THE CONTRIBUTION OF MOTOR VEHICLES AND OTHER SOURCES TO AMBIENT AIR POLLUTION
Report # 16 in the series: The Annualized Social Cost of Motor- Vehicle Use in the United States, based on 1990- 1991 Data
UCD- ITS- RR- 96- 3 ( 16) rev. 2
Mark A. Delucchi1
Donald R. McCubbin2
1Institute of Transportation Studies
University of California
Davis, California 95616
madelucchi@ ucdavis. edu
www. its. ucdavis. edu/ people/ faculty/ delucchi/
2Abt Associates, Inc.
Bethesda, Maryland
August 1996
updated October 2004 ( rev. 1)
reformatted and edited February 2006 ( rev. 2) ACKNOWLEDGMENTS
This report is one in a series that documents an analysis of the full social- cost of motor- vehicle use in the United States. The series is entitled The Annualized Social Cost of Motor- Vehicle Use in the United States, based on 1990- 1991 Data. Support for the social- cost analysis was provided by Pew Charitable Trusts, the Federal Highway Administration ( through Battelle Columbus Laboratory), the University of California Transportation Center, the University of California Energy Research Group ( now the University of California Energy Institute), and the U. S. Congress Office of Technology Assessment.
Many people provided helpful comments and ideas. In particular, we thank David Greene, Gloria Helfand, Arthur Jacoby, Bob Johnston, Charles Komanoff, Alan Krupnick, Charles Lave, Douglass Lee, Steve Lockwood, Paul McCarthy, Peter Miller, Steve Plotkin, Jonathan Rubin, Ken Small, Brandt Stevens, Jim Sweeney, Todd Litman, and Quanlu Wang for reviewing or discussing parts of the series, although not necessarily this particular report. Of course, we alone are responsible for the contents of this report.
i
REPORTS IN THE UCD SOCIAL- COST SERIES
There are 21 reports in this series. Each report has the publication number UCD- ITS- RR- 96- 3 (#), where the # in parentheses is the report number.
Report 1: The Annualized Social Cost of Motor- Vehicle Use in the U. S., 1990- 1991: Summary of Theory, Methods, Data, and Results ( M. Delucchi)
Report 2: Some Conceptual and Methodological Issues in the Analysis of the Social Cost of Motor- Vehicle Use ( M. Delucchi)
Report 3: Review of Some of the Literature on the Social Cost of Motor- Vehicle Use ( J. Murphy and M. Delucchi)
Report 4: Personal Nonmonetary Costs of Motor- Vehicle Use ( M. Delucchi)
Report 5: Motor- Vehicle Goods and Services Priced in the Private Sector ( M. Delucchi)
Report 6: Motor- Vehicle Goods and Services Bundled in the Private Sector ( M. Delucchi, with J. Murphy)
Report 7: Motor- Vehicle Infrastructure and Services Provided by the Public Sector ( M. Delucchi, with J. Murphy)
Report 8: Monetary Externalities of Motor- Vehicle Use ( M. Delucchi)
Report 9: Summary of the Nonmonetary Externalities of Motor- Vehicle Use ( M. Delucchi)
Report 10: The Allocation of the Social Costs of Motor- Vehicle Use to Six Classes of Motor Vehicles ( M. Delucchi)
Report 11: The Cost of the Health Effects of Air Pollution from Motor Vehicles ( D. McCubbin and M. Delucchi)
Report 12: The Cost of Crop Losses Caused by Ozone Air Pollution from Motor Vehicles ( M. Delucchi, J. Murphy, J. Kim, and D. McCubbin)
Report 13: The Cost of Reduced Visibility Due to Particulate Air Pollution from Motor Vehicles ( M. Delucchi, J. Murphy, D. McCubbin, and J. Kim)
ii
Report 14: The External Damage Cost of Direct Noise from Motor Vehicles ( M. Delucchi and S. Hsu) ( with separate 100- page data Appendix)
Report 15: U. S. Military Expenditures to Protect the Use of Persian- Gulf Oil for Motor Vehicles ( M. Delucchi and J. Murphy)
Report 16: The Contribution of Motor Vehicles and Other Sources to Ambient Air Pollution ( M. Delucchi and D. McCubbin)
Report 17: Tax and Fee Payments by Motor- Vehicle Users for the Use of Highways, Fuels, and Vehicles ( M. Delucchi)
Report 18: Tax Expenditures Related to the Production and Consumption of Transportation Fuels ( M. Delucchi and J. Murphy)
Report 19: The Cost of Motor- Vehicle Accidents ( M. Delucchi)
Report 20: Some Comments on the Benefits of Motor- Vehicle Use ( M. Delucchi)
Report 21: References and Bibliography ( M. Delucchi)
There are two ways to get copies of the reports.
1). Most reports are posted as pdf files on Delucchi’s faculty page on the UC Davis ITS web site: www. its. ucdavis. edu/ people/ faculty/ delucchi/
2). You can order hard copies of the reports from ITS:
A. fax: ( 530) 752- 6572
B. e- mail: itspublications@ ucdavis. edu
C. ITS web site: http:// www. its. ucdavis. edu
D. mail: Institute of Transportation Studies, University of California, One Shields Avenue, Davis, California 95616 attn: publications
For general information about ITS, call ( 530) 752- 6548.
ITS charges for hard copies of the reports. The average cost is $ 10 per report. You can get a cost list before hand, of course. Or, you can have them send the reports with an invoice.
iii
LIST OF ACRONYMS AND ABBREVIATIONS AND OTHER NAMES
The following are used throughout all the reports of the series, although not necessarily in this particular report
AER = Annual Energy Review ( Energy Information Administration)
AHS = American Housing Survey ( Bureau of the Census and others)
ARB = Air Resources Board
BLS = Bureau of Labor Statistics ( U. S. Department of Labor)
BEA = Bureau of Economic Analysis ( U. S. Department of Commerce)
BTS = Bureau of Transportation Statistics ( U. S. Department of Transportation)
CARB = California Air Resources Board
CMB = chemical mass- balance [ model]
CO = carbon monoxide
dB = decibel
DOE = Department of Energy
DOT = Department of Transportation
EIA = Energy Information Administration ( U. S. Department of Energy)
EPA = United States Environmental Protection Agency
EMFAC = California’s emission- factor model
FHWA = Federal Highway Administration ( U. S. Department of Transportation)
FTA = Federal Transit Administration ( U. S. Department of Transportation)
GNP = Gross National Product
GSA = General Services Administration
HC = hydrocarbon
HDDT = heavy- duty diesel truck
HDDV = heavy- duty diesel vehicle
HDGT = heavy- duty gasoline truck
HDGV = heavy- duty gasoline vehicle
HDT = heavy- duty truck
HDV = heavy- duty vehicle
HU = housing unit
IEA = International Energy Agency
IMPC = Institutional and Municipal Parking Congress
LDDT = light- duty diesel truck
LDDV = light- duty diesel vehicle
LDGT = light- duty gasoline truck
LDGV = light- duty gasoline vehicle
LDT = light- duty truck
LDV = light- duty vehicle
MC = marginal cost
MOBILE5 = EPA’s mobile- source emission- factor model.
MSC = marginal social cost
iv
MV = motor vehicle
NIPA = National Income Product Accounts
NOx = nitrogen oxides
NPTS = Nationwide Personal Transportation Survey
OECD = Organization for Economic Cooperation and Development
O3 = ozone
OTA = Office of Technology Assessment ( U. S. Congress; now defunct)
PART5 = EPA’s mobile- source particulate emission- factor model
PCE = Personal Consumption Expenditures ( in the National Income Product Accounts)
PM = particulate matter
PM10 = particulate matter of 10 micrometers or less aerodynamic diameter
PM2.5 = particulate matter of 2.5 micrometers or less aerodynamic diameter
PMT = person- miles of travel
RECS = Residential Energy Consumption Survey
SIC = standard industrial classification
SOx = sulfur oxides
TIA = Transportation in America
TSP = total suspended particulate matter
TIUS = Truck Inventory and Use Survey ( U. S. Bureau of the Census)
USDOE = U. S. Department of Energy
USDOL = U. S. Department of Labor
USDOT = U. S. Department of Transportation
VMT = vehicle- miles of travel
VOC = volatile organic compound
WTP = willingness- to- pay
v
TABLE OF CONTENTS
ACKNOWLEDGMENTS............................................................................................................. i
REPORTS IN THE UCD SOCIAL- COST SERIES.................................................................... ii
LIST OF ACRONYMS AND ABBREVIATIONS AND OTHER NAMES.......................... iv
TABLE OF CONTENTS............................................................................................................. vi
16. THE CONTRIBUTION OF MOTOR VEHICLES AND OTHER SOURCES TO AMBIENT AIR POLLUTION....................................................................... 1
16.1 MODELING AIR POLLUTION AND THE CONTRIBUTION OF MOTOR VEHICLES...................................................................................................................... 1
16.1.1 Background................................................................................................ 1
16.1.2 Modeling pollution formation and estimating the contribution of motor vehicle emissions to ambient pollution....................................................................................................... 2
16.2 ESTIMATES OF EMISSIONS: THE EPA’S OFFICIAL EMISSIONS INVENTORY ( OEIP’, I, C), AND OUR CORRECTIONS TO THE EPA ESTIMATES ( ECP’, I).................................................................................................... 13
16.2.1 Background.............................................................................................. 13
16.2.2 Estimates of VOCs, NOx, and CO emissions from mobile sources ( MOBILE5A model)...................................................... 14
16.2.3 Estimates of PM and SOx exhaust emissions from mobile sources ( PART5 model).............................................................. 21
16.2.4 Estimates of PM dust from paved roads ( AP- 42 Volume 1, and PART5 model)................................................................ 37
16.2.5 Estimates of PM dust from unpaved roads ( AP- 42, Volume 1)................................................................................................... 53
16.2.6 Estimates of PM emissions from construction, including road construction ( AP- 42, Volume 1).................................. 54
16.2.7 Summary of correction factors.............................................................. 55
16.3 THE DISPERSION OF EMISSIONS FROM SOURCE TO AMBIENT AIR- QUALITY MONITOR.................................................................................................... 55
16.3.1 Conceptual approach to air- quality modeling.................................... 55
16.3.2 The Gaussian model............................................................................... 57
16.3.3 The results of the model......................................................................... 81
16.3.4 Comparison with other estimates......................................................... 81
16.3.5 Long- range transport.............................................................................. 83
16.4 ATMOSPHERIC CHEMISTRY: THE CONTRIBUTION OF MOTOR VEHICLES TO OZONE................................................................................................ 83
16.4.1 Background.............................................................................................. 83
16.4.2 Alternative simple methods for estimating the contribution of precursors to ozone formation.................................... 84 vi
16.5 ATMOSPHERIC CHEMISTRY: THE FORMATION OF SECONDARY SULFATE AND NITRATE PARTICULATES FROM EMISSIONS OF NOX, SO2, AND NH3.......................................................................................................... 92
16.5.1 Background............................................................................................... 92
16.5.2 Formation of ammonium sulfate from SO2 and NH3 emissions.................................................................................................... 92
16.5.3 Formation of ammonium nitrate from NOx and NH3 emissions.................................................................................................... 98
16.5.4 Other contributors to secondary particulate formation.................. 101
16.5.5 Secondary organic aerosols ( SOA)..................................................... 102
16.5.6 Size distribution of ammonium sulfate, ammonium nitrate, and organic aerosols................................................................. 103
16.5.7 Formal model of ambient particulate levels after a change in emissions................................................................................ 105
16.6 COMPARISON OF OUR MODELING RESULTS WITH THE SOURCE- APPORTIONMENTS FROM CHEMICAL MASS- BALANCE STUDIES.......................... 108
16.7 REFERENCES.......................................................................................................... 111
ABBREVIATIONS USED IN TABLES IN THIS REPORT................................................................... 127
TABLES
TABLE 16- 1. CORRECTIONS TO THE EMISSIONS INVENTORY: THE RATIO OF OUR ESTIMATE OF EMISSIONS TO THE EPA’S ( 1995D) OFFICIAL ESTIMATES................................................................................................................ 128
TABLE 16- 2. PM AND OTHER EXHAUST EMISSIONS FROM HIGH- MILEAGE, IN- USE LIGHT- DUTY GASOLINE VEHICLES COMPARED TO PART5 MODEL EMISSIONS................................................................................................... 129
TABLE 16- 3. PM EXHAUST EMISSIONS FROM IN- USE HEAVY- DUTY VEHICLES TESTED OVER ON A CHASSIS DYNAMOMETER....................................................... 130
A. TESTS OF PRE- 1980 VEHICLES OVER THE HDTC................................................................. 130
TABLE 16- 3. PM EXHAUST EMISSIONS FROM IN- USE HEAVY- DUTY VEHICLES TESTED OVER ON A CHASSIS DYNAMOMETER....................................................... 131
B. PM EMISSIONS FROM 1980S AND 1990S IN- USE HEAVY- HEAVY DIESEL VEHICLES, TESTED ON THE WEST VIRGINIA UNIVERSITY PORTABLE CHASSIS DYNAMOMETER..................................................................... 131
TABLE 16- 4. COMPARISON OF MOTOR VEHICLE PM EXHAUST EMISSIONS BACK- CALCULATED FROM FIELD STUDIES AND EMISSIONS CALCULATED BY THE PART5 MODEL ( GRAMS/ MILE)......................................... 132
TABLE 16- 5. CALCULATION OF TRAVEL FRACTIONS AND AVERAGE VEHICLE WEIGHTS, FOR USE IN THE PART5 MODEL APPLIED IN TABLE 16- 4 AND TABLE 16- 6...................................................................................................... 136 vii
TABLE 16- 6. CALCULATION OF TOTAL PM EMISSIONS FROM TRAFFIC, USING PART5/ AP- 42........................................................................................................ 138
TABLE 16- 7. COMPARISON OF EMFAC7F AND MOBILE5A ESTIMATES OF PM EMISSIONS......................................................................................................... 139
TABLE 16- 8. MOTOR- VEHICLE AND FUGITIVE- DUST EMISSIONS OF PM IN URBAN AREAS OF THE U. S. IN 1990, ACCORDING TO THE OFFICIAL EPA EMISSION INVENTORY ( MILLION TONS)....................................................... 140
TABLE 16- 9. SOURCE CONTRIBUTIONS TO AMBIENT PM10, AS ESTIMATED BY CHEMICAL MASS- BALANCE STUDIES..................................................................... 141
TABLE 16- 9 ( CONTINUED).......................................................................................................... 142
TABLE 16- 9 ( CONTINUED).......................................................................................................... 143
TABLE 16- 10. SOURCE CONTRIBUTIONS TO AMBIENT PM2.5, AS ESTIMATED BY CHEMICAL MASS- BALANCE STUDIES..................................................................... 146
TABLE 16- 11. THE RATIO OF ROAD- DUST PM TO MOTOR- VEHICLE EXHAUST PM: CMB SOURCE APPORTIONING VERSUS THE EMISSIONS INVENTORY.............................................................................................................. 149
TABLE 16- 12. ATMOSPHERIC RESIDENCE TIME AS A FUNCTION OF PARTICLE SIZE........................................................................................................................... 150
