Fang Yan a, Ekbordin Winijkul a,aDepartment of Civil and Environmental Engineering, UbDivision of Decision of Information Sciences, Argonne

a r t i c l e i n f o

Article history:Received 18 October 2010Received in revised form15 April 2011Accepted 7 June 2011

Keywords:EmissionProjection

1. Background and rationale

Projections of future emissions have many applications.Frequently, they are used to determine the types of management oremission control measures that must be implemented in order to

maintain air quality at an acceptable level in an urban environmentor a region. However, emissions of air pollutants affect atmospherichealth beyond their airsheds; two of the most notable impacts areclimate change (Chung et al., 2002; Menon et al., 2002; Schulz et al.,2006; Streets et al., 2009) and hemispheric or intercontinentaltransport (Jaffe et al., 1999; Fiore et al., 2002; Cooper et al., 2005).Global emission projections have recently been identied as criticalelements in understanding these large-scale impacts (Levy et al.,2008). Such projections are required to understand the climate

* Corresponding author. Tel.: 1 217 244 5277; fax: 1 217 333 6968.

Contents lists availab

Atmospheric E

lse

Atmospheric Environment 45 (2011) 4830e4844E-mail address: yark@illinois.edu (T.C. Bond).global totals, from 32% in 2030 to around 22% in 2050. Total emissions from normal vehicles candecrease 1.3e2% per year. However, superemitters have a large effect on emission totals. They canpotentially contribute more than 50% of global emissions around 2020, which suggests that they shouldbe specically addressed in modeling and mitigation policies. As new vehicles become cleaner, themajority of on-road emissions will come from the legacy eet. This work establishes a modelingframework to explore policies targeted at that eet.

2011 Elsevier Ltd. All rights reserved.TransportationOn-road vehicleParticulate matter (PM)Technology1352-2310/$ e see front matter 2011 Elsevier Ltd.doi:10.1016/j.atmosenv.2011.06.018Soonkyu Jung a, Tami C. Bond a,*, David G. Streets b

niversity of Illinois at Urbana-Champaign, Urbana, IL 61801, USANational Laboratory, Argonne, IL 60439, USA

a b s t r a c t

We present global emission projections of primary particulate matter (PM) from exhaust of on-roadvehicles under four commonly-used global fuel use scenarios from 2010 to 2050. The projections arebased on a dynamic model of vehicle population linked to emission characteristics, SPEW-Trend. Unlikeprevious models of global emissions, this model incorporates more details on the technology stock,including the vehicle type and age, and the number of emitters with very high emissions (super-emitters). However, our estimates of vehicle growth are driven by changes in predicted fuelconsumption from macroeconomic scenarios, ensuring that PM projections are consistent with thesescenarios. Total emissions are then obtained by integrating emissions of heterogeneous vehicle groups ofall ages and types. Changes in types of vehicles in use are governed by retirement rates, timing ofemission standards and the rate at which superemitters develop from normal vehicles. Retirement ratesare modeled as a function of vehicle age and income level with a relationship based on empirical data,capturing the fact that people with lower income tend to keep vehicles longer. Adoption dates ofemission standards are either estimated from planned implementation or from income levels.We project that global PM emissions range from 1100 Gg to 1360 Gg in 2030, depending on the

scenario. An emission decrease is estimated until 2035 because emission standards are implementedand older engines built to lower standards are phased out. From 2010 to 2050, fuel consumptionincreases in all regions except North America, Europe and Pacic, according to all scenarios. Globalemission intensities decrease continuously under all scenarios for the rst 30 years due to the intro-duction of more advanced and cleaner emission standards. This leads to decreasing emissions from mostregions. Emissions are expected to increase signicantly in only Africa (1.2e3.1% per year). Because wehave tied emission standards to income levels, Africa introduces those standards 30e40 years later thanother regions and thus makes a remarkable contribution to the global emissions in 2050 (almost half).All Asian regions (South Asia, East Asia, and Southeast Asia) have a decreasing fractional contribution tofrom on-road vehicles

Global emission projections of particulate matter (PM): I. Exhaust emissionsjournal homepage: www.eAll rights reserved.le at ScienceDirect

nvironment

vier .com/locate/atmosenv

vironresponse to combined emissions of greenhouse gases, aerosols, andother trace species in the next 30e50 years.

Consistent emission projections on multinational scales arerequired to understand future impacts on air quality and climate. Byconsistent,wemean that assumptions about emission causes andresponses to economic growth should be treated similarly in allregions. This paper discusses global projections of primary partic-ulate matter (PM) emissions from on-road vehicles, based on a newmodel of technology that responds to socioeconomic conditions indifferent scenarios.

Of necessity, a globally consistent inventory has less detail thana local emission inventory. An emission inventory for one city, state,or province might rely on a model that incorporates survey data onvehicle and travel characteristics. This level of detail cannot pres-ently be represented in global inventories, because the data aresimply unavailable in many locations. The challenge is to createprojections in which assumptions are in harmony with the mostcurrent understanding of emissions at local or regional scale andwhich can utilize the kinds of socioeconomic data that are availableglobally and for world regions.

In the remainder of this section, we discuss previous work anda brief introduction to emission modeling. We present a basicvehicle eet model in Section 2, including a discussion of simplesensitivities. In Section 3, we describe the development of ourmodel (Speciated Pollutant Emission Wizard Trend model orSPEW-Trend). This model contains details about technology stock,including vehicle vintage, and uses exogenous data from economicscenarios to choose new technologies and retire old ones. Weoutline choices for model parameters: growth, retirement, emis-sion standards, emission factor and degradation rates, and thefraction of vehicles with extremely high emission factors (termedsuperemitters in this paper). Section 4 presents global modelresults, including projected emissions of total exhaust primary PMto 2050 and regional contributions. These results include projec-tions under four economic scenarios.

Many of the factors that govern emissions are not well known,and the SPEW-Trend model was designed specically to examinethe implications of uncertainties. This paper is the rst presenta-tion of the relationships and parameters underlying SPEW-Trend,and thus provides extensive model description and a brief discus-sion of the implications for PM emission projections. Future workwill use this framework to explore uncertainties and projections ofother atmospheric pollutants. Our goal in this sequence of papers isto represent the implications of particular policies explicitly. Thus,we make business-as-usual assumptions for vehicle eet dynamicsin this initial paper. We do not assume any air quality regulationsother than the implementation of emission standards. The value ofadditional emission reduction programs, such as inspection andmaintenance regimes, will be explored in future work. As describedin Section 1.2, this paper also takes exogenous projections of totalfuel consumption from macroeconomic models. Policies thatinvolve fuel-switching to meet greenhouse-gas targets are set inthose models, and our work gives PM emissions that would beconsistent with those consumption rates.

1.1. Previous work

Early global emission projections (Intergovernmental Panel onClimate Change (IPCC), 2001) recognized that emission factors(quantity of emission per mass of fuel used or per kilometer driven)would change with time, but did not develop any relationshipsbetween technological change and emission factor. Streets et al.(2004) used changes in emission factors to represent technolog-ical evolution for black and organic carbon, which are components

F. Yan et al. / Atmospheric Enof PM. The emission factors were based on a simple technologystock model, and representations of transitions relied mainly onexpert judgment. Rao et al. (2005) developed future projections byassuming that emission coefcients would decrease as grossdomestic product (GDP) increased. Uherek et al. (2010) gave anoverview of past, present, and future emissions from land trans-port, and they built emission projections, supposing that emissionstandards will be further tightened in OECD regions, as well as inother world regions, notably in Asia.

We have previously described a global emission inventory ofblack and organic carbon particles (Bond et al., 2004) whichdetermines emissions by apportioning fuel use among differenttechnologies. This approach has been used for past emissions (Bondet al., 2007) and we now extend it to future projections by con-necting this dynamic technology representation with the SPEWemission tabulation.

1.2. Approaches to modeling

Emissions depend strongly on combustion or manufacturingprocesses or end-of-pipe controls. Emission estimates shouldreect technology choice at different times and in different regions.When emission rates are strongly heterogeneous, large emissionfractions might be attributable to small fractions of the technologybase.

There is currently a dichotomy between economic (topedown)models, which seek supply-demand equilibrium and model broadeconomic sectors, and engineering (bottomeup) models, whichspecify the physical components of technological change. Currenteconomic models, or Computational General Equilibrium (CGE)models, reduce emission intensities (Rao et al., 2005; Tao et al.,2007) with time, but do not fully account for technologies withvarying emission factors. Thus, they do not reect the policy andaction levers available to decision-makers. Recently, models thathybridize the two approaches have been applied to electricitygeneration (Wing et al., 2008) and transportation (Kim et al., 2006),but even these efforts do not contain the technological detail thatwe espouse. SPEW-Trend is considered a step toward hybridizationfrom the engineering side. It is an engineering-based, technology-rich model driven by activity provided by economic models.

For two major reasons, we rely on exogenous inputs from CGEmodels, instead of predicting fuel consumption and emissions bysumming the contributions of different technologies. First, CGEmodels are used to project future emissions of CO2 and othergreenhouse gases for use in climatemodeling. Using theCGEoutputsto drive fuel consumption allows the projections developed here tobe fully consistent with their partner greenhouse-gas scenarios.Second, CGE models represent broad trends in efciency, and theyoften disagree with bottom-up models, being less optimistic (SueWing et al., 2008). We allow the bottomeup technological changemodel to be driven by the big-picture economic model.

Unlike models that move toward hybridization from theeconomic side, SPEW-Trend does not seek optimal solutions.Market allocation models (e.g. MARKAL (ETSAP, 2001)) are oftenused to identify future technology choices that will meet emissionstandards at the lowest cost. However, these rational decisions maynot govern the small emitters that are responsible for a large fractionof the emissions. The purpose of SPEW-Trend is to project emissionsbased on historical consumer behavior rather than lowest cost.

