Particulate Matter from Gasoline Light Duty Vehicles Based on KansasLight Duty Vehicles Based on Kansas

City and Other Studies

Edward Nam

United States Environmental Protection AgencyUnited States Environmental Protection Agency Office of Transportation and Air Quality

Mobile Source Technical Review Subcommittee Meeting May 8 2008May 8, 2008

AcknowledgmentsAcknowledgments• Study Sponsors ($4 million)

U S EPA– U.S. EPA– U.S. Department of Energy

• via National Renewable Energy Laboratory– Coordinating Research Councilg– NACAA (Emission Inventory Improvement Program)– U.S. Department of Transportation

• U.S. EPA Carl Fulper Carl Scarbro Richard Baldauf James Warila Harvey– Carl Fulper, Carl Scarbro, Richard Baldauf, James Warila, Harvey Michaels, John Koupal, Richard Rykowski, Joseph Somers, Chad Bailey, Tony Fernandez, David Brzezinski, Gwo Shyu, Joe McDonald, Gene Tierney, Bruce Cantrell, Kitty Walsh, Robin Moran

• U C Riverside George Scora• U.C. Riverside - George Scora• University of Michigan – Erika Roesler• Eastern Research Group (ERG)

– Subcontractors: BKI, DRI, NuStats, Sensors, ESP

2

Subcontractors: BKI, DRI, NuStats, Sensors, ESP• DRI: Eric Fujita, Dave Campbell, Pat Arnot

IntroductionIntroduction• Why do we need to study PM from gasoline sources?

Q ti b t th d i f li di l i i t– Questions about the dominance of gasoline vs diesel in inventory– Update outdated emissions models for regional State Implementation

Plans and conformity analyses made timely by new PM NAAQS stds– Develop baseline PM inventories to apply fuel effects (eg. Ethanol).

• Kansas City Results– Study Design– Preliminary Results

KC PM rates are largely representative– KC PM rates are largely representative– High Emissions Analysis– How does temperature affect PM emissions?– How can we model deterioration in cars? (model year vs age: technology

d t i ti )vs deterioration)• Two models introduced: additive and multiplicative

– Preliminary inventory comparison for 2002• What requires more work

3

q– Longitudinal study of vehicles to quantify deterioration

BackgroundBackground

• Does Gas or Diesel dominate the PMDoes Gas or Diesel dominate the PM inventory? – Many past source apportionment studies have y p pp

had conflicting results on whether diesel or gasoline PM dominateE i i i t i ll i di t di l– Emission inventories generally indicate diesel dominatesThis study only addresses the gasoline– This study only addresses the gasoline portion (and not the diesel, which also requires an update)

4

2009 National PM2.5 BreakdownThese are estimations based on MOBILE modeling (bottom up)

onroad gas sources contribute ~1% of PM from

2009 Mobile Source PM25

Other Nonroad Non-Diesel

2009 Total PM25 (mobile+stationary) Mobile Source PMcontribute 1% of PM from all sources and ~10% of mobile sources

Non Diesel37%Highway Gasoline

1%Highway Diesel

2%

f d ? Highway Ga10%

Other Nonroad

Manmade Stationary

86%

C1/C2 Commercial

Marine

Locomotives0.3%

confounders?

Highway Die14%

Locomotives0 6%C1/C2 Commercial

Other Nonroad Diesel27%Other Nonroad

Diesel4%

0.2%

5

0.6%C1/C2 Commercial Marine0.4%

Other Nonroad Non-Diesel

5% These fractions are being updated

NFRAQS DRI StudyNFRAQS DRI Study• Study done in 1996-1997 Welby Site (Denver)y

in Denver• Study focused on

wintertime ambient and

Fine PM apportionment results

wintertime ambient and emission data

• Vehicle emission testing performed locallyperformed locally

• Extensive ambient PM sampling and speciation

• Study concludes that Gas PM dominates diesel PM 3:1

6Can’t compare pie to previous chart due to denominator discrepancy

Questions continue• Source Apportionment Studies “Top-Down”