TABLE 16- 13. COMPARISON OF MOTOR VEHICLE PM EMISSIONS BACK- CALCULATED FROM FIELD STUDIES AND EMISSIONS CALCULATED BY PART5/ AP- 42-- STUDIES OUTSIDE OF THE U. S. MIDWEST ( GRAMS/ MILE)........................................................................................................ 152
TABLE 16- 14. SIZE DISTRIBUTION OF PARTICLES FROM VARIOUS SOURCES.......................... 154
TABLE 16- 15. ESTIMATES OF CONTRIBUTION TO AIR QUALITY, RELATIVE TO CONTRIBUTION OF LDVS, PER KG OF EMISSIONS, BASED ON SIMPLE DISPERSION MODELING: ASSUMED VALUES OF INPUT PARAMETERS........................................................................................................... 156
TABLE 16- 15 ( CONTINUED)........................................................................................................ 159
TABLE 16- 15 ( CONTINUED)........................................................................................................ 161
TABLE 16- 16. STATISTICS REGARDING AQCRS AND COUNTIES WITHIN AQCRS..................................................................................................................... 163
TABLE 16- 17. STATISTICS FOR MAJOR POINT SOURCES............................................................ 164
TABLE 16- 18A. DEPOSITION VELOCITY OF PARTICLES AND GASES ( CM/ SEC)...................... 166
TABLE 16- 18. OUR ASSUMPTIONS AND CALCULATIONS REGARDING SETTLING AND DEPOSITION VELOCITY AND REACTION RATES OF PARTICLES AND GASESA............................................................................................................ 167
TABLE 16- 19. MODEL RESULTS: ESTIMATED VALUES FOR DNP’, I, C, AND DNP’, I, OC, THE CONTRIBUTION TO AMBIENT POLLUTION PER U NIT OF EMISSION, FOR EACH POLLUTANT AND EMISSION- SOURCE CATEGORY, RELATIVE TO THE CONTRIBUTION OF LIGHT- DUTY MOTOR- VEHICLES.................................................................................................... 168 viii
A. URBAN MONITORS, EMISSION SOURCES WITHIN THE COUNTY, LOW- COST CASE...................................................................................................................... 168
B. URBAN MONITORS, EMISSION SOURCES WITHIN THE COUNTY, HIGH- COST CASE...................................................................................................................... 169
C. URBAN MONITORS, EMISSIONS OUTSIDE THE COUNTY, SMALL AQCRS, LOW- COST CASE............................................................................................................ 170
D. URBAN MONITORS, EMISSIONS OUTSIDE THE COUNTY, SMALL AQCRS, HIGH- COST CASE............................................................................................ 171
E. URBAN MONITORS, EMISSIONS OUTSIDE THE COUNTY, LARGE AQCRS, LOW- COST CASE............................................................................................................ 172
F. URBAN MONITORS, EMISSIONS OUTSIDE THE COUNTY, LARGE AQCRS, HIGH- COST CASE........................................................................................................... 173
G. AGRICULTURAL MONITORS, EMISSION SOURCES WITHIN THE COUNTY, LOW- COST CASE............................................................................................................ 174
H. AGRICULTURAL MONITORS, EMISSION SOURCES WITHIN THE COUNTY, HIGH- COST CASE........................................................................................... 175
I. AGRICULTURAL MONITORS, EMISSIONS OUTSIDE THE COUNTY, SMALL AQCRS, LOW- COST CASE............................................................................................. 176
J. AGRICULTURAL MONITORS, EMISSIONS OUTSIDE THE COUNTY, SMALL AQCRS, HIGH- COST CASE............................................................................................ 177
K. AGRICULTURAL MONITORS, EMISSIONS OUTSIDE THE COUNTY, LARGE AQCRS, LOW- COST CASE............................................................................................. 178
L. AGRICULTURAL MONITORS, EMISSIONS OUTSIDE THE COUNTY, LARGE AQCRS, HIGH- COST CASE............................................................................................ 179
TABLE 16- 20. EPA- ESTIMATED EXPOSURE FACTORS FOR DIFFERENT PM EMISSION SOURCES ( EPA, 1994B).......................................................................... 180
TABLE 16- 21. DIESEL ENGINES IN THE SOUTH COAST AIR BASIN, 1982: FUEL USE, EMISSIONS, AND CONTRIBUTION TO TOTAL PARTICULATE POLLUTION.............................................................................................................. 182
TABLE 16- 22. OZONE SENSITIVITY TO VOC AND NOX EMISSIONS....................................... 184
TABLE 16- 23. EMISSIONS, POCP- WEIGHTED EMISSIONS, AND POCP- ADJUSTMENT FACTORS FOR VARIOUS VOC- EMISSION SOURCES........................ 186
TABLE 16- 24. ADJUSTED SALES OF DISTILLATE FUEL OIL IN ARIZONA, CALIFORNIA, AND NEVADA IN 1993, BY TYPE OF END USE ( 103 GALLONS)................................................................................................................ 187
TABLE 16- 25. SOURCE- SPECIFIC FACS BY LAND COVER TYPE................................................ 188
TABLE 16- 26. COMPARISON OF SOURCE- APPORTIONMENTS FROM CHEMICAL MASS- BALANCE STUDIES ( CMB) WITH MODELING RESULTS -- PERCENTAGES OF PM10 ATTRIBUTABLE TO FOUR SOURCES................................ 189
FIGURES
ix
FIGURE 16- 1. MOTOR- VEHICLE EMISSION SOURCES, OTHER EMISSION SOURCES, AND RECEPTOR SITES IN COUNTIES IN AN AIR- QUALITY CONTROL REGION................................................................................................... 191
FIGURE 16- 2. MODELED REPRESENTATION OF MOTOR- VEHICLE EMISSION SOURCES, OTHER EMISSION SOURCES, AND RECEPTOR SITES IN COUNTIES IN AN AIR- QUALITY CONTROL REGION............................................... 192
FIGURE 16- 3. DISPERSION OF POLLUTION FROM A POINT SOURCE....................................... 193
x
16. THE CONTRIBUTION OF MOTOR VEHICLES AND OTHER SOURCES TO AMBIENT AIR POLLUTION
16.1 MODELING AIR POLLUTION AND THE CONTRIBUTION OF MOTOR VEHICLES
16.1.1 Background
In this Report, we explain how we model the contribution of motor- vehicles and other emissions sources to ambient air pollution.
In Reports 11, 12, and 13 of this social- cost series ( see the list at the beginning of this report), we develop dose- response functions that estimate changes in human health, crop production, and visibility as a function of changes in ambient air pollution:
ΔE= fΔP, O()= fPI, PP, O( [ 0]
where:
ΔE = the change in the effect of interest ( human health, crop production, or visibility)
ΔP = the change in ambient air pollution
O = other variables ( such as population or incidence rate)
PI = the initial pollution level
PP = the pollution level after the change in pollution -- in this social- cost analysis, the level after removing all anthropogenic emissions, or 10% or 100% of motor- vehicle related emissions
The initial pollution level, PI, is the actual ambient air quality in each county in the U. S. These data, and the data for any of the other variables O, such as population, are discussed in Reports 11, 12, and 13. In this report, we discuss how we estimate PP, the pollution level after removing anthropogenic emissions, or 10% or 100% of motor- vehicle related emissions.
Note that, when we estimate the pollution level after removing motor- vehicle related emissions, we estimate the effects of a specific, “ marginal” change in pollution: the difference between actual pollution ( PI) and, what pollution would have been had motor- vehicle- related emissions been reduced by 10% or 100% ( PP). We did consider as an alternative estimating the effect of all anthropogenic air pollution and then assigning a fraction of this total effect to motor vehicles, but for two reasons rejected this alternative. First, some of our dose- response functions ( in Reports 11, 12, and 13) are nonlinear, which means that the change in effects ( the responses) depends not only on the difference between PI and PP ( the “ doses”), but on the absolute magnitudes of PI and PP as well. A decrease in pollution from 15 units to 10 units does not necessarily
1
result in the same change in effects as does a decrease from 10 units to 5 units or from 5 units to zero units. If all of the dose- response functions were linear, then effects would be a function only of the difference between PI and PP, and one would have to specify only this difference, and not the absolute values of PI and PP. But as this is not the case, we must specify the absolute magnitudes of PP and PI.
Second, because ozone formation is a nonlinear function of two precursor pollutants, NOx and VOCs, the only way to model the real nonlinear effect on ozone of motor- vehicle ozone- precursor emissions is to model actual ozone levels with and without motor vehicle precursor emissions. It simply is not meaningful to model the elimination of all anthropogenic pollution and then use some ad- hoc rules or “ apportioning” factors assign a fraction of this eliminated pollution to motor vehicles.
In short, we perform a “ with/ without” analysis: we estimate the health, agriculture, or visibility effects of the difference between total air pollution ( with motor- vehicle- related emissions) and air pollution with 10% or 100% of motor- vehicle- related emissions eliminated. To estimate the difference in pollution due to motor- vehicle emissions, we use data on ambient air quality, a detailed emissions inventory, emissions correction factors, and a simple air- quality dispersion model.
16.1.2 Modeling pollution formation and estimating the contribution of motor vehicle emissions to ambient pollution
Recall that our task in this report is to estimate PP, the pollution level without motor- vehicle related emissions ( equation 0). In each county, we estimate PP on the assumption that the ratio of PP to PI ( initial pollution in each county) is equal to the ratio of the modeled PP to modeled PI:
Assume: PPPI= PP* PI* PP= PI⋅ PP* PI* ( 1a, 1b)
where:
PP = the estimated actual pollution level after the change in pollution ( eliminate all anthropogenic emissions, or eliminate 10% or 100% of motor- vehicle- related emissions)
PI = the actual total ambient pollution level ( data from air- quality monitors [ EPA, 1993]; discussed in Reports 11, 12, and 13)
PP* = the modeled level of pollution after the change in pollution
PI* = the modeled level of total ambient pollution.
Thus, in order to estimate PP, we must develop a model of ambient pollution, and estimate the ratio of PP* to PI* in each county.
2
In general, ambient air pollution at particular time and place is a function of the amount of pollutants emitted per unit time, the physical dispersion of the emissions from the emissions source to the site where the ambient pollution is being measured, and chemical transformations of pollutants. Dispersion and chemical transformations are a function of topography, meteorology, the mix of pollutants, and other factors. Formally:
PIP*= fEP', i; DP', id, h, m, t...(); CP'→ Ps, m, t...()() PPP*= fEP'^, i; DP', id, h, m, t...(); CP'→ Ps, m, t...()() ( 2a, 2b)
where:
PIp* = the modeled initial level of ambient pollution P, at a particular time and place
PPp* = the modeled level of pollution P at a particular time and place, after the change in emissions
P = the ambient pollutant, measured at the ambient air- quality monitors and included in health, crop, or visibility damage functions: carbon monoxide ( CO), ozone ( O3), nitrogen oxides ( NOx), total suspended particulate matter ( TSP), particulate matter less than 10 microns in aerodynamic diameter ( PM10), and particulate matter less than 2.5 microns ( PM2.5) 1
Ep’, i = emissions of P’ from source i, over some time period
p’ = the emitted pollutant: CO’ (--> CO), PM2.5- 10’ ( also called “ coarse” PM10) (- -> PM10), PM2.5’ (--> PM2.5, PM10), NOx’ (--> NO2, O3, PM10, PM2.5); volatile organic compounds ( VOCs’; --> O3, PM2.5), SO2’ (--> PM10, PM2.5), ammonia ( NH3’ --> PM10, PM2.5)
Dp’, i( d, h, m, t...) = the dispersion of emissions P’ from source i, as a function of distance ( d), height ( h), meteorology ( m; e. g., wind, temperature), topography ( t), and other factors
Cp’--> p( s, m, t...) = the chemical transformation of emissions of P’ to ambient pollutant P, as a function of the mix of pollution ( s), meteorology ( m), topography ( t), and other factors
1We do not include sulfur dioxide ( SO2) as an ambient polluta nt because we do not attribute any health, visibility, or agricultural effects to SO2 per se. However, we do account for the contribution of SOx emissions to ambient particulate levels.
In Report # 11, we also estimate the health effects of toxic air pollutants, but the method of estimating the motor- vehicle contribution to toxic air pollution is different from the met fhod, outlined in this report, of estimating the motor- vehicle contribution to other ambient pollution. The analysis of the damage cost of motor- vehicle toxics is presented in Report # 11.
3
Ep’^, i = emissions of P’ from source i over some time period, minus the emissions that are presumed to be eliminated; in other words, the emissions of P’ from source i that remain after the hypothetical change in emissions has occurred
Note that we distinguish between ambient air pollutants ( P), measured at air- quality monitors, and emitted pollutants ( P’), which disperse, and in some cases participate in chemical reactions, to become ambient, measured pollutants. Emitted pollutants can be the same chemical compounds as ambient pollutants ( e. g., carbon monoxide [ CO] is emitted, and also is an ambient pollutant), or can be involved in chemical reactions that produce ambient pollutants ( e. g., volatile organic compounds [ VOCs] are emitted, and are involved in the atmospheric formation of ozone).
To model the link between emissions and ambient air pollution we make several simplifications:
I). We assume that in each county c, the ambient pollution measured at the air- quality monitors is a function of:
i) emissions in county c, and
ii) emissions from other counties in the same Air Quality Control Region2 ( AQCR) as county c.
In essence, we model emissions from two source areas, or bands: the county of the monitor, and the band of counties around the county of the monitor. As explained next, we do this as a compromise between the impossible task of modeling emissions from every individual source and the oversimplification of having only one set of emission sources per air basin.
Recall that we estimate ambient air quality, as measured at EPA- ambient air- quality monitors, in each county. Ideally, we would model air quality in each county as a function of emissions from every source that contributes in any way to air quality in the county. This would require that we formally locate and characterize every individual emissions source, define air basins and pollution transport regions, and model air quality as a function of all effective emissions sources. Unfortunately, we do not have the data or resources to be able to do such detailed modeling for every county and air basin the U. S.
Rather than model the effect on air quality of every individual emissions source, one can define bands or regions of emissions, each with an effective “ center” of emissions, and model the effect on air quality of emissions from these bands. The greater the number of bands or regions ( as aggregations of emissions sources), the greater the precision, but the greater the data and analytical requirements. Our balance is to choose two emissions “ bands,” or areas: the county of the air- quality monitor in question, and the counties outside of this county but within the same AQCR. Within the county, we will estimate the actual effective location of different source categories
2Air quality control regions are defined in the Code of Federal Regulations ( Section 40: Part 81).
