United Stales Air and Radiation EPA420-R-96-008 Environmental Protection March 1996 Agency oEPA Methodology for Estimating Basic Emission Rates for Use in the MOBILE Emission Factor Model > Printed on Recycled Paper ------- METHODOLOGY FOR ESTIMATING BASIC EMISSION RATES FOR USE IN THE MOBILE EMISSION FACTOR MODEL FINAL REPORT Prepared for: Ms. Connie Radwan Office of Mobile Sources U.S. Environmental Protection Agency 2565 Plymouth Road Ann Arbor, Ml 48105 Prepared by: E.H. Pechan & Associates, Inc. 5537-C Hempstead Way Springfield, VA 22151 March 8,1996 EPA Contract No. 68-D3-0035 Work Assignment II-68 Pechan Report No. 96.03.003/1768 THIS DOCUMENT HAS NOT BEEN PEER OR ADMINISTRATIVELY REVIEWED WITHIN EPA AND IS FOR AGENCY USE/DISTRIBUTION ONLY. DO NOT QUOTE, CITE, OR DISTRIBUTE. MENTION OF TRADE NAMES OR COMMERCIAL PRODUCTS DOES NOT CONSTITUTE ENDORSEMENT OR RECOMMENDATION FOR USE. ------- CONTENTS Page ACRONYMS AND ABBREVIATIONS iii CHAPTER I INTRODUCTION/BACKGROUND 1 A. PURPOSE OF PROJECT 1 B. DISCUSSION OF INTERACTION BETWEEN MOBILE AND TECH MODELS 1 C. SUMMARY AND EVALUATION OF COMMENTS ON CURRENT METHODOLOGY 2 CHAPTER II IM240-TO-FTP CORRELATION 4 A. ELIMINATION OF SUSPICIOUS RECORDS 4 B. STRATIFICATION OF HAMMOND DATA BASE 5 C. DEVELOPMENT OF IM240 LANE-TO-FTP BAG 3 CORRELATION 6 D. DEVELOPMENT OF FTP CORRELATIONS 6 E. CONVERSION OF IM240 DATA BASE TO FTP EQUIVALENTS 7 CHAPTER III DEVELOPMENT OF BER EQUATIONS 9 A. ACCOUNTING FOR I/M REPAIR EFFECTS 9 B. DETERIORATION EFFECTS 14 CHAPTER IV EVALUATION OF PROCEDURE 16 A. STRENGTHS OF PROCEDURE 16 B. WEAKNESSES OF PROCEDURE 16 C. APPLICABILITY OF PROCEDURE TO LIGHT-DUTY GASOLINE TRUCKS . . 16 CHAPTERV RECOMMENDATIONS FOR FURTHER ACTION 18 A. SUGGESTIONS FOR FUTURE SAMPLING 18 B. RECOMMENDATIONS FOR INCORPORATING BAG 4 OF THE FTP 18 C. RECOMMENDATIONS FOR FURTHER STUDY 20 REFERENCES 21 11 ------- ACRONYMS AND ABBREVIATIONS API American Petroleum Institute BERs basic emission rates CO carbon monoxide DR deterioration rate EPA U.S. Environmental Protection Agency FTP Federal Test Procedure g/mi grams per mile HC hydrocarbon I/M inspection and maintenance LDGTs light-duty gasoline trucks LDGV light-duty gas vehicle LDVs light-duty vehicles NOX oxides of nitrogen QMS Office of Mobile Sources RVP Reid vapor pressure SFTP Supplemental FTP VINs Vehicle Identification Numbers ZML zero mile level ------- CHAPTER I INTRODUCTION/BACKGROUND A. PURPOSE OF PROJECT In the development of the U.S. Environmental Protection Agency's (EPA) MOBILES emission factor model, emission data from the Federal Test Procedure (FTP) were supplemented with emission data collected from the IM240 test procedure. A statistical correlation was performed between the FTP and IM240 data sets so that the larger data set of IM240 tests could be used in developing basic emission rates (BERs). Historically, the BERs in the MOBILE model had been based primarily on FTP data, with correction factors applied to the emissions measured at the standard conditions of the FTP. The purpose of this report is to assist EPA in improving the data and resulting emission rates used as input to the TECH and MOBILE models. This will be accomplished by developing a methodology for estimating motor vehicle exhaust BERs based on FTP and IM240 data. The work performed in preparing this report was limited in scope. The purpose of the assignment was to focus on presenting alternatives to the methodologies used in the development of MOBILESa, and did not include any data analysis. Changes to technology type groupings and emitter category groupings were outside the scope of this report. B. DISCUSSION OF INTERACTION BETWEEN MOBILE AND TECH MODELS EPA has developed two models that are used in the calculation of exhaust emission rates from highway vehicles - the TECH model and the MOBILE model. The TECH model is primarily used in-house by EPA whereas the MOBILE model is used almost universally, outside of California, for calculating in-use emission factors from highway vehicles. Input to the TECH model consists of BER equations by model year group, technology type, emitter category, and FTP bag. The TECH model then applies emitter category growth rates and creates BER equations by model year. Output from the TECH model is used in the block data section of the corresponding version of the MOBILE model (currently MOBILESa). The output from the TECH model of concern in this report is the set of BER equations by model year and pollutant (hydrocarbon [HC], oxides of nitrogen [NOJ, and carbon monoxide [CO]). Users of the MOBILE model then input information regarding speed, temperature, fuel Reid vapor pressure (RVP), inspection and maintenance (I/M) program parameters, and other local area information. The MOBILE model then applies correction factors to the BERs based on these data and weights the model-year specific emission factors based on default or user-supplied registration distributions and mileage accumulation rates by vehicle type. The MOBILE model then outputs fleet average emission rates by vehicle type. ------- C. SUMMARY AND EVALUATION OF COMMENTS ON CURRENT METHODOLOGY EPA provided two reports that evaluated the methodology that EPA used in developing the BER equations for MOBILESa (Sierra, 1994; SAI, 1994). Both reports were sponsored by the American Petroleum Institute (API). Both the Sierra (Sierra, 1994) and SAI (SAI, 1994) reports spent considerable time evaluating the methodology developed by EPA to estimate light-duty gasoline vehicle (LDGV) exhaust BERs. This methodology essentially consists of taking fuel specific emission rates for targeted technology group-vehicle model year-emitter types resulting from IM240 lane testing with tank fuel and converting them (via various adjustments and a resulting regression equation) to corresponding group FTP emission rates using Indolene. The resulting regression equations, logarithmic for HC and CO pollutants and linear for NOX, estimate the FTP emission rate minus the cold start offset from the IM240 emission rate. These rates are then used as inputs to the TECHS preprocessor model which calculates inputs for the MOBILES model. The