Science of the Total Environment 334 335 (2004) 287 298

www.elsevier.com/locate/scitotenv

An uncertainty analysis of air pollution externalities from road

transport in Belgium in 2010
L. Int Panis *, L. De Nocker, E. Cornelis, R. Torfs
Vito, Flemish Institute for Technological Research Boeretang 200, B-2400 Mol, Belgium
Accepted 1 April 2004

Abstract
Although stricter standards for vehicles will reduce emissions to air significantly by 2010, a number of problems will
remain, especially related to particulate concentrations in cities, ground-level ozone, and CO2. To evaluate the impacts of
new policy measures, tools need to be available that assess the potential benefits of these measures in terms of the
vehicle fleet, fuel choice, modal choice, kilometers driven, emissions, and the impacts on public health and related external
costs.
The ExternE accounting framework offers the most up to date and comprehensive methodology to assess marginal
external costs of energy-related pollutants. It combines emission models, air dispersion models at local and regional scales
with dose response functions and valuation rules. Vito has extended this accounting framework with data and models
related to the future composition of the vehicle fleet and transportation demand to evaluate the impact of new policy
proposals on air quality and aggregated (total) external costs by 2010.
Special attention was given to uncertainty analysis. The uncertainty for more than 100 different parameters was
combined in Monte Carlo simulations to assess the range of possible outcomes and the main drivers of these results.
Although the impacts from emission standards and total fleet mileage look dominant at first, a number of other factors
were found to be important as well. This includes the number of diesel vehicles, inspection and maintenance (high-emitter
cars), use of air conditioning, and heavy duty transit traffic.
D 2004 Elsevier B.V. All rights reserved.
Keywords: Environmental external costs; Monte Carlo; Uncertainty analysis; Modal split; Air pollution impacts; Scenario

1. Introduction
In the most recent ExternE project, the impact
pathway methodology for the calculation of energyrelated externalities was updated and extended spe-

* Corresponding author. Tel.: +32-14-33-58-87; fax: +32-1432-11-85.

E-mail address: luc.intpanis@vito.be (L. Int Panis).
0048-9697/$ - see front matter D 2004 Elsevier B.V. All rights reserved.
doi:10.1016/j.scitotenv.2004.04.073

cifically for the application to transportation problems (Friedrich and Bickel, 2001; Int Panis and De
Nocker, 2001). Several studies have already used
this new framework for the calculation of marginal
costs of air pollution for different transportation
modes and vehicles (e.g. Int Panis et al., in
press,2000b, 2001a,b). Studies of marginal externalities (on a per kilometre or per litre of fuel basis),
have provided information that proved useful to
specific policy options (fuel choice, technological

288

L. Int Panis et al. / Science of the Total Environment 334 335 (2004) 287298

comparisons, excises). But policy makers generally

prefer information at a higher level of aggregation
(e.g., for an entire fleet or at the national level).
Aggregated results are more useful to assess the
significance of air pollution and compare it to other
monetary data (e.g., GDP, avoidance costs, budgets,
etc.).
Studies providing aggregated estimates of the
total air pollution costs from transport at the national
level are scarce because of the lack of necessary
data for the aggregation (e.g., mobility and traffic
statistics). In addition, lack of comparable data
makes international comparisons very difficult (Friedrich and Bickel, 2001). Uncertainty about the data
and implicit assumptions can induce apparent differences between countries (artefacts) with respect to
the height of the total costs (Int Panis et al., in
press).
In addition, comparisons have been made of the
evolution of total externalities based on historical data
sets (Int Panis et al., 2001a,b). This has revealed that
subtle changes in fleet composition and mobility
statistics can significantly affect the observed trend
of externalities.
Despite these problems, policy makers often demand that historical data sets are extrapolated to the
future in order to study the evolution of mobility
statistics, emissions (VMM, 2000), or externalities
(Int Panis et al., in press). When building scenarios
to estimate future impacts, it is common to make
assumptions about the evolution of a range of
statistics. This is no problem if this directly concerns
the statistic to be studied and when assumptions are
clearly stated. But in any study, lack of data or time
dictates that a number of parameters are kept constant. These parameters are often considered unimportant, but this is an implicit assumption that should
be tested. Preliminary Monte Carlo testing for passenger cars in Belgium has demonstrated that uncertainty may be larger than anticipated and can be
attributed to parameters that were overlooked previously or thought to be unimportant (Int Panis et al.,
2001b). In the study presented here, we quantify
how large the uncertainty on estimates of road
transport externalities is for the year 2010 and which
parameters are most important for the different
modes (passenger cars, heavy duty trucks, buses,
and motorcycles).