4
( highway vehicles, power plants, off- road vehicles, construction, and so on). In the outside counties, we will assume a single effective location for all emission sources. We discuss this in more detail in Section 16.3.
II). We ignore the transport of pollution from one AQCR to another, and assume that pollution within an AQCR is a function only of emissions within the AQCR. This assumption obviates the difficulties of analyzing long- range pollutant transport, and hence greatly simplifies our analysis. Of course, as discussed a bit further in Section 16.3, on dispersion modeling, we recognize that in some areas, such as the Northeastern U. S., long- range transport is important, and ideally should not be ignored.
III). We assume that emissions of precursor pollutants P’ disperse as P’ from the source to the receptor ( the ambient air- quality monitor), and then aft the receptor undergo any chemical transformations to produce ambient pollutant P. For example, we assume that VOC and NOx emissions disperse as such from anywhere in the AQCR to the receptor in the county of interest, and at the receptor then are converted into ozone ( O3). We make this assumption because we cannot easily model chemical transformations as a function of the distance from the source.
IV). In equation 1, we estimate the ratio PP*/ PI*; we do not estimate PI* and PP* individually in units of concentration ( μg/ m3). We do this because there is less uncertainty in modeling dispersion from one source relative to another than in modeling dispersion in absolute terms. Our model estimates the dispersion of emissions from non- motor- vehicle sources relative to dispersion of emissions from light- duty motor vehicles. With this relative model of dispersion, we can estimate the ratio PP*/ PI*, but not PP* and PI* individually. We discuss this more below and in section 16.3.
V). In the cases where we model the chemical transformation of precursor emissions to ambient pollutants ( VOCs, NOx --> O3; NOx, SOx, NH3, VOCs --> PM10, PM2.5), we ignore meteorology and topography and assume that the ambient pollution is a function only of the amount precursor emissions at the site of the monitor.
A simple model of pollutant formation
With these assumptions, we consider a simple model of pollutant formation:
5
PIP, c*= CP'→ PEP1', i, c⋅ DP1', i, c+ EP1', i, oc⋅ DP1', i, oc() iΣ, EP2', i, c⋅ DP2', i, c+ EP2', i, oc⋅ DP2', i, oc() iΣ, ... ⎛ ⎝ ⎜ ⎜ ⎜ ⎜ ⎜ ⎞ ⎠ ⎟ ⎟ ⎟ ⎟ ⎟ PPP, c*= CP'→ PEP1'^, i, c⋅ DP1', i, c+ EP1'^, i, oc⋅ DP1', i, oc() iΣ, EP2'^, i, c⋅ DP2', i, c+ EP2'^, i, oc⋅ DP2', i, oc() iΣ, ... ⎛ ⎝ ⎜ ⎜ ⎜ ⎜ ⎜ ⎞ ⎠ ⎟ ⎟ ⎟ ⎟ ⎟
( 3a, 3b)
where:
EP', i, oc= EP', i, oo ∈ RcΣ
PIp, c* = the modeled level of total ambient pollution P “ received” or formed at air- quality monitors in county C, in a year, given the baseline emissions
PPp, c* = the modeled level of total ambient pollution P “ received” or formed at air- quality monitors in county C, in a year, after the change in emissions
subscript P = the ambient pollutant, measured at ambient air- quality monitors
subscript C = the county of interest ( i. e., the county for which air quality and the cost of air pollution are estimated)
subscript P’ = the emitted pollutants
subscript Rc = the AQCR that contains county C
subscript OC = all counties other than county C in AQCR Rc
subscript O = a county other than C in AQCR Rc ( all O together make OC)
Cp’--> p = the chemical transformation of emissions of precursor pollutants P’ ( P1’, P2’,...) to ambient pollutant P ( discussed below; this transformation function is assumed to be the same in every county, and to be independent of the source of the emissions)
Ep1’, i, c, Ep2’, i, c ... = yearly baseline emissions of precursor pollutants P1’, P2’... from emissions source i in county C
Ep1’, i, oc, Ep2’, i, oc ... = yearly baseline emissions of precursor pollutants P1’, P2’... from emissions source i in all counties except C in AQCR R
6
Dp1’, i, c, Dp2’, i, c = the fraction of emissions of precursor pollutants P1’, P2’... from source i in county C that reaches the ambient air- quality monitor in county C
Dp1’, i, oc, Dp2’, i, oc = the fraction of emissions of precursor pollutants P1’, P2’..., from source i in all counties except C in AQCR R, that reaches the ambient air- quality monitor in county C
Ep1’^, i, c, Ep2’^, i, c ... = yearly emissions of precursor pollutants P1’, P2’... from source i in county C, after the change in emissions
Ep1’^, i, oc, Ep2’^, i, oc ... = yearly emissions of precursor pollutants P1’, P2’... from source i in all counties except C in AQCR R, after the change in emissions
Ep’, i, o = emissions of pollutant P’ from source i in county O in AQCR Rc ( for simplicity, we leave the notation for P’ general, and do not write out separate equations for P1’, P2’, P1’^, and P2’^)
Now, recall that we will model pollution with 100% of anthropogenic emissions eliminated, and with 10% and 100% of emissions related to motor- vehicle use eliminated. Emissions “ related” to motor- vehicle use comprise direct emissions, such as evaporative, tailpipe and road dust emissions, and “ indirect” emissions from sources such as the production of motor fuel at refineries, the assembly of motor vehicles, the servicing of motor vehicles, the manufacture of materials used in motor vehicles, road construction, and so on. Because so many sources are related to motor- vehicle use in one way or another, we incorporate formally into our model a motor- vehicle share factor, which is the share of emissions, from each source in the emissions inventory, that is related to motor- vehicle use. From some of the sources in the inventory ( such as highway construction, and of course motor- vehicles themselves), all of the emissions are attributable to motor- vehicle use; from other sources ( such as agricultural operations), none of the emissions are attributable to motor vehicle use; and from still other sources ( such as petroleum refineries), some portion of the emissions are attributable to motor- vehicle use. Thus, for the cases in which we eliminate 10% or 100% of motor- vehicle- related emissions:
7
EP^, i= EP', i− EP', i⋅ k⋅ MSP', i= EP', i⋅ 1− k⋅ MSP', i() andPPP, c*= CP'→ PEP1', i, c⋅ DP1', i, c+ EP1', i, oc⋅ DP1', i, oc() iΣ⋅ 1− k⋅ MSP1', i(), EP2', i, c⋅ DP2', i, c+ EP2', i, oc⋅ DP2', i, oc() iΣ⋅ 1− k⋅ MSP2', i(), ... ⎛ ⎝ ⎜ ⎜ ⎜ ⎜ ⎜ ⎞ ⎠ ⎟ ⎟ ⎟ ⎟ ⎟ 4a, 4b
where:
MSp, i = the motor- vehicle- related fraction of emissions of precursor pollutant P’ ( P1’, P2’...) from emissions source i; that is, of the emissions of pollutant P’, from source i, MSp, i is the fraction that is related to motor- vehicle use ( e. g., all tailpipe emissions from motor- vehicles are related to motor- vehicle use; some fraction of refinery emissions is related to motor- vehicle use, and no fraction of emissions from agricultural tillage is related to motor- vehicle use) ( estimated in Report # 10 of this social- cost series)
k = 1.0 in the case in which 100% of motor- vehicle- related emissions are removed, and 0.10 in the case in which 10% of motor- vehicle- related emissions are removed
i = sources of emissions of P’ ( includes all sources in the emissions inventory: motor vehicles, power plants, industries, businesses, farms, and so on).
In the case in which we eliminate 100% of anthropogenic emissions, Ep^, i is equal to emissions from natural sources.
Now, with two more adjustments, our model of pollutant formation will be complete. First, note that in equations 2, 3, and 4, we have a term for annual county- level emissions of pollutant P’ from source i: Ep’, c, i ( for the county C with the air- quality monitor of interest) or Ep’, oc, i ( for all counties except C in AQCR Rc). Now, the emissions data that we have are the EPA’s ( 1995d, 1995e) official inventory of emissions in every county of the U. S., in 1990. ( We discuss these estimates below.) Let us designate the official EPA county- inventory estimate of emissions of pollutant P’ from source i as: OEIp’, c, i, or OEIp’, oc, i. It appears that most of these official inventory estimates -- the OEI -- are reasonably accurate. However, we do know that the official inventory ( OEI) over- or under- estimates emissions of some pollutants from some sources. Therefore, in general, we will assume that the true county- level emissions of pollutant P’ from source i ( Ep’, c, i, Ep’, oc, i) are equal to the official estimate of emissions multiplied by a correction factor:
8
EP', i, c= ECP', i⋅ OEIP', i, cEP', i, oc= ECP', i⋅ OEIP', i, oo ∈ RcΣ ( 5)
where:
OEIp’, i, c = the EPA’s official emission- inventory estimates of emissions of pollutant P’ from source i in county C ( data from EPA, discussed below)
OEIp’, i, o = the EPA’s official emission- inventory estimates of emissions of pollutant P’ from source i in county O ( any county other than C in AQCR Rc) ( data from EPA, discussed below)
ECp’, i = our emissions- inventory correction factor, equal to the ratio of our estimate of true emissions of pollutants P’ from source i to the EPA’s official estimate ( discussed below; this factor is 1.0 for most sources i, and is assumed to be the same in every county).
Second, we will normalize the dispersion terms in equation 4, Dp’, i, c and Dp’, i, oc, to the dispersion of direct emissions of fine PM from light- duty motor- vehicles in county C. We define a normalized dispersion, DN:
DNP', i, c= DP', i, cDfPM', LDV, cDNP', i, oc= DP', i, ocDfPM', LDV, c
where:
DNp’, i, c = the fraction of emissions of precursor pollutants P’ from source i in county C that reach the ambient air- quality monitor in County C, relative to the fraction of direct emissions of fine PM from light- duty motor- vehicles in county C that reach the ambient air quality monitor in county C
DNp’, i, oc = the fraction of emissions of precursor pollutants P’ from source i in all outside counties OC ( all counties except C in AQCR Rc) that reach the ambient air- quality monitor in county C, relative to the fraction of direct emissions of fine PM from light- duty motor- vehicles in county C that reach the ambient air quality monitor in county C
Dfpm’, LDV, c = the fraction of direct emissions of fine PM from light- duty motor- vehicles in county C that reach the ambient air quality monitor in county C
9
Note that the dispersion term always is normalized with respect to LDV emissions of fine PM in County C. That is, even the dispersion of emissions in all outside counties, OC, is normalized to the dispersion of LDV fine PM emissions in County C. Because every DN term -- for every pollutant, from every source and location -- is normalized with respect to the same Dfpm’, LDV, c, we properly may add together any product of emissions ( E) normalized dispersion ( DN). Thus, the pollution contribution of emissions outside county C is additive with the contribution of emissions in County C, because both contributions are estimated with respect to the same baseline ( Dfpm’, LDV, c). Similarly, with all DN estimated relative to Dfpm’, LDV, c, we may add up the contributions of fine PM, coarse PM, sulfate PM, and nitrate PM, where each contribution is estimated as the product of normalized dispersion and emissions, in order to determine the total contribution of different sources to total ambient PM10 ( which consists of directly emitted fine PM, directly emitted coarse PM, and nitrates and sulfates).
We now have our final general model of ambient pollution, shown here for the case in which we eliminate 10% or 100% of motor- vehicle- related emissions:
PIP, c*= DfPM', LDV, c⋅ CP'→ PPT1' iiΣ, PT2' i,... iΣ⎛ ⎝ ⎜ ⎞ ⎠ ⎟ PPP, c*= DfPM', LDV, c⋅ CP'→ PPT1' i⋅ 1− k⋅ MSP1', i() iΣ, PT2' i⋅ 1− k⋅ MSP2', i(),... iΣ⎛ ⎝ ⎜ ⎞ ⎠ ⎟ PPP, c* PIP, c* = CP'→ PPT1' i⋅ 1− k⋅ MSP1', i() iΣ, PT2' i⋅ 1− k⋅ MSP2', i(),... iΣ⎛ ⎝ ⎜ ⎞ ⎠ ⎟ CP'→ PPT1' iiΣ, PT2' i,... iΣ⎛ ⎝ ⎜ ⎞ ⎠ ⎟ ( 6)
PT1'= ECP1', i⋅ DNP1', i, c⋅ OEIP1', i, c+ DNP1', i, oc⋅ OEIP1', i, oo ∈ RcΣ⎛ ⎝ ⎜ ⎞ ⎠ ⎟ PT2'= ECP2', i⋅ DNP2', i, c⋅ OEIP2', i, c+ DNP2', i, oc⋅ OEIP2', i, oo ∈ RcΣ⎛ ⎝ ⎜ ⎞ ⎠ ⎟
where all terms are as defined above.
10
In the case in which we eliminate 100% of anthropogenic pollution, there are two changes to the numerator of the PPp, c*/ PIp, c* ratio: the ( 1- k. MSp1’, i) term is dropped, and the OEIp1’, i, c become emissions of pollutant i from natural sources in county C.
Notice that the Dfpm’, LDV, c terms will cancel out when we take the ratio of PP to PI, in equation 1. Thus, we do not have to estimate any “ absolute” dispersion factors; rather, we need estimate only dispersion factors relative to light- duty motor- vehicle dispersion factors ( the DN terms). This is important because there is less uncertainty in estimating pollution dispersion from one source relative to another than in estimating dispersion per se.
In this most general form, the model applies to ambient pollutants, such as ozone ( O3) and secondary particulates ( PM2.5 and PM10) , that form via chemical reactions that involve emissions of precursor pollutants P’. However, in the case of ambient pollutants CO, NO2, and “ direct” PM10 and PM2.5, we ignore atmospheric chemistry. In these cases, the ambient pollutants are the same as the emitted pollutants, and the model simplifies to:
PIP, c*= DfPM', LDV, c⋅ ECP', i⋅ DNP', i, c⋅ OEIP', i, c+ DNP', i, oc⋅ OEIP', i, oo ∈ RcΣ⎛ ⎝ ⎜ ⎞ ⎠ ⎟ ⎛ ⎝ ⎜ ⎞ ⎠ ⎟ iΣPPP, c*= DfPM', LDV, c⋅ ECP', i⋅ 1− k⋅ MSP', i()⋅ DNP', i, c⋅ OEIP', i, c+ DNP', i, oc⋅ OEIP', i, oo ∈ RcΣ⎛ ⎝ ⎜ ⎞ ⎠ ⎟ ⎛ ⎝ ⎜ ⎞ ⎠ ⎟ iΣPPP, c* PIP, c* = ECP', i⋅ 1− k⋅ MSP', i()⋅ DNP', i, c⋅ OEIP', i, c+ DNP', i, oc⋅ OEIP', i, oo ∈ RcΣ⎛ ⎝ ⎜ ⎞ ⎠ ⎟ ⎛ ⎝ ⎜ ⎞ ⎠ ⎟ iΣECP', i⋅ DNP', i, c⋅ OEIP', i, c+ DNP', i, oc⋅ OEIP', i, oo ∈ RcΣ⎛ ⎝ ⎜ ⎞ ⎠ ⎟ ⎛ ⎝ ⎜ ⎞ ⎠ ⎟ iΣ
( 7a, 7b)
There are sophisticated models of emissions, dispersion, and atmospheric chemistry. However, it is time consuming and expensive to run all of the best models for every region in the U. S. To keep our task manageable, we will:
• use the results from the best available emissions models;
• treat dispersion very crudely;
11
• use an extremely simple nonlinear model of tropospheric ozone chemistry;
• greatly simplify tropospheric aerosol chemistry.