variety of factors (and their interactions) to be considered and the multitude of steps to be taken to perform these tasks are very complex. Many of Pechan's concerns are in line with both the Sierra and SAI reports. In particular: • Weighing the foreign manufactured cars introduces potential bias - it might be preferred to obtain a better sample or to not stratify the data set by this factor. • Eliminating data because the Weather Bureau's Chicago monthly average temperatures were exceeded by more than 25 degrees and were therefore classified as statistical outliers is probably unnecessary. • The value of applying a fuel-temperature "seasonal" adjustment factor to convert IM240 lane-with-tank-fuel data to IM240 lab-with-Indolene data is questionable. Recent research (performed after MOBILESa had been released) suggests that fuel and temperature both individually and interactively may not play a strong role in exhaust emissions modelling, although the temperature range for this testing was limited to 50°F to 100°F (Chou, 1995). Moreover, as pointed out in the Sierra report, when converting emission rates from IM240 lane-with-tank-fuel to FTP lab-with-Indolene, a direct correlation can be made (IM240 lane tank fuel - FTP lab Indolene) rather than introducing another correlation (IM240 lane tank fuel - IM240 lab Indolene and IM240 lab Indolene - FTP lab Indolene). This is the methodology discussed in this report. • Since the IM240 test is a subset of the FTP cycle, but does not include the cold start, the effects of Bag 1 must be accounted for when regressing the FTP on the IM240 emission rates. Although it is reasonable to subtract out the cold start effects (adjusted for mileage) from the FTP rates, it is not reasonable to substitute the IM240 rates if there is a negative difference. The results might be biased in these situations, since the IM240 rate was being regressed on itself. ------- • In the MOBILE5a methodology used for correlating IM240 emissions data to FTP emissions data, residuals were added to the correlation equation to achieve randomized scatter. EPA used residuals to restore a distribution of predicted FTP values for a given IM240 score. (In other words, the use of residuals indicates that for a given IM240 score, more than one FTP score might be predicted.) However, there appear to be inconsistencies in the methodology that are not satisfactorily explained. Using residuals in this manner did not improve the correlation. Pechan recommends eliminating the use of residuals in developing the FTP correlations. • The TECHS model uses the FTP emission data to create model year (there are two) - technology type (there are four) emission rates by multiplying the emission rate of each of the defined emitter groups (four for HC and CO and two for NOJ by the proportion that the emitter category is of that mileage-vehicle age fleet. Different model year groupings were used in developing the individual emitter category emission rates and the emitter category growth rates. It might be better to use consistent groupings of the data in calculating both emission rates and growth rates, assuming there are sufficient data points in each subgroup to accomplish this. ------- CHAPTER El IM240-TO-FTP CORRELATION The current methodology used to develop the BER equations input to TECHS and subsequently used by MOBILESa is discussed below. These equations were derived from two data sets. The first is the Hammond, Indiana Lane Data Base, which includes IM240 test data for 6,597 light-duty vehicles (LDVs) from 1981 and later model years. The second data base consists of 400 FTP tests performed on a subset of the Hammond vehicles at an EPA contractor's lab in South Bend, Indiana. The basic steps that were followed in developing the BER equations provided as input to TECHS are as follows: • Adjust Hammond data base for foreign manufacturers, missing or suspicious odometer mileage, seasonal outliers, and fuel effects; • Correlate data from IM240 lane-tested vehicles in Hammond data base that were also tested at the lab for fuel and temperature differences to determine lab- equivalent IM240 scores; • Develop regression equation between IM240 lab scores and corresponding FTP scores taking cold starts into account; • Apply correlation to entire Hammond IM240 data base to determine FTP- equivalent emissions; and • Develop BER equations from FTP-equivalent data base. Our proposed methodology for correlating the IM240 data to FTP data consists of the following basic steps: Eliminate suspicious records; Apply appropriate stratification to Hammond data base; Develop correlation between IM240 lane data base and bag 3 of the FTP; Develop correlation between bag 3 of the FTP and each of the other bags as well as the total FTP; Apply correlation by bag to the IM240 data base; Develop BER equations from FTP-equivalent data base; and Apply correction factor to eliminate effects of I/M from BER equations. Each of these steps is discussed individually below along with the rationale for any deviations from the previous methodology. A. ELIMINATION OF SUSPICIOUS RECORDS The procedure used in developing the BERs for MOBILES excluded records with a mileage listed as "0" or greater than 300,000 miles. The upper bound of 300,000 was ------- reasonable to apply to the Hammond data base as the mileage in the data base jumps from 238,925 to 541,418, with several records having mileage greater than 800,000. When the new test data that EPA will be using for developing MOBILES BERs are available, the data should be analyzed by plotting a histogram to see if there is an obvious point after which the data significantly drops off (i.e., a number of consecutive bins with a frequency of 0). Another check that should be done to validate the odometer reading is to evaluate the distribution of mileage accumulation. Again, there may some obvious outliers where the data should be carefully scrutinized before the record is accepted into the final data base. Vehicle Identification Numbers (VINs) can also be checked for validity if VINs are collected. This can be done by computing the check digit as described in the Federal