2. Methodology
2.1. The ExternE methodology
The European ExternE project provides an accounting framework based on an impact pathway
methodology. It basically follows a pollutant from its
emission until it causes an impact or damage. It
allows monetization of environmental damage costs
from air pollution for a specific technology and
trajectory. It estimates in detail the generation of
emissions, their atmospheric dispersion, the exposure
of man and environment to the pollutants, and the
resulting impacts on public health, agriculture, and
buildings. More detailed descriptions of this methodology can be found in Friedrich and Bickel
(2001).
Marginal externalities are dominated by the public
health impacts from emissions of PM2.5, especially
important for diesel-fuelled vehicles and in urban
traffic and from nitrate aerosols (following emissions
from NOx). There is a large uncertainty on these
results, as uncertainties in different steps are accumulated. This uncertainty is documented in Rabl and
Spadaro (1999, 2001) and studied in detail in Int
Panis et al. (2002) but is not further discussed here.
In this paper, we focus on the additional uncertainty
resulting from the aggregation of the site and trajectory specific data in a total air pollution cost for an
entire country.
2.2. Uncertainty calculations
Monte Carlo analysis is a common name for a
group of iterative statistical techniques. In this paper,
we study the propagation of errors through complex
calculations. The crisp estimates of every parameter
in the calculation are replaced by a probability
distribution that describes the range of values that
the parameter can take, as well as the probability
that a certain value will actually occur. This procedure is then repeated a large number of times
(typically between 1000 and 10,000 trials) so that
a large number of combinations of different input
parameters occur. All calculations for this study
were performed with the commercially available
Crystal Balln software of Decisioneering on a
desktop computer.

L. Int Panis et al. / Science of the Total Environment 334 335 (2004) 287298

The result of a Monte Carlo run is an equally large

number of estimates for the result, which can also be
displayed as a probability distribution. This highlights
the fact that the final estimate is uncertain. The same
technique can also be used to identify the relative
importance of the different input parameters. This was
achieved by looking at the rank-correlation between
individual parameters and the calculation results.
The fundamental problem is therefore the selection
of probability distributions for all input parameters
(Int Panis et al., 2001b). It may look impossible at first
to describe the type of distribution, the standard
deviation, or even the range when hardly any information on the central values is available. For most
parameters, this is indeed a very daunting task, which
involves a careful study of the literature and the
available estimates. In some cases, the selection of a
particular distribution may seem arbitrary but is always well documented, and (paradoxically) it provides more information than the point estimates.
In this study, we use a Monte Carlo technique to
study the parametric uncertainty of aggregating marginal external costs of all motorized road transportation modes to the national level. This does not include
the uncertainty of the marginal estimates on which it
is based. For a discussion of these uncertainties, the
reader is referred to Rabl and Spadaro (1999, 2001)
and Int Panis et al. (2002).

289

(De Vlieger I., personal communication). These distributions also reflect the business as usual scenario
which takes into account all policy options and
technical evolutions that had been decided on 1/1/
2002. Evidently, results can change significantly
when other scenarios are applied.
2.3.1. Fleet size of motorcycles
In the minimal scenario, we assume a yearly 10%
reduction of new motorcycles (this is an extrapolation
of the 1999 2000 evolution). The average scenario
keeps the number of new motorcycles sold constant at
the 2000 level. The maximal scenario extrapolates the
increase from the observed 1991 2000 trend.
2.3.2. Fleet size of mopeds (two-stroke engines)
The minimal scenario extrapolates the yearly decrease from the observed historical (1990 1998)
trend. The central estimate is an average growth equal
to the 1993 1998 average. The maximal scenario
assumes a continuous growth extrapolated from scooter sales in the 1990 1998 period.

2.3. Selection of probability distributions

2.3.3. Fleet size of passenger cars and trucks

Experts at Vito think that it is unlikely that our
estimates for fleet size (in 2010) would be wrong by
more than 5% for passenger cars and small trucks and
7% for large trucks. Therefore, standard deviations
(S.D.) of 2.5% and 3.5% were adopted to describe
uncertainty.

Input parameters for motorcycles and mopeds were

derived from the papers by Cornelis (2001) and
Cornelis et al. (2002). Input parameters for passenger
cars are the same as in Int Panis et al. (2001b), except
for some modifications detailed below. In addition,
input parameters for buses and trucks were constructed based on information from experts at Vito.
All input parameters, their central value, range, and
assumed distribution are listed in Tables 1 3 for
motorcycles, buses, and trucks, respectively. For each
parameter, a source or reference is given as well as a
short motivation for the selection of the range and/or
distribution. Some of the most important parameters
and assumptions are discussed below. The distributions for trucks are only first estimates that will be
improved in ongoing studies when the parameter
under study turns out to have a significant effect