As we stated in the beginning of this report, we will use our air- quality model to estimate the change in air quality for our dose- response functions for human health ( Report # 11), crop damages ( Report # 12), and visibility ( Report # 13). The application of the model is virtually identical in all three cases ( human health, crops, and visibility). In the case of human health and visibility, we model pollution at urban air- quality monitors, because health and visibility costs are greatest in urban areas ( broadly defined, to include suburban areas). In the case of crop damage, we model pollution at agricultural monitors. As we shall see in section 16.3, this dichotomy ( urban or agricultural) affects but one parameter in the entire model -- the distance from the emissions source to the receptor ( the air- quality monitor).
In the remainder of this report, we present our analysis of emissions, emission- correction factors, dispersion, and atmospheric chemistry. As a check, we will compare our estimates of the motor- vehicle contribution to ambient pollution with analyses of the chemical composition of pollution captured at ambient air- quality monitors.
12
16.2 ESTIMATES OF EMISSIONS: THE EPA’S OFFICIAL EMISSIONS INVENTORY ( OEIP’, I, C), AND OUR CORRECTIONS TO THE EPA ESTIMATES ( ECP’, I)
16.2.1 Background
The EPA ( 1995d, 1995e) has produced a detailed, county- by- county emission inventory, which provides estimates of emissions of all criteria pollutants, from a wide variety of biogenic and anthropogenic sources, for every county in the U. S. ( The 1995d report has the inventory for PM, VOCs, NOx, and SOx [ biogenic emissions excluded], and the 1995e report has the inventory for biogenic emissions of VOC and NOx.) We use these estimates as our starting point in estimating the motor- vehicle contribution to ambient air pollution. However, even though these official estimates are the best that have been published, many of them are very uncertain, and a few are thought to be seriously in error.
Consequently, we examined the uncertainty of some of the emissions estimates in the EPA inventory. If an official estimate of emissions of some pollutant, P’, from source i seemed accurate, or if we did not have any reason to question it, we used it as is in equations 6 or 7 above -- that is, we implicitly assigned a value of 1.0 to the correction factor, ECp’, i, for that pollutant from that emissions source. Otherwise, we estimated a correction factor ( other than 1.0) to apply to the official estimate to make it, in our view, more accurate.
In the official inventory, emissions calculated as the product of an emission factor, which is given in grams of emission per unit of activity ( e. g., grams per mile of travel by light- duty cars), and total activity ( e. g., miles by light- duty cars):
Emissions = emission factor ( grams emitted/ unit activity) * units of activity.
Uncertainty in emissions estimates, then, is related to uncertainty either in the emission factors or in the activity levels.
It appears that most total activity levels are known reasonably well. For example, estimates of total vehicle miles of travel ( VMT) -- the activity which is multiplied by gram/ mile emissions ( from a computer model called MOBILE5a) to produce total grams of emission -- probably are accurate to within better than 10%, although the uncertainty in the estimates of VMT by heavy- duty trucks might be greater than this ( Guensler et al., 1991).
The emission factors, however, can be very uncertain. Emission factors for stationary sources ( such as petroleum refineries) and area sources ( such as road construction activities) are documented in the EPA's voluminous emission- factor handbook, known as AP- 42 Volume 1 ( EPA, 1995a). Emission factors for VOCs, CO, and NOx for the various classes of motor vehicles are estimated in grams/ mile by an EPA computer model, called MOBILE5a. ( California has its own version, called
13
EMFAC7F.) Emission factors of PM and SOx are estimated by a separate EPA computer model, similar to the MOBILE model, called PART5.
Our investigation of the uncertainty of emission factors used to estimate OEIp’, i, c led us to the following conclusions.
• First, it is likely that the MOBILE5a model underestimates real- world gram/ mile emissions of VOCs, CO, and NOx from light- duty gasoline- powered motor vehicles.
• Second, it is possible that the PART5 model underestimates real- world PM emissions from heavy- duty diesel vehicles, although there is little evidence one way or the other.
• Third, it is very likely that AP- 42 overestimates emissions of PM10 road dust and substantially overestimates emissions of PM2.5 road dust.
• Finally, it is likely that AP- 42 overestimates emissions of PM10 and PM2.5 from road construction.
In the following sections we detail these conclusions, and develop the correction factors that we apply to the official emissions estimates ( EPA, 1995d, 1995e) to produce what we believe are more accurate estimates.
16.2.2 Estimates of VOCs, NOx, and CO emissions from mobile sources ( MOBILE5A model)
Background
The MOBILE5a computer model estimates gram/ mile emissions of VOCs, CO, and NOx from several classes of gasoline and diesel- fuel vehicles. The model calculates emissions for a particular year, as a function of the mix of vehicles in the fleet, VMT by vehicle class, vehicle speed, ambient temperature, fuel characteristics, characteristics of inspection and maintenance programs, and other factors. The model is built on the basis of emissions tests of vehicles in use, which are tested mainly but not exclusively over a standardized drive cycle known as the Federal Test Procedure ( FTP). MOBILE 5A, which is the version used to produce the county- by- county emissions estimates in the official inventory we used ( EPA, 1995d, 1995e), was released in 1993. ( A major update to MOBILE, MOBILE6, has been released since the original writing of this report in 1996.)
Shortcomings of the MOBILE model
By the late 1980s, evidence had accumulated that the then- current version of the EPA's emission- factor model, MOBILE3, greatly under- predicted emissions of VOCs and CO from light- duty gasoline vehicles. In 1991, a seminal report by the National Research Council ( 1991) concluded that “ measurements from roadside tests, tunnel studies, and remote- sensing of in- use vehicles provide consistent and compelling evidence that vehicles on the road have substantially higher CO and VOC emissions than current emissions models predict” ( p. 288). Analyses of the relative abundance of VOCs, CO, and NOx in the atmosphere, and of the composition of ambient VOCs, also
14
indicated that emissions of VOCs and CO from mobile sources were underestimated. The models appeared to underestimate VOC and CO emissions by a factor of 2 or 3.
The MOBILE3 performed poorly for several reasons ( NRC, 1991; EPA, 1995b):
1). It underrepresented the proportion of vehicles with extremely high emissions ( called “ super emitters”).
2). It did not include running- loss and resting- loss evaporative emissions of VOCs.
3). It underestimated the rate at which emissions increase as a vehicle accumulates mileage and its emission control systems deteriorate.
4). It did not account for or properly represent the significant increase in emissions during high speeds, hard accelerations, and steep climbs, mainly because the official emissions test, the FTP, does not run vehicles at high engine loads. Because these emissions result from loads “ outside” the official test regime, they usually are called “ off- cycle” emissions.
5) It probably underestimated the total number of starts that occurred with a cool or cold catalyst.
6) It did not represent well the effect of air conditioning on emissions ( the use of air conditioning greatly increases NOx emissions).
In the late 1980s and early 1990s, the EPA conducted extensive testing of in- use vehicles, and revised subsequent versions of the model. Compared with MOBILE3, the most recent version of the model, MOBILE5a, has a more accurate representation of super- emitters, includes running and resting- loss emissions ( MOBILE3 did not), and assumes that emissions increase much more rapidly with mileage ( EPA, 1995b). As a result, the current version of the EPA's emission- factor model, MOBILE5a, predicts much higher emissions than did the previous versions, and appears to predict real- world emissions much more closely ( EPA, 1995b; Auto/ Oil Air Quality Improvement Program, 1995).
However, MOBILE5a still suffers from shortcoming 4) to 6) in the list above: it does not properly represent “ off- cycle” emissions, it probably underestimates the total number of cold starts, and it does not represent well the effects of air conditioning ( Cadle et al., 1997a; EPA, 1995b; German, 1995) 3. As a result, MOBILE5a still apparently
3There may be other problems as well. In a study in Sacramento, California, data from remote sensing indicated that the vehicle- weighted average age was 16% older than is assumed in the CaliforniaMotor Vehicle Emission Inventory Version 7G ( MVEI 7G), and records from Inspection and Maintenance programs indicated that the real- world mileage accumulation rate was higher than assumed in MVEI 7G ( Cadle et al., 1997a). As a result, mobile- source emissions in California might be underestimated substantially. If the MOBILE model similarly mis- estimates the age distribution and mileage accumulation, then it too will underestimate emissions on this score.
15
underestimates CO, VOC, and perhaps NOx emissions from light- duty gasoline vehicle, although not by nearly as much as did MOBILE34.
Off- cycle emissions. In the official emissions test, the FTP, the load on the engine is light: the highest acceleration rate is 3.3 mph/ sec ( equivalent to 0 to 33 mph in 10 seconds), and the highest speed is 57 mph, both on level ground ( Ross et al., 1998). In the real world, the load on the engine often is much higher: people often accelerate must faster than 3.3 mph/ sec, very often drive much more than 57 mph, and ocassionally drive up steep grades, or with heavy loads in the vehicle ( Ross et al., 1998, 1995; German, 1995). This high- power, “ off- cycle” driving can significantly increase emissions of all pollutants, especially if the load is so great that the microprocessor that controls the fuel and engine system instructs the fuel injectors to introduce excess fuel. ( This is called “ command enrichment,” and it occurs in most current vehicles.) For example, Fernández et al. ( 1997) measured on- road emissions from a CARB research vehicle driven in Los Angeles up grades of up to 7%, and found that HC emissions increased about 0.04 grams per mile ( g/ mi) per 1% grade increment, and CO emissions 3.0 g/ mi per 1% grade increment. For a 3% grade, the incremental emissions would be 0.12 g/ mi HC, and 9.0 g/ mi CO.
Ross et al. ( 1998) estimate that in high- power driving with command enrichment, tailpipe g/ sec emissions of CO are 500 times greater than CO emissions over the FTP cycle, mainly because the fuel enrichment increases engine- out emissions of CO and renders the oxidation catalyst almost completely ineffective. Emissions of HC are about 100 times higher, and emissions of NOx about 20 times higher. They estimate that over the life of a properly functioning 1993 model- year vehicle, excess emissions from high- power command enrichment amount to 2.8 g/ mi CO, 0.05 g/ mi HC, and 0.09 g/ mi NOx.
Ross et al. ( 1998) also note that “ excess” emissions can occur at engine loads less than the level sufficient to trigger command enrichment but still more than the highest load in the FTP. They estimate that such moderately high- power driving ( including air conditioning, which we discuss separately below) causes incremental NOx emissions of 0.15 g/ mi.
Number of starts with cooled down catalyst. A cold catalytic converter does not catalyze reactions well, and hence does a poor job of reducing engine- out emissions. As a result, the tailpipe emissions from a cold vehicle are quite high, but drop fairly rapidly as the engine warms the catalytic converter to its effective operating temperature.
4Cadle et al. ( 1998a, 1997a) provide an excellent discussion of real- world emissions from vehicles, including mobile source contributions to the emissions inventory, emissions factor models and activity data, model comparison and development, emission reduction programs, remote sensing, offcycle emissions, and PM emissions. Ross et al. ( 1998, 1995) also provide a good discussion of real- world emissions from passenger cars, although they do not explicitly estimate the extent to which the MOBILE5A model mis- estimates emissions. Fox et al. ( 1994) discuss a variety of possible deficiencies in MOBILE5A, and assess the importance of uncertainty in key input parameters in estimates of fleet- wide emissions.
16
When an engine is turned off, the catalytic converter, which is heated by exhaust gases, begins to cool immediately, and is cold within 45 to 60 minutes ( German, 1995). This behavior, combined with the poor performance of the catalyst when cold, means that five 1- mile trips one hour apart will produce much more pollution than does one 5- mile trip. In other words, gram/ mile emissions -- which is what MOBILE5a estimates -- are a function of the total number of times a vehicle is started with a cold or cool catalyst.
It appears that in reality there are more cold or cool starts than is assumed in MOBILE5a. The emission factors in MOBILE5a are based on the FTP, which is 7.5 miles long and assumes that 43% of all vehicle starts are “ cold” starts. Recent limited data on trip patterns indicate that the fraction of trips that are begun with a cool or cold catalyst might be accurate, but that the average trip length is much less than 7.5 miles ( EPA, 1995b; German, 1995). Assuming that total VMT is correct, this means that there are more starts, and hence many more starts with cold or cool catalysts, than is assumed in MOBILE5a. This, in turn, means that the average emissions per mile are higher than estimated by MOBILE5a, because as mentioned above during cold start and cold- transient driving the catalytic converter is cold and relatively ineffective at reducing engine- out emissions.
Air conditioning. In most FTP tests, the vehicle's air conditioning is not on, and consequently the MOBILE5a emissions model, which is based largely on FTP emissions data, does not account for the effect of air conditioning on emissions. In tests reported by EPA ( 1995b), the use of air conditioning increased VOC emissions by 25%, CO emissions by 51%, and NOx emissions by 92%, over the full FTP, albeit under extreme conditions of high temperature and high humidity. Cadle et al. ( 1997a) report that air conditioning at 95o F and 40% relative humidity had only a minor effect on HC and CO emissions, but increased NOx emissions by 75%. Fernández et al. ( 1997) found that air conditioning at the full setting increased HC emissions by 0.07 g/ mi, and CO emissions by 31.9 g/ mi when driving up steep grades. ( They did not measure NOx emissions.) The large increase in NOx emissions has come as something of a surprise, and by itself suggests that MOBILE5a might significantly underestimate drive- cycle, year- round average emissions.
So how much is MOBILE5a off?