Motor Vehicle Safety Standard No. 115. In addition, the VIN can be used to check the validity of the model year. The tenth digit of the VIN indicates a vehicle's model year. B. STRATIFICATION OF HAMMOND DATA BASE In developing the MOBILE5 BERs, records from certain foreign manufacturers were replicated two to four times in the data base because it was felt that vehicles from foreign manufacturers were underrepresented in the Hammond data base. To avoid the problems of having a test fleet that does not represent the in-use fleet very well, the sample of vehicles being recruited for testing should be selected to better match real world conditions. This stratification of the sample is discussed further in another section of the report. The fact that a manufacturer is foreign or domestic should not alone have a significant impact on emissions. Today, many foreign auto manufacturers build cars in the United States. The more important determinant in emissions would be the manufacturer itself rather than whether the manufacturer is foreign or domestic. This is evidenced in vehicle recalls for certain engine groups of a particular manufacturer. Other characteristics of the vehicle that would affect emissions include engine size, engine type, location, age, mileage, and maintenance record. Stratification of vehicles by engine type is already considered in the current methodology as the TECH model weights technology type by the fraction of vehicles within each technology type by model year. Vehicle age and mileage are also factored into the current methodology, although mileage is currently used as the dominant factor in determining emissions. In actuality, age may be a more important factor in determining a vehicle's emission rate than mileage. This would be the case if emission component failures are more a function of weather than of accumulated mileage. It would be interesting if EPA could devise a test of weather and other aging effects versus mileage on failures of emission components. If it can be shown that weather/aging account for more emission component failures, Pechan would recommend revising current modeling procedures to replace mileage with vehicle age. Engine size, however, is not currently considered. This could have an impact on emissions as a range of engine sizes are used in automobiles. The location of the vehicle is factored into the current methodology somewhat by the temperature, RVP, registration distribution, I/M program, and other local inputs to the MOBILE model. The maintenance record of a vehicle is one factor that has not been accounted for in the past and could have a major impact on emissions. A well-maintained vehicle that is old and has high mileage ------- may emit less than a poorly maintained newer vehicle. If adjustments are to be made to the data base of test records, some of these other factors should be considered as well. C. DEVELOPMENT OF IM240 LANE-TO-FTP BAG 3 CORRELATION After probable outliers are removed from the IM240 data base and the data base is adjusted to better reflect the actual fleet of in-use vehicles, the IM240 test scores need to be correlated to the FTP scores. Based on comments in both the Sierra (Sierra, 1994) and SAI (SAI, 1994) reports evaluating the methodology used to derive the BERs used in MOBILESa, Pechan recommends eliminating the fuel and temperature adjustments. Instead, Pechan recommends developing a direct correlation between IM240 lane test data and the corresponding FTP Bag 3 data, since the IM240 driving cycle was developed from the Bag 3 FTP driving cycle. As such, a good correlation should be achieved between IM240 and FTP Bag 3 emissions. Although variations in fuel RVP and temperature can impact emissions, in this instance, the temperature and fuel corrections appeared to make little difference when compared to the unadjusted lane data. A number of other factors may also influence differences in IM240 scores between lane and lab such as (1) differences in preconditioning between the two tests; (2) inconsistent dynamometer settings; (3) vehicle changes between tests; and (4) operator impacts. As illustrated in the Sierra report (Sierra, 1994), several different approaches to the lane versus lab temperature and fuel adjustments yielded almost no differences from the approach used for MOBILESa. With no clear trends in the data by month or season, the small sample sizes available for some months in combination with the variability added by applying the seasonal adjustments, Pechan recommends that no lane to lab adjustments be applied to the IM240 data. Equation (1) illustrates the regression equation. Bag 3 = a + b * FTP (1) D. DEVELOPMENT OF FTP CORRELATIONS The next step of the proposed methodology is to regress the FTP Bag 3 data against the FTP Bag 2 data, developing a relationship that will then be applied to the IM240 data that has been correlated to the FTP Bag 3. Preliminary analysis using EPA's FTP data base as a whole (i.e., not stratified) indicated a relatively good correlation between Bag 3 and Bag 2. Using linear regression, an Rz value of 0.86 was obtained for HC, 0.88 for CO, and 0.91 for NOX. Presumably, these correlation coefficients would improve even more if the data base is first stratified to obtain separate correlation equations for each technology group, model year group, and emitter class combination. A log fit of the data should also be investigated to determine whether a better fit of the data is obtained with a log regression. Finally, the Bag 3 FTP data need to be related to Bag 1, the cold start bag. Since both Bag 1 and Bag 3 follow the same cycle, differences between these two bags should be attributable to the cold start. For MOBILESa, the cold start offset was calculated as the mean value of the difference between the FTP and the IM240 for the set of normal emitters with an FTP value greater than the IM240 value. For the vehicle-specific correlations, a cold start offset equation was developed where the cold start offset was calculated as a ------- function of mileage, assuming that as a vehicle ages the cold start emissions will increase because the catalyst takes