2. Vehicle eet model

2.1. Model equations

In this section, we review the basic structure of a vehicle eet

ment 45 (2011) 4830e4844 4831model. We use this model with constant rates of retirement and

emission change to demonstrate sensitivities to these parametersin Section 2.2.

New vehicles emit less pollution than old ones because ofdesign changes and improvements in air pollution control devices,such as catalytic converters. On the other hand, emissions mayincrease with aging because of degradation in engine performanceand air pollution control equipment. It is thus important torepresent the vehicle age distribution, as well as any changes invehicle type such as implementation of emission standards. Forthis purpose, we use a simple technology stock model (Zachariadiset al., 1995). At the end of year t, the number of vehicles (V) of typei and age s is given by:

Vi;st Vi;st 11 Ri;st

Ni;0t (1)where N is the number of new vehicles purchased, and R is thefraction of existing vehicles retired or scrapped during year t. Totalemissions E(t), in g, can then be calculated as:

Et X X

Vi;stFi;sEFi;s (2)

To which model parametersdgrowth rate, retirement rate, rateof new-vehicle emission decrease, and emission standard imple-

Bond et al., 2007), in which emissions are reduced by a factor of 5over about 20 years.We also use a simple logistic retirementmodel,which is discussed more fully and justied in Section 3.2. Briey,the equation for survival rate (Su) as a function of age s is

Sus 11 exparet1 s=L50;ret (3)

where L50,ret is the age at which 50% of the vehicles have retired(which is different from the median age within the entire vehiclepopulation); aret is a shape factor related to the onset of signicantretirement.

For this simple sensitivity study, we vary the following param-eters, with base-case values given in parentheses: median life(L50,ret, 10.4 years); shape factor (aret, 4.8); and rate of decrease inPM emission rate (21 years for four-fold reduction).

Fig. 1 shows the effects of standard introduction, retirementparameters, and growth rates on total eet emissions, normalizedto the initial year. (The effect on eet-average emission factors isalso given for these sensitivity studies in Fig. S1.) Retirement rate

ate a

F. Yan et al. / Atmospheric Environment 45 (2011) 4830e48444832mentation timedare emissions most sensitive? It may seem intu-itive that they all affect emissions, but we would like to determinewhether any factor is relatively unimportant in emission projec-tions. For this demonstration, we use a simple model that does notreect the more complex relationships explored later in the paper.Here, the annual growth rate of fuel consumption is constant at 2%.Emission factor decreases with year to simulate the effect ofincoming standards. We use an approximation to a trend in PMemission rates for heavy-duty vehicles (Yanowitz et al., 2000, t by

Fig. 1. Effect of model assumptions on total emission. Left: Changes due to retirement ri s

where Fi,s and EFi,s are the fuel consumption (in kg vehicle1) andemission factor (in g kg1) for vehicles of type i and age s.

The eet model in Equations (1) and (2) requires estimates ofnew vehicles introduced during a particular year and fuelconsumption by these new vehicles, the retirement rate R orsurvival rate of old vehicles, and emission factors. Before we searchfor the values of such parameters, we rst explore the importanceof each factor to total emissions.

2.2. Sensitivity of emissions to model parametersgrowth rates. L50 is the half-life of vehicles; a is a shape parameter in the logistic functioreduction for new vehicles. Five-year lag indicates that emission reduction for new vehicplays an important role in determining total emissions. The tworetirement rate parameters (aret and L50,ret) have comparableeffects on total emissions and average emission factor (the latter isgiven in Supplemental information). Doubling the annual growthrate to 4% increases total emissions more rapidly than changes inretirement parameters, but several years after the introduction ofemission standards, a doubled growth rate has about the sameeffect as a halved retirement rate. The right panel in Fig. 1demonstrates how the timing of emission standards affects totalemissions. When emission factors change more slowly (rate half asrapid), there is no reduction in total emissions after 30 years. A5-year lag in beginning to reduce emission factors is of equal orgreater signicance. If regulations are delayed, a greater stock ofpolluting vehicles remains in use. This is even more importantwhen growth is rapid, as shown in the 5-year lag, 4% growth rate.Such fast growth is typical of rapidly developing countries, wherethe introduction of standards is now under consideration. Theintroduction of standards early in the period of rapid growth iscritical for future air quality trends.

This sensitivity study shows that growth rates, retirement rates,and introduction of emission standards have approximately equalimportance in estimating future emissions. These factors are notseparable. Economic development enables growth, but it might alsoencourage tighter emission standards and enable implementation,

nd growth rates assumptions. Right: changes due to timing of emission standards and

n (see Equation (3)); growth is the eet growth rate; EF red is the rate of emissionles is initiated at 5 years; otherwise, it occurs at time 0.

(WDI) (World Bank, 2008) and International Energy Agency (IEA)(2009a, b) databases, respectively. For the projected future cases,we use four scenarios (A1B, A2, B1, and B2) developed for theSpecial Report on Emission Scenarios (IPCC, 2001; Nakicenovic,

vironment 45 (2011) 4830e4844 4833as well as promote more rapid retirement. A projection modelshould consider the response of each factor to economic conditions.

3. Model approach and parameters

For the remainder of this paper, we describe the development ofand results from SPEW-Trend, an object-oriented eet model thattracks vehicles of each type and age using the simple equationsgiven in Section 2.1. This model contains more explicit detail thanthe model used for the sensitivity study in Section 2.2. Instead ofa prescribed trend of emission factors, we represent changes inemission factors with the introduction of vehicles built to particularstandards. We include emissions by light-duty gasoline vehicles(LDGV), light-duty diesel vehicles (LDDV), and heavy-duty diesel(HDDV) vehicles. Within these broad categories, sub-types reectthe emission standard under which the vehicle is manufactured. Inthis model, we do not represent other vehicles such as two- andthree-wheelers, whichmay contribute about 10% of emissions (IEA/SMP). Because we apportion the entire on-road fuel use among ourthree modeled types, neglecting the other vehicles affects totalemissions only if the emission factors are very different.

Some important and relatively novel aspects of SPEW-Trendinclude: (1) Interface with CGE models, which are used to setnew demand; (2) ability to dene emission standards based oneither desired implementation dates or economic conditions; and(3) situation-specic service, retirement, and emission degradationrates. By situation-specic, we mean that each technology objectcan respond to its economic surroundings. For example, a retire-ment rate can be based on vehicle age, regional income (repre-sented by GDP per capita), or global GDP per capita.

The simple growth and retirement rates discussed in thesensitivity analysis (Section 2.2) are for illustrative purposes andare not the ones used for our projections. In the following sections,we discuss how we derive inputs for SPEW-Trend. Section 3.1describes how we determine new demand from CGE outputs.Section 3.2 reviews literature regarding retirement rates anddiscusses our choice of parameters. We discuss our choice ofemission factors in Section 3.3, which includes timing of emissionstandards, degradation rates, and the rate at which superemittersare allowed to develop from normal vehicles.

3.1. Demand for new vehicles

We drive changes in vehicle eet composition with predictedfuel consumption. For each year, fuel consumption by existingvehicles is calculated; this is usually lower than consumption in theprevious year because older vehicles are driven less (Section 3.4).New vehicles are introduced to account for the difference betweentotal fuel consumption for the year and fuel consumption byexisting vehicles. The number of vehicles added contains someuncertainty due to the amount of fuel consumed per vehicle, whichmay change with demand for either smaller, more efcient vehiclesor larger, more powerful vehicles. While this uncertainty affects thenumber of vehicles in the eet, it does not greatly affect emissionsthat are normalized to mass, which is the approach in SPEW-Trend.Potential improvements in fuel efciency are handled by themacroeconomic models, and SPEW-Trend produces emission esti-mates that are consistent with those models. Although our vehicleeet model estimates the number of vehicles, its primary purposeis to project the amount of fuel used by vehicles of a particularvintage, and this latter estimate is not affected by changes in vehicleefciency.

In this work, historical data (before 2005) for socioeconomicvariables (GDP and population) and fuel consumption (light and

F. Yan et al. / Atmospheric Enheavy oil for road sector) are from World Development Indicators2000) as formulated by the Integrated Model to Assess the GlobalEnvironmental (IMAGE) group at world-region level (RIVM, 2001).These scenarios are not marker scenarios except for B1 and they areidentical to those chosen in previous work on black carbon (BC) andorganic carbon (OC) emission forecasting (Streets et al., 2004) andrepresent a range of economic conditions as described in Table 1.New scenarios are presently under development, and the methoddescribed here is easily applicable to any kind of economicprojections. The IPCC scenarios, although frequently used, arerather outdated now; they rely on real data until the year 1990 andmake projections thereafter, so that the period 1990e2005 doesnot match actual data. We use the fuel consumption data for roadvehicles in IEA (2009a, b), plus GDP and population in WDI (WorldBank 2008) as historical estimates until 2005. After that, theyfollow the trend of those in IMAGE.

IMAGE scenarios give aggregated fuel consumption for thewhole transportation sector. These CGE projections considerneither technological nor sectoral detail, nor do they apportion fuelconsumption among on-road and off-road uses. Our task is toapportion the projected fuel consumption into on-road use andthen among different vehicle types to develop air pollutant emis-sion scenarios that are consistent with these projections.

We make the following assumptions to project future fuelconsumption based on the growth rates in IMAGE after 2005. Theyare broadly consistent with present-day fuel consumption shares.(1) Growth rates of total on-road fuel use will be the same as thoseprojected by IMAGE for the total transportation sector. (2) Gasolineconsumption has the growth rate of transportation light oil inIMAGE. (3) Total on-road diesel consumption has the growth rate oftransportation heavy oil in IMAGE. (4) For light-duty vehicles, thefuel consumption of LDDV compared with LDGV is assumed tomaintain a constant but regionally-dependent ratio given in theIEA/SMP TransportModel1 for year 2000. (5) HDDVs account for theremainder of the projected on-road diesel consumption. Thecomposition of the light-duty eet may in fact change, but torepresent this in the model world, we would shift projected fuelconsumption between HD and LD vehicles, and total emissionswould change little. (As we shall see, this decision contributes onlya small uncertainty, because emissions are dominated by HDvehicles.)