2.3.4. High-emitting passenger cars

Evidence has been accumulating over the last
years that a small number of cars in the fleet may
have very high emissions caused by some kind of
malfunction (e.g., in the catalytic converter; Sjodin et
al., 2000). Unfortunately, we have no data on the
number of malfunctions in the Belgian fleet or the
effect on emissions. Therefore, it was never taken into
account in earlier estimates of externalities. Nevertheless, it seems clear that even a small number of
high-emitter cars can significantly increase the total
emission of the fleet. A sensitivity analysis is therefore warranted. The fraction of high-emitter cars was
estimated between 0.2% and 0.5% of the fleet based
on European and U.S. studies. A lognormal distribution was chosen to avoid negative numbers. Correction factors for VOC emissions from petrol cars

290

L. Int Panis et al. / Science of the Total Environment 334 335 (2004) 287298

Table 1
The central estimate and distribution of all statistics used in the 2010 aggregation for motorcycles
Parameter

Estimate

Source

S.D.; 90%-ile; range

Distribution

Motivation

431.742
532.000
36%

[*]
[*]
[*]

300261 538981
350.874 750.988
23 44%

Triangular
Triangular
Custom

47%
17%
4%
14%

[*]
[*]
[**]
[**]

41 56%
15 20%
2 6% (90%-ile)
14 19% (90%-ile)

Custom
Custom
Normal
Uniform

35%

[**]

35 40% (Range)

Uniform

See text
See text
Survival curves compiled
from observations

No data. sensitivity analysis

Tends to increase. sensitivity
analysis, max 5% increase
Tends to increase. sensitivity
analysis, max 5% increase

0%

[*]

0.1% (S.D.)

Normal

No data. sensitivity analysis

4.580 km/year
5.176 km/year

[**]
[*], [**]

345 (S.D.)
3952 6400

Normal
Uniform

60%

[**], [***]

55 65%

Triangular

70%

[**], [***]

65 75%

Triangular

125%

[**], [***]

120 130%

Triangular

Cornelis 2002
Every expert opinion has the
same probability
No data sensitivity analysis,
max 5% deviation
No data sensitivity analysis,
max 5% deviation
No data sensitivity analysis,
max 5% deviation

50% Urban

[*],[**]

40 60% (Range)

Triangular

No data. sensitivity analysis

50% Rural
30% Urban

[*], [**]

20 35%

Triangular

No data. sensitivity analysis

Highway peak traffic

Rural peak traffic
Urban peak traffic

50% Rural
0%
0%
0%

[*], [**]
[*], [**]
[*], [**]

40 60%
0 20%
0 10%
0 30%

Triangular
Triangular
Uniform

No data. sensitivity analysis

No data. sensitivity analysis
No data. sensitivity analysis

Emissions factors
CO

6%

[**]

94 106%

Normal

VOC
PM
NOx
Fuel consumption (CO2 SO2)

10%
15%
20%
8%

[**]
[**]
[**]
[**]

90 110%
85 115%
80 120%
92 108%

Normal
Normal
Normal
Normal

Rough approximation based

on sensitivity to different speed
scenarios of Cornelis (2001)

Fleet parameters
Fleet-size four-stroke vehicles
Fleet-size two-stroke vehicles
Fleet size uncontrolled
(scrappage of old vehicles)
Fleet size stage 1
Fleet size stage 2
Fraction of two-stroke >50 cc
Fraction < 250 cc
Fraction >750 cc

Mileage parameters
Average mileage growth
(individual motorcycle)
Average mileage of two strokes
Average mileage of four strokes
Two-stroke >50 cc mileage
fraction of average
Four-stroke < 250 cc mileage
fraction of average
Four-stroke >750% mileage
fraction of average
Mobility parameters
Geographical mileage distribution
two-stroke
Geographical mileage distribution
four-stroke

[*] Cornelis E., 2002, personal communication; [**] Cornelis (2001), Cornelis et al. (2002); [***] Andre, 1999.

(times 10 to times 40; uniform distribution) and PM

emissions were compiled from a range of available
measurements. Possible increases range from 250% to
19,300%) for petrol cars and from 200% to 8800%

for diesels. The lognormal distribution was chosen to

avoid negative solutions, and because we feel that the
higher emission ranges are probably very rare even
among high-emitter cars. The addition of these

L. Int Panis et al. / Science of the Total Environment 334 335 (2004) 287298

291

Table 2
The central estimate and distribution of all statistics used in the 2010 aggregation for buses
Parameter

Estimate

Source

S.D.; %-ile; range

Distribution

Motivation

Fleet parameters
Fleet-size city buses
Fleet-size coaches

8.306
5.962

[*]
[*]

208 (S.D.)
149 (S.D.)