longer to reach light-off. Our proposed procedure includes the incorporation of mileage in the calculation of the cold start offset. For the proposed procedure, the difference between the FTP Bag 1 and FTP Bag 3 would be calculated for all vehicles in the FTP data set. This cold start offset would then be regressed against mileage for each technology group, model year group, and emitter category. Emissions computed using the resulting regression equation would then be added to the calculated Bag 3 results to give an estimated Bag 1 for each of the vehicles in the IM240 data base. These steps are illustrated in equations (2) through (4): Bag 2 = c + d * Bag 3 (2) CS = Bag 1 - Bag 3 = e + / * ODOM (3) Bag 1 =Bag 3 + e + / * ODOM (4) where: CS = cold start offset, and ODOM = odometer reading (miles) Once the regression coefficients a through f have been calculated, FTP emissions by bag can be estimated for each of the IM240 lane test data points. It should be noted that the above equations do not imply that only linear regressions should be used. The actual data should be examined using logarithmic or log-linear regressions, as well, and the regression type yielding the best fit to the data should be selected. It should be noted that after the FTP data are stratified into each of the subcategories (i.e., model year group, technology type, emitter group), some of the subcategory sample sizes may be too small to use in these regressions. In this case, some of the subcategories may need to be combined, or alternative assumptions may be needed. E. CONVERSION OF IM240 DATA BASE TO FTP EQUIVALENTS The correlations developed in the previous section would be applied to each vehicle in the IM240 data base. The procedure for adjusting each IM240 record is summarized below. • Calculate the FTP Bag 3 equivalent for each record using the technology group, model year group, and emitter category regression coefficients developed above in equation (1). • Estimate FTP Bag 2 emissions by applying the FTP Bag 3-to-FTP Bag 2 correlation to the FTP Bag 3 emissions that were calculated in the previous step, again according to technology type, model year group, and emitter category (i.e., calculate Bag 2 emissions using equation (2)). ------- • Estimate Bag 1 emissions by adding the appropriate cold start offset to the calculated Bag 3 emissions as a function of the vehicle mileage, with different equations applying to each technology group, model year group, and emitter category combination (equation (4)). • Calculate the total FTP-equivalent emissions using the following equation (i.e., weight each of the bags according to the fraction of total FTP mileage accumulated in each mode): FTP = 0.206 * Bag 1 +• 0.521 * Bag 2 + 0.273 * Bag 3 (5) 8 ------- CHAPTER III DEVELOPMENT OF BER EQUATIONS Once FTP-equivalent emissions have been calculated for each vehicle in the IM240 data set, a single BER equation, including zero mile level (ZML) and deterioration rate (DR) as a function of mileage, needs to be calculated for each of the groupings of vehicles. These equations would then be input to the TECH model. As with each of the previous steps, it is assumed that the vehicles would be grouped by technology type, model year group, and emitter category and that a single BER equation would be developed for each of these groupings. The ZML and DR for each BER equation would be calculated by performing a linear regression of the FTP-equivalent emissions against mileage, where the y-intercept would represent the ZML and the slope would be the DR. As was done previously for MOBILESa, emission caps should be applied to the normal emitters. A. ACCOUNTING FOR I/M REPAIR EFFECTS New IM240 data collected by EPA after that used for developing the emission rate equations input to MOBILESa includes additional data from the Indiana test site as well as from testing done in Phoenix. Both sets of data now include vehicles that have been through more than one cycle of I/M testing. Thus, it can be assumed that some of the vehicles tested would have lower emissions than they would otherwise have had in the absence of an I/M program (i.e., some vehicles would have failed a previous I/M test and then have been repaired to pass the test). Therefore, in order to use the BER equations developed as discussed above in the TECH and MOBILE models, where all BER equations are assumed to be FTP-equivalent, the effects of I/M need to be estimated and removed from the BER equations. The BER equations used by MOBILE need to be FTP-equivalent since all temperature, fuel, I/M, and other correction factors were calculated based on FTP conditions. The existence of an I/M program for more than one test cycle would cause IM240 emissions to be different from IM240 emissions calculated at the first test cycle because a certain number of vehicles failing previous I/M tests will have been repaired to pass the test. Vehicles that have never failed an I/M test would presumably have emissions equivalent to those of vehicles that are not subject to an I/M test, except for the tampering deterrence effect. This section discusses a possible methodology that could be incorporated to estimate BER equations with the effects of the I/M programs removed. The I/M adjustments should be calculated as a fleetwide effect, since the effect on the individual vehicles tested will be unknown, at least with the current data set. In the future, it should be possible to track which vehicles have failed an IM240 test and, as a result, have been repaired to pass the IM240 test. To calculate the effects of an I/M program on vehicle emissions, EPA maintains a data set showing the before and after repair FTP emissions of vehicles that have failed idle or IM240 tests. This is a subset of the vehicles in the Hammond data set. The average of the before and after repair FTP emissions, by emitter category, are entered into the TECHS model, which then estimates ------- I/M program emission reductions by model year, and accounts for such program characteristics as program network type, program frequency, emission cutpoints, etc. These by-model-year emission reduction effects are included