The growth in total fuel consumption, plus the fuel consumptionby retired vehicles, sets the number of new vehicles deployed eachyear. We assume that the new vehicles chosen are those con-forming to the emission standard in force at the time.

In the IMAGEwork, countries are grouped into 17world regions:Canada, U.S., Central America, South America, Northern Africa,Western Africa, Eastern Africa, Southern Africa, OECD Europe,Eastern Europe, Former USSR, Middle East, South Asia, East Asia,Southeast Asia, Oceania, and Japan. Our projection model usesthese regions individually, but we aggregate some of them forpresentation purposes: North America (Canada U.S.), LatinAmerica (Central South America), Africa (Northern Western Eastern Southern Africa), Europe (OECD EasternEurope), Former USSR, Middle East, South Asia, East Asia, SoutheastAsia, and Pacic (Oceania Japan).

1 http://www.wbcsd.org/plugins/DocSearch/details.asp%3ftypeDocDet&ObjectIdMTE0Njc, accessed May 2007.

parameters. The goodness of t might not be distinguishablebetween any two distributions (Manski and Goldin, 1983;Zachariadis et al., 1995) even suggested that the logistic modelmight provide a better t. Second, the logistic function iscommonly-used by econometricians to represent factors governingdiscrete choice (Hausman and Mcfadden, 1984). In projectingfuture global emissions under different scenarios, we will portraythe response of technology to economic and social conditions. Themodel must represent the factors responsible for vehicle retire-ment, not just the rate of retirement. The Weibull distributioncannot be modied to do this.

Zachariadis et al. (1995) suggested that their Weibull parame-ters b (failure steepness) and T (service life) could be approximatedby the 50th and 95th percentiles of the lifetime, although theyrecognized that this relationship was not predictive. However, wefound that this approximationwas poor even for the data presentedby Zachariadis. Retirement rates estimated using vehicle lifetimeparameters did notmatch observed data for many countries, exceptfor West Germany.

Table 1Four economic scenarios used for projections.

A1B A2 B1 B2

Economic growth Rapid Lower High ModeratePopulation growth Low High Low ModerateEfcient technology

introductionRapid Slow,

fragmentedEmphasis onglobal solutions

Less rapid

Globalization More Less More LessGlobal GDP (trillion, 1995 $US)2000 34 34 34 342030 101 64 87 792050 211 97 160 128

On-road transportation fuel consumption (Mtonne)2000 1350 1350 1350 13502030 2710 2050 2100 18602050 2840 2380 1930 1530

F. Yan et al. / Atmospheric Environment 45 (2011) 4830e484448343.2. Retirement rates

3.2.1. Functional form of retirement ratesTable 2 summarizes relationships that represent scrappage, or

permanent retirement of vehicles. These relationships provideeither the fraction of vehicles retired (R) or the fraction of vehiclessurviving beyond a specic age s. The retirement rate R duringa time period Dt that begins at time t can be determined from thesurvival rate (Su):

Rt;Dt 1 Sut DtSut (4)

Of the models summarized in Table 2, that of Walker (1968)examined changes in the U. S. vehicle eet before and afterWorld War II. Zachariadis et al. (1995) t a modied Weibullfunction to vehicle trends in several European countries. The U.S.EPA (2005a) used a transformed normal model, which is a specialcase of the Weibull distribution. Each model allows differences inthe onset of signicant retirement and results in an S-shapedsurvival curve, a form we also used in work describing historicalemissions (Bond et al., 2007). Retirement data suggest that a smallfraction of very old vehicles persists well beyond their expectedretirement age (Greenspan and Cohen, 1999; Walker, 1968). Themodel of Walker (1968) includes a third parameter that accountsfor this effect. Total emissions are not very sensitive to the long tailin the age distribution curve, so we do not retain the term.

For several reasons, we choose the logistic function to representretirement, even though some previous, well-accepted work hasused the Weibull distribution (Zachariadis et al., 1995; U.S. EPA,2005a). First, the logistic function can closely match many

Weibull-like distributions with an appropriate choice of

Table 2Some published scrappage models. R is retirement rate, Su is survival rate, s islifetime in years.

Citation Equation Key parameters

Walker, 1968Logistica

Rs 1B expa1 s=L50

B, tail representingold vehiclesb

a, shape factorL50, median lifetime

Zacharidiset al., 1995

Sus exp

s bT

bb, failure steepnessT, characteristicservice life

Modied WeibullU.S. EPA, 2005a Sus 1

2p exp

s s02s2

2s0, age at onset ofsignicant retirement

Transformed normal s, steepness of change

a Parameters transformed from those given in paper.b Not retained in this analysis.3.2.2. Factors affecting retirement ratesFactors governing vehicle scrappage have been extensively

investigated, as summarized in Table 3. Other than vehicle age ormeasure of service, economic considerations are the most signi-cant factors in permanent vehicle retirement. Parks (1979) andGreenspan and Cohen (1999) agree that scrappage is favored whenthe cost of repairs is high relative to the price of new cars, whileManski and Goldin (1983) nd longer lifetimes when the price ofnew cars is high. Greenspan and Cohen (1999) also nd that theprice of gasoline inversely affects scrappage, possibly becausevehicles are driven less.

In addition, studies have examined vehicle replacement(DeJong, 1996; Gilbert, 1992) and new vehicle choice (Brownstoneet al., 2000) as a function of socioeconomic variables. Thesestudies, reviewed by Chen and Niemeier (2005), reveal heteroge-neity in vehicle ownership decisions depending on household andvehicle characteristics. However, such studies do not predict whenvehicles are actually removed from operation. Furthermore,because data are lacking, we cannot represent much of thisheterogeneity in a global emission model.

We settle for representing two of the main factors in retirementdecisions: vehicle age and the balance between vehicle cost andrepair. The former is retained in our eet model. The price and labor

Table 3Literature review of factors affecting vehicle scrappage.

Citation Parameters found tobe signicanta

Other parametersexamined

Parks, 1977;Parks, 1979

Vehicle age; price of repairsrelative to price of newautomobiles

Make; price of scrapautomobiles; interestrates

Manski andGoldin, 1983

Age, new vehicle price Engine size; continentof vehicle originb

Greenspan andCohen, 1999

Vehicle age; price of gasoline;price of repairs relative toprice of new automobiles

Emission standard;number of teenagers;unemployment rate

Chen andNiemeier, 2005

Vehicle age; vehicle agesquared; odometer reading;passes emission test; needsrepair; gross polluter; vehiclemake

Locations

Chen and Lin, 2006 Vehicle age; number of previousrepairs; some vehicle type

Vehicle age; numberof repair calls; fuel type;makes

a t-statistic < 2.b Results did indicate signicant effects of engine size and vehicle origin, but

paper did not emphasize these results.

been used to examine the timing of environmental decisions suchas lead reduction in reneries (Kerr and Newell, 2003), KyotoProtocol ratication (Zahran et al., 2007) and state policy adoption(Jones and Branton, 2005). As the latter countries can be consideredtechnology-forcing for emission standards and other countriesare technology-following (Faiz et al., 1995), the ceteris paribusassumptions of the all-country Cox regression are violated. Whilethe coefcients of per-capita GDP were signicant when all coun-tries were included, they were not signicant when U.S. andEuropean countries were excluded. Thus, other factors, such as

vironment 45 (2011) 4830e4844 4835indices used in retirement rate studies are not available worldwide,so we make a rather bold assumption, namely, that the prices ofnew vehicles, which can be produced and sold anywhere, are set bythe global market, while repair costs are governed by local laborrates. We use the ratio between regional GDP per capita and globalGDP per capita as a factor in the retirement decision. Thisassumption reects the ndings of the studies in Table 3, and isconsistent with the conventional wisdom that vehicles are retainedlonger in lower-income countries. The next section will discusshow we obtained the dependence from observations.

3.2.3. Observed retirement ratesTo the best of our knowledge, studies on retirement rates are not

available in most world regions. Data most commonly available are(1) total vehicles registered or in use for different years, and (2)vehicle age distribution at a specic point in time. Both newdemand and retirement affect total population. In principle, thetwo independent data sets (registrations and age distribution)should be able to constrain the retirement relationship. Weacknowledge, however, that the recorded data may have incon-sistencies (Greenspan and Cohen, 1999). For example, the rela-tionships developed will be poor if old vehicles are not removedfrom registration records.

We used an iterative method, depicted in Fig. S2, to estimateretirement rate parameters when these two data sets were avail-able. On-road vehicle trends came primarily from InternationalRoad Federation (2003), via a global compilation of vehicle data(Bond et al., 2007). First, we estimate the parameters of a survivalcurve as a function of age. Applying this to the observed agedistribution, the previous population and retirement at any point intime can be inferred. These estimates are then used to infer thenumbers of new cars purchased each year, determine survival rates,and then re-estimate parameters for the survival rate curve untilconvergence is obtained. Table S1 summarizes the ts to retirementrate model for a few selected countries. In general, cars retire moreslowly in developing countries. Values of L50,ret for cars are smallerthan those for trucks, indicating that trucks take longer to beginretiring. The derived data show a loose linear relationship betweenaret/L50,ret and the value of local-to-global GDP per capita. There-fore, we use a linear regression between the two quantities toestimate the parameters governing survival rates in differentregions. When the vehicle eet was modeled with these relation-ships, the average age within the entire eet in Europe was slightlylower than reported ages (European Environmental Agency (EEA),2011).

3.3. Emission factors

3.3.1. Implementation of regulationsAs demonstrated in Section 2.2, the timing of standard intro-

duction affects total emissions. Many countries either have alreadyadopted progressively tighter regulations governing vehicle emis-sions, or have set timelines for implementation of those standards.U. S. emission standards are termed Tiers I, Tier II, etc., whileEuropean emission standards are termed Euro I, Euro II, etc. Manyother countries, such as India and China, follow the Europeanstandards; thus, the Tier and Euro sequences capture most of thetransitions throughout the world. In this work, we representemissions from vehicles without regulation (None), from vehiclesin regions where rudimentary inspection using opacity or smoke-meters may lead to some reduction in emissions (Opacity), andboth the Tier and Euro sequences.