Normal
Normal

Maximum deviation 5%
Maximum deviation 5%

Mileage parameters
Average mileage growth (per vehicle, city bus)
Average mileage growth (per vehicle, coach)
Yearly mileage of city bus (basis 1995)
Yearly mileage of coach (basis 1995)

0%
0.7%
46.240 km
23.750 km

[*]
[*]
[*]
[*]

0.1 % (S.D.)
0.1 % (S.D.)
45,315 47,165 km
22,563 24,938 km

Normal
Normal
Normal
Normal

Sensivity analysis
Sensivity analysis
Relative error 2%
Relative error 5%

Mobility parameters
Geographical mileage distribution (city bus)

61% Urban

[*], [**]

56.6 65%

Normal

Sensitivity analysis based

on two different studies

[*], [**]

20 30%

Triangular

Sensitivity analysis

[*]
[*]
[*]

15 45%
10 30%
5 20%

Triangular
Triangular
Uniform

Sensitivity analysis
Sensitivity analysis
Sensitivity analysis
(same as PC)
Sensitivity analysis
(same as PC)
Sensitivity analysis

Urban peak traffic (large cities)

Urban peak traffic (small cities)
Highway peak traffic

38% Rural
2% Highway
25% Urban
38% Rural
38% highway
30%
20%
12.5%

Rural peak traffic

10%

[*]

2 10%

Uniform

Cold-start fraction

1%

[*]

5% 15%

Uniform

Geographical mileage distribution (coach)

[*] De Vlieger and Cornelis TEMAT BeBAUHDP scenario, [**] Smer Trend HDP scenario.

parameters was the only change that was made to the

parameters given in Int Panis et al. (2001b).
2.3.5. Foreign trucks
Belgium is a small country with a central location in
Europe and several harbours and industrial clusters.
The amount of transit freight transport and the number
of foreign trucks on Belgian roads is therefore very
high. Apparently, no statistics are kept on the numbers
or origin of foreign trucks, let alone on their technical
attributes or route choice. To provide a first estimate of
the magnitude of impacts from foreign trucks, we have
used European statistics on the total haulage on Belgian
territory and the haulage by vehicles registered in
Belgium. This provided us with a very variable estimate for the 1994 1998 period, in which about half of
the ton
kilometers were attributed to foreign trucks.
Additionally, we have assumed that all transport by
foreign vehicles was made with large trucks, that the
fleets composition is the same otherwise, and that they
only use highways and rural roads. These assumptions

may not be correct [e.g., arguments made by Int Panis

et al. (in press,2000b)], but they are unlikely to overestimate the estimated impacts (http://europa.eu.int/
comm/energy_transport/etif/list_of_tables.html).

3. Results
3.1. The total external cost per vehicle class
3.1.1. Motorcycles
The mean impact from mopeds and motorcycles is
estimated at 41.6 million Euro and 42.4 million Euro,
respectively, for the year 2000. The 95% confidence
interval for the total (84 million) is between 66 and
104 million Euro.
In 2010, this impact could rise to 102 (71 137)
million Euro. The result from the Monte Carlo procedure is shown in Fig. 1 and Table 4. This mode is
the only one in road transport for which externalities
are expected to increase. This increase suggests that

292

L. Int Panis et al. / Science of the Total Environment 334 335 (2004) 287298

Table 3
The central estimate and distribution of all statistics used in the 2010 aggregation for Heavy-duty vehicles (freight)
Parameter

Estimate

Fleet parameters
Fleet size
3.5 7.5
7.5 16
16 32
32 40

27,909
34,120
92,242
143

Mileage parameters
Average mileage growth
(per vehicle, until 2005)
Average mileage growth
(per vehicle, after 2005)
Fraction foreign trucks in
haulage on national roads
Yearly mileage (basis 1995)

3.5 7.5
7.5 16
16 32
32 40

Source

S.D.; %-ile; range

Distribution

[***]

90%-ile
26514 29304
32414 35826
85785 98699
133 153

Normal

Motivation

Maximum
Maximum
Maximum
Maximum

deviation
deviation
deviation
deviation

2.2984%

[*]

0.5 % (S.D.)

Normal

Sensitivity analysis

1.7984%

[*]

0.5 % (S.D.)