Even though the EPA has gathered data on these problems to be able to improve the subsequent version of the model, MOBILE6 ( EPA, 1995b, we still we face the question of the extent to which the mobile- source emission inventory developed with MOBILE5a still underestimates emissions of VOCs, CO, and perhaps NOx from light- duty gasoline vehicles. Unfortunately, there are few quantitative estimates of the extent of the underestimation. The discussion above suggests that emissions of CO are substantially underestimated, and that emissions of VOCs are underestimated less than are emissions of CO. There is some evidence that under some conditions NOx actually is overestimated ( EPA, 1995b; Auto/ Oil Air Quality Improvement Program, 1995), but
17
when all of the factors discussed above ( off- cycle emissions, cold starts, and especially air conditioning use) are considered, it is more likely that NOx emissions will be found to be underestimated than overestimated.
The following five studies help us quantify the extent to which the MOBILE5a model might be in error:
1). A comparison of ambient ratios of CO: NOx and VOCs: NOx with emission ratios of CO: NOx and VOCs: NOx indicates that the 1991 version of California's emission model, EMFAC7E, underestimates mobile- source CO emissions by a factor of 1.5 and mobile- source VOC emissions by a factor of about 2.2 ( Fujita et al., 1992). The subsequent version of EMFAC7F, similar to EPA’s MOBILE5a, estimates higher VOC emissions than does EMFAC7E, but according to a recent study ( Fujita et al., 1995) still underestimates mobile- source emission factors for VOCs. Fujita et al. ( 1995) used VOC profiles of motor- vehicle VOC exhaust and other VOC emissions sources to estimate the motor- vehicle contribution to measured ambient VOC concentrations in seven urban areas in the San Francisco Bay Area and San Joaquin Valley. They compared this estimated ambient contribution with the ratio of motor- vehicle VOC emissions ( estimated using EMFAC7F) to total estimated VOC emissions in each area. Exhaust and evaporative emissions contributed 70 to 74% of the measured ambient VOCs in the seven urban areas ( excluding biogenic VOCs and acetone), but only 43% of the estimated primary anthropogenic VOC emissions. ( See also Magliano et al., 1993.)
There are three reasons why the ambient chemical- mass- balance source apportionment to motor vehicles might exceed the emissions- inventory apportionment to motor vehicles: 1) in the source apportionment of ambient concentration, the portion attributed to motor- vehicles actually might include some non- vehicular sources that have a VOC profile similar to the motor- vehicle profile; 2) the ambient monitors used in the source apportioning might capture a greater percentage of motor- vehicle emissions than of other emissions, most likely because the monitors are closer to motor vehicles; 3) the motor- vehicle VOC emission factors might be underestimated. However, if underestimation of VOC emissions accounts for all of the discrepancy estimated by Fujita et al. ( 1995), then EMFAC7F underestimated VOC emissions by a factor of 3.4 (!), because motor- vehicle emissions would have to have been 3.4 times higher in order for their share of total emissions to have been 72% ( assuming that all other sources in the inventory were correctly estimated). We believe, however, that part of the discrepancy between the 72% ambient share and 43% estimated emission- inventory share was due to the second possibility, that the monitors generally captured a larger fraction of motor- vehicle emissions than of other emissions. Thus, this study suggests that EMFAC7F underestimates VOC emissions by less than a factor of 3.4
2). German ( 1995) of EPA has made preliminary estimates of the extent to which in- use emissions from a low- emitting vehicle in the year 2020 will exceed the levels predicted by the current model, MOBILE5a. He estimates that VOC emission will be 1.15 times higher than predicted by MOBILE5a, that CO emissions will be 1.47 times higher, and that NOx emissions will be 1.35 times higher.
18
3). The Ross et al. ( 1998) estimates of the excess emissions due to “ offcycle” driving -- 2.8 g/ mi CO, 0.24 g/ mi NOx, and 0.05 g/ mi HC over the life of a properly functioning 1993 model year vehicle -- are 20% of the total estimated in- use CO, 16% of total estimated in- use NOx, and 3% of the total estimated in- use HC. If MOBILE5 accounts for all of their “ in- use” emission sources except off- cycle emissions, then underestimates CO by 25%, NOx by 20%, and HC by 3%.
5). Finally, a comparison of California’s updated emission- factor model, EMFAC7G, with the EMFAC7F version gives some indication of the extent to which MOBILE5a underestimates real- world emissions. EMFAC7F is similar to EPA’s MOBILE5a. The updated version, EMFAC7G, accounts better for high- emitting vehicles, real- world driving patterns, inspection and maintenance programs, and the distribution of starts than does EMFAC7F. In other words, EMFAC7G accounts for many of the factors that cause MOBILE5a to underestimate real- world emissions. The ratios of EMFAC7G to EMFAC7F estimates of emissions from all vehicles in the South Coast Air Basin in summer 1990 are: VOCs 1.29, CO 1.81, and NOx 1.41 ( California Air Resources Board, 1995).
German’s ( 1995) preliminary estimates pertain to a low- emitting vehicle in the year 2020. Because we are working with the 1990 emission inventory, we are interested in the extent to which MOBILE5a under- predicted emissions from a “ fleet average” vehicle in 1990. We expect that generally, MOBILE5a under- predicts emissions from a fleet average vehicle in 1990 by at least as much as it under- predicts emissions from a low- emitting vehicle in the year 2020, because the fleet average vehicle in 1990 will be have higher baseline emissions, and greater variation in emissions as a function of the drivecycle and the number of cold starts. In support of this, we note that the difference between EMFAC7G and EMFAC7F decreases from the year 1990 to the year 2000.
With these considerations, we assume, in our low- cost case, that actual emissions of CO from light- duty gasoline cars and trucks are 1.5 times higher than estimated in the official MOBILE5a- based inventory, that actual emissions of VOCs are 1.1 times higher, and that NOx emissions are 1.2 times higher. In our high– cost case, we assume that actual emissions of CO from light- duty gasoline cars and trucks are 1.8 times higher than estimated in the official MOBILE5a- based inventory, that actual emissions of VOCs are 1.3 times higher, and that actual emissions of NOx are 1.4 times higher. These adjustments are summarized in Table 16- 1 below.
Corrections to VOCs, NOx, and CO emissions from diesel vehicles or heavy- duty gasoline vehicles? For two reasons, we believe that the MOBILE5a estimates of VOC and CO emissions from diesel vehicles and HDGVs are not seriously in error, and consequently do not make any corrections to the official inventory estimates of these emissions.
First, the MOBILE5a model underestimates CO and VOC emissions from LDGVs mainly because the emission control system of LDGVs is not very effective under certain conditions that are not well represented in the database underlying the MOBILE5a model. However, because diesel vehicles do not have catalytic converters,
19
computer- controlled air/ fuel ratios or evaporative control systems ( because diesel fuel is not volatile), one would expect that CO and VOC emissions from conditions not represented in the MOBILE5a model would not be as radically different from emissions under modeled conditions as is the case with LDGVs when the emission control system essentially stops working.
Second, the available data show that diesel vehicles do not produce significant emissions of CO or VOCs anyway.
The situation with NOx is less clear. On the one hand, the recent tunnel studies indicate that MOBILE5a predicts NOx emissions from diesel vehicles reasonably well ( Auto/ Oil Air Quality Improvement Program, 1995), and a recent study of heavy- duty truck emissions on Interstate 20 near the Georgia- Alabama border showed that heavy- duty NOx emissions were within 25% of the values predicted by MOBILE5 ( Cadle et al., 1997a) On the other hand, with the new electronic engine control systems, manufacturers can control to fuel injection to maximize fuel economy but increase NOx emissions, and it appears that some manufacturers of heavy- duty engines have been programming the on- board engine control computer to have late fuel injection, and hence low NOx emissions but also low fuel economy, when the EPA heavy- duty emissions test is being run, but early fuel injection, and hence high fuel economy but also high NOx emissions, when the vehicle is in use ( Walsh, 1997,1998). The difference between the in- use and test cycle NOx emissions can be substantial -- up to 80% ( Walsh, 1997, 1998). However, we are interested in the difference between MOBILE5a estimates and in- use emissions in 1990, not the difference between certification test results and in- use emissions in 1997, and it is by no means obvious that the HDVs used to establish the MOBILE5a emission factors were tuned differently than were the vehicles in- use in 1990, especially since most if not all vehicles in- use in 1990 were not programmed to “ cheat” in the manner described above.
Therefore, we assume that the MOBILE5a model accurately predicts VOC, CO, and NOx emissions from diesel vehicles, and make no correction to the diesel- vehicle emissions inventory of these pollutants. We assume also that MOBILE5a model accurately predicts emissions of these pollutants from heavy- duty gasoline vehicles, and so make no correction to that inventory either.
20
16.2.3 Estimates of PM and SOx exhaust emissions from mobile sources ( PART5 model)
The EPA's PART5 model, similar in structure to the MOBILE5a model, calculates g/ mi exhaust emissions of PM and SOx from 12 classes of motor vehicles ( the same classes of vehicles included in the MOBILE5a model discussed above). It also calculates g/ mi emissions of road dust and particles from tire wear and brake wear5. The g/ mi emission factors of PART5 can be multiplied by estimates of VMT in a particular region to produce a total inventory of mobile- source PM emissions for the region. Because there are relatively few light- duty diesel vehicles and heavy- duty gasoline vehicles, virtually all on- road mobile- source PM comes from light- duty gasoline cars and trucks, and heavy- duty diesel vehicles ( EPA, 1998b):
Contribution of different vehicle classes to total on- road mobile source PM:
LDGVs
LDGT
HDGV
LDDV
LDDT
HDDV
total 103 tons
1987
18%
10%
3%
2%
1%
65%
360
1997
21%
15%
3%
2%
1%
58%
267
In this section, we argue that PART5 may under- estimate exhaust emissions of PM from light- duty gasoline cars and trucks, and heavy- duty diesel vehicles. In the following section ( 16.2.4), we argue that PART5 and AP- 42 overestimate road- dust emissions. Because tirewear and brakewear emissions are much smaller than exhaust and road- dust emissions, we do not analyze the accuracy of the emission factors.
Note that while the EPA has updated MOBILE5 to MOBILE6, as of this writing ( October 2004) is has not updated PART5.
Overview of PART5 estimates of exhaust PM
Formally, PART5 calculates exhaust emissions of PM from each vehicle class, in a target year designated by the user:
EXPMFV, T= EXPMM, V⋅ TFM, V, TMΣ ( M1)
where:
subscript V = the twelve classes of motor vehicles ( light- duty and heavy- duty gasoline or diesel vehicles, two classes of light- duty gasoline trucks, light-
5PART5 also estimates the amount of “ indirect” sulfate, formed in the atmosphere from SO2 emissions, on the assumption that 12% of the sulfur emitted as SO2 becomes sulfur in ammonium sulfate or ammonium bisulfate ( EPA, 1995c). However, indirect sulfate emissions are not counted as PM emissions in an emissions inventory. We treat them separately here, too.
21
duty diesel trucks, 3 classes of diesel vehicles between light- and heavy- duty, buses, and motorcycles)
subscript M = model year of vehicle ( PART5 goes back 25 years from the target year T)
EXPMFV, T = the exhaust- PM emission factor for the fleet of vehicles of class V in user- designated target- year T ( g/ mi)
EXPMM, V = emissions from model year M of vehicle class V ( g/ mi)
TFM, V, T = of total vehicle- miles of travel by vehicle class M in target- year T, the fraction that is done by model- year M
In the case of gasoline vehicles, the total exhaust PM, EXPM in equation M1, is calculated as the sum of lead, direct sulfate, and carbon PM exhuast:
EXPMM, GV= EXPBM, GV+ EXSO4M, GV+ EXCM, GV ( M2)
where:
EXPBM, GV = exhaust emissions of lead from model- year M of gasoline- vehicle class GV ( g/ mi)
EXSO4M, GV = direct sulfate emissions from model- year M of gasoline- vehicle class GV ( g/ mi)
EXCM, GV = exhaust emissions of particulate carbon from model- year M of gasoline- vehicle class GV ( g/ mi)
The parameter EXC is given in a table of g/ mi emission rates organized by vehicle class, model year, and type of fuel and emission control equipment. The parameter EXSO4 is given in g/ mi by type of emission control equipment and vehicle speed.
The calculation of the lead emission factor, EXPB in equation M2, is fairly complex ( EPA, 1995c). However, in 1986 the lead content of “ leaded” gasoline was decreased to 0.1 grams per gallon, and by 1991, sales of leaded gasoline were only 3% of total gasoline sales anyway ( EPA, 1992a), with the result that from 1991 on, lead emissions from on- highway vehicles have been essentially zero ( EPA, 1998b). Consequently, we do not discuss lead- particulate emissions further.
In the case of light- duty diesels, the parameter EXPM is given in a table of g/ mi emission rates organized by vehicle class ( light- duty diesel vehicles, and light- duty diesel trucks) and model year. However, as indicated above, in the summary of the EPA’s Emission Trends estimates, there are so few light- duty diesel vehicles and trucks in the U. S. that presently, it is not worth analyzing the pertinent PART5 emission factors. We do not discuss them further here.
For other diesel- vehicle classes, the g/ mi emission factor EXPM is calculated as:
22
EXPMM, DV= EXPMBM, DV⋅ CFM, DV ( M3)
where:
EXPMBM, DV = emissions from model- year M of diesel- vehicle class DV ( g/ brake- horsepower- hour [ bph- hr])
CFM, DV = bhp- hr/ mi conversion factor for model- year M of diesel- vehicle class DV
The parameter EXPMB is given in a table of g/ bhp- hr emission rates organized by vehicle class ( class 2B of heavy- duty, light- heavy, medium- heavy, heavy- heavy, and buses) and model year6.
Note that in the case of diesel vehicles, the total exhaust PM emission rate ( EXPM or EXPMB), which comprises direct sulfate and carbon PM, is not a calculated value, but rather is a basic g/ mi or g/ bhp- hr number in a data table, whereas in the case of gasoline vehicles the total exhuast PM ( EXPM) is calculated as the sum of separately estimated components ( lead, sulfate, and carbon).
As mentioned above, the fleet emission factors produced by PART5 are multiplied by total fleet travel to produce an estimate of total emissions:
EXPMTT= EXPMFV, T⋅ VMTV, TVΣ ( M4)
where:
EXPMTT = total exhaust emissions of PM from motor vehicles in year T ( grams)
VMTV, T = total vehicle miles of travel by vehicle class V in year T
We can see from equations M1- M4 that there are four potential general sources of error in the calculation of an emissions inventory: the basic emission factors by model year ( EXPMB [ heavy- duty diesel vehicles], EXSO4 [ light- duty gasoline vehicles], and EXC [ light- duty gasoline vehicles]), the bph- hr/ mi conversion factor ( CF [ heavy- duty diesel vehicles]), the travel fractions by model year ( TF), and the total travel by vehicle class ( VMT) 7. In the following sections we discuss the accuracy of the basic emission factors. Recently, Browning ( 1998a, 1998b) has analyzed and updated the bhp- hr/ mi
6The values shown in Table 2 of the EPA’s ( 1995c) User’s Guide are for diesel vehicles that burn the high- sulfur fuel in use prior to 1993. To represent emissions from diesel vehicles that use the low- sulfur fuel mandated beginning in 1993, the EPA makes “ appropriate adjustments” to the high- sulfur values.