We rst investigated whether a Cox proportional-hazardregression (Cox, 1972) could be used to project the onset of stan-

F. Yan et al. / Atmospheric Endard adoption dates based on GDP per capita. Such a model hastrends in neighboring countries or air quality problems in additionto income, must govern the introduction of standards, and wechose a more empirical method of introducing standards for pastand future years.

For regions inwhich a single country has the highest population(Canada, U.S., Former USSR, South Asia, East Asia, and Oceania), weuse that dominant country to provide the timing of standardimplementation. Other regions are quite heterogeneous in terms ofstandard adoption (Middle East, Southeast Asia, South America,Eastern Europe). For these regions, we use an average of theimplementation year in each country to represent the region.Heterogeneity within the region is impossible to represent ina large regional model. For modeling country-level air quality,a more detailed representation of individual countries would berequired; our model demonstrates only broad regional trends. Athird type of region (Central America, Northern Africa, SouthernAfrica) contains large countries that committed to standardimplementation shortly after 2000, but the remaining countries inthe region have not committed to such standards even now, 10years later. For these regions, we assume that average imple-mentation timing is that of the leading country plus 10 years.Finally, two regions (Eastern and Western Africa) have no currentplans for standards. We assume that they will adopt standardswhen they reach a level of GDP per capita similar to the average ofother technology-following world regions. This value (excludingthe technology-forcing regions of North America, Europe andJapan) is $3600 per capita (1995 U.S. constant dollars). Thus,Western and Eastern Africa adopt Euro standards in the 2040s inA1B and even later in other scenarios. To represent technologywithin each region more precisely, more detailed modeling ofindividual countries is needed.

More stringent emission standards, such as Euro II and thosefollowing, are not independent of the rst emission standard. Oncethe rst standard is adopted, countries tend to accept increasinglystringent standards more readily in the future. We use constanttime intervals between emission standard introduction instead ofattempting to predict them based on economic conditions. Table 4summarizes the lag years between different Euro emission stan-dards from 30 countries (CONCAWE, 1997; DELPHI, 2009, 2010;Dieselnet; UNEP, 2008) giving 4 years on average. We use thisvalue to estimate the succession of more advanced standardsfollowing the rst one. The assumptions above are obviously criticalin determining the trajectory of emissions, because they affect not

Table 4Lag years (Date of newer standard minus date of old one) between different Euroemissions standards used in model.

Delta (years) LDGV/LDDV HDDV

Euro II-Euro I 5 4Euro III-Euro II 4 4Euro IV-Euro III 3 4Euro V-Euro IV 4 3Euro VI-Euro V 5 5

Average 4 4

virononly current emissions, but also the vehicle eet acquired duringperiods of high growth. Compliance with emission standards is alsoimportant.

3.3.2. Emission factors for new vehiclesEmission factors can be presented as mass-based (e.g. grams of

pollutant per kilogram of fuel consumed), or service-based emis-sion factors (e.g. grams of pollutant per kilometer driven). We usethe mass-based emission factor because global energy projectionsare usually given in terms of fuel mass.

Actual emission rates may differ from prescribed standards, someasured emission factors are important. However, there may beinsufcient measurements to predict emission factors for vehiclesbuilt to very new or forthcoming standards. Ntziachristos andSamaras (2001) introduced the concept of reduction factor (RF).Rather than assuming that vehicles conform to a desired emissionstandard, they assumed that the ratio between two standardsrepresents an achievable reduction. The reduction factor is:

RFx 1 ESxESbaseline(5)

where ES is Emission Standard x (e.g. Tier II or Euro II) in g km1 org kWh1; ESbaseline is a baseline emission standard (Tier I and Euro Iare used here).

RFx in Equation (5) can be used for service-based emissionfactors. Since we use mass-based emission factors in this study,Equation (5) needs to account for the fuel economy of higherstandard engines. Future emission factors under standard x (EFx)can be calculated from the following equation:

EFx

ESxFExESbaselineFEbaseline

EFbaseline (6)

where FE Fuel Economy in km L1 or kWh L1.The advantage of the reduction factor approach is its ability to

incorporate measured data on emission factors, which may differfrom the actual standard. In this section, we determine EFx for newvehicles. Degradation, or the increase in emissions with age, will bediscussed in the next section. We dene new vehicles as those withlow mileage according to the denition of Ubanwa et al. (2003)(less than 40,000 km for gasoline and less than 80,500 km fordiesel vehicles). Although many studies provide measured vehicleemission factors, we chose studies that provided odometerreadings.

Baseline emission factors for U.S. and European standards arebased on measurements in the appropriate region. For LDGV, weaveraged emission measurements for eight vehicles reported inMaricq et al. (1999) as 0.01 g kg1-fuel for Tier I and Euro I. Emis-sion factors for new LDDV are 1.3 g kg1-fuel (Ntziachristos andSamaras, 2001) for Euro I, and averaged from three vehicles as0.9 g kg1-fuel (Ubanwa et al., 2003) for Tier I. For HDDV, weaveraged emission factors as 1.7 g kg1-fuel from data reported byEEA (2002) for Euro I, and as 1.3 g kg1-fuel from 12 emission datapoints reported in Yanowitz et al. (2000) for 1988 U.S. standard.

We apply Equation (6) and baseline emission factors to calculateemission factors for most types of vehicles except LDGV, which donot have an emission standard for PM in either the U.S. or Europe.We assumed that the baseline emission factor (0.01 g kg1-fuel) ofa gasoline vehicle applies to all the tighter gasoline emissionstandards (all Tier IIs and Euro IIeEuro VI). Another exception is thecase of HDDV in the U.S.; for themwe averaged emission factors as1.3 g kg1-fuel (28 samples), 0.7 g kg1-fuel (6 samples), 0.8 g kg1-fuel (20 samples), 0.6 g kg1-fuel (31 samples) for 1991, 1993, 1994,and 1996 standards, respectively, by using emission data from low-

F. Yan et al. / Atmospheric En4836mileage vehicles reported in Yanowitz et al. (2000).We rely on measurements for vehicles built without standardsor for pre-baseline standards. We chose 0.05 g kg1-fuel reportedfrom testing of 659 vehicles in Ubanwa et al. (2003) for pre-baselinegasoline vehicles. For LDDV, we use 1.5 g kg1-fuel, the average ofmore than 50 vehicles from Ntziachristos et al. (2001) and Ubanwaet al. (2003) for early regulations. We averaged the emission factorsof more than 26 HDDV from Yanowitz et al. (2000) and EEA (2002)and use 2.9 g kg1-fuel for early regulation vehicles.

Finally, these measured values reect vehicles built to someregulation, butweneed emission estimates for vehicleswithout anyregulation. We assume that emission factors before regulation aregreater than those for early regulationby the same factor as betweenearly regulation and baseline emission (Euro I or Tier I). Estimates ofemission factors for vehicles without standards are 1.9 and4.2 g kg1-fuel for LDDV and HDDV, respectively. These emissionfactors may be underestimated, however; Kirchstetter et al. (2008)showed that diesel vehicle emission factors could have decreasedby a factor of 10 since 1960, based on ambient measurements.

3.3.3. Degradation ratesDegradation rate (Ubanwa et al., 2003; U.S. EPA, 2005b) is the

increase in emission factor with time (or usage). Factors causingthis increase include normal wear, failure of components, poormaintenance, misuse, improper fuel, and control systemtampering. Degradation rates could be different between engines,regions, and usage patterns. These factors have not been clearlyisolated in studies to date. We determine degradation rates basedon studies that include more than 300 vehicles, but these studiesare limited to the U.S. and Europe.

The general pattern of degradation rate is modied fromUbanwa et al. (2003). In our study, we separate degradationpatterns into three phases, as shown in Fig. 2a. During the new-engine phase, emissions will be constant for one year for gasolineand two years for diesel vehicles (Ubanwa et al., 2003). Then,emissions increase linearly with a rate depending on technologyduring the second phase (degradation phase). In the last stabilizedphase, emissions maintain a maximum level. For gasoline vehicles,three studies provide degradation rates. Two of them (Cadle et al.,1999; Durbin et al., 1999) are close, while the third (Ubanwa et al.,2003) is quite different.We use the averages of the rst two studies,which are 0.04 g kg1-fuel/year for pre-baseline standards and0.005 g kg1-fuel/year for post-baseline standards. For other type ofvehicles, we use PM degradation rates from Ubanwa et al. (2003).

Our new-vehicle emission factors are based on measured dataand are lower than the standard. Therefore, the applied degrada-tion rates result in emissions more than 10% above the emissionstandard only after seven years of lifetime.

3.3.4. SuperemittersFor many air pollutants, a small fraction of vehicles using the

poorest technology or with the worst maintenance contributesignicantly to total emissions (Hansen and Rosen, 1990; Lawson,1993; Zhang et al., 1995). Bond et al. (2004) estimated super-emitter fractions based on a review of sparse literature (Sections5.4.1 and 5.4.3 in that paper). Central values of superemitter frac-tions chosen from this literature tabulationwere 5% for the U.S. andWestern Europe; 10% for Eastern Europe; and 20% for Asia and LatinAmerica. We interpret more recent studies using the superemitterdenition given in Subramanian et al. (2009). Emission data ofblack carbon, a component of particulate matter, for 251 trucks inCalifornia indicate that superemitters are about 13% of the dieseleet (Ban-Weiss et al., 2009). In Beijing, diesel superemitters wereabout 17% with regard to both black carbon and PM0.5 (Wang et al.,2011). These vehicles produced about 45% (United States) and

ment 45 (2011) 4830e484450e60% of emissions (Beijing). Although carbon monoxide (CO) is

b) ann

vironnot a direct proxy for PM emissions, remote-sensingmeasurementssupport the existence of a small, high-emitting fraction of vehicles.Fifteen percent of vehicles produce 50% of CO emissions in Madrid(Pujadas et al., 2004); the same fraction cause 45% of the emissionsin Beijing (Zhou et al., 2007); 20% of vehicles produce 80% ofemissions in Tianjin (Oliver, 2008); 10% of vehicles produce 75% ofCO inMichigan (ESP, 2007) and 70% in Australia (Bluett et al., 2008).Clearly, emissions from these vehicles will signicantly affectmodeled emissions.