Normal

Sensitivity analysis

128%

[*]

99 156%

Normal

[*]

90%-ile

Normal
(10% deviation
possible but unlikely)

Derived from European

statistics
Possible deviation 5 10%

31.138 km
35.290 km
40.674 km
38957 km

29581 32695
33526 37055
38640 42708
37009 40905

km
km
km
km

Mobility parameters
Parameter

Estimate

Source or reference S.D., 90%-ile or range Distribution

[**]

Motivation

Urban mileage fraction

3.5 7.5

30%

7.5 16
16 32
32 40
Urban peak traffic (large cities)

20%
15%
10%
30%

[*]

5 30%

Triangular

Highway peak traffic

20%

[*]

5 20%

Triangular

Rural peak traffic

10%

[*]

2 10%

Triangular

Cold-start fraction urban

4%

[*]

3 5%

Uniform

Cold-start fraction rural

Cold-start fraction highway

11%
2%

[*]
[*]

7 15%
0 3%

Uniform
Triangular

Sensitivity analysis
(same as PC)
Sensitivity analysis
(same as PC)
Sensitivity analysis
(same as PC)
Based on 10 20
km/working day

Sensitivity analysis

90%
0.1%

80 95%
0 0.1%

Triangular
Triangular

Sensitivity analysis
Sensitivity analysis

2.5%

0 5%

Triangular

Sensitivity analysis

10%

5 15%

Traingular

Sensitivity analysis

25%

20 30%

Triangular

Sensitivity analysis

Other parameters
Efficiency of PM filters
Penetration of advanced PM
reduction techniques in Euro2
Penetration of advanced PM
reduction techniques in Euro3
Penetration of advanced PM
reduction techniques in Euro4
Penetration of advanced PM
reduction techniques in Euro5

5%
5%
7%
7%

Normal
Sensitivity analysis
(cfr., passenger cars)
10% points

[*] De Vlieger and Cornelis, personal communication, [**] SMER, Trend scenario HDG, [***] TEMAT BeBAU HDG 2010.

L. Int Panis et al. / Science of the Total Environment 334 335 (2004) 287298

293

Fig. 1. Frequency distribution of motorcycle and moped externalities in 2010.

technical improvements of the fleet cannot keep up

with the expected growth of the fleet. Nevertheless,
uncertainty is so large that better estimates must
become available. Without them, reliable studies of
(the ongoing) modal shift towards motorcycles are
impossible. In addition, we have not taken into
account the recent EC decision of new emission limits
to come into effect by 2006. Nevertheless, we anticipate that even by 2010, externalities will still be
dominated by older uncontrolled and stage 1 motorcycles (about 75% of the total), so that the effect of
more stringent emission standards for new vehicles
will only reach is desired effect well after 2010.

fraction of high-emitter vehicles in the Belgian fleet

increases the average only slightly, but the upper
confidence limit is clearly much higher.
Compared to an estimate of 1.5 million Euro for the
year 2000, this implies a predicted decrease of externalities of more than 50% over the current decade. This
is mainly due to the replacement of uncontrolled cars
(which still dominate impacts in the year 2000) by
more advanced vehicles complying with the more
stringent set of Euro emission standards. This trend
is counteracted by the mileage growth and the increasing number of passenger cars, especially the diesel
segment of the fleet (Int Panis et al., 2001a).

3.1.2. Passenger cars

Earlier preliminary estimates for passenger cars
yielded an average outcome for the year 2010 of
685 (619 754) million Euro (Int Panis et al.,
2001b). The current estimate is 721 (637 937) million Euro. Taking into account the presence of a small

3.1.3. Buses
The total air pollution externalities amount to 39
million Euro for Belgian buses in the year 2010, which
is fairly small compared to other modes. The 95%
confidence interval on the calculation is very narrow at
36 43 million Euro. Most of the externalities (85%)
can be attributed to city buses. Externalities from
coaches are much lower, because they are used more
often on highways and even in rural areas where
exposure of people to their emissions is low. City buses
on the other hand have higher emissions because of
their highly dynamic driving pattern. They are mostly
used in urban or densely populated places so that their
effect on public health impacts is much larger.
We can compare the 2010 total with the estimate
for 2000 (138 million Euro) and calculations for

Table 4
External costs per vehicle class in million Euro
Vehicle class

2000

2010

95% Confidence
interval for 2010

Motorcycles
Passenger cars
Buses
Trucks
Total

84
1541
138
999
2821

102
721
39
603
1468

71 137
637 937
36 43
516 702
1340 1680

Change
2000/2010
+ 21%
53%
72%
40%
48%

294

L. Int Panis et al. / Science of the Total Environment 334 335 (2004) 287298

Fig. 2. Relative contribution of different vehicle classes to road transport externalities.

historical data (170 million Euro in 1998; Int Panis et

al., in press,2000b, 2001a,b). Damages from bus
transport are therefore thought to decrease dramatically over the next 10 years. This can easily be
explained by the fact that in the BAU scenario on
which our calculations are based, fleet size and
average mileage both show a slow decreasing trend.
Technological innovations are therefore not offset by
traffic growth as in motorcycles or passenger cars.
The most important innovation is the recent decision
of the Flemish government to retrofit most public
buses with particulate filters. This approximately
reduces externalities with 4% in 2010. Most beneficial
effects will however occur later in time, after uncontrolled and Euro1 buses are phased out.
On the other hand, government measures have
succeeded in dramatically increasing ridership on city
buses in Flanders over the last year. This recent
evolution is not reflected in the unlying scenario
assumptions and cannot be maintained for a long time
without increasing the supply (additional rides) and
eventually adding new vehicles. Whereas, this recent
evolution may potentially increase externalities from
buses, the contribution of buses to total transport
externalities is so low that its effect on total transport
externalities will be small.
3.1.4. Trucks
Heavy duty vehicles for freight transport will become even more important in road transport externalities by 2010. Damages will amount up to 603 million
Euro, but the 95% uncertainty interval is fairly wide