as an input file used by MOBILESa. A similar approach could be implemented that reverses the current procedure for estimating I/M repair effects. Due to the limited size of EPA's repair data base, it may be advisable to incorporate similar data from other sources. Preliminary investigation into the availability of additional repair data bases has turned up a number of potential sources. A drawback to using these other sources is that in most cases, emissions were estimated with remote sensing devices or IM240 tests, and do not have corresponding FTP emission results. Nonetheless, the additional information that these other sources would provide could be a useful supplement to EPA's current repair data base. The MOBILE model currently uses the implicit assumption that an I/M program does not change the rate of deterioration of a vehicle's emission system. Only the zero-mile level is assumed to change. This is evidenced in the fact that changing the start year of an IM240 program in MOBILESa does not change the emission factors, as long as one full testing cycle has passed. This makes intuitive sense in that, within each test cycle, each vehicle needs to return to the level of the emission cutpoint. If each vehicle is tested every two years and each vehicle that fails the test is repaired to meet the cutpoint level, then this repair has only two more years of deterioration before it will need to be repaired again. The difficulty of applying a correction to account for this is that the effect of I/M will differ by emitter category (as super emitters will need a much higher percentage reduction in emissions to reach the cutpoint than will high emitters). To counteract the effects of successive I/M cycles beyond the first test cycle, the populations of all four emitter categories will need to be adjusted. Also, the BER equations representing each group may be different than they would otherwise be in the absence of an I/M program. The assumption of 2-year deterioration of emission components implies that a vehicle that is a high emitter in the first test cycle would continue to be a high emitter in successive test cycles. When the high emitter fails an I/M test and is then repaired, emission levels should first drop to close to the test cutpoint and would then continue to deteriorate, presumably at the same rate as before the repair. Although the validity of this assumption may be questionable on an individual vehicle basis, it is more likely to hold true for the fleet as a whole. For this reason, in this methodology, adjustments for I/M are made to the BER equations rather than to the individual vehicle data points. This methodology combines BER adjustments with adjustments to the emitter category populations and growth rates. This approach involves segregating the FTP-equivalent data by I/M test cycle and by area (e.g., separate Indiana data from Phoenix data) to compute a new set of BER equations. For example, the Indiana data used for MOBILESa was collected from 1990 to 1992. Data from the next test cycle (presumably 1992 to 1994) would be analyzed separately. Assuming all conditions are relatively equal over the separate test cycles, such as fleet turnover, mileage accumulation rates, etc., the primary difference between the two sets of BER equations would be the effect of I/M repairs, which would be seen in the ZML. 10 ------- Figure III-l illustrates how an I/M program would affect the BER for a single vehicle. The line labeled No I/M shows how the vehicle would deteriorate in the absence of an I/M program. The x-intercept represents the ZML. The line labeled After Repair Level illustrates the emission rate that the vehicle will achieve to pass the I/M test. It is assumed that this emission rate is slightly lower than the IM240 emission rate cutpoint. The first time the vehicle is tested, it fails the test and is repaired to meet the emission cutpoint. Over the next test cycle, the vehicle deteriorates at the same rate that it did before the test, but with a different ZML. This pattern is repeated for each successive test. Thus, in each test cycle, the ZML for the vehicle is effectively lowered by the amount of the repair. When the vehicle is initially tested in the third cycle, its ZML has effectively been lowered by two times the repair reduction. Figure 111-1 Effect of I/M on a Typical Vehicle Emission Rate (g/.-ni) Mileage To compensate for the effects of I/M on the second and later test cycles, EPA can develop BER equations for each test cycle/model year group/technology type/emitter category combination. Then, adjust the ZML for the second, third, and subsequent test cycles by multiplying each ZML by the effective repair rate for that emitter category, and by the number of test cycles after the first cycle. (For example, the ZML representing the third test cycle should be reduced multiplied by two times the repair rate.) The following equations illustrate this adjustment: = ZML(x) + DR * ODOM (6) = ZML(\) - (x - 1) * REP (7) 11 ------- where: EF(x) = Emission rate from I/M cycle x {grams per mile [g/mi]) ZML(x) = ZML from I/M cycle x (g/mi) DR = Deterioration rate (g/mi/10,000 miles) ODOM = Accumulated mileage REP = Emission reductions from repairs (g/mi) Next, test data from all cycles at a given site will need to be combined, with the I/M effects removed. First, all data from a single site should be recombined. This can be done by first determining the fraction of vehicles tested in each test cycle. The fraction of vehicles in the first test cycle {i.e., those without an I/M repair effect) would be multiplied by the ZML and DR of the first cycle. The fraction of vehicles in the second test cycle would be multiplied by the adjusted ZML for the second test cycle and the DR for this cycle. The same procedure would be applied for vehicles in the third and later test cycles. The result would be a new BER equation adjusted to eliminate the effects of I/M repairs. If data from three test cycles had been collected, the following equation would be used: * FRAC(\) + ZML(2) * FRAC(2) + ZML(3) * FRAC(3)+ (8) where: ZML(x) = ZML from I/M cycle x FRAC(x) = Fraction of