7As noted above, we have dropped light- duty diesel vehicles and trucks, and emissions of lead, from the analysis. We also drop emissions from heavy- duty gasoline vehicles, because they contribute so little to total PM emissions from motor vehicles ( EPA, 1998b).
23
conversion factors, so we do not consider them further here. Guensler et al. ( 1991) discuss the accuracy of travel statistics for heavy- duty vehicles in California.
Sulfate PM emissions from gasoline vehicles.
The sulfate emission rates in PART5 are based on relatively old data, and are given independent of the sulfur content of gasoline. They probably do not account fully for emissions from very old or malfunctioning vehicles, or from vehicles driven “ off cycle”. As a result, PART5 might overestimate sulfate emissions.
In PART5, LDGVs that have catalytic converters with air emit 16- 25 mg/ mi sulfate, and all other LDGVs emit 1- 5 mg/ mi sulfate ( EPA, 1995c). The calculated LDGV fleet- average emission rate for the 1990s is on the order of 10 mg/ mi sulfate. These rates are identical to those in the 1985 4th edition of EPA’s Compilation of Air Pollutant Emission Factors for mobile sources ( EPA, AP- 42, vol. 2, 1985), which, in turn, come from the 1981 version of AP- 42, and from a 1983 EPA report on particulate emissions from motor vehicles. It therefore is likely that the emission rates in PART5 are based on tests of late- 70s vintage vehicles with late- 70s gasoline. If so, the PART5 emission factors might not be accurate for 1990s vehicles and fuel.
There is some evidence that PART5 overestimates sulfate emissions from LDGVs. Sagebiel et al. ( 1996) measured exhaust emissions from 23 high- mileage, in- use LDGVs ( model years 1976- 1990), over the IM240 emissions test, and found an average sulfate ( anion) emission rate of only 0.17 mg/ mi8. There was no appreciable trend with respect to model year. This average implies that less than 0.5% of the sulfur in the gasoline was converted to sulfur in SO4. Watson et al. ( 1994c) measured the composition of PM2.5 from approximately 600 LDGVs tested in 1989- 1990 at an I& M facility in Phoenix, Arizona, and found that was only 2.3% of the total mass of PM2.5. Pierson and Brachaczek ( 1983) measured emissions from vehicles in the tunnels in Pennsylvania in 1975- 1979, and found sulfate () emissions of 5 mg/ mi ( 7% of total PM) for LDGVs and 68 mg/ mi ( 5% of total PM) for HDDVs. By comparison, PART5 reports that direct sulfate emissions from LDGVs are more than 50% of total exhaust PM in the 1990s. Finally, emissions of total PM from late- model, new, properly functioning LDGVs are in the range of 2- 3 mg/ mi ( Cadle et al., 1998b; Mulawa et al., 1997; EPA, 1993c) -- less than the PART5 sulfate emission rate alone. SO42- SO42-
Another, related line of reasoning suggests that PART5 overestimates sulfate emissions from LDGVs. The PART5 Users Guide implies ( probably mistakenly) that 2% of the sulfur in gasoline is converted to sulfur in SO4 ( EPA, 1995c, p. 53), and clearly assumes that 2% of the sulfur in diesel fuel is converted to SO4 ( EPA, 1995c, p. 57). Assuming a sulfur content of 340 ppm by weight ( EPA, 1995c) and a fuel economy of 22
8For one of the vehicles, the measured sulfate emission was greater than what could have been produced if all of the sulfur in the gasoline had been converted to sulfate. The authors speculate that some material had “ built up over time and was dislodged during the test” ( p. 81). We have excluded this vehicle from our averaging.
24
mpg, a conversion of 2% of S- fuel to S- SO4 results in a sulfate emission rate of 0.003 g/ mi considerably lower than the rate reported by PART5. With reformulated “ phase II” gasoline, which the EPA ( 1995c) assumes has a sulfur content of 138 ppm, the emission rate at 2% conversion would be 0.001 g/ mi -- an order of magnitude lower than the rate reported by PART59.
Drive- cycle effects. How might differences between real- world driving and the test cycle affect emissions? In the sections that follow, we argue that the PART5 emission factors do not fully reflect emissions from old or malfunctioning vehicles, or from vehicles driven in ways not represented in the emission test cycles. Old vehicles, malfunctioning vehicles, and vehicles driven “ off cycle” ( e. g., with very hard accelerations) generally burn fuel less completely, on account of lower combustion temperatures, less oxygen, or poisoned catalysts, and as a result emit more organic PM. However, it is not immediately clear how lower temperatures and oxygen levels, or poisoned catalysts, would affect emissions of particulate sulfate. Essentially all particulate sulfate comes from the fuel sulfur, which is a fixed quantity that is apportioned at the tailpipe between H2SO4, SO2, H2S, and other sulfur compounds. A decrease in the amount of oxygen available, or a reduction in the efficiency of the catalytic converter, might reduce the formation of the more oxidized species, such as H2SO4, and increase emissions of H2S. If so, then on account of this effect, the “ in- use” fleet of LDGVs, driven in the real world, would emit less sulfate then PART5 predicts.
The foregoing data analysis suggests to us that PART5 might overestimate direct sulfate emissions from LDGVs, especially LDGVs of model year 1981 and later. More clearly, the data indicate that the ratio of sulfate PM to total PM in PART5 is much too high. To resolve this, we need measurements of H2S, H2SO4, and other sulfur emissions from a wide range of vehicle types, vintages, and ages, driven under a wide range of conditions.
Emissions of nitrate, salt, and metal PM.
9In its calculations of S- SO2 emissions, as the difference between total fuel- S and sulfate- S, PART5 assumes that the sulfate “ particles” are droplets of sulfuric acid dissolved in water H2O: H2SO4 [ 7: 1, v/ v]). This implies that the basic sulfate emission factors in PART5 ( e. g., 16- 25 mg/ mi for vehicles with catalytic converters with air emit) include the weight of 7 water molecules and H2 for every SO4 group. If this is correct -- if the basic sulfate emission factors do include this weight -- then, for the purpose of comparing the PART5 “ suflate” emission factors with the “ sulfate” emissions data presented here, we should multiply emissions of SO4 ( which is what we present) by the ratio of the weight of the sulfuric acid droplet to the weight of SO4, 2.33.
It is not clear whether the basic sulfate emission factors are for SO4, or sulfuric acid droplets H2O: H2SO4 [ 7: 1, v/ v]. The 4th edition of AP- 42, which is the source of the PART5 factors, does not speak to the matter. We note, though, that all of the PM data we have seen report the weight of S or SO4, not the weight of droplets of sulfuric acid.
25
As indicated in equation M2, PART5 estimates emissions of lead, sulfate, and organic PM. It apparently does not include emissions of direct nitrate or salts, such as chloride. In their tests of 23 in- use LDGVs, Sagabiel et al. ( 1996) ( see the discussion above) measured an average nitrate emission rate of 0.04 mg/ mi, and an average chloride emission rate of 0.10 mg/ mi. Although these rates obviously are quite small, they are together comparable to the sulfate emissions measured by Sagabiel et al. ( 1996). More significantly, Watson et al. ( 1994c) measured the composition of PM2.5 from approximately 600 LDGVs and 80 HDDVs tested in 1989- 1990 at an I& M facility in Phoenix, Arizona, and found the following contributions to the PM2.5 mass:
LDGVs
HDDVs
carbon
43.6%
73.0%
NO3-
3.9%
0.3%
SO42-
2.3%
2.4%
NH4+
1.7%
0.9%
silicon
1.6%
0.5%
sulfur
1.0%
1.2%
other metals
~ 3- 4%
~ 1- 2%
hydrogen, oxygen, nitrogen..
remainder ( not measured)
remainder ( not measured)
These results show clearly that LDGV emissions of nitrate, ammonium, and metal10 PM, which PART5 does not count, are together several times larger than emissions of sulfate PM, which PART5 does count. This omission might cause PART5 to significantly underestimate total PM emissions from LDGVs11.
Organic PM and total PM from gasoline vehicles.
The PART5 emission factor. As mentioned above, organic PM emissions from gasoline vehicles are presented in a table of g/ mi emission rates organized by vehicle class ( LDGVs, LDGT I, LDGT II, and HDGV), model year, and type of fuel and emission control equipment ( leaded gasoline, unleaded gasoline and no catalyst, unleaded
10Cadle et al. ( 1997b) and Pierson and Brachaczek ( 1983) also report emissions of metals.
11Recall that for HDDVs, the basic emission factor in PART5 is for total PM. Thus, as long as the tests upon which the PART5 factor is based did indeed measure all PM, there is no problem of omission. However, PART5 also apportions the total exhaust PM into two components: direct sulfate PM and organic PM. For this apportioning, PART5 assumes that total PM = sulfate PM + organic PM. The results of Watson et al. ( 1994c) indicate that it would be better to apportion the total to sulfate PM, organic PM, and “ other,” which would be some 4% of the total.
26
gasoline and catalyst without air, and unleaded gasoline and catalyst with air). We may ignore the emission factors for vehicles using leaded gasoline, vehicles without a catalytic converter, and heavy- duty gasoline vehicles, because PM emissions from these sources are minor ( EPA, 1998b). We thus focus on on the emission factors for light- duty vehicles and trucks equipped with a catalytic converter.
PART5 assumes that all light- duty, catalyst- equipped cars and trucks of model year 1981 and later emit 4.3 mg/ mi organic PM ( EPA, 1995c). This emission factor is invariant with respect to user- specifiable inputs for the drive cycle ( cruising or transient), vehicle speed, altitude ( high or low), and inspection & maintenance ( I& M) ( in force or not) ( EPA, 1995c). It is not a function of the age of the vehicle. For any scenario for the year 1990 or later, for any region of the country, light- duty gasoline vehicles and trucks will emit nearly or exactly 4.3 g/ mi organic PM.
According to the EPA’s ( 1995c) User’s Guide, the organic- PM emission factors for gasoline vehicles were determined on the basis of the factors in AP- 42, volume 2 ( EPA, 1985) and the “ updated information” in the EPA’s ( 1993a) Motor- Vehicle Related Air Toxics Study. Comparing the factors in PART5 with the data and factors in the other EPA ( 1985, 1993b) reports, it appears that the PART5 factors for vehicles using leaded gasoline and vehicles without catalytic converters come from AP- 42, volume 2 ( EPA, 1985), and that factors for vehicles with catalytic converters come from the Motor- Vehicle Toxics study ( EPA, 1993a). Appendix H of the latter study ( EPA, 1993a) summarizes the results of nine studies of PM emissions from light- duty gasoline cars and trucks. Three of these studies were published after the 4th edition of AP- 24 ( EPA, 1985) and present emissions data for cars of model year 1981 and later. The average emission rate in all three studies was 5 to 10 mg/ mi, depending on how one does the averaging, and whether the highest emitting vehicle is included. However, in the study that the EPA ( 1993a) gives the most weight to, the average emission rate was 2 mg/ mi. Given that studies in the EPA ( 1993a) apparently report total PM, it is not clear how the how the PART5 organic- PM emission factors were derived from them. Presumably, all of the measurements in the three studies were taken over the FTP.
Now, given this, how might the PART5 emission factor for organic PM ( and total PM) be in error? In general, there are thre ways: 1) the vehicles tested in the three studies from which the PART5 emission factor apparently was derived might not be representative of the in- use vehicle fleet, in regards to characteristics that affect g/ mi emissons; 2) driving in the real world might differ from the driving in the FTP, in ways that affect g/ mi emissions of PM; and 3) future vehicles might have emissions different from those used as the basis of the PART5 estimates.
We believe that there are more high- emitting vehicles in the real world than were tested in the PM emission tests, and that there is more high- emitting driving in the real world than in the FTP, but that PM emission rate for new vehicles generally has been declining, and will continue to decline, with model year.
Were the vehicles tested representative of the in- use fleet, with regards to characteristics that affect g/ mi emissions? We believe that the most serious problem with the PART5 emission factor is that it is based on emissions from properly
27
functioning, well- maintained, and in most cases new vehicles. In the real world there are malfunctioning, poorly maintained, old vehicles, and although there are only a small number of them, they emit so much more than do properly functioning new vehicles that they can raise the fleet- average emission rate appreciably. There is by now considerable evidence that a small number of vehicles emit large amounts of PM, and cause the in- use fleet- average PM emission rate to exceed that assumed in PART5.
Recently, the Desert Research Institute ( Sagebiel et al. 1996) measured exhaust emissions from 23 high- mileage, in- use light- duty gasoline vehicles ( model years 1976- 1990), over the IM240 emissions test, and found that PM exhaust emissions: A) varied by over two orders of magnitude, and B) generally were much higher than predicted by PART5 ( Table 16- 2). These results are important because they pertain to high- mileage in- use vehicles, pulled off of the road and tested without modification. Six of the vehicles smoked visibly, and emitted about ten times more PM than did vehicles that didn’t smoke. Even the non- smoking vehicles, however, emitted considerably more PM than predicted by PART5 ( 50 mg/ mi in the tests versus 20 mg/ mi predicted by PART5 -- see Table 16- 2).
Several other studies report similar results for light- duty gasoline vehicles. Hanson and Rosen ( 1990) measured aerosol black carbon in the exhaust of gasoline vehicles driving up a hill in Berkeley in 1985, and found that emissions varied by more than two orders of magnitude, and that 20% of the vehicles -- the “ high emitters” -- accounted for 65% of the emissions. Miguel et al. ( 1998) measured emissions of particulate PAH and solid carbon ( carbon black) from vehicles in the Caldecott Tunnel in the San Francisco Bay Area in 1996, and estimated an average emission rate of 17 mg/ mi for LDGVs -- much higher than the PART5 predictions about 4 mg/ mi, for all organic PM, in 1996. ( See also Table 16- 4).
In a study of smoking light- duty vehicles in Los Angleles, researchers found that the PM mass emission rate ranged from 29 to 1,651 mg/ mi, with many emission rates one to two orders of magnitude above the EMFAC- prediction of 10 mg/ mi ( Cadle et al., 1997a). Similarly, a fleet of 103 in- use, high- emitting light- duty vehicles in Orange County, California, tested in 1995 on a transportable dynamometer, emitted an average of 138 mg/ mi ( Cadle et al., 1997b) -- about an order of magnitude higher than the PART5 prediction for total PM. The average emission rate for smoking vehicles was 395 mg/ mi. The vehicles averaged 12.3 years old, and had an average of 126,000 miles. Another recent study in the South Coast Air Basin found that 1.1 to 1.8% of the light- duty vehicles emitted visible smoke, in the range of 64 to 2,3223 mg/ mi, with an average of 399 mg/ mi, over the FTP ( Durbin et al., 1999). In a related study, Durbin et al. ( 1999a) found that high- emitting but not smoking vehicles emitted 5 to 10 times as much PM as normal emitting vehicles ( 11 – 80 mg/ mi vs. 2 – 30 mg/ mi). Cadle et al. ( 1997b) conclude that “ it is clear that the current in- use, high- mileage, older vehicles can have significantly higher PM- 10 emission rates than new vehicles, and higher than the rates used in the EPA... model” ( p. 3408).