(516 702 million Euro). This result is not directly

comparable to older estimates, because none of these
have taken into account the impacts caused by foreign
trucks on Belgian roads. The current estimate for the
year 2000 is 999 million Euro. Although the estimate
presented here should be regarded as a sensitivity
analysis only, the potential effect is overwhelming
(up to 1/3 of impacts may be caused by foreign trucks).
3.2. Total external costs for Belgian road transport
Total externalities for the Belgian road transport
sector in 2010 are estimated to add up to 1.5 billion
Euro with a 95% confidence interval for the aggregation of 1.3 1.7 billion Euro. The main damages come
from passenger cars and trucks, with the relative share
of the latter increasing over time (Fig. 2). The contribution of motorcycles, mopeds, city buses, and
coaches is relatively unimportant and much smaller
than the uncertainty interval on the total estimate.
Older studies estimated the total for the year 1998
at 2.4 billion Euro, but these did not take into account
the effect of high-emitter cars, foreign trucks, or
motorcycles. With the methodology presented here,
the estimate for 2000 stands at 2.8 billion Euro.

4. Discussion
To study which of the parameters are the most
important contributors to the variation, we have calculated rank correlations which are a measure of the

L. Int Panis et al. / Science of the Total Environment 334 335 (2004) 287298

sensitivity of the result to changes in one of the

parameters. 10,000 trials were used to obtain a stable
ranking of the contribution of the variables.
4.1. Motorcycles
Three issues dominate the large total uncertainty
range for motorcycles. The very poor knowledge of
mileages is the most important factor. The uncertainty about the average speeds has a large effect on the
PM emissions and hence on the total uncertainty.
The third factor is the poorly known number of
mopeds. Official registration was ended in 1987,
and sales have known important fluctuations in
recent years, which makes extrapolation a perilous
task. Powered two-wheelers are therefore the most
uncertain mode in this analysis, but this can also be
partly attributed to the painstakingly detailed study
of literature by Cornelis (2001) and Cornelis et al.
(2002).
4.2. Passenger cars
Uncertainty for passenger cars is dominated by
three parameters that are not generally taken into
account when constructing scenario extrapolations:
high-emitting vehicles, the evolution of the diesel
fleet segment, and the additional emissions from air
conditioning use.
We have no data on the number of high-emitting
passenger cars in the Belgian fleet. The data for
passenger cars were compiled from Swedish and
U.S. estimates. No data were found on high emitters
for other vehicle classes. Despite their very low
numbers, there is a general awareness that better
inspection and maintenance may be a very costeffective measure in lowering emissions in technologically advanced fleets (De Vlieger et al., 2002).
Emissions rates have become very low, and malfunctions can potentially increase emissions by an order
of magnitude or even more. The average effect
(multiplying a small number of vehicles with a large
increase in emissions) is significant in the total sum,
but because both are inherently poorly known, the
contribution to uncertainty is very high. In response
to this finding, we see that attempts are being made
to better quantify the parameter, e.g., by the use of
OBD.

295

Although this is very difficult to predict (Lantz, in

press), an accurate prediction of the petrol/diesel ratio
in the 2010 fleet is very important. It is the second
most important parameter in the estimate for passenger cars. This is related to the fact that external costs
of PM2.5 in ExternE are dominated by severe health
endpoints. Diesels have much higher PM2.5 emissions
than petrol-fuelled cars. Calculations that do not take
the expected increase of the number of diesel cars into
account give much lower estimates for PM2.5 emissions and hence for health-related impacts. The high
uncertainty of this parameter is typical of Belgian
conditions and may be less important elsewhere.
Extra emissions generated when air conditioning
systems are switched on seem to be potentially very
important. Although interest is growing in this field,
few data are available. On the basis of two small-scale
studies, the most recent MEET report concluded that
there remain too many uncertainties to propose quantitative factors that take the use of air conditioning
into account. Some pollutants are emitted more or less
depending on the conditions (EC, 1999).
Nevertheless, if we include the extra emissions
based on Van den Brink et al. (2000), it turns out that
these are more important (in determining uncertainty)
than any of the parameters describing mobility. The
main effect is probably due to the supposedly increased PM emissions from diesels in urban driving.
Interestingly, the correlation is small with the number
of cars that actually have an air conditioning system.
Although this is obviously an important parameter,
estimates of air conditioning penetration in the fleet
are much more reliable than estimates of air conditioning usage and its possible effect on emissions.
Other important parameters for passenger cars are
discussed in Int Panis et al. (2001b).
Most of the approximately 100 variable parameters
in the analysis are unimportant. They have a very low
correlation with the final result. For some parameters,
the low correlation may come as a surprise at first.
The result for passenger cars for example is
nearly independent of the number of cars built before
2000 (i.e., already on the road today). Despite the
fact that this fleet segment contains some very old
cars (with very high emissions), their numbers (and
mileage) in 2010 will be so low that even if our
prediction is very poor, this will not affect the
outcome too much.