We use a modied logistic function to represent the rate atwhich normal vehicles become superemitters:

frs gain1 expasup1 s=L50;sup (7)

where s is vehicle age, as before; fr is the fractional rate at whichnormal vehicles become superemitters (fraction per year); asupdetermines the slope of this curvewith age, L50,sup is the vehicle lifeat which the rate becomes half the maximum, and gain is themaximum rate of superemitter transition for the oldest vehicles.

The parameters we used (asup 5.5; L50,sup 5.0; gain 0.032)were chosen so that equilibrium values of superemitters areapproximately the same as those in the 2004 inventory. While theequation depends only on vehicle age, the relationship betweenvehicle retirement and income means that regions with lower GDP

0 5 10 150

0.5

1

1.5

2

2.5

3

3.5

4a

Vehicle Age (year)

Emis

sio

n F

act

or

(g/kg

-fue

l)

Degradation Rate

LDDV-EUROI

New

en

gine

phas

e

Deg

rada

tion

Stab

iliza

tion

Fig. 2. a) General pattern of degradation rates (Light duty diesel, Euro I is shown); bZachariadis et al. (2001), Van Wee et al. (2000) and Davis and Diegel (2008)).

F. Yan et al. / Atmospheric Enper capita and more slowly retiring vehicles have a higher equi-librium rate of superemitters. For scenario A1B in 2010, the esti-mated superemitter fractions are around 4% and 18% for NorthAmerica and Asia, respectively.

Yanowitz et al. (2000) reported a large database of emissionmeasurements in the United States. We grouped the 166 emissiontests performed on the same driving cycle by engine standards. Asexpected, emission at the 50th percentile decreased as standardstightened. Emission at the 90th percentile also decreased withtighter standards, especially after 1993. We therefore assume thatengine technologies improved greatly after 1993, and that thisdifference affects superemitters as well. We use two superemitteremission factors in this paper: old-engine superemitters (up to1993 in the United States) and new-engine superemitters (post-1993 models). We do not have such extensive emission studiesfrom Europe, but the Euro I standard is similar to the U.S. 1993-standard. We therefore assume that this is the separationbetween old and new-engine superemitters.

For gasoline vehicles, we averaged emission factors fromsmoker and poor maintenance vehicles reported in Durbin et al.(1999) and Cadle et al. (1999). These are 3.2 g kg1-fuel (old-engine) and 0.3 g kg1-fuel (new-engine). For LDDV and HDDV, weuse averages of 11 vehicles from McCormick et al. (2003) as8.3 g kg1-fuel for old-engines, which agrees with estimates fromSubramanian et al. (2009) for Bangkok. Because McCormick et al.(2003) does not provide a signicant number of post-1993 tests,we estimate new-engine emission factors from the Yanowitz et al.(2000) database. We found that averaging the highest 2% of 326old-engine tests in this database gave values close to superemittersfrom McCormick et al. (2003) and Subramanian et al. (2009).Assuming that old and new-engine tests have similar emissiondistributions in this database, we average the highest 2% of 78 new-engine tests and use 2.9 g kg1-fuel as the new-engine super-emitter emission factor.

More recent studies do not yield direct results for PM super-emitter emission factors, but they are consistent with ourassumptions. Ban-Weiss et al. (2009) measured only black carbon,but average emissions from the 13% we identied as superemitterswere about six times higher than normal vehicles. The differencebetween normal vehicles and superemitters in the Beijing study ofdiesel engines was a factor of 10 for black carbon and 5 for PM0.5(Wang et al., 2011). Depending on the base technology, our factorsrange from 2e6 for old-engine superemitters. For new-enginesuperemitters compared with Euro III technology, the factor is 3.5.

0 5 10 150.4

0.5

0.6

0.7

0.8

0.9

1

Vehicle Age (year)

Ann

ua

l mile

age

pe

r ve

hicl

e (no

rma

lize

d to

Ag

e=

0)

Vehicle Activity

LDGVHDDV

ual mileage per vehicle (normalized to age 0) as a function of age (Data sources:

ment 45 (2011) 4830e4844 48373.4. Vehicle activity

Average annual distance driven decreases with vehicle age. Newcars are often more reliable, comfortable, and energy-efcient thanold cars and their owners may also take greater pleasure in drivingthem. We use data from Zachariadis et al. (2001), Van Wee et al.(2000) and the Davis and Diegel (2008) to represent vehicleactivity by age. Fig. 2b shows the vehicle activity change with agefor LDGV and HDDV. For LDDV, the vehicle activity trend is treatedthe same as LDGV. We have found no information on how vehicleactivity varies with age in developing countries, and therefore usethe same relationship worldwide. Recall that fuel consumption isset by the macroeconomic scenarios, so the age dependence ofvehicle activity affects only the prole of cars on the road, not thetotal fuel consumption.

4. Results and discussion

Here, we discuss global emission scenarios from simulationsunder the four economic scenarios. We rst discuss overall trendsin emissions versus fuel consumption (Section 4.1). Section 4.2

discusses the varying regional contributions, and Section 4.3analyzes how vehicles with different technology contribute toglobal emissions. Section 4.4 returns to emission intensities in lightof the regional and technological discussions.

4.1. Global emissions

Fig. 3 shows the magnitudes of fuel use and emissions for thefour scenarios modeled in this work. Values normalized to the year2005 are also presented along the right axis to demonstratecomparative changes. Emissions are identical for all scenarios until2005 when historical data end and projections begin. Fuelconsumption estimates for the four scenarios reect underlyingassumptions about different economic trajectories from IMAGE.The emission trajectories also reect the diversity of scenarios, butthe patterns differ from those of fuel consumption due to changesin emission factors. Emissions in 2030 range from 1100 Gg in B2,which has the lowest fuel consumption, to 1360 Gg in A1B, whichhas the highest. Emissions in B1 and A2 lie between these two cases

from Africa. First, fuel consumption grows consistently in Africa,from 70 Mt in 2010 to 430 Mt in 2050, while global and regionalfuel consumption in all other regions peak in the middle 2040s orearlier. Second, looser emission controls and late introduction ofemission standards cause higher emission intensity in Africa.Therefore, global emissions increase when the fast growth in Africabegins to surpass the reductions in other regions.

In A2, fuel consumption increases monotonically, while emis-sions have a similar pattern to that in A1B, showing rst a decreaseand then a slow increase after 2040. The explanation for the laterincrease is also the notable growth of fuel consumption and inef-cient emission controls in Africa. In B1 and B2, lower growth ratesof fuel consumption allow incoming standards to offset the growthof fuel consumption. Emissions under B1 and B2 are continuouslydecreasing. Despite the fast growth in Africa, emission decreasecaused by a dramatic reduction of fuel consumption and imple-mentation of advanced emission standards in all other regionsdrives the emission trend in B1 and B2. Thus, there is no laterincrease under these two scenarios.

b

F. Yan et al. / Atmospheric Environment 45 (2011) 4830e48444838at around 1200 Gg. All scenarios show emission decreases by 2035.The difference between the two extreme cases (A1B and B2)becomes greater with time, being about a factor of two different in2050 (1500 Gg and 800 Gg, respectively).

HDDVs contribute 86e90% of emissions after 2010 (shown inFig. S3 in A1B), so the dynamics of HDDVs are critical to the overalltrend. There is some uncertainty in our projections due to divisionof diesel into light-duty and heavy-duty engine use. However, mostof the emissions come from heavy-duty engines, which generallyuse diesel fuel because of their higher power requirements.

We discuss changes in normalized consumption and emissionshere; for example, normalized fuel consumption of two means thatconsumption is twice the 2005 value. The highest fuel consumptiongrowth occurs in A1B, which describes a future world of very rapideconomic and population growth. In this scenario, normalized fuelconsumption peaks at 2.1 in the early 2040s. Despite this increasein fuel consumption, emissions decrease continuously until 2035,reaching a value of 0.8. Then, they begin to increase to a factor of 0.9in 2050. The sharp decrease before 2035dwhich is also seen in theother scenariosdis partly due to the population reduction of old-engine superemitters (developed from vehicles before Euro I orTier I), whose emission factors are 2e30 times higher than newengines. The implementation of new and more advanced standardsalso contributes to this decrease in emissions. The increase after2035 can be explained by the signicant emission contribution

2010 2020 2030 2040 20501400

1600

1800

2000

2200

2400

2600

2800

3000

3200

Year

Fue

l Co

nsu

mpt

ion

(Mto

nn

e/yr

)

A1BA2B1B2

1

1.2

1.4

1.6

1.8

2

Fue

lCo

nsu

mpt

ion

(No

rma

lize

d to

20

05)

aFig. 3. a) Global fuel consumption and b) PM emission from on-road vehicles: absolute magbegin above zero.Overall, global emission trends do not follow the trends of fuelconsumption, due to varying trends of emissions caused by tech-nology change in different regions, each of which representsa complex interplay of factors. Therefore, an examination ofregional emission intensity is needed in order to understand theunderlying factors (see Section 4.4).