Cadle et al. ( 1998b) measured PM10 emissions from a sample of in- use light duty gasoline and diesel vehicles tested over the FTP in the Denver, Colorado area.
28
New light- duty gasoline cars and trucks ( MY 1991- 1996) emitted only 2.8 mg/ mi PM10 in the summer, but 24.9 mg/ mi in the winter. Older gasoline LDVs emitted considerably more; for example, MY 1981- 1985 vehicles emitted about 48 mg/ mi in all seasons. Smoking vehicles emitted 330 mg/ mi. Most of the PM emissions were attributed to the cold- start phase of the driving cycle. With a series of assumptions that they acknowledge “ could result in a low estimate of real- world PM emissions” ( p. 136), the authors estimate a fleet- average year- round emission rate of about 36 mg/ mi, including emissions from smoking gasoline vehicles and a few light- duty diesel vehicles. ( The most critical assumption is that smoking gasoline vehicles contribute only 0.1%.) By contrast, PART5, specified for the year 1996, an altitude of 5500 feet, I& M, and reformulated gasoline, estimates that light- duty gasoline cars and trucks emit a VMT- weighted average of 16 mg/ mi, and that the entire light- duty fleet, including light- duty diesels, emits 17 mg/ mi. Thus, PART5 underestimates a “ conservative” estimate of in- use total PM10 emissions from LDVs in Denver by at least a factor of two. If smoking gasoline vehicles contribute more than 0.1% of VMT, then the underestimation by PART5 is considerably worse.
The findings of Mulawa et al. ( 1997) are similar to those of Cadle et al. ( 1998b). Mulawa et al. ( 1997) tested 10 in- use LDGVs, model years 1977 to 1994, and found that PM emissions increased with decreasing temperature, and that virtually all of PM emissions in the FTP occurred during the cold- start phase of the test, due, they assume, to enrichment. Recent model- year vehicles ( 1987, 1989, and 1994) with low mileage emitted averaged 2.5 mg/ mi at 75o F, but 11.7 mg/ mi at 20o F. Earlier model- year vehicles with higher mileage generally emitted more PM.
Hammerle et al. ( 1992) measured PM emissions from four 1991 Ford Escorts, and four 1991 Ford Explorers, at 5,000, 20,000, 55,000, 85,000, and 105,000 miles. ( These were not “ in- use” vehicles, but rather “ test” vehicles driven almost exclusively at highway speed over their life and presumably maintained by Ford.) They found that vehicles tended to emit more PM as they aged, and more PM in cold- start tests than in hot- or warm- start tests.
Williams et al. ( 1989a, 1989b) measured PM emissions from “ in- use” gasoline and diesel vehicles in Australia. The light- duty gasoline and diesel vehicles were tested over an urban cycle equivalent to the U. S. FTP. ( The tests on HDDVs are discussed below.) Most of the vehicles were model years from the late 1970s to the mid 1980s. PM emissions from LDGVs ranged from 50 to 290 mg/ mi ( average 113 mg/ mi), and PM emissions from LDDVs ranged from 290 mg/ mi to 1,400 mg/ mi ( average of 595 mg/ mi). PM emissions from LDGVs were correlated with NMHC emissions, and PM emissions from diesel vehicles were correlated with NMHC and CO emissions. Emissions were higher in the cold- start portion of the drive cycle.
Do vehicles emit more PM in real- world driving than in the FTP? As discussed in section in 16.2.2, the FTP has three shortcomings: it does not include accelerations hard enough to induce “ command enrichment,” it underestimates the number of cold starts, and it generally is performed with the air conditioning off.
29
During a hard acceleration, the air/ fuel ratio is reduced, to increase the charge density and hence power output. With less oxygen available, less of the fuel is completely oxidized to H2O and CO2, and more is only partially oxidized or not oxidized at all, and emitted as HC, CO, and organic particulate. Similarly, during a cold start, the air/ fuel ratio is reduced, and the catalyst is cold and relatively inefficient at oxidizing HC, CO, and organic particulates. And the use of air conditioning places an additional burden on the engine that can increase the likelihood of command enrichment.
Recent evidence supports the proposition that PM emissions are higher during hard accelerations and cold start than over the entire FTP. The tests by Hammerle et al., ( 1992), Mulawa et al. ( 1997), and Cadle et al. ( 1998b), cited above, found that PM emissions increased with decreasing temperature, and that virtually all of PM emissions in the FTP occurred during the cold- start phase of the test.
The correlation between HC and PM emission ( Mulawa et al, 1997; Sagabiel et al., 1996; EPA, 1993a; Williams, 1989a, 1989b), and the evidence that HC emissions increase under enrichment ( section 16.2.2), suggest that PM emissions increase under enrichment. In direct support of this, Fanick et al. ( 1996) found that a 1994 Ford Taurus using reformulated gasoline emitted almost 4 times more PM under fuel- rich driving conditions ( such as occur during hard accelerations) than under FTP/ stoichiometric conditions. Mulawa et al. ( 1997) conclude that “ rich- operating, high- emitters can be expected to have high PM emissions” ( p. 1302).
Will PM emissions change in the future? As noted above, PART5 assumes that all catalyst- equipped LDGVs of model- year 1981 and later, and all catalyst- equipped LDGTs of model- year 1987 and later, emit 4.3 mg/ mi organic PM, everywhere, all the time. However, the studies cited above indicate clearly that relatively new, properly functioning LDGVs of about model year 1990 and later, tested over the FTP at low altitude and warm temperatures, emit on the order of 2- 3 mg/ mi total PM, and hence slightly less organic PM ( Durbin et al., 1999a; Cadle et al., 1998b; Mulawa et al., 1997; EPA, 1993c). Furthermore, if PM emissions remain correlated with HC emissions, then future decreases in HC emissions can be expected to be result in decreases in [ organic] PM emissions.
At a mininum, PART5 should have more model- year categories, perhaps corresponding to years in which the HC standards change, with progressively lower “ base” organic PM emission rates. As discussed below, it would be best if this were done as part of an overhaul of PART5 to make it function more like MOBILE6.
Light- duty gasoline vehicle summary.
The foregoing analysis indicates the following problems with PART5, and possible solutions:
• PART5 may overestimate sulfate emissions, and probably overestimates the ratio of sulfate to total PM -- especially for more recent vehicle model years. PART5
30
should estimate sulfate emissions as a function of the sulfur content of the fuel, and the age and model- year of the vehicle.
• PART5 does not include emissions of nitrate or metal PM. These should be added.
• The PART5 emission factors for organic and total PM do not account for high- emitting vehicles, or high- emitting driving or conditions. On the other hand, they do not account for reductions in PM emissions related incidentally to reductions in HC emission standards. PART5 should estimate organic PM emissions as a function of the age and model year of the vehicle ( accounting for changes in the HC standard), the ambient temperature, the drive cycle ( accounting for “ off- FTP” driving), and malfunctions and poor maintenance practices that lead to unusually high emissions.
We believe that the most significant problem with PART5 is its failure to account for high- emitting vehicles and driving conditions, and that as a result of this, PART5 underestimates real- world, in- use emissions. Cadle et al. ( 1998b) agree:
.. the failure [ of PART5] to include high emitters will result in a significant underestimation of the light- duty fleet average PM- 10 emission rate ( p. 3).
If we assume that some of the fleet are old or malfunctioning vehicles (“ super- emitters”), then the total levels of emissions are much higher than those predicted by PART5. About 10% of the fleet are super- emitters ( the results from Sagebiel et al. suggest that the fraction of super- emitters could be higher) 12, and super- emitters emit roughly five to ten times more than normal vehicles. If we start with the assumption that the “ normal” vehicles emit about 15 mg/ mi g/ mi, as assumed by PART5 for 1990 calendar years, we end up with LDGV fleet emissions being 1.4 to 1.9 times higher than predicted by PART5.
PM emissions from heavy- duty diesel vehicles.
The PART5 emission factors. As explained above ( equation M3), PART5 contains a table of total PM emission factors, in g/ bhp- hr, for HDDV vehicles. These factors, and the corresponding PM emission standards ( from Davis, 1998) for four classes of HDDVs are as follows ( g/ bhp- hr):
2B heavy
light- heavy
medium- heavy
heavy- heavy
PM standard
pre- 1987
0.52
0.52
0.69
0.64
none
1988- 1990
0.51
0.51
0.48
0.44
0.60
1991- 1993
0.29
0.29
0.27
0.27
0.25
1994 +
0.10
0.10
0.09
0.08
0.10
12 Regarding CO emissions, Ross et al. ( 1995) classify vehicles in two groups: 90% of the vehicles emit CO at about the normal FTP- measured rate, and 10% emit at a much higher rate.
31
Note that the emission rates for the years 1988 on follow the emission standards: the three model- year categories in PART5 are the same as the model- year groups for the emission standards, and the PART5 emission rates are close to the corresponding PM standards. Apparently, the PART5 emission factors for the years 1988 on are estimated on the basis of the engine- certification tests submitted by manufacturers to demonstrate compliance with the standards ( EPA, 1993c). The use of the certification data implies an assumption that heavy- duty diesel engines maintained and driven in the real world will, over their entire lives, have the same emissions as new engines tested for compliance over the heavy- duty transient cycle ( HDTC) ( Walsh, 1995). Needless to say, we will want to examine this assumption.
The emission rates for pre- 1987 vehicles apparently are based on the few available tests of in- use engines prior to 1987 ( Guensler et al., 1991). In 1983 and 1984, the EPA tested 30 in- use heavy- duty diesel engines. The engines were removed from their chassis, and tested “ as is” ( i. e., without being tuned up) over the HDTC for new engines, on an engine dynamometer. The results for eight of the engines were problematic, and discarded. The results13 for the remaining 22 engines were ( Guensler et al., 1991):
9 medium - heavy engines
13 heavy- heavy engines
0.62 - 0.89 g/ bhp- hr
0.58 - 2.14 g/ bhp- hr
After these initial tests of the 22 engines “ as received”, the EPA tuned up and re- tested 7 of the medium- heavy and 6 of the heavy- heavy engines. After this tune up, the engines emitted more NOx but less HCs ( Guensler et al., 1991). Because PM emissions generally change in the same direction as do HCs, and in the opposite direction from NOx, we can presume that the PM emissions also decreased after tune- up.
It is not clear which set of test results -- before tune up, or after tune up -- the EPA used to establish its baseline emission factor. Guensler et al. ( 1991) speculate that the official emission factors are based on the results of the tests conducted after the engines were tuned up. In support of this, we note that PART5 factors shown above ( 0.69 g/ bhp- hr for medium- heavy, and 0.64 g/ bhp- hr for heavy- heavy), and the emission factor used for all heavy engines in the 4th edition of AP- 42 ( 0.70 g/ bhp- hr) ( EPA, 1985), are at the low end of the range of results from the tests on the engines “ as received”.
Problems with the PART5 PM emission factors for HDDVs. Our analysis here considers the same issues analyzed with regards to LDGVs. First, we ask whether the tests from which the PART5 factors are derived included vehicles representative of the in- use fleet. Then, we discuss the reality of the test cycle, the HDTC. Finally, we briefly discuss emissions from future vehicles.
13It is not clear if this is TSP or PM10.
32
It seems clear that the in- use vehicles emit more PM than do the new, properly tuned vehicles that are tested for engine certification. In fact, the 1983/ 1984 EPA tests mentioned above showed that in- use vehicles tested “ as received” emitted more PM than the same vehicles tested after being tuned up. Moreover, none of the vehicles tested for engine certification, and apparently none of the vehicles tested in the 1983/ 1984 tests, were high emitters: even the highest level measured in the EPA tests, 2.14 g/ bhp- hr, is less than one would expect from a badly smoking engine. Given that the small amount of super- emitters that one typically observes in a fleet can significantly raise fleet- average emissions, the omission of super- emitting engines from the emissions tests will result in emission factors that significantly underestimate real- world emissions.
The 22 engines tested in 1983 and 1984 had accumulated from 29,000 to 410,000 miles at the time of testing ( Guensler et al., 1991). It is not clear, however, if the mileage distribution was representative of the fleet average at the time, or if the EPA accounted for the effect of mileage in establishing its baseline emission factors ( Guensler et al., 1991). In fact, in general, it is not clear if the vehicles selected were broadly representative of the in- use fleet.
Chassis dynamometer tests. Chassis dynamometer tests of heavy- duty vehicles also suggest that base emission factors in PART5 pertain to relatively new, properly functioning vehicles. The EPA has measured PM exhaust emissions from in- use heavy- duty diesel vehicles ( HDDVs) and heavy- duty gasoline vehicles ( HDGVs), driven over the transient test cycle on a chassis dynamometer ( Black et al., 1984; Dietzmann et al., 1980). The test results, and the corresponding predictions from PART5, are shown in Table 16- 3, part A. One perhaps can infer that PM emissions from the in- use HDDVs vehicles increase with increasing mileage, although so few vehicles were tested that inferences might not be reliable. At only 60,000 miles -- well below the midpoint of the life of an HDDV -- emissions already were at or above the level predicted by PART5. This suggests to us that a fleet of HDDVs, which on average has more than 100,000 miles of travel per vehicle, emits more exhaust PM than is predicted by PART5. Of the five HDGVs tested, four emitted close to the amount predicted by PART5, but three of these had new or nearly new engines. The fifth HDGV emitted several times more PM than predicted by PART5. Thus, we expect, again, that a real in- use HDV fleet, with a substantial proportion of high- mileage vehicles ( in the case of HDDVs, over 400,000 or 500,000 miles), and a few high- emitting vehicles, will emit considerably more PM than is predicted by PART5.
Williams et al. ( 1989b) tested 12 HDDVs, model years 1974- 1985, over a multi- model steady- state drive cycle on chassis dynamometer, in Australia. PM emissions ranged from 1.3 g/ mi to 11.5 g/ mi, with an average of 3.4 mg/ mi, or 2.6 g/ mi without the highest emitter. PM emissions were correlated with NMHC and CO emissions. Because the HDDVs tested were not built for the U. S. market, and were not tested over the HDTC ( although the Williams et al. [ 1989b] found that the vehicles had similar emission rates over a transient cycle), it probably is not sensible to compare the measured emissions with the predictions of PART5. Still, two conclusions can be
33
drawn: first, the fleet- average emissions are quite high, and second, the single “ super emitting” vehicle ( 11.5 g/ mi) significantly raised the fleet average emission rate, from 2.6 g/ mi to 3.4 g/ mi.