– Gas >> Diesel• NFRAQS, LADCO (OC), SCAB Split (Schauer), PAQS (using Schauer

diesel profile)– Gas ~= Diesel

• SCAB Split (Fujita), AtlantaSC Sp ( uj a), a a– Diesel >> Gas

• SCAB (’82), SEARCH, PAQS (using NFRAQS diesel profile)– Major uncertainties of source apportionment studies

• Secondary aerosol formation accounts for large portion of ambient PM• Dependent on uncertain (and scarce) profiles

– Emissions profiles from small nonroad engines (including 2-stroke cycle engines) are unknown and potentially significant

• Emissions profiles from other man-made sources are unknown• EPA (OTAQ) will report on this issue soon

– Using Kansas City + recent Diesel testing data using a “bottom-up” approach

7

approach– KC also has new source profiles to add to the SA databases

KC PM Results CaveatKC PM Results Caveat• We are presenting PM data gathered:

– At set point in time• These results should not be confused with

inventory impacts in the future• The definition of what exactly PM is and how

d PM l t t bi t PM i tillmeasured PM relates to ambient PM is still being studied

sampling temp dilution location measurement media– sampling temp, dilution, location, measurement media• Contract report and Results report are now on

EPA/OTAQ website (emission factor research)8

EPA/OTAQ website (emission factor research)

Major Objectives of KC ProgramMajor Objectives of KC Program• Address the need for representative gasoline PM from

the current fleetthe current fleet– Improve basis for modeling PM

• MOBILE6 derived from PART5, which is based on testing conducted in 1970’s & 1980’s

• No effect of temperature, deterioration or speed– First large random sample methodology used in recruiting

gasoline LDVs/LDTsQuantify distribution of PM emissions in the fleet– Quantify distribution of PM emissions in the fleet

– Quantify effect of ambient temperature on PM emissions– Develop PM and Toxics speciation profiles

• Other GoalsOther Goals– Evaluate alternative PM measurement techniques

• E.g. Collect the first real-time gas PM data on a fleet– Evaluate performance of PEMS

9

p– Real world data on emissions, activity & fuel economy

KC study backgroundKC study background• Address need for representative gasoline PM emission

estimates from current fleetestimates from current fleet• Evaluate alternative measurement techniques• Study conducted in Kansas City 2004-05• 496 gasoline light-duty cars and trucks tested

– Model Years 1968-2005• Summer and winter testingg

– Vehicles tested at ambient temperature – About ½ of the vehicles tested each season– 43 vehicles tested in both winter and summer

• Representative sampling a critical objective• We believe that the vehicles represent an accurate

cross-section of KC demographics and that sufficient hi h itt t d

10

high-emitters were captured

KC Test Facility -I t t tiInstrumentation

Twin-Roll Dyno

Emission & Dyno Control Trailer

PM & Unregulated Sampling Systems

PEMS UnitCVS Flow

11Vehicles driven on LA92 drive cycle under ambient temperatures

Preliminary Results• Kansas City study suggests that PM emissions

are:are:– largely consistent with past gas PM measurement

programs– correlated with HC more than any other pollutant, but

correlation is far from perfect– Higher from trucks than carsg– Higher from older vehicles than newer vehicles – Coming from a small number of high emissions

vehicles but many were not identified as “smokers”vehicles, but many were not identified as smokers– can be highly variable from test to test– Higher (exponentially) in winter than summer

12

g ( p y)– We have found strong evidence that some

vehicles have high PM levels due to oil burning

Results by VehicleResults by Vehicle400

450 CarsTruckPM10 Std

300

350

g/m

i)

200

250

Em

issi

ons

(mg

100

150PM

E

0

50

1965 1970 1975 1980 1985 1990 1995 2000 2005 2010Model Year

13

Model Year

Each point is a vehicle result on an ambient LA92 testSo cannot compare directly to emissions standard

Results by Strata and Season

100

1000 Summer Winter

10

100

g/m

ile

1

PM2.