We compare global PM emission projections between our workand that by IEA/SMP (Fulton and Eads, 2004), GAINS model andUherek et al. (2010), shown in Fig. 4. All studies assume thatexhaust emission standards will be further tightened in the fastdeveloped regions or adopted in the rest of the world and projecta decrease of PM emission. The decrease rates in our work are lowerthan those in the other three references because we considersuperemitters and introduce emission factors late in Africa (Section4.4). The IEA/SMP model shows extremely high emissions in therst 20 years, because their model assumes that LD gasoline vehi-cles and diesel vehicles have the same emission factors, while weuse measured emission factors. That model is more optimistic withregard to emission standard implementation, even in Africa, wherestandards are assumed to initiate by 2015. Future emissions fromthe GAINS model and historical emissions from the work of Borkenet al. (2007) are closer to our estimates of global PM emissions ifsuperemitters are not considered, especially in year 2000, 2005 and2030 where the difference is less than 12%. On the regional level,we estimate higher emissions in most regions except North

2010 2020 2030 2040 2050600

800

1000

1200

1400

1600

1800

Year

PM Em

issi

on

(Gg/

yr)

0.4

0.5

0.6

0.7

0.8

0.9

1

PM Em

issi

on

(No

rma

lize

d to

20

05)nitudes (left y axis) and normalized values to 2005 (right y axis). Note that the y-axes

America and Europe (detailed comparison is given in Table S2).With similar fuel consumption between this work and Borken et al.(2007), the large difference (over 80%) of emissions in LatinAmerica, Africa and Middle East, Former USSR, and Pacic regionsmust result from differences in emission factors. In Borken et al.(2007), a weighted emission factor is assumed by each vehicletype and region, while the emission factors in this work reectvehicle eet composition by technologies and ages, especiallysuperemitters.

4.2. Regional contributions

Regional emissions respond differently within each economicscenario, as Figs. 5, 6 and 7 illustrate. As shown in Fig. 6a and b,although fuel consumption increases in all regions except NorthAmerica, Europe and Pacic, emissions increase signicantly inonly Africa, due mainly to rapid growth of vehicles with less thanoptimal emissions performance (Streets et al., 2004) and slowscrappage of old vehicles. With the later implementation ofadvanced emission standards, emissions from Africa are consis-tently positive (1.2e3.1%/year) from 2010 to 2050 under allscenarios.

Fig. 7 shows the fractional contribution of fuel consumption andemissions from different regions in scenario A1B. The otherscenarios have similar relative behavior among regions, althoughthe growth rates differ, so the fractional contributions are similarunder all scenarios. Africa increases its share of global emissions,resulting in a sizeable percentage contribution (over 46%) in 2050.Although Latin America (17%) is a dominant contributor to globalemissions in 2010, emission controls more than compensate for

2000 2010 2020 2030 2040 20500

500

1000

1500

2000

2500

3000

Year

PM Em

issi

on

(Gg/

yr)

This work-A1BThis work-A2This work-B1This work-B2This work w/o Superemitters-A1BThis work w/o superemitters-B1IEA/SMP modela

Uherek et. al 2010b

GAINS(BL WEO 2009)cBorken et.al 2007d

Fig. 4. Comparisons of total PM exhaust emissions from on-road transportation from2000 to 2050. (a) Total on-road emissions from IEA/SMP (Fulton and Eads, 2004);(b) exhaust emissions from on-road mobile sources in GAINS model (scenarioBL_WEO_2009); (c) Road transportation from Uherek et al. (2010, Fig. 26); Roadpassenger and freight transportation in year 2000 (Borken et al., 2007).

Fig. 5. Regional primary PM emissions (Gg yr1) fr

F. Yan et al. / Atmospheric Environment 45 (2011) 4830e4844 4839increased fuel consumption, as shown the negative emissionom on-road vehicles under the four scenarios.

-2 0 2 4

North America

Latin America

Africa

Middle East

Europe

Former USSR South Asia

East Asia

Southeast Asia Pacific

Global

Annual Change (%)

Fuel

-2 0 2 4

North America

Latin America

Africa

Middle East

Europe

Former USSR South Asia East Asia

Southeast Asia Pacific

Global

Annual Change (%)

PM Emission

A1B A2 B1 B2

a b

d b)

F. Yan et al. / Atmospheric Environment 45 (2011) 4830e48444840change rates in Fig. 6b. Thus, the percentage contribution declinesby 9% by 2050 in this region (Fig. 7b). The total fractional contri-bution of Asian regions (South Asia, East Asia, and Southeast Asia)decreases from 32% in 2030 to around 22% in 2050, also because ofemission controls.

4.3. Emission composition by engine type

Our forecasts are based on identifying activity by vehicles builtto specic emission standards. The contribution of each technologytype is shown in Fig. 8. Emission trends in A1B and B1 are similar tothose in A2 and B2, respectively, despite differing magnitudes, soonly A1B and B1 are presented here.

U.S. emission standards (e.g. Tier I) apply only in North Americaand some countries in Latin America. Emissions from vehicles builtto these standards account for less than 10% of the global emissions.For that reason, wewill use terminology from Euro standards in thefollowing discussion and group U.S. standards with the closest Eurostandards. Our model has not accounted for advanced emissionstandards that might be implemented after the currently plannedEuro VI or Tier II. However, PM emissions from these vehicles arevery low and do not contribute signicantly to total emissions.

Fig. 6. Average annual change of a) fuel consumption anThe most striking aspect of Fig. 8 is the contribution of super-emitters (top graphs), which contribute about half of global

0 10 20 30 40 50

North America

Latin America

Africa

Middle East

Europe

Former USSR

South Asia

East Asia

Southeast Asia

Pacific

Fractional Contribution (%)

Fuel Consumption

201020302050

N

F

S

a

Fig. 7. Fractional contribution of fuel consumption and emissions by differeemissions. To focus on the contribution of normal vehicles andthe effects of emission standards, the bottom graphs presentemissions in simulations where no superemitters were allowed todevelop. Our model assumes that most regions except Africa willhave implemented Euro VI standards by 2025. The trend of emis-sions from normal vehicles only (Fig. 8c and d) conrms that, ingeneral, the global eet will be cleaned up by emission standards,despite fuel growth. Some old vehicles built to poorer standardsremain after adoption of Euro VI, but these phase-out in the nexttwo decades. The bottom gures show the declining contribution ofthese older vehicles, made more visible without the contribution ofsuperemitters. In these lower graphs (Fig. 8c and d), the consistentemissions from vehicles without regulations (None) is mostlyattributable to Africa where no emission standards are imple-mented until the late 2040s.

As global eet emissions are reduced by emission standards,normal vehicles become a progressively smaller fraction of emis-sions. In 2050, total emissions without the effects of superemittersare 54e56% of those with superemitters, with global emissiondecreases averaging 1.3e2.0% per year. Emission projections arethus highly dependent on the behavior of superemitters.

Superemitters begin producing more than 50% of global emis-

PM emission between 2010 and 2050 in four scenarios.sions around the year 2020; the exact year varies by region andscenario (detailed information is shown in Table S3). In 2050, the

0 10 20 30 40 50

orth America

Latin America

Africa

Middle East

Europe

ormer USSR

South Asia

East Asia

outheast Asia

Pacific

Fractional Contribution (%)

PM Emissionb

nt regions in scenario A1B; other scenarios are similar to that in A1B.

vironF. Yan et al. / Atmospheric Enfuel consumption contribution of superemitters is only about 14%of total, yet their emission contribution is about 62%. Asia, theMiddle East, and Latin America contribute about half of the globalsuperemitter emissions. In these developing regions, the lowerincome level results in a large fraction of superemitters, and growthin road transport means that there are more and more super-emitters on the road. Thus, our model depicts a progressivelycleaner eet of normal vehicles, but a growing eet that has a largerabsolute number of superemitters. The implication is that tighterstandards for new vehicles alone cannot satisfactorily protectenvironmental quality; the potential proliferation of superemittersmust be addressed specically. Another implication is that globalemission trends will remain highly uncertain until the behavior ofsuperemitters is better understood.

4.4. Global and regional emission intensities

The preceding sections have shown that regions are diverse interms of technological dependence and evolution. It is instructiveto examine the evolution of global and regional emission intensitiesunder different scenarios. With the understanding of how differentregions and different vehicle types contribute to global emissions,we can now explain the different trends in fuel consumption andemissions that were shown in Fig. 3. Here, we use the termintensity to mean the ratio between emissions and fuelconsumption. Global and regional emission intensities are shownin Figs. 9 and 10.

Fig. 8. PM emission contribution by vehicles under different emission standards: a) globald) emission without superemitters in B1. In the bottom graphs (c and d), fuel consumptionvehicles to superemitters is zero. The black lines in a) and b) indicate the years when supstandards; NONE, no regulation; Super 1, superemitters from older engines; Super 2, superment 45 (2011) 4830e4844 4841Emission intensities decrease continuously for the rst 30 yearsof the study period, as stringent emission standards are applied andolder vehicles, especially old-engine superemitters, are retired. Thisexplains the observation in Fig. 3 that fuel consumption increasesalong all trajectories, but PM emissions decrease consistently in therst few years. Emission intensities increase slightly in late 2030sand 2040s under A1B and B1, respectively. This is caused by thenotable emission increase in Africa where vehicles without regu-lations compose the majority of the eet, as discussed in Sections4.1 and 4.2. Growth in regions with lower incomes and highersuperemitter fractions also contributes to this trend. Anotherindirect reason for this increase is the convergent economic growthpatterns among regions under these two scenarios, which will beexplained later.

Emission intensities in North America, Europe, and the Pacicare 3e5 times lower than those in less-developed regions, becauseof the earlier adoption of advanced emission standards in thosetechnology-leading countries. Fig.10 indicates that the difference ofemission intensities between regions is greater than that amongscenarios for a single region. Therefore, total fuel consumption andthe timing of standard introduction are the most important factorsrequired to represent overall emissions, as long as the compositionof vehicle eets is considered. Other economic factors playa secondary but important role.