296

L. Int Panis et al. / Science of the Total Environment 334 335 (2004) 287298

The annual average mileages of the different vehicle classes, such as mopeds, motorcycles, small
trucks, and LPG passenger cars are generally thought
to be poorly known. But their contribution to air
pollution is low, either because of their low emissions
and/or marginal share in the (total) fleet of road
vehicles. A similar argument can be made for petrol
cars. Although they form a major part of the fleet, the
range of different mileage estimates from literature is
much smaller (typically less than 10%), and their
contribution to external costs is marginal compared
to diesel-fuelled vehicles.
4.3. Buses and coaches
Uncertainty in the estimate for buses is dominated
by lack of data on city bus traffic. The main factor is the
mileage fraction which is travelled in densely populated areas. There are no usable statistics available that
distinguish between public transport in rural or urban
areas. The second and third most important factors are
the uncertainty about the fleet size and mileage in 2010,
which clearly depend on policy decisions and are
therefore difficult to model. Finally, cold-start emissions prove to be especially important. On the other
hand, knowing the efficiency of particulate filters

precisely is not very important, because the introduction scenario for public buses is exactly known.
4.4. Trucks
Two parameters dominate the uncertainty on the
estimate for heavy-duty vehicles: the unknown growth
rate of the haulage industry and the contribution of
foreign trucks.
The expected increase in mileage by trucks is very
hard to predict, even from available historical data. It
was already demonstrated before that, even small,
differences in assumptions make international comparisons very difficult (Int Panis et al., in press). In
scenario analyses, this poses no problem because this
is usually one of the parameters that determine the
different policy options. The effect of foreign trucks
on the other hand is much harder to correct for. It is a
problem that is probably more important in the
Benelux region than in other European countries.
Despite the conservative assumptions, impacts may
be large. It is absolutely necessary that statistics are
collected on the use of Belgian roads by foreign
trucks, especially those that may not conform to
European emission legislation (cfr., the high emitter
argument made for passenger cars).

Fig. 3. Rank correlations between all parameters and the total external costs in a Monte Carlo run with 10,000 trials.

L. Int Panis et al. / Science of the Total Environment 334 335 (2004) 287298

4.5. Uncertainty on the total damage costs

Because the total external cost is dominated by
impacts from passenger cars and trucks, parameters
important in these estimates will also be important for
the total. Impacts from buses and powered two-wheelers are much smaller and, although they may be very
uncertain, contribute only marginally to the uncertainty of the total. The 10 most important parameters are
shown in Fig. 3.
Quite surprisingly, we find that despite our very
conservative assumptions, two parameters that are
new in this study are very important in the overall
uncertainty: the PM emission rates of high-emitter
diesel passenger cars and the contribution of foreign
trucks. Some of the other parameters had already been
identified in previous studies: the uncertain growth
rate of road haulage, the fraction of diesels in passenger car sales, and additional emissions from the use of
air conditioning systems.

5. Conclusions
Monte Carlo analysis is a convenient tool to
analyse the propagation of errors in the aggregation
of marginal impacts.
Even if accurate data are not available for the
construction of probability distributions, the analysis
reveals which estimates should preferably be improved
to obtain a better estimate of the dependant variable.
The most important factors that contribute to the
uncertainty of our estimate for 2010 are the fraction
high-emitter diesel passenger cars, the possible contribution of air conditioning, and the impacts of
foreign trucks. All three have been neglected in a
number of recent studies.
Additional studies that provide more accurate estimates for these parameters can reduce the uncertainty
of aggregated external cost estimates.
The additional uncertainty that arises from the
aggregation of marginal externalities seems small
when compared to the uncertainties associated with
emissions, atmospheric modelling, epidemiology, and
monetary valuation (Int Panis et al., 2002; Rabl and
Spadaro, 1999, 2001).
Additional uncertainty is large when compared to
the expected effect of local policies. Most of the

297

decreasing trend can be attributed to European standards adopted during the previous decade.
Despite the uncertainty of the value of environmental externalities, we conclude that the amount is high
and will remain significant in 2010. Reducing impacts
will therefore remain important to policy makers.