5, m

0 01

0.1

P

0.011 2 3 4 5 6 7 8

Trucks Stratum CarsOlder Newer Older Newer

14Emissions are higher in the winterEmissions are higher for older vehicles

High Emissions AnalysisHigh Emissions Analysis3020

nt

20

10

Log Histogramsco

un

10

0

Std. Dev = 1.27

Mean = 2.30

N = 261.000

Std. Dev = 1.39

Mean = 1.55

N = 202.00

LNPMCOMP

5.75

5.25

4.75

4.25

3.75

3.25

2.75

2.25

1.75

1.25

.75.25-.250 N 261.00

LNPMCOMP

5.75

5.25

4.75

4.25

3.75

3.25

2.75

2.25

1.75

1.25

.75.25-.25

-.75

-1.25

summer winter

Ln(PM) Ln(PM)

• Emissions distribution skewness of sampled fleet confirms presence of high emitting vehicles

• Emissions are roughly log-normally distributed

summer winter

15

• 50% of emissions coming from 13% of vehicles• Look at sec by sec data to get an idea of what is causing high emissions

Typical results: PM spike at ~840 sec

16

With a relatively prominent cold start

e.g. High Emitter due to hi cold start

HC ill i ldHC still in cold start

17Cold Start bleeding into bag 2 Not a very cold temperature test, Not common

Broad PM humps (likely OC, or oil)

PM hPM humpNo EC hump

18Not cold start – probably Organic PM from oil or stored in measurement systemOccurs appx on 40 tests with avg PM ~ 50mg/mi (skewed)

Most likely an oil burner

19Mostly OC, intermittent, not necessarily load dependentnot necessarily reflected in HC

Old Hi Emitter (poor fuel control)

20limited fuel control & may be burning oil

Test-to-test variability• Out of 24 repeat test vehicles (consecutive days)

– There was considerable variability – 1 moderate flipper (flipped from moderate to lo)pp ( pp )– 1 severe flippers (flipped from hi to lo-emitter)…

1994 Chevy APV / 38F0.014 70

first test, 189mg/misecond test 5mg/mi

/ PFI / 124,000 miles

0.01

0.012

50

60second test, 5mg/mifspeed

0.006

0.008

Dus

trak

(g/s

)

20

30

40

spee

d (m

ph)

0.002

0.004

0

10

20

210

0 200 400 600 800 1000 1200 1400 1600

time

-10

TOR sees 6 fold increase in OC, but not flipperNo significant difference in H/C/N

Hi-emitter variability• Variability is more common for PM compared to H/C/N

1987 F-150 / 46F

0.16 70first test 232mg/mi

PFI

0.12

0.14

50

60

first test, 232mg/misecond test, 99mg/mifspeed

0.08

0.1

PM

(mg/

s)

30

40

d (m

ph)

0.04

0.06Dus

trak

10

20

spee

d

0

0.02

-10

0

22

0 200 400 600 800 1000 1200 1400 1600time

QCM doesn’t see it, but TOR does – even instruments are variable

Assessment of Temperat re EffectsTemperature Effects

• Vehicle tested at ambient temperaturesVehicle tested at ambient temperatures• Winter Temperatures:

– avg ~ 45ºF, min ~ 20ºFavg 45 F, min 20 F

• Summer Temperatures:– avg ~76ºF max ~100ºFavg 76 F, max 100 F

• 43 vehicles tested summer and winter (33 remain after QA)remain after QA)

• Correlation vehicle tested 24 times over a range of temperatures

23

p

Note: Based on 43 vehicles Results from vehicles tested in repeated in both Rounds 1 & 2Results from vehicles tested in both summer and winter show higher emissions in winter for virtually every vehicle 10 times

higher in winter

24

Variability & temperature effect f 33 h d ifrom 33 matched pairs

1000

100

10

PM

(mg/

mi)