Fig. 10 also compares regional emission intensities amongscenarios for each region. In 2030, emission intensities in NorthAmerica, Europe, and the Pacic are 20e30% lower under A2 thanunder A1B, while other regions have higher emission intensities

emission in A1B; b) global emission in B1; c) emission without superemitters in A1B;is the same as top graphs under each scenario, except that the transition from normaleremitters contribute over 50% to the global emissions. Abbreviations: OPAC, Opacityemitters from newer engines. Numerical values given in Table S4 and S5.

F. Yan et al. / Atmospheric Environ4842Fig. 9. Global emission intensities under four scenarios (g PM/kg fuel) (left) andemission factors for new engines under different emission standards (except super-emitters) (right). Emission factors for some standards are not shown, including HDDVwith opacity standards (2.9 g kg1-fuel) and without regulation (4.2 g kg1-fuel). ForLDGV, emission factors for superemitters are 0.3 g kg1-fuel (new engine) and

1under A2. This nding has a perverse cause. A2 is a less globalizedscenario, so income levels in different regions becomemore diversewith time. Because of this regional diversity, the price of newvehicles in the more-developed regions is lower relative to income,so people can afford new cars and old vehicles are replaced morequickly. Therefore, the fraction of relatively dirtier vehicles (oldvehicles and superemitters) is lower. In A1B, a more globalized

3.2 g kg -fuel (old engine); For LDDV and HDDV, emission factors for superemittersare 2.9 g kg1-fuel (new-engine) and 8.3 g kg1-fuel (old engine). Abbreviations: OPACmeans Opacity; NONE means no regulation.

0 0.5 1 1.5 2 2.5

North America

Latin America

Africa

Middle East

Europe

Former USSR

South Asia

East Asia

Southeast Asia

Pacific

Global

Emission Intensity (g/kg-fuel)

A1BA2B1B2

Fig. 10. Regional emission intensities in 2030 (unit: g PM kg fuel1).scenario, income levels are more similar between regions. Thisconvergence means relatively expensive vehicles, longer vehiclelifetimes, and a greater fraction of polluting vehicles in developedregions. Beyond 2030 (not shown), the trend continues: regionalincome distribution becomes more diverse in A2 and moreconvergent in A1B. Emission intensities under A2 continue todecrease compared with A1B, even for some less-developedregions. This explains the lower global emission intensity for A2compared with A1B, which is apparent in Fig. 9 for 2030e2050.Regions in B2 are not as economically diverse as in A2, so thedifference of emission intensities between the two B scenarios isnot as signicant as that in the A series.

Traditionally, convergence of income between developed anddeveloping regions is seen as desirable, but in this case, relativelycheap labor might make clean technology less expensive, facili-tating its implementation in areas of rapid growth. Environmentalpolicies should give special attention to vehicle retirement ratesand the role of superemitters, because of their inherent importance,but also because they may be affected by regional income diversityin unexpected ways. Finally, we obtained this result because of theway we expressed retirement rates. While this formulation isplausible, it needs further study to conrm the implications forincome convergence.

5. Conclusions and recommendations

This paper has outlined global projections of primary PMemissions from on-road vehicle exhaust, based on a new tech-nology model that responds to socioeconomic conditions indifferent future scenarios. The dynamic populationmodel allows usto investigate the role of technology selection, eet turnover andhigh emitters in future emissions with a consistent framework.

The trend of emissions represents the combination of emissionintensity and fuel consumption. In the rst 20e30 years, PMemissions decrease in all scenarios because lower emission inten-sity dominates increasing fuel consumption. We examined theevolution of global and regional emission intensities, and analyzedthe underlying factors that affect the trend of those emissionintensities. Globally-averaged emission intensity will decrease inthe near-term but may increase again as growth occurs in regionswith looser emission standards. Vehicles without regulations orlate introduction of standards, as well as a possible dependence onconvergent economic growth patterns act to increase emissionintensity.

If our assumptions about superemitters are valid, they havecontribute over 50% of emissions in all years after 2020 and drivethe trend of PM emissions. If they are eliminated, global emissionsdecline at an average rate of 1.3%e2.0% per year. However, ourmodel representation of superemittersdand hence theseresultsdare based on very little information about the causes andprevalence of such high-emitting vehicles. Because superemittersplay such a signicant role in determining emission trends, it isnecessary to understand their behavior in more detail. Factors thatcause high-emitting vehicles, their levels of service, their retire-ment rates and their emission rates should be better explored witheet observations. Remote-sensing measurements or other statis-tical characterizations of the vehicle population in different worldregions would be particularly valuable.

What causes countries to adopt emission standards? Adoptiondoes not have a signicant relationship with income if technology-forcing regions are excluded from analysis. Later in our analysisperiod, low-income regions have a disproportionate contribution toglobal emissions. Many of these regions already have severe urbanair quality problems, so the factors that lead to acceptance of

ment 45 (2011) 4830e4844emission standards for the health of urban populations must be

vironexamined in order to determine the fate of the global and regionalatmosphere.

There is large uncertainty in the connection between techno-logical changes and income levels. Many of the relationships rep-resented in SPEW-Trend, such as retirement rates, are only looselyconstrained by observations. While future economic trajectoriesare always unknown, uncertainties in emission intensity also resultfrom a lack of knowledge about retirement rates, emission degra-dation rates, and the contribution of superemitters. Ongoing work,which will be presented in a forthcoming paper, uses sensitivityanalysis and Monte Carlo simulations to identify importantassumptions and explore the effect of uncertainties on global andregional emissions. The results presented here are our rst estimateof future global emission trends but must be understood in thecontext of these uncertainties.

The model results given here describe emission trajectoriesunder different scenarios for light-duty and heavy-duty vehiclesthat use liquid fuel. They do not yet include some importantconsiderations: (1) emissions from two- and three-wheelers, whichare important in Asian and African countries; (2) gaseous emis-sions, which affect ozone concentrations; and (3) non-exhaustemissions, which largely affect coarse particulate matter. Fora full understanding of future air quality, these factors should beincluded.

Nevertheless, the model results presented here illustrate howvehicle populations, and hence emissions, respond to generaltrends in fuel usage and introduction of emission standards. Urbanand national air quality agencies are planning a cleaner vehicle eeteither by tightening emission standards or by switching to vehicleswithout end-use emission, including those powered by electricityor hydrogen.When these changes are implemented, themajority ofon-road emissions will come from the legacy eet, which willgovern the rate at which total emissions can be reduced. A goal ofthis work is therefore to explore how inertia in vehicle populationsleads to persistence of emissions. Once a vehicle is placed in service,the total emission corresponding to the service life of the vehicle isalmost guaranteed. Emissions may be delayed by reduced vehicleusage, but cannot be avoided unless the eet is subjected toaccelerated retirement or improved maintenance. Future work willevaluate potential emission reductions by policies that target thelegacy eet, in light of the uncertainties in population dynamics.

Acknowledgments

This workwas funded by the U.S. Department of Energy throughits operating contract with Argonne National Laboratory (DE-AC02-06CH11357) and by the Clean Air Task Force. We thank K. G. Duleepfor providing vehicle age distributions from six regions (Africa, Asia,EU, FSU, North America, and South America).

Appendix. Supplemental information

Supplemental information related to this article can be foundonline at doi:10.1016/j.atmosenv.2011.06.018.

References

Ban-Weiss, G.A., Lunden, M.M., Kirchstetter, T.W., Harley, R.A., 2009. Measurementof black carbon and particle number emission factors from individual heavy-duty trucks. Environmental Science and Technology 43, 1419e1424.

Bluett, J., Dey, K., Fisher, G., 2008. Assessing Vehicle Air Pollution Emissions. NIWAClient Report: CHC2008-001 available on-line at. http://www.environment.gov.au/atmosphere/airquality/publications/pubs/vehicle-pollution.pdf.

Bond, T.C., Bhardwaj, E., Dong, R., Jogani, R., Jung, S.K., Roden, C., Streets, D.G.,Trautmann, N.M., 2007. Historical emissions of black and organic carbon aerosol

F. Yan et al. / Atmospheric Enfrom energy-related combustion, 1850e2000. Global Biogeochemical Cycles 21,GB2018.Bond, T.C., Streets, D.G., Yarber, K.F., Nelson, S.M., Woo, J.H., Klimont, Z., 2004.A technology-based global inventory of black and organic carbon emissionsfrom combustion. Journal of Geophysical Research e Atmospheres 109, D14203.

Borken, J., Steller, H., Mertei, T., Vanhove, F., 2007. Global and country inventory ofroad passenger and freight transportation: fuel consumption and emissions ofair pollutants in year 2000. Transportation Research Record, 127e136.

Brownstone, D., Bunch, D.S., Train, K., 2000. Joint mixed logit models of stated andrevealed preferences for alternative-fuel vehicles. Transportation Research PartB e Methodological 34, 315e338.

Cadle, S.H., Mulawa, P.A., Hunsanger, E.C., Nelson, K., Ragazzi, R.A., Barrett, R.,Gallagher, G.L., Lawson, D.R., Knapp, K.T., Snow, R., 1999. Composition of light-duty motor vehicle exhaust particulate matter in the Denver, Colorado area.Environmental Science & Technology 33, 2328e2339.

Chen, C., Niemeier, D., 2005. A mass point vehicle scrappage model. TransportationResearch Part B e Methodological 39, 401e415.

Chen, C., Lin, J., 2006. Making an informed vehicle scrappage decision. TransportReviews 26, 731e748.

Chung, C.E., Ramanathan, V., Kiehl, J.T., 2002. Effects of the south Asian absorbinghaze on the northeast monsoon and surface-air heat exchange. Journal ofClimate 15, 2462e2476.

CONCAWE,1997. Motor Vehicle Emission Regulations and Fuel Specications, Part 2:Detailed Information and Historic Review (1970e1996) No. 6/97 Brussels, March.

Cooper, O.R., Stohl, A., Eckhardt, S., Parrish, D.D., Oltmans, S.J., Johnson, B.J.,Nedelec, P., Schmidlin, F.J., Newchurch, M.J., Kondo, Y., Kita, K., 2005.A springtime comparison of tropospheric ozone and transport pathways on theeast and west coasts of the United States. Journal of Geophysical Research eAtmospheres 110, 2462e2476.