total number of tested vehicles that were tested in I/M cycle x Finally, data from all sites would need to be combined. Both the ZMLs and deterioration rates need to be combined. As when weighting by cycle, EPA should first determine the fraction of total vehicles tested by site. Weight both the ZMLs and deterioration rates from each site by these fractions to determine a fleetwide BER equation with I/M effects removed. One of the effects of an I/M program is to mask the true size of the non-normal emitter categories. Ideally, an I/M program would cause all non-normal emitters to emit at normal emitter rates. However, assuming the life of the repairs made is two years, by the time these repaired vehicles are tested in the next test cycle, the vehicles would have emission rates approximately equal to their emission rates as tested in the previous cycle. Thus, the overall effect of repairs is to hold the deterioration rates of repaired vehicles at zero (when evaluated at identical points in successive inspection cycles). For example, a vehicle in the high emitter category that fails an I/M test would be repaired, lowering its emissions to the point that it initially becomes a normal emitter. Over the next two years, however, the vehicle again begins to deteriorate, and by the time the vehicle is tested again, its emissions will be approximately the same as they were at the previous test (i.e., in the high emitter category). Thus, when emissions are only reported for each initial inspection in a test cycle, the vehicle's emissions are relatively unchanged. Without being repaired in each test cycle, the vehicle would eventually become a super emitter. The resultant effect on the overall emissions data base is that the emitter category growth rates would essentially be halted at the levels achieved by the end of the first inspection cycle. Thus, in combination with the BER equation adjustments discussed above, the effects of I/M can be eliminated by calculating the emitter category growth rates 12 ------- using only test data from the first inspection cycle. Since the BER equations are separated by emitter category, all data collected could still be used in developing the BER equations by model year group, technology type, emitter category, and test cycle as discussed above. Using the emitter category growth rate functions developed with data from the first test cycle only would then roughly eliminate the effects of I/M program repairs. This procedure assumes that the vehicle deterioration rates before and after repairs are the same (at least at the subgroup level). This assumption could be investigated by tracing a sample of vehicles through several I/M cycles. Chapter V discusses this issue in more detail. B. DETERIORATION EFFECTS Currently, the data input to TECHS includes a ZML and a single deterioration rate for each model year group, emitter category, technology type, and bag. Within the TECH model, BER equations are developed for each model year that include a ZML and two deterioration rates - one representing deterioration below 50,000 miles and another representing deterioration above 50,000 miles. The equations for each model year are determined based on assumptions regarding the growth rates for each of the emitter classes as a function of average accumulated mileage by age. The reports provided by EPA evaluating the MOBILES BER equation development are critical of the deterioration rate kink applied at 50,000 miles (Sierra, 1994; SAI, 1994). Pechan has investigated the possibility of using alternative methodologies for developing BER equations that do not incorporate a kink at 50,000 miles. EPA's intention with the current methodology was to incorporate both mileage and age in determining these deterioration rates. An alternative approach that considers both factors is discussed below. The issue of whether vehicle age might be more important than or at least as important as mileage in determining a vehicle's emission rate has not been carefully evaluated. This issue is worth investigating, and the results of such an analysis could be used in determining a better method for determining by-model-year emission rates that do not include a kink. To analyze this issue, each of the BER equations developed for input to TECHS as a function of mileage should also be calculated as a function of vehicle age. (This would have to be calculated by subtracting the vehicle model year from the year of the emissions test and adding one.) If a statistical analysis shows a better fit of the data when vehicle age is used rather than mileage, then perhaps new BER equations could be developed that take the age-based BER equations and convert them to mileage-based BER equations. (Because activity data collected for calculating motor vehicle emissions is primarily in the form of VMT, at least in the near future, VMT will need to continue to be used as the activity factor for determining vehicle emissions.) This might be done by first determining the average mileage accumulated by vehicles of every age (e.g., average mileage accumulated by 1-year old vehicles, 2-year old vehicles, etc.). These mileage values would then be substituted for the corresponding vehicle ages along the x-axis of the BER plots. The BER equations would then be recalculated as a function of mileage, substituting in the average mileage accumulated at each age. This would most likely yield a non-linear equation. 