110 20 30 40 50 60 70 80 90 100

• Test to test variability should average out with sufficient

0.1Temperature

25

• Test to test variability should average out with sufficient matched pair testing

• What remains is temperature effect

Comparison of Temperature Effects to Recent StudiesRecent Studies

6

y = -0.0334x + 3.10

y = -0.0385x + 3.7822R2 = 0.8935 2

4

y = -0.0415x + 3.6049-2

0-40 -20 0 20 40 60 80 100 120

lnPM

yR2 = 0.6632

y = 0 0353x 4 2332-6

-4Correlation vehicleKC SW Matched pairsMSAT Test vehicles y = -0.0353x - 4.2332

R2 = 0.7284

-10

-8

T t

ORD Test vehiclesLinear (Correlation vehicle)Linear (MSAT Test vehicles)Linear (KC SW Matched pairs)Linear (ORD Test vehicles)

26

TemperatureLinear (ORD Test vehicles)

Kansas city PM emissions doubles per 20°F drop- This is fairly consistent with other studies

Temperature Effects• PM increases exponentially as temperature decreases

160

140

160

KC SW Matched pairsfit to KC w/ correction

100

120

/mi)

ORD Test vehicles

60

80

PM

(mg/

20

40

270

-20 0 20 40 60 80 100Temperature

PM Rates at a Snapshot in Time

120

Temperature adjusted results- Data adjusted to 75F

100

60

80

PM

(mg/

mi)

KC measured

40

P

0

20

1975 1980 1985 1990 1995 2000

28

1975 1980 1985 1990 1995 2000

Model Year

2002 KC Rates were put into MOVES & d i h MOBILE/NMIM& compared with MOBILE/NMIM

• This is a preliminary simple model (no modal• This is a preliminary simple model (no modal rates, speed effects, deterioration rates, etc) and does not predict actual emission inventory effectsM thl f h t t ( ti l l)• Monthly runs for each state (aggregation level)

• Nationally, MOVES is about 1.6x higher than MOBILE/NMIMMOBILE/NMIM– But these will vary significantly by region

National 2002 LDG PM2.5 Emission Factors

0.040.0450.05

* This is a preliminarycomparison

0.010.0150.02

0.0250.03

0.035

g/m

i NMIMMOVES

290

0.0050.01

1 2 3 4 5 6 7 8 9 10 11 12

Month

Regionally, there can be large differencesbe large differences especially in winter

(examples)

Vermont 2002 LDG PM2.5 Emission Factors

0.040.0450.05

(examples)• Vermont (2.61)

0.010.0150.02

0.0250.03

0.035

g/m

i NMIMMOVES

00.0050.01

1 2 3 4 5 6 7 8 9 10 11 12

Month

• Florida (0 76)

Florida 2002 LDG PM2.5 Emission Factors

0.040.0450.05

Florida (0.76)

0 010.0150.02

0.0250.03

0.035

g/m

i NMIMMv2002t

300

0.0050.01

1 2 3 4 5 6 7 8 9 10 11 12

MonthBut what will these do in the future?

Deterioration ModelDeterioration Model• Only way to definitely quantify deterioration is to

conduct a large longitudinal study (none exist)conduct a large longitudinal study (none exist)• Another option is to establish new car rates

(ZML or Zero Mile Levels) based on past studies(ZML or Zero Mile Levels) based on past studies• Then use KC as a guide for how PM may

deteriorate to present levelsdeteriorate to present levels• Two potential models were explored

– AdditiveAdditive– Multiplicative– Both are consistent with the data, but predict different

31

pinventories into the future

Deterioration ModelDeterioration Model• We have chosen to use the multiplicative model p

for MOVES LD Gas PM for the following reasons:– It is more consistent with trends in HC

F l t l d t i ti• Fuel control deterioration• Aftertreatment deterioration• Longer useful life standards (120K vs 50K)