Cox, D.R., 1972. Regression models and life-tables. Journal of the Royal StatisticalSociety Series B e Statistical Methodology 34, 187e220.

Davis, S.C., Diegel, S.W., Oak Ridge National Laboratory, 2008. Transportation EnergyData Book, twenty eighth ed. Oak Ridge National Laboratory, Oak Ridge, TN.

DeJong, G., 1996. A disaggregate model system of vehicle holding duration, typechoice and use. Transportation Research Part B e Methodological 30, 263e276.

DELPHI, 2009. Worldwide Emission Standards: Heavy Duty & Off-road Vehicles.http://delphi.com/pdf/emissions/Delphi-Heavy-Duty-Emissions-Brochure-2009.pdf accessed March 2010.

DELPHI, 2010. Worldwide Emission Standards: Passenger Cars & Light Duty Vehi-cles. http://delphi.com/pdf/emissions/Delphi-Passenger-Car-Light-Duty-Truck-Emissions-Brochure-2010-2011.pdf accessed March 2010.

Dieselnet, http://www.dieselnet.com/standards/, accessed May 2010.Durbin, T.D., Smith, M.R., Norbeck, J.M., Truex, T.J., 1999. Population density,

particulate emission characterization, and impact on the particulate inventoryof smoking vehicles in the South Coast Air Quality Management District. Journalof the Air & Waste Management Association 49, 28e38.

ESP, 2007. 2007 High Emitter Remote Sensing Project, Prepared for SoutheastMichigan Council of Governments available on-line at. http://library.semcog.org/InmagicGenie/DocumentFolder/HighEmissionsReport.pdf.

ETSAP, 2001. http://www.etsap.org/markal/main.html accessed May 2010.European Environment Agency (EEA), 2002. EMEP/CORINAIR Emission Inventory

Guidebook, third ed. January, available on-line at. http://www.eea.europa.eu/publications/technical_report_2001_3 accessed August 2010.

European Environment Agency (EEA), 2011. Average Age of Passenger Cars, Lightand Heavy Duty Trucks, Buses/Coaches and Two-Wheelers, 1995e2009 fromTremove V3.1 available on-line at. http://www.eea.europa.eu/data-and-maps/indicators/average-age-of-the-vehicle-eet/average-age-of-the-vehicle-3accessed March 2011.

Faiz,A.,Gautam,S., Burki, E.,1995.Air-pollution frommotor-vehiclese issuesandoptionsfor Latin-American countries. Science of the Total Environment 169, 303e310.

Fulton, L., Eads, G., 2004. IEA/SMP Model Documentation and Reference CaseProjections. http://www.wbcsd.org/web/publications/mobility/smp-model-document.pdf last time accessed in January, 2011.

Fiore, A.M., Jacob, D.J., Bey, I., Yantosca, R.M., Field, B.D., Fusco, A.C., Wilkinson, J.G.,2002. Background ozone over the United States in summer: origin, trend, andcontribution to pollution episodes. Journal of Geophysical Research e Atmo-spheres 107, 4275.

GAINS model, http://gains.iiasa.ac.at/index.php/home-page/241-on-line-access-to-gains, last time accessed in January, 2011.

Gilbert, C.C.S., 1992. A duration model of automobile ownership. TransportationResearch Part B e Methodological 26, 97e114.

Greenspan, A., Cohen, D., 1999. Motor vehicle stocks, scrappage, and sales. Reviewof Economics and Statistics 81, 369e383.

Hansen, A.D.A., Rosen, H., 1990. Individual measurements of the emission factor ofaerosol black carbon in automobile plumes. Journal of the Air & WasteManagement Association 40, 1654e1657.

Hausman, J., Mcfadden, D., 1984. Specication tests for the multinomial logit model.Econometrica 52, 1219e1240.

Intergovernmental Panel on Climate Change (IPCC), 2001. Climate Change 2001:The Scientic Basis. University Press, Cambridge.

International Energy Agency (IEA), 2009a. Energy Statistics of Non-OECD CountriesCD-ROM Paris, France.

International Energy Agency (IEA), 2009b. Energy Statistics of OECD Countries CD-ROM Paris, France.

InternationalRoadFederation,2003.WorldRoadStatistics.http://econ.worldbank.org/

ment 45 (2011) 4830e4844 4843WBSITE/EXTERNAL/EXTDEC/EXTRESEARCH/0,,contentMDK:20699572wpagePK:64214825wpiPK:64214943wtheSitePK:469382,00.html accessed June 2011.

Jaffe, D., Anderson, T., Covert, D., Kotchenruther, R., Trost, B., Danielson, J.,Simpson, W., Berntsen, T., Karlsdottir, S., Blake, D., Harris, J., Carmichael, G.,Uno, I., 1999. Transport of Asian air pollution to North America. GeophysicalResearch Letters 26, 711e714.

Jones, B.S., Branton, R.P., 2005. Beyond logit and probit: Cox duration models ofsingle, repeating, and competing events for state policy adoption. State Politics& Policy Quarterly 5, 420e443.

Kerr, S., Newell, R.G., 2003. Policy-induced technology adoption: evidence from theUS lead phasedown. Journal of Industrial Economics 51, 317e343.

Kim, S.H., Edmonds, J., Lurz, J., Smith, S.J., Wise, M., 2006. The (OECTS)-E-bjframework for integrated assessment: hybrid modeling of transportation.Energy Journal, 63e91.

Kirchstetter, T.W., Agular, J., Tonse, S., Fairley, D., Novakov, T., 2008. Black carbonconcentrations and diesel vehicle emission factors derived fromcoefcient of hazemeasurements in California: 1967e2003. Atmospheric Environment 42, 480e491.

Lawson, D.R., 1993. Passing the test e human-behavior and California smog checkprogram. Journal of the Air & Waste Management Association 43, 1567e1575.

Levy II, H., Shindell, D., Gilliland, A., Horowitz, L.W., Schwarzkopf, M.D. (Eds.), 2008.Climate Projections Based on Emissions Scenarios for Long-lived and Short-lived Radiatively Active Gases and Aerosols. U.S. Climate Change ScienceProgram Synthesis and Assessment Product 3.2.

Manski, C.F., Goldin, E., 1983. An econometric-analysis of automobile scrappage.Transportation Science 17, 365e375.

McCormick, R.L., Graboski, M.S., Alleman, T.L., Alvarez, J.R., Duleep, K.G., 2003.Quantifying the emission benets of opacity testing and repair of heavy-duty

Streets, D.G., Bond, T.C., Lee, T., Jang, C., 2004. On the future of carbonaceous aerosolemissions. Journal of Geophysical Research e Atmospheres 109, D24212.

Streets, D.G., Yan, F., Chin, M., Diehl, T., Mahowald, N., Schultz, M., Wild, M., Wu, Y.,Yu, C., 2009. Anthropogenic and natural contributions to regional trends inaerosol optical depth, 1980e2006. Journal of Geophysical Research 114,D00D18.

Subramanian, R., Winijkul, E., Bond, T.C., Thiansathit, W., Oanh, N.T.K., Paw-Armart, I., Duleep, K.G., 2009. Climate-relevant properties of diesel particulateemissions: results from a piggyback study in Bangkok, Thailand. EnvironmentalScience & Technology 43, 4213e4218.

Tao, Z.N., Williams, A., Donaghy, K., Hewings, G., 2007. A socio-economic method forestimating future air pollutant emissions e Chicago case study. AtmosphericEnvironment 41, 5398e5409.

Ubanwa, B., Burnette, A., Kishan, S., Fritz, S.G., 2003. Exhaust particulate matteremission factors and deterioration rate for in-use motor vehicles. Journal ofEngineering for Gas Turbines and Power e Transactions of the ASME 125,513e523.

Uherek, E., Halenka, T., Borken-Kleefeld, J., et al., 2010. Transport impacts onatmosphere and climate: land transport. Atmospheric Environment 44,4772e4814.

UNEP, 2008. Middle East, West Asia and North Africa Vehicle Standards and Fleets.http://www.unep.org/pcfv/PDF/MatrixMENAWA-VEHSJun08.pdf accessedMarch 2010.

United States Environmental Protection Agency (U.S. EPA), 2005a. Calculation ofAge Distributions in the Nonroad Model: Growth and Scrappage. EPA 420-R-

F. Yan et al. / Atmospheric Environment 45 (2011) 4830e48444844diesel vehicles. Environmental Science & Technology 37, 630e637.Menon, S., Hansen, J., Nazarenko, L., Luo, Y.F., 2002. Climate effects of black carbon

aerosols in China and India. Science 297, 2250e2253.Maricq, M.M., Podsiadlik, D.H., Chase, R.E., 1999. Gasoline vehicle particle size

distributions: comparison of steady state, FTP, and US06 measurements. Envi-ronmental Science & Technology 33, 2007e2015.

Nakicenovic, N., 2000. Special Report on Emissions Scenarios: a Special Report ofWorking Group III of the Intergovernmental Panel on Climate Change. Cam-bridge University Press, Cambridge, New York.

Ntziachristos, L., Samaras, Z., 2001. An empirical method for predicting exhaustemissions of regulated pollutants from future vehicle technologies. Atmo-spheric Environment 35, 1985e1999.

Oliver, H.H., 2008. In-use Vehicle Emissions in China e Tianjin Study. Discussionpaper 2008-08 Harvard Kennedy School.

Parks, R.W., 1977. Determinants of scrapping rates for postwar vintage automobiles.Econometrica 45, 1099e1115.

Parks, R.W., 1979. Durability, maintenance and the price of used assets. EconomicInquiry 17, 197e217.

Pujadas, M., Nunez, L., Plaza, J., Bezares, J.C., Fernandez, J.M., 2004. Comparisonbetween experimental and calculated vehicle idle emission factors for Madrideet. Science of the Total Environment 334, 1