Most recently, West Virginia University ( WVU) has been testing heavy- duty diesel and alternative- fuel vehicles on a portable chassis dynamometer. The vehicles are tested on- site, over a variety of test cycles, including the Truck Central Business District Cycle, a 5- mile truck route, and WVUs own truck cycle. All of the vehicles are in the heavy- heavy class ( the average gross vehicle weight is over 60,000 lbs). There is a relatively wide range of makes and ages. Results from 1993 and early tests are published in Wang et al. ( 1993); results from later tests are available on the web ( see Table 16- 3, part B). Nearly 100 PM emission results are available.
Table 16- 3, part B, summarizes the results of the WVU tests, and compares the in- use emissions with the pertinent PART5 emission factor. We see that PART5 slightly overestimates emissions for model years 1988- 1990, slightly underestimates emissions for model years 1991- 1993, and significantly underestimates emissions from model years 1994 and later. Assuming that WVU did not test any super- emitters -- the highest emission rate in all the tests was only 2.74 g/ mi, well below what a badly smoking vehicle emits -- we can infer that PART5 significantly underestimates in- use emissions from a fleet with small percentage of high- emitting vehicles.
Finally, Yanowitz et al. ( 2000) provide a comprehensive summary of emissions tests of heavy- duty diesel vehicles, including chassis dynamometer studies, tunnel studies, and remote- sensing studies. Their review of chassis dyno studies includes all of the studies reviewed here, plus several not reviewed here. Yanowitz et al. ( 2000) show PM emissions in g/ gal by model year; these are on the order of 5- 6 g/ gal for the 1988- 1993 fleet, and 2 g/ gal for the 1994- on fleet. The average fuel economy of the tested vehicles was 4 mpg, so their results are roughly 1- 1.5 g/ mi for the 1988- 1993 fleet, and 0.5 g/ mi for the 1994+ fleet. These results are similar to the those shown in Table 16- 3 B, and hence offer further evidence that PART5 underestimates emissions from model years 1991 and later.
Measurements of on- road emissions. We have found four studies of on- road emissions from HDDVs. In 1983, Pierson and Brachaczek measured the ambient airborne PM at the exit of the Allegheny and Tuscarora Mountain Tunnels on the Pennsylvania Turnpike, and with these and other data, back- calculated the HDDV emission rate14. More recently, Whittorf et al. ( 1994) and Gertler et al. ( 1995) reported the results of a similar experiment at the Fort McHenry Tunnel in Baltimore, Maryland. Balogh et al. ( 1993) measured the PM concentration along a university road that had heavy bus traffic, and back- calculated the bus emission rate. Finally, Miguel et al. ( 1998) measured emissions of particulate PAH and solid carbon ( carbon black) from gasoline
14Pierson and Brachaczek ( 1983) summarize the method: “ Known traffic and air fluxes are combined with net ( tunnel minus intake) tunnel- air pollutant concentrations to derive mg/ km emission rates of the various species observed. Correlation against the changing traffic composition gives emission- rate estimates resolved as to vehicle type” ( p. 2).
34
and diesel vehicles in the Caldecott Tunnel in the San Francisco Bay Area in 1996. ( Yanowitz et al. [ 2002] also tabulate the Fort McHenry, Tuscarora, and Caldecott studies, plus a study in Vancouver Canada and a study in Zurich Switzerland.)
In Table 16- 4, we compare the results of these studies with the estimates of the PART5 model specified for the same conditions. In all cases except two ( gasoline vehicls in Pierson and Br., and diesel vehicles in Whittorf et al.) PART5 underestimates the “ adjusted” on- road PM exhaust emission rate. ( Details of the adjustments are given in the notes to Table 16- 4.) Now, because the majority of emissions from super- emitters occur during transient driving, not during the high- speed cruising of the on- road tests, our adjustments of the reported on- road cruising emissions to levels that would have occured in an on- road transient test do not include any “ excess” emissions from super- emitting vehicles in the transient cycle. We believe that in the real world, with high- emitting vehicles in transient driving, the fleet average emission rate is even higher than indicated by the “ adjusted” results of Table 16- 4.
The ratio of exhaust PM to road- dust PM in the emissions inventory versus the same ratio measured at ambient air- quality monitors. As discussed below, the ratio of emissions of road dust to exhaust emissions from highway vehicles, in the EPA’s ( 1995d) emissions inventory, is many times higher than the ratio of dust to motor- vehicle exhaust at ambient air- quality monitors. If the ambient ratios are accurate, and if the differences between the ambient ratios and the emissions ratios cannot be explained entirely by differences in emissions dispersion ( which, it seems, they cannot), then the AP- 42- based estimates of road- dust emissions are too high, or the PART5- based estimates of highway- vehicle PM emissions are too low, or, most likely, both.
PART5 versus EMFAC7F. One basis, albeit still a weak one, for quantifying the degree to which PART5 underestimates exhaust emissions from HDDVs is a comparison of the PM emission factors from PART5 with the PM emission factors from California’s emission- factor model, EMFAC 7F. We ran PART5 and EMFAC7F for the year 1990, and got the results shown in Table 16- 7. The EMFAC7F estimates of exhaust PM from HDDVs are about 1.8 times as high as the PART5 estimates. Although the EMFAC7F tirewear estimates are at least an order of magnitude higher than the PART5 estimates, this does not qualitatively affect the results since tirewear is a small fraction of emissions. 15
15In a study of ambient particulate matter associated with motor- vehicles in an expressway tunnel, Pierson and Brachaczek ( 1983) estimated that tires contributed only 1% of the total motor- vehicle PM emission rate of about 0.30 g/ mi. Rogge et al. ( 1993) estimated that tire wear particles constituted at most 1.6% of total PM2.0 road dust. In a CMB analysis of sources of particulate matter at four sites in Los Angeles, Schauer et al. ( 1996) estimated that tire wear debris was less than 10% of PM2.0 road dust, less than 5% of PM2.0 vehicle exhaust, and less than 3% of total PM2.0 road dust and vehicle exhaust.
In any event, we suspect that neither PART5 nor EMFAC7F is correct about tirewear: PART5 assumes that tirewear emissions are proportional simply to the number of wheels, so that a bus is predicted to have the same emissions as does a car, and only twice the emissions of a motorcycle. It is inconceivable that a bus emits only twice as much tire PM as does a motorcycle. EMFAC7F is more
35
Why are CARB’s EMFAC7F estimates higher than the EPA’s PART5 estimates? According to Guensler et al. ( 1991), CARB had used the EPA’s estimates until 1988, when CARB modified the EPA emissions factors to reflect inspection and maintenance practices in California. CARB developed its new estimates for EMFAC7F on the basis of a report by Radian Corporation, which reviewed the original data used to establish the EPA ( PART5) factors, plus additional information. The Radian report apparently estimated a factor to adjust the EPA’s estimates upwards to account for high emissions from poorly maintained vehicles ( Guensler et al, 1991). This adjustment factor might partially explain why the EMFAC7F estimates are so much higher than the PART5 estimates.
The drive cycle. Guensler et al. ( 1991) note that the trucks in the real world may idle more than is assumed in the HDTC, and that the emissions inventory apparently does not account for emissions from truck engines being run to provide auxiliary power for refrigeration and other purposes. If this is so, then the PART5 emission factors, which are based on HDTC tests, underestimate real- world emissions.
On the other hand, the EPA ( 1993a) cites a 1988 study by the University of Michigan that found that class VIIIB ( heavy- heavy) trucks accumulated 73% of their mileage on freeways when in large urban areas -- much more than the 25% assumed in the HDTC. To the extent that PM emissions arise more from transients than from steady- state operation, and that freeway driving involves less transients, the underestimation of freeway driving will overestimate real- world emissions. However, it is not clear to what extent the freeway driving estimated by the University of Michigan is steady state. In many large urban areas, freeways are congested for many hours a day, and cause trucks to spend a lot of time idling and stopping and starting. These are conditions that increase g/ bhp- hr emissions. Hence, it is not immediately clear to what extent, if any, the possible underestimation of freeway driving results in an overestimate of PM emissions.
Heavy- duty diesel vehicle summary
In summary, the HDDV PM emission factors in PART5 probably underestimate real- world emissions, most likely because the test database from which the PART5 factors were derived does not include a representative number of old, malfunctioning, poorly tuned, or inherently high emitting vehicles. In addition, the HTDC might not be representative of real driving conditions in the country; for example, there might be a lot more idling and hard accelerating in the real world than is present in the HDTC.
realistic in this respect, in that it estimates the same tirewear emissions for buses as for HDDVs. However, for two reasons, the EMFAC estimates appear to us to be too high all the way around.
First, back- of- the envelope calculations of the total amount of tire material worn away from tires suggests that the wear rate per mile is much less than is estimated by EMFAC7F. Second, the EMFAC7F estimates of tirewear TSP ( Table 16- 7) are a much higher percentage of tailpipe and road dust TSP emissions than seems reasonable on the basis of the studies cited in the first paragraph to this note.
36
Our conclusion
The data reviewed above suggest that PART5 underestimates emissions from real on- road vehicles, primarily because PART5 seems to be based on low- mileage, properly functioning vehicle, and takes little, if any, account of super- emitters. In our low- cost case, we assume that the PART5 model underestimates PM emissions by a factor of only 1.5. In our high- cost case, we assume that PART5 underestimates emissions by a factor of 2.0.
16.2.4 Estimates of PM dust from paved roads ( AP- 42 Volume 1, and PART5 model)
Motor vehicle traffic kicks up the dust on the road16. Some of this “ emitted” road dust is small enough to be suspended in the air as particulate matter. Surprisingly, such “ re- entrained road dust,” as it is called, is by far the largest source of particulate matter in the official U. S. emissions inventory that we used in our analysis: in 1990, road dust from paved and unpaved roads accounted for nearly half of all PM10 emissions in the U. S. emissions inventory ( EPA, 1995d)
Because road dust apparently is such a large source of emissions, it is important to determine if the emission factors used to calculate road- dust emissions are accurate. In this section, we present evidence that the current EPA ( 1995a) AP- 42 emission factors, used in the PART5 model, substantially overestimate emissions of PM10 and especially PM2.5 from paved roads. ( We briefly discuss emissions from unpaved roads in the following section)
16Rogge et al. ( 1993) describe the processes well:
Urban street surfaces act as repositories fo rparticulate matter... particulate automobile exhaust, lubricating oil residues, tire wear particles, weathered street surface particles, and brake lining wear particles ar direct contributors to the paved road dust. Biogenic materialsuch as leaf detritus ( e. g., from street trees, shrubs, lawns)... and garden soil organicsalso contribute to the street dust. Indirectly, via atmospheric transport and fallout, practically any anthropogenic or biogenic source can add to the dust accumulation on road surfaces.. Roads and streets also can function as a source of airborne particulate matter and likewise as a source for toxic compounds washed into drainage sytems or delivered to aquifiers. Resuspended by wind and vehicle- induced turbulences, road dust particles are injected into the atmosphere. In fact, resuspension, fallout, street sweeping, rain, and generation of new particles ( e. g, vehicle exhuast) drive a dynamic source and sink relationship which can contribute appreciable amounts of particulate matter and toxic substances to the atmosphere and hydrosphere ( p. 1900).
37
The paved road- dust equations
In the official U. S. emissions inventory, emissions of road dust from paved roads ( RDP) are calculated with the following formulas17:
TP = RDP + Ta + Ti + B ( D1)
= k ( sL/ 2) 0.65 ( W/ 3) 1.5 ( D2)
RDP = TP - Ta - Ti - B ( D3)
where:
TP = total PM emissions due to motor vehicles on paved roads: tailpipe PM + road- dust PM + tire- wear PM + brake- wear PM.
RDP = emissions of road- dust particulate matter from paved roads ( g/ mi)
Ta = tailpipe emissions of PM ( grams/ mile; calculated from the PART5 model, discussed above)
Ti = tire- wear emissions of PM ( grams/ mile; given in grams/ mile for various vehicle classes, in the PART5 model)
B = brake- wear emissions of PM ( grams/ mile; assumed to be zero in the application of equations D1 and D2
k = multiplier to obtain different PM size classes ( EPA, 1995a; e. g., to get emissions of TSP, k = 38; to get emissions of PM10, k = 7.3; to get emissions of PM2.5, k = 3.3)
sL = the silt loading on the surface of the road ( grams/ meter2) ( based on an equation that relates silt loading to average daily traffic ( ADT) volume [ EPA, 1997a] 18)
W = the average weight of vehicles on the roadway ( tons).
17Xueli et al. ( 1993) estimate the following formula for emissions of road dust on a road in Shanghai:
E = 0.000501 V0.823 U0.139( T/ 4)
where E is kg/ km/ vehicle, V is vehicle speed in m/ s, U is windspeed in m/ s, and T is the vehicle load in tonnes.
18Beginning with the 1996 inventory, the EPA changed the method used to estimate sL for the years after 1990. Instead of estimating sL as a continuous function of ADT, the EPA estimated sL for ADT categories: 1 g/ m2 for local roads, 0.2 g/ m2 for non- local roads with ADT < 5000, and 0.04 g/ m2 for other roads ( EPA, 1997a, 1998). These values resulted in lower total sL estimates than did the the sL vs. ADT function. More recently the EPA ( 2003) has recommended even lower default values:
ADT < 500
ADT = 500 to 5,000
ADT = 5,000 to 10,000
ADT > 10,000
freeways
0.6
0.2
0.06
0.03
0.015
38
Equation D2 is presented in AP- 42 ( EPA, 1995a), and equation D3 ( without the brakewear term B) is given in the PART5 model. In the estimation of the national emissions inventory, the emission factors obtained from PART5 ( equation D3) are multiplied by the fraction of days in a month with less than 0.01 inches of precipitation, on the assumption that more than 0.01 inches of precipitation in a day is sufficient to keep the dust on the road ( EPA, 1997a, 1997b) 19.
It is important to note that dust emissions from paved roads are calculated by subtracting tailpipe ( Ta), tirewear ( Ti), and brakewear ( B) emissions from empirically estimated total emissions from motor- vehicle traffic on paved roads ( TP). Contrary to the implication in the emission- factor handbook, AP- 42 ( EPA, 1995a), equation D2 does not predict road- dust emissions per se; rather, it predicts total motor- vehicle- related emissions. ( AP- 42 is misleading because it presents equation D2 but not equation D3, and states that equation D2 predicts “ dust emissions from vehicle traffic on a paved road