– Oil consumption and burning may be less of a problem with newer technologies than older

• This implies that PM inventories from light dutyThis implies that PM inventories from light duty gas are likely to decrease in the future compared to the new baseline

32

ZML: 15 Test Programs from 1979-2006 tested new vehicles2006 tested new vehicles

35

40 allGibbs et al., 1979Cadle et al., 1979Urban&Garbe, 1980Lang et al., 1981VW, 1991CARB, 1982

25

30

)

Ford, 1992E24-1 DenverE24-2 UC RiversideE24-3 San AntonioChrysler/Ford/GMSwRi/NRELKC-Summer LA92MSAT-Tier2mean for each programFit

15

20

25

PM

10 (m

g/m

i) Fit

10

15P

PM = 15.8exp(-.0886 MY)

0

5

0 5 10 15 20 25 30

33

Model Year (+1975)

PM emissions from new cars have dropped by a factor of 10 in 30 years.This is likely from: improved fuel control, & aftertreatment.

PM LDV ZML+ Deterioration

100

197519801985

80

1990199520002005KC measured in 2004KC d l i 2004

KC data is plotted for comparison

60

M (m

g/m

i)

KC model in 2004

19851980

40

P

1975

0

20

2005 20001995

1990

MY Future rates same as 2002

34

0 5 10 15 20 25 30Age

Future rates: there is no evidence in data that ZMLs continue to drop in Tier2+

Model compared to data in 2004Model compared to data in 200490

100

KC measuredmodel in 2004ZML

70

80

ZMLKC 5 yr measured avg

50

60

PM (m

g/m

i)

Aging model in Cal yr 2004

20

30

40Aging model in Cal yr 2004

0

10

20

1975 1980 1985 1990 1995 2000

ZML

35

1975 1980 1985 1990 1995 2000

Model Year

Conclusions• Direct tailpipe PM emissions are are largely affected by:

– ambient conditions (temperature) – vehicle type: model year/age of vehicle & car vs truck

• Modal analysis will help us understand PM formation, deterioration, and measurement– PM spikes in rich and cold start eventsPM spikes in rich and cold start events– PM seems to increase exponentially with load – Organic PM humps emitted after storage, or possibly oil burning on

some high emittersg– Variability is a natural aspect of any PM testing program

• Comparing new vehicle rates with aged vehicle rates from KC we present two deterioration models:KC, we present two deterioration models:

• Rates from KC data (without deterioration) were run with (a preliminary version of) MOVES and compared to MOBILE/NMIM Direct tailpipe PM emissions in MOVES

36

MOBILE/NMIM. Direct tailpipe PM emissions in MOVES are about 1.6 times larger nationwide, but colder regions and climates have larger differences

Workgroup CommentsWorkgroup Comments

• Awaiting comments on 2nd PM presentation onAwaiting comments on 2 PM presentation on deterioration and real-time PM

• First round of comments:– Most comments were detail oriented– There were two comments on the “conservative”

assumptions made in the additive deterioration model.

• Upon further analysis and research the EPA is now seriouslyUpon further analysis and research, the EPA is now seriously considering the multiplicative approach, which assumes lower deterioration rates for late model year vehicles than in the past.

37

p

Next steps & Issues for MSTRS• Next Steps for MOVES

– Determine modal rates – Finalize emission rates in MOVES for LD as well as

heavy-duty– Compare the relative contribution of inventory from p y

gas vs diesel (bottom up)– Quantify the fuel effects of ethanol on PM– PM SpeciationPM Speciation

• Issues for MSTRS– Given time constraints of modeling needs

I l d d i i d l– Issues related to deterioration model– Issues with any of the analysis approaches or

measurements

38

– Recommendations for further research? Partners?• High emitters, deterioration, measurement techniques, on-

road emissions, oil burning, etc.

Appendixpp

39

What causes higher gasoline PM?What causes higher gasoline PM?