Acknowledgements
This paper is based on work cofinanced by the
Belgian Federal Office for Scientific, Technical and
Cultural Affairs (OSTC). The authors want to thank
Mrs. I. De Vlieger for commenting on an earlier
version of this paper.

References
Andre M. Driving statistics for the assessment of pollutant emissions from road transport. INRETS France: Bron, 1999.
Cornelis E. Feasibility of reporting environmental impacts of motorized two-wheelers (De Haalbaarheid van het in kaart brengen
van milieu-impacts van brom- en motorfietsen). Eindwerk, gespecialiseerde studies milieudeskundige Brussel: Vrije Universiteit, 2001 (86 pp; in Dutch).
Cornelis E, De Vlieger I, Int Panis L. Emissions of mopeds and
motorcycles in Belgium. In: Sucharov L, Brebbia C, Benitez
F, editors. Urban transport VIII, urban transport and the environment for the 21st century. Southampton: WIT press, 2002.
pp. 491 9.
De Vlieger I, Cornelis E, Int Panis L. Evaluating transport policies,
a comparison between two approaches. In: Sturm P, editor. Mitteilungen institut fur verbrennungskraftmaschinen und therodynamik. Heft 81 vol. 1. Austria: Technische Universitat Graz,
2002. pp. 257 64.
EC. MEET methodology for calculating transport emissions and
energy consumption. Luxembourg: Office for official publications of the European Communities; 1999.
Friedrich R, Bickel P, editors. Environmental external costs of
transport. Heidelberg: Springer Verlag, 2001 (326 pp.).
Int Panis L, De Nocker L. Marginal costs. In: Friedrich R, Bickel P,
editors. Environmental external costs of transport. Heidelberg:
Springer Verlag, 2001. pp. 169 73.
Int Panis L, Watkiss P, De Nocker L, Torfs R. A comparative
analysis of trends in environmental externalities of road transport (1990 2010) in Belgium and the UK. Proceedings of the
TERA2K Conference. Milan, Italy: Scuola Enrico Mattei, ENI.
(in press).
Int Panis L, De Nocker L, Torfs R, Wuyts H. External costs of
heavy-duty vehicles for goods transport and buses. In: Sucharov
L, Brebbia CA, editors. Proceedings of the 6th International
conference on Urban Transport and the environment for the
21th century. WIT press, 2000b. pp. 211 20.

298

L. Int Panis et al. / Science of the Total Environment 334 335 (2004) 287298

Int Panis L, De Nocker L, De Vlieger I, Torfs R. Trends and

uncertainty in air pollution impacts and external costs of Belgian
passenger car traffic. Int J Veh Des 2001a;27(1 4):183 94.
Int Panis L, De Nocker L, Torfs R. A Monte Carlo analysis of air
pollution impacts from Belgian passenger cars in 2010. In:
Sucharov L, Brebbia C, editors. Urban transport VI. Urban
transport and the environment for the 21st century. Southampton: WIT press, 2001b. pp. 441 53.
Int Panis L, Rabl A, De Nocker L, Torfs R. Diesel or petrol? An
environmental comparison hampered by uncertainty. In: Sturm
P, editor. Proceedings of 11th International Symp. Transport and
Air Pollution. Mitteilungen Institut fur Verbrennungskraftmaschinen und Therodynamik. Heft 81 vol. 1. Austria: Technische
Universitat Graz, 2002. pp. 48 54.
Lantz F. An econometric analysis of the diesel car penetration in
France. Proceedings of the TERA2K Conference. Milan, Italy:
Scuola Enrico Mattei, ENI. (in press).

Rabl A, Spadaro J. Damages and costs of air pollution: an analysis

of uncertainties. Environ Int 1999;25(1):29 46.
Rabl A, Spadaro J. Uncertainty. In: Friedrich R, Bickel P, editors.
Environmental external costs of transport. Heidelberg: Springer
Verlag, 2001. pp. 147 59.
Sjodin A, Westerholm R, Almen J, De Serves C. Emissions of
regulated and non-regulated compounds from high- and lowemitting gasoline light-duty vehicles. Actes INRETS 2000;70:
135 42 [Arceuil France].
Van den Brink R, Gense R, Klein J. Nieuwe berekenings-methodiek
emissies wegverkeer, Bijdrage aan het colloqium Verkeer, Milieu, en Techniek. RIVM. 2000;275 89 [356pp.].
VMM Vlaamse Milieu Maatschappij J. Milieuen natuurrapport
Vlaanderen scenarios, MIRA-S, Leuven: 2000 VMM and
Garant publishers, 2000 (637 pp.; in Dutch).