13 ------- CHAPTER IV EVALUATION OF PROCEDURE A. STRENGTHS OF PROCEDURE One of the strengths of the procedure described in this report is that it provides a methodology for incorporating IM240 data samples that includes test data beyond the first test cycle. This allows for a much larger database of emission testing to be included in determining emission rates to be used by MOBILE6. In addition, this procedure eliminates some of the shortcomings of the methodology used to develop BERs for MOBILESa, as documented by Sierra (Sierra, 1994) and SAI (SAI, 1994). For example, eliminating the step to weight data from certain foreign manufacturers removes some bias from the data sample. Another strength of the procedure is a more direct correlation of IM240 emissions to FTP emissions. The steps of determining an IM240 lane to IM240 lab correlation and then determining IM240 lab to FTP lab correlations is simplified to one step. This eliminates the variability added by performing the temperature/fuel adjustments, which apparently had little overall effect on emissions. B. WEAKNESSES OF PROCEDURE Weaknesses of the procedures documented in this report stem primarily from the inability to perform the data analyses described in this report on the actual set of data that will be used in developing BERs for MOBILES. Without having the actual data available, it is sometimes difficult to determine what the best statistical approach to incorporating the data might be, and in what categories data there might be insufficient data to obtain a statistically significant data strata. C. APPLICABILITY OF PROCEDURE TO LIGHT-DUTY GASOLINE TRUCKS The procedure discussed in this report for converting IM240 emissions to FTP emissions should be applicable to light-duty gasoline trucks (LDGTs). However, care must be taken when applying the correlations to LDGTs. For example, a mapping of LDGV technology types, model year groups, and emitter categories to the appropriate LDGT technology types, model year groups, and emitter categories should be developed. This should be done to ensure that the LDGV groups that are representing LDGT groups have similar emission control component performance. Because the emission standards for LDGTs are generally higher than the LDGV emission standards for the same model year, LDGTs should be assigned to emitter categories based on the LDGT standards rather than LDGV standards. (For example, the definition of a normal HC/CO emitter is a vehicle that emits up to twice the HC standard and up to 3 times the CO standard. The LDGT standards rather than LDGV standards should be used in making this determination, 14 ------- rather than an emission rate limit derived from the LDGV standard.) If there is not a one- to one match between LDGV and LDGT technology type/model year group combinations, it may be best to combine the LDGV data from several technology type/model year groupings to be applied to a smaller number of LDGT groupings. ------- CHAPTER V RECOMMENDATIONS FOR FURTHER ACTION A. SUGGESTIONS FOR FUTURE SAMPLING Several points should be considered when recruiting the set of vehicles that EPA will use for lab tests (both IM240 and FTP) in the future to allow for more representative IM240-to-FTP correlation equations. First, the recruited subset of vehicles should represent the national fleet of vehicles to the extent possible. The percentage of vehicles in the United States by manufacturer, engine size, technology type, mileage emitter group, model year, and any other factor deemed important to vehicle emission rates should be estimated. These characteristics should be prioritized in terms of their importance on vehicle emission rates. The recruited vehicles should match the national population relatively well, with vehicle characteristics of the highest priorities selected first. Next, it is important that all four emitter categories are well represented, and that at least half of the recruited vehicles be normal emitters so that there will be sufficient data for developing the correlations in all emitter categories. Finally, where possible, a number of vehicles that previously underwent lab testing should be targeted so that emission tracing over the vehicle's life can be examined to assist in evaluating the validity of deterioration assumptions. B. RECOMMENDATIONS FOR INCORPORATING BAG 4 OF THE FTP A new Supplemental FTP (SFTP) has been proposed that addresses driving modes that occur in practice, but have not been represented previously by the FTP. This unrepresented driving is higher speeds and higher accelerations, for the most part. In total, though, the SFTP addresses the following five items: 1. Aggressive driving behavior; 2. Rapid speed fluctuations; 3. Start driving behavior; 4. Intermediate soak times; and 5. Air conditioner operation. The above five items are examined through including three new single bag cycles via the supplementary FTP. Bag 4 is a hot stabilized 866 cycle run with a new simulation of in-use AC operation. Bag 5 is a new start control cycle simulating driving with the new simulation of in-use AC operation and proceeded by a soak period. Bag 6 is a new aggressive driving cycle run in the hot stabilized condition. A certification test fuel is used for emissions tests performed using these new cycles. Manufacturers compliance with the applicable emission standards is computed using the following weighting factors: 16 ------- THC/NMHC CO and NO, Bag 1 .21 .15 Bag 4 .24 .37 Bag5 .27 .20 Bag 6 .28 .28 Standards apply to 40 percent of 1998 model year vehicles, 80 percent of 1999 model year, and 100 percent of 2000 model year and later. For the time being, Pechan recommends that the Bag 4 results be used to modify the speed correction factor routine in MOBILESa. The composite scores should be used to calibrate the net results after the BERs and correction factors have been applied. The most significant problem associated with including the Bag 4 results in the TECH model is that only a limited number of vehicles will be tested on the new cycles. If EPA wants to use these data to modify the TECH model, then sampling on both the FTP and SFTP series is needed for a wide range of model years. Important issues in long range planning for the Office of Mobile Sources (OMS) related to SFTP results are (1) how manufacturers respond to these new requirements, and how emissions under the current cycles might be affected; and (2) how best to design post- MOBILE6 versions of MOBILE to integrate both the results of SFTP testing and state of the art of transportation planning models that are being used for conformity determinations and other urban scale emissions assessments. Potential steps in developing new emission relationships using Bag 4 data include the following: 1. Include high speed operations in speed correction factor subroutine. Other subroutines (correction factors) may be needed to allow other emissions influencing factors (aggressive driving) to be included in the model. 