• Over-fueling • Fuel Propertiesg– Cold start– High load (WOT)

p– T# performance– Aromatics

– Sensor failures– Fuel system failures

• Component wear

– Sulfur• Lubricating Oil

PCV: Positive• Component wear– Leaky injectors– Valve seal

– PCV: Positive Crankcase Ventilation

– Direct Leak into li d– Piston rings…

• Other malmaintenance

cylinder– Oil Composition– New oil change

40

New oil change

Option 1: Additive Deterioration model• The difference between the temperature adjusted

Kansas City Results and the ZML (by age) is the remaining Age Effect in 2004remaining Age Effect in 2004

80

90

carstruckscar age modelt k d l

60

70

mg/

mi)

truck age model

30

40

50

Adju

sted

PM

(m

10

20

41

00 5 10 15 20 25 30 35

Age

Steps originate from the MOVES binning structure

Additive model combined with ZMLAdditive model combined with ZML• ZML trends account for ~20% of Emission Rate,

D t i ti t f th t• Deterioration accounts for the rest90

100carstruckscar new modeltruck new model

60

70

80

mi)

40

50

60

Adj

uste

d PM

(mg/

m

20

30

420

10

0 5 10 15 20 25 30

Age

Additive Age Effect CommentsAdditive Age Effect Comments• Comparing ZML with Age:

– There is very little deterioration for young vehicles (<9 yrs old) y y g ( y )• Additive model implies that 20 years from now today’s

vehicles will have their emissions increase from about 1 5 mg/mi to about 50 mg/mi1.5 mg/mi to about 50 mg/mi

• Advantages:– More environmentally conservative estimate of future projections

D ib h PM f il ti d t i ti b– Describes how PM from oil consumption deterioration may be the same now as it was in past (i.e. oil consumption for 10 yr old MY 2000 vehicle is same as 10 yr old MY 1980 vehicle)

– Describes how tightly controlled newer vehicles have much more– Describes how tightly controlled newer vehicles have much more room for deterioration than older vehicles

– Less sensitive to very low ZMLs• Alternative model - multiplicative

43

• Alternative model - multiplicative

Option 2: Multiplicative D i i M d lDeterioration Model

• A Multiplicative Deterioration model is easierA Multiplicative Deterioration model is easier analyzed in log-scale since the model is “additive in log-transformed space”

• We also rely on hydrocarbon trends to guide us (which seem largely multiplicative in deterioration)

• Multiplicative model assumes that newer h l i ill l l h h ldtechnologies will last longer than the older ones

44

HC Deterioration from AZ 2004HC Deterioration from AZ, 2004• Random tests (warmed up) at(warmed up) at nominal temperatures• Nearly 2000 yLDGV tests• Emissions average rates• Parallel lines in• Parallel lines in log space• ZMLs (Zero mile levels) offset from each other• DFs level off in 10-14 age bin

Variability due to Fewer tests for newer vehicles

45

10 14 age bin

PM log slopes from KC• We expect log-transformed PM rates to follow similar

“railroad tracks” (as HC) • Level off at around 20 years (instead of 12)• We assume future vehicles follow same trend

3.5

4

4.5

2

2.5

3

ln (P

M)

Model year

1

1.5

2

1975198019851990199520002005

ygroups

460

0.5

0 5 10 15 20 25

Age

KC model in 2004

Multiplicative deterioration model commentsMultiplicative deterioration model comments• Multiplicative model implies that 20 years from

now today’s vehicles will have their emissionsnow today s vehicles will have their emissions increase from about 2 mg/mi to about 10 mg/mi

• Advantages:g– Fuel control deterioration is likely a multiplicative

Deterioration phenomenon– Statistically, H/C/N deterioration tends to be y,

consistent with multiplicative approach – DFs should maintain HC/PM ratio over time (as

observed in data))• Disadvantage: Much more sensitive to low (and

uncertain) ZMLs• Multiplicative model tentatively planned for use

47

• Multiplicative model tentatively planned for use in MOVES