conditions on freeways, ramps, arterials, and local roadways (10). The cycles were developed using chase-car data from three cities, Los Angeles, Spokane, and Baltimore, and assume that the driving behavior and conditions in these three cities “are not dependent on the city in which the driving was performed” (ref 10, p 10). CARB has the UC and the Unified Correction Cycles, all designed to account for driving throughout the state but 10.1021/es0109747 CCC: $22.00
2002 American Chemical Society Published on Web 01/16/2002
TABLE 1. Summary Statistics for Chase-Car Data average speed (mph) trip length (mi) trip time (min) nonzero power (mph2/s) total idle (% of data) total cruise (% of data) total acceleration (% of data) total deceleration (% of data)
TABLE 2. Testing Cycles
Los Angeles
Spokane
26.3 7.9 17.9 57.1 14.7 32.8 28.8 23.8
27.0 6.0 13.3 49.6 15.2 40.1 23.8 20.9
based solely on driving data collected in Los Angeles. Research indicates that driving patterns in different U.S. cities are dissimilar enough to suggest significantly different emission rates (11-13). However, a key limitation to these studies is that the evaluation is based on driving data alone. While driving differences are important, they are not by themselves conclusive with regard to the actual emissions generated; a difference in average modal activity (i.e., acceleration, cruise, and deceleration) over the course of a trip does not necessarily translate to significant differences in tailpipe emission rates. Although both the EPA and ARB cycles are based on the same data collection technique (i.e., using chase-car information), we cannot directly compare the laboratory-generated emission rates generated by each cycle because the methodologies used to construct the actual cycles are slightly different. However, the spatial representativeness of any base cycle can be examined with laboratory dynamometer experiments comparing the emissions generated by cycles developed by region. These regionally based cycles can be derived from representative chase-car driving data collected throughout a specifically defined geographic area. In this study, we compare the actual differences between dynamometer emission rates in cycles constructed to represent driving in Spokane, WA and those for UC representing driving in Los Angeles.
Methods Driving Cycles. To conduct the experiment, we created several new driving cycles using the same methodology used to create UC. Briefly, when creating a new driving cycle, the procedure begins by dividing chase-car data into segments of similar driving, and randomly selecting and linking segments on the basis of performance criteria that minimize the difference between the speed-acceleration frequency distribution (SAFD) exhibited by the cycle, which uses only a portion of the data, and SAFD using all of the chase-car data. The stringing together of segments continues until a predefined cycle length or time is achieved. The procedures used to create UC are fairly complex, and additional details can be found in Austin et al. (6). To create the new cycles for Spokane, we began by reviewing the summary statistics for chase data collected in Spokane as part of earlier EPA work on the MOBILE6 update. When the summary statistics between Los Angeles and Spokane are compared, it is clear that although average speeds are very similar, both trip length and time differ (see Table 1). The modal activity also varies, with the Los Angeles chase data exhibiting less overall cruise time and more acceleration/deceleration events than that of Spokane. To fully explore whether the UC-generated emission rates are applicable for estimating emissions in other regions, at least three different Spokane cycles must be created and tested under laboratory conditions. The first cycle, described in Table 2 as C1, was distance-matched. That is, thousands of Spokane-based candidate cycles were generated using the UC cycle-development methodology with cycle lengths matching the average trip distance observed in the Spokane
cycle UC C1 C2 S-Spd FTP
average average average max max nonzero total time distance speed speed accel- power power (min) (mi) (mph) (mph) eration (mph2/s) (mph2/s) 24.1 14.5 24.9 25.5 22.9
9.8 6.4 11.7 10.5 7.4
24.9 26.6 28.0 24.6 19.5
67.2 58.0 63.4 61.8 56.7
6.9 5.7 5.4 5.8 3.3
65.94 53.57 58.26 66.47 38.82
24.98 19.81 21.53 25.79 15.39
chase data (approximately 6 mi). The best-fit cycle (i.e., minimizing the differences between SAFD generated for the cycle and one generated for the data as a whole) was selected and denoted C1. As Table 2 shows, the average cycle speed, trip time, and distance for C1 are very close to that exhibited in the overall chase data. The C1 cycle should be the most representative cycle for predicting emissions generated on an average trip in Spokane. If the average emission rates generated by the C1 cycle differ significantly from those produced by UC, it may be that the differences are an artifact of cycle length (UC is approximately 24 min long while C1 is about 15 min long). Because there are more modal events represented in a longer cycle, the probability of incorporating higher (or lower) emitting acceleration/deceleration combinations into the cycle also increases. Consequently, although cycles are used to produce emission rates (i.e., gm/mi), it is still possible that emission rates generated from longer cycles will be significantly different from those produced by shorter cycles. Our second cycle, C2, is based on Spokane driving but forced to match the UC cycle length (in minutes). It is important to recognize that when the Spokane chase data is used to develop a cycle, it means that the speed-acceleration distribution of modal events and the maximum and minimum accelerations/decelerations will be reflective of those contained in the region’s chase-car data. If the average emission rates produced by C2 and UC were not significantly different from one another (but the C1 and UC emission rates were significantly different), this would tend to indicate that it might be possible to use a simple multiplier to factor UC emission rates for representation to other regions. Finally, the last driving cycle, also based on the Spokane chase data, was forced to match the average speed of UC (24.9 mph). This cycle, denoted S-Spd, allows us to compare emission rates produced by differences in regional driving given the same average speed. For example, if Spokane drivers tended to accelerate more frequently at higher speeds than Los Angeles drivers, while still maintaining the same average trip speed, this would be reflected in the cycle. Jourmard et al. (14) performed a similar experiment by comparing the emission rates generated by different cycles having more or less the same average speed. The interpretation of their findings is complicated, however, because the different cycles that were tested were also generated by different cycle construction methodologies. Andre (15) found that the methods used to construct the driving were highly influential in the magnitude of the resulting emission rates. Consequently, it is hard to determine if differences in emission rates were primarily a function of the cycles or of the way in which the cycles were constructed. As can be seen, from Table 2, while the S-Spd average cycle speed is about the same as that of UC, the maximum acceleration and speed are similar to C1 and C2 (the Spokane driving). The speed-time traces for each of the cycles are shown in Figure 1. Vehicle Recruitment and Testing. Sixteen vehicles were randomly procured for the experiment. Each vehicle was VOL. 36, NO. 4, 2002 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
737
FIGURE 1. Speed-time traces: (a) unified cycle, (b) Spokane C1 cycle, (c) Spokane C2 cycle, and (d) S-Spd cycle. picked up in mid-afternoon and brought to the facility for inspection, refueling, and a hot LA4 preconditioning cycle. An FTP test was conducted the following morning, and the remainder of the day was spent running each of the five hot-stabilized tests shown in Table 2. The last cycle served as a preconditioning for the second cold-start FTP the following morning. A total of 214 tests, with repetitions, were conducted on the 16 vehicles. Vehicles were tested on a chassis dynamometer whose calibration results correlated well against CARB’s El Monte facility in southern California. The testing facility was equipped with a Clayton model DC-100 dynamometer with 738
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 36, NO. 4, 2002
a system IV controller that provided both mechanical and electric inertia simulation sufficient to test vehicles ranging from 250-9000 lbs. A drivers aid accepted ASCII files defining each of the desired driving cycles. Exhaust was collected using a Horiba CVS system, modified for tailpipe collection of dilution air during modal emissions testing. A modified Horiba analytical bench provided emissions analysis. Modifications were critical for accurate modal emission measurements and included automatic range changing and supplemental high-range analyzers for accurately measuring the peaks in CO and CO2 emissions that occurred during transient testing. Analyzers
read CO concentrations up to 10% and CO2 concentrations up to 15%. Statistical Analysis. We hypothesized that there were no significant differences in the mean emission rates produced by different cycles (i.e., different cycles would not produce significantly different corresponding emission rates). Under this framework, the vehicles used in the testing were not of intrinsic interest but rather were assumed to be a random sample from the entire population of vehicles. That is, the vehicle was a random factor, and the respective cycle was considered a fixed factor. We used a random effects analysis of variance to statistically examine the results. Employing standard notation, the analysis of variance (ANOVA) mixed model to assess mean differences in emissions rates may be written as
Yijn ) µ + Ri + βj + (Rβ)ij + ijn, i ) 1, 2; j ) 1, ..., 17; n ) 1, ..., N (1) where Yijn is the observed emission rate by pollutant, µ is the grand mean, Ri represents the random effects due to individual vehicles (i ) vehicle), βj represents the fixed effects for the specified cycle (j ) cycle), (Rβ)ij represents a random interaction effect between the cycle and the vehicle, and ijn is the random error (n ) number of observations). The random error is assumed to be independently and identically distributed with ijn ∼ N(0, σ2). The model also imposes the additional constraints that the sums of both the fixed and interaction effects are constrained to zero
∑β ) 0, j
j
∑(Rβ)
ij
)0
TABLE 3. Age, Mileage, and FTP Results model odomTHC CO NOx yr eter (gm/mi) (gm/mi) (gm/mi)
vehicle Toyota Tercel GEO Metro Saturn SC Chevy Caprice Ford Explorer Jeep Cherokee (2WD) Taurus Wagon Honda Accord Nissan Altima Toyota Camry Chevy Cavalier Toyota 4Runner Ford Escort Ford Mustang Honda Civic Pontiac Grand AM
86 89 91 91 92 93 93 94 96 96 97 97 98 98 98 98
96 171 77 802 87 225 61 482 81 162 79 722 60 326 44 473 55 882 55 377 48 575 49 381 37 648 25 330 18 928 37 237
Summary Statistics (Averages) model odoyr meter THC tier 0 (N ) 5) tier 1 (N ) 8) trucks (N ) 3)
1990 76 601 0.47 1997 40 431 0.09 1994 70 088 0.24
Ri and (Rβ)ij are random variables with Ri ∼ N(0, σR2), (Rβ)ij ∼ N(0, σRβ2), and variance components σR2, σβ2, σRβ2, and σ2. The variance components were computed using a restricted maximum likelihood method and represent the usual unbiased analysis of variance components. Under the null hypothesis for this model, (1) the variances for the vehicle effects sum to zero (σR2 ) 0), (2) the cycle effects sum to zero (β1 ) β2 ) 0), and (3) the interaction effect variances are equal to zero (σRβ2 ) 0).
Results and Discussion The vehicle characteristics and FTP test results (averaged over three repetitions) are shown in Table 3. Five tier 0 vehicles were tested, ranging in model year from 1986 to 1993. The average odometer mileage for the tier 0 group was approximately 77 000 mi. Note that the average emissions for the FTP cycle testing for the tier 0 vehicles were close to the tier 0 standards, with the exception of CO. For CO, two vehicles tested as high emitters, generally defined as twice the FTP standard. These vehicles were the Toyota Tercel and the Geo Metro; the Tercel also tested high for total hydrocarbons (THC). Eight tier 1 vehicles and three SUVs were also tested. The average model year for the tier 1 vehicles was 1997, and all of these vehicles tested within the normal emitter standards relative to the FTP-based standards. Comparison of UC to Distance-Matched Spokane Cycle (C1). The results of the random effects model comparison between UC and the distanced-matched Spokane cycle are shown in Table 4. These comparisons suggest significant differences in measured emissions for all three pollutants (CO, THC, and NOx) between the two cycles. For each pollutant, the Spokane cycle resulted in a smaller emissions rate than that which was produced on UC. For THC, the Spokane cycle emission rate was about 14% less, for CO about 29%, and nearly 32% less for NOx. Comparison of UC to UC Time-Matched Spokane Cycle (C2). The results of the random effects model comparison
CO
NOx
7.57 1.06 2.81
0.53 0.16 0.41
TABLE 4. ANOVA Results: C1 and UC Cycle Comparison mean emissions (gm/mi)
(2)
i
0.821 19.407 0.270 0.576 7.357 0.580 0.436 4.368 0.989 0.275 3.943 0.505 0.253 4.644 0.567 0.327 2.016 0.500 0.231 2.771 0.304 0.100 0.992 0.117 0.090 1.525 0.320 0.132 1.181 0.144 0.143 1.484 0.266 0.152 1.766 0.159 0.025 0.261 0.058 0.073 0.533 0.187 0.039 0.702 0.052 0.129 1.792 0.104
pollutant
cycle comparison
UC
C1
THC CO NOx
sig (p ) 0.07) sig (p ) 0.00) sig (p ) 0.00)
0.129 3.512 0.357
0.111 2.502 0.246
TABLE 5. ANOVA Results: C2 and UC Cycle Comparison mean emissions (gm/mi) pollutant
cycle comparison
UC
C2
THC CO NOx
sig (p ) 0.00) sig (p ) 0.00) sig (p ) 0.00)
0.129 3.512 0.357
0.087 2.220 0.268
between UC and the time-matched Spokane cycle also indicate significant differences for all pollutants (Table 5). In each case, the Spokane C2 cycle resulted in a smaller emissions rate than was produced for UC, with differences of about 32% for HC, about 37% for CO, and 25% for NOx. These results indicate that simply factoring UC emission rates up or down for application to another region is not an alternative. That is, the measured differences in emission rates are not a function of cycle length but rather reflect inherent variations in the driving behavior and operations in Spokane. Comparison of UC to UC-Matched Speed Spokane Cycle (S-Spd). We also tested a subset of vehicles (N ) 8) on UC and the UC-matched speed cycle (Table 6). Although these results should be considered exploratory because of the relative small sample size, they are suggestive. It appears that cycles of the same average speed, but governed by regionally based driving behavior (i.e., acceleration-speed distribution that is “local” in nature), can result in essentially equivalent emission rates to those produced in UC. Discussion. The data comparing the UC cycle emissions rates to emission rates generated using data from Spokane suggest that the UC cycle-generated emission rates cannot VOL. 36, NO. 4, 2002 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
739
TABLE 6. ANOVA Results: S-Spd and UC Cycle Comparison mean emissions (gm/mi) pollutant
cycle comparison
UC
S-Spd
THC CO NOx
not sig (p ) 0.23) not sig (p ) 0.43) sig (p ) 0.00)
0.141 5.169 0.367
0.125 4.881 0.320
be considered spatially representative. In other words, the emissions rates generated by UC may reflect driving in Los Angeles but not other regions of the U.S., where driving patterns are different. This finding also implies that every region needs to develop its own suite of cycles and emission rates in order to properly estimate regional mobile source inventories. This is clearly a very expensive and timeconsuming proposition. To better understand the relationship between the results generated across cycles, several regression analyses were also conducted. Regressing the UC-measured emissions against the C1 emissions, averaged over three repetitions, suggested a strong trend of lower measured emissions for the C1 cycle across all vehicles. In Figure 2a-c, the actual data, the regression line, and the 95% confidence interval are plotted for each of the pollutants. For HC vehicle testing (Figure 2a), results were more variable than for CO and NOx testing, as evidenced by the wider 95% confidence interval and the number of vehicles testing outside the confidence interval. Using the 1:1 (dashed) match line, HC results for both the UC and C1 cycles fall within the confidence interval up to about 0.2 gm/mi. That is, for lower emissions there may not be much difference, on average, between the UC and C1 results. However, above 0.2 gm/mi (recall the FTP standard is 0.41), the C1 cycle generates significantly lower emission rates (based on our sample data) than would be generated by UC (this can be seen because the 1:1 line falls outside the 95% confidence interval). The C1 cycle HC emission rates increase by only about 0.81 for each unit increase in the HC emission rate for UC. For CO, the results on average would be about the same between the two cycles only for measured emission rates lower than about 1.7 gm/mi, and for NOx at about 0.2, both levels are substantially below the federal standards. Almost all of the vehicles tested lower on the C1 cycle than on UC. On average, both the C1 cycle CO emission and NOx emission rates increased by only about 0.69 for each unit increase in measured rates for UC. For the C2 cycle comparison to UC (see Figure 3), HC emission rates are comparable only below about 0.03 gm/mi (this compared to 0.2 gm/mi for the C1 cycle). The 95% confidence intervals are also narrower than the C1 regression, indicating that as the cycle length increases and produces a closer match to the SAFD performance measure, the trend is for even greater reductions in Spokane emission rates when compared to the UC generated rates. For all three pollutants, a unit increase in measured UC rates resulted in approximately 0.68-0.75 increases in the C2 emission rates. Note here that the same vehicles as C1 comparison produce higher emissions on both cycles. The important element of this analysis is that it implies that unit emission rates can be adjusted linearly to better represent regional driving differences. This method would allow regions to develop smaller datasets and to construct the adjustment formula specific to their region. Several potential consequences of these observations affect (1) mobile source inventory estimation, (2) research results based on the comparison of driving cycle results, and finally, (3) driving cycle construction theory and methodology. 740
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 36, NO. 4, 2002
FIGURE 2. C1-UC regressions. With respect to inventory preparation, it is important to understand how emission rates, such as UC and FTP, are used in mobile source inventory preparation. After a basic emission rate has been derived from the cycle testing, the rates are then adjusted to represent nontest conditions (e.g., temperature or fuel corrections), weighted to reflect fleet technology and model year, and multiplied by an appropriate travel activity factor. In the case of running stabilized emissions, the travel activity is vehicle miles of travel. Thus, differences in the basic emission rates are likely to carry through the entire inventory estimation. This study’s results suggest that California inventories prepared using the UC base cycle may be off by as much as 30% for each region where traffic congestion and roadway networks are significantly different from that of Los Angeles. For example, in the San Joaquin Valley or Sacramento, both roadway networks and almost certainly driving behavior will
conclusion (although the actual amount might differ) can be drawn for inventories prepared using FTP, which reflects one driver in Los Angeles, or using the new MOBILE6 emission rates, which reflect drivers in Spokane, Baltimore, and Los Angeles. Without additional testing, the application of the MOBILE6 emission rates for estimating inventories in, for example, Atlanta or Houston, where the roadway network and driving characteristics are likely to differ, is questionable. Regions depend on the inventories for identifying appropriate control strategies that will enable them to meet federal standards. Inventories that are overestimated result in significant costs for implementation of control measures that are unwarranted, while underestimation can result in too few control measures to meet federal standards. Either way, more testing and an increased understanding of how regional variability in driving behavior and driving conditions can affect emission rates are needed.
Acknowledgments The author wishes to thank Prof. Britt Holme´n (University of Connecticut) for her critical reading and suggestions on early manuscript drafts. This research was partially funded by the UC Davis-Caltrans Air Quality Project. The statements and conclusions in this study are those of the author and not necessarily those of the California Department of Transportation.
Literature Cited
FIGURE 3. C2-UC regressions. differ from that of Los Angeles. Accurate measurement of emission rates is critical for the preparation of mobile source inventories, of which running stabilized emissions account for roughly between 60% (organic gases) and 90% (nitrogen oxides) of estimated total mobile source emissions inventories (16). This study has shown that, under experimental conditions and controlling for cycle construction methodology, there are serious questions regarding the spatial representativeness of UC, the base cycle used in California to construct mobile emissions inventories. By conceptual extension, the same
(1) Environmental Protection Agency; Facility specific speed correction factors; Report Draft EPA 420-P-99-002 M6.SPD.002; Office of Mobile Sources, Office of Air and Radiation, U.S. Environmental Protection Agency: Ann Arbor, MI, 1999. (2) California Air Resources Board; Methodology for Estimating Emissions from On-road Motor Vehicles, Vol II: EMFAC7G; Technical Support Division, Mobile Source Emission Inventory Branch, ARB; CalEPA: Sacramento, CA, 1996. (3) Lyons, T. J., Kenworthy, J. R., Austin, P. I., Newman, P. W. Transportation Research 1986, 20A, 447-62. (4) Marico, M. M., Podsiadlik, D. H., Chase, R. E. Environ. Sci. Technol. 1999, 33, 1618-26. (5) Cadle, S. H., Mulawa, P., Groblicki, P., Laroo, C., Raggazzi, R. A., et al. Environ. Sci. Technol. 2001, 35, 26-32. (6) Austin, T. C., DiGenova, F. J., Carlson, T. R., Joy, R. W., Gianolini, K. A., Lee, J. M. Characterization of Driving Patterns and Emissions from Light-Duty Vehicles in California; Sierra Research, Inc., Prepared for California Air Resources Board, Sacramento, CA, 1993. (7) Environmental Protection Agency; Review of Federal Test Procedure: Modification Status Report. Report EPA 420-R-93007; Office of Mobile Sources, Office of Air and Radiation, U.S. Environmental Protection Agency: Ann Arbor, MI, 1993. (8) Barth, M., An, F., Norbeck, J., Ross, M. Transportation Research Record 1520; 1996. (9) Morey, J. E., Limanond, T., Niemeier, D. A. Journal of Transportation Statistics 2000, 3 (2), 1-14. (10) Environmental Protection Agency; Development of Speed Correction Cycles; Report Draft M6.SPD.001; U.S. EPA Assessment and Modeling Division, National Vehicle Fuel and Emissions Laboratory: Ann Arbor, MI, 1997. (11) LeBlanc, D. C., Saunders: F. M., Meyer, M. D., Guensler, R. Transportation Research Record 1472; 1995. (12) Milkins, EaW., H. SAE Technical Paper Series, No. 830939; 1983. (13) Kuhler, MaK., D. SAE Technical Paper Series, No. 780650; 1978. (14) Joumard, M. A., Vidon, R., Tassel, P., Pruvost, C. Atmos. Environ. 2000, 34, 4621-8. (15) Andre, M. SAE Technical Paper Series; 1996; 1-13. (16) Approval and Promulgation of State Implementation Plans; Californias South Coast. 40 CFR Part 52, 2000, pp 18903-6.
Received for review May 15, 2001. Revised manuscript received October 22, 2001. Accepted October 23, 2001. ES0109747
VOL. 36, NO. 4, 2002 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
741