2. Identify emissions that were not captured in current emission factors for new vehicles being tested on the SFTP. Adjust later model year vehicle emission rates in proportion to the differences observed for new vehicles. 3. Methods used by manufacturers to meet emission standards with SFTP has a control effect that affects new vehicle emission performance. 4. Older vehicles have higher emission factors and continue to emit at current level of control unless affected by an in-use control program. 5. Anticipate how new transportation modeling methods can be used to weight new cycle emission results to capture area-by-area differences in travel behavior. 6. MOBILESa procedures for including corrections for air conditioning usage, extra loads, trailer towing, et al. date back to MOBILE2. If air conditioning usage is included as a standard condition for emissions testing, then, at a minimum, the 17 ------- correction factor routines for air conditioning use and extra loads need to be reformulated. Longer term decisions about whether to include Bag 4 data {with air conditioning in operation) in the basic emission rates involves a policy decision about whether the primary emphasis of MOBILE6 is providing ozone season daily emission factors. If ozone season emission factors are the primary use, then incorporating Bag 4 data with air conditioning operation in the BERs is more desirable than it would be if evaluating regulatory options, estimating annual emissions, or wintertime emissions was the primary focus. Some sampling for the new cycles should be performed without the air conditioning in operation in order to have the opportunity to remove this effect on emissions where warranted. C. RECOMMENDATIONS FOR FURTHER STUDY Since the methodology described in this report for eliminating the effects of I/M is heavily dependent upon assumptions involving the effectiveness of repairs, effort should be invested in investigating these assumptions. First, the repair rates themselves should be updated to include as large a sample size as possible, including a broad coverage of vehicle age and mileage accumulation. Secondly, the deterioration rate of individual vehicles before and after emission component repairs should be investigated. Since this would involve the tracking of specific vehicles over a period of years, such investigation may be best undertaken using a well-selected sample (including both normal and non-normal emitters) of vehicles that have participated in I/M programs for several cycles. Emission data both before and after any necessary repairs have been made could be evaluated vehicle-by-vehicle, to determine whether deterioration rates were affected by the repairs. By tracking vehicles over several test cycles, it might also be possible to determine whether deterioration (excluding the effects of any repairs) is linear or nonlinear. In addition, for vehicles in this sample that needed repairs to pass the IM240 test, the effectiveness of these repairs from test cycle to test cycle could be evaluated. The data from these samples could also be used to investigate whether age or mileage is the more appropriate determinant of vehicle emission rates. 18 ------- REFERENCES Chou, 1995: Chou, David and Jeffrey Long, Determination of the Effects of Speed, Temperature, and Fuel Factors on Exhaust Emissions, presented at the Fifth CRC On- Road Vehicle Emissions Workshop, San Diego, California, April 3-5, 1995. SAI, 1994: Systems Applications International, "Investigation of MOBILE5a Emission Factors - Assessment of Exhaust and Nonexhaust Factor Methodologies and Oxygenate Effects," prepared for American Petroleum Institute, February 7, 1994. Sierra, 1994: Sierra Research, Inc., "Investigation of MOBILESa Emission Factors - Evaluation of IM240-to-FTP Correlation and Base Emission Rate Equations," prepared for American Petroleum Institute, June 1994. 19 ------- TECHNICAL REPORT ABSTRACT REPORT TITLE: Methodology for Estimating Basic Emission Rates for Use in the MOBILE Emission factor Model - Final Report REPORT DATE: March 8, 1996 CONTRACT NO.: 68-D3-0035 PRIME CONTRACTOR: E.H. Pechan & Associates, Inc. WORK ASSIGNMENT NO./DELIVERY ORDER NO. (if applicable): 11-68 PROJECT OFFICER: Mary Wilkins PROJECT OFFICER ADDRESS: MD-12, RTF, NC 27711 TEL.: (919) 541-5229 PROGRAM OFFICE: Office of Mobile Sources NO. OF PAGES IN REPORT: 24 DOES THIS REPORT CONTAIN CONFIDENTIAL BUSINESS INFORMATION YES NO X REPORT ABSTRACT - Include a brief (200 words or less) factual summary of the scope and nature of the work performed and referenced in the report. In the development of EPA's MOBILE5 emission factor model, emission data from the Federal Test Procedure (FTP) were supplemented with emission data collected from the IM240 test procedure. A statistical correlation was performed between the FTP and IM240 data sets so that the larger data set of IM240 tests could be used in developing basic emission rates (BERs). Two reports were provided that detail shortcomings in the approach and statistical analyses used to develop the MOBILES exhaust BERs. The objective of this work assignment is to assist EPA in improving the use of IM240 data used in developing the MOBILE model exhaust BERs (for LDGVs only). Since the time that the MOBILES BERs were developed, additional FTP and IM240 emission test data have been collected, which EPA plans to incorporate into the development of exhaust BERs for MOBILES. This report documents a proposed methodology for improving the current procedure used to estimate motor vehicle exhaust BERs based on FTP and IM240 data. One focus of the report is to determine a method for accounting for I/M program effects, since vehicles in the new data set may have been through multiple I/M program test cycles! KEY WORDS/DESCRIPTORS - Select the scientific or engineering terms that identify the major concept of the research and are sufficiently specific and precise to be used as index entries for cataloging. MOBILES; MOBILE6; IM240; basic emission rates; FTP